Effects of Land Use Change on Juvenile Fishes and Brown Shrimp
Abundance in North Carolina’s Estuarine Nursery Habitats
Fisheries Resource Grant
Final Report
Grant EP-06-08 2004-1773-41
Feb 5, 2010
Submitted by
Joseph J. Luczkovich 1,2
Gregory F. R. Meyer 1,3
Mark M. Brinson 1
Terry T. West 1
Jason Hassell 4
East Carolina University
Greenville, North Carolina 27858
1
2
3
4
Department of Biology
Institute for Coastal Science and Policy
Coastal Resources Management PhD Program
Hassell Seafood, 321 Hassell Avenue, Washington, NC 27889
Table of Contents
List of Figures................................................................................................................................................................3
List of Tables .................................................................................................................................................................5
Acknowledgements .......................................................................................................................................................6
Executive Summary .......................................................................................................................................................7
List of abbreviations .................................................................................................................................................... 13
Introduction ................................................................................................................................................................. 15
LIFE HISTORY FOR THE SELECTED SPECIES .............................................................................................................. 18
Spot (Leiostomus xanthurus).............................................................................................................................. 18
Southern Flounder (Paralichthys lethostigma) .................................................................................................. 19
Atlantic croaker (Micropogonias undulatus) ..................................................................................................... 20
Atlantic menhaden (Brevoortia tyrannus).......................................................................................................... 20
Pinfish (Lagodon rhomboides) ........................................................................................................................... 21
Brown Shrimp (Farfantepenaeus aztectus) ........................................................................................................ 21
OBJECTIVES ............................................................................................................................................................. 22
Methods ....................................................................................................................................................................... 23
LAND USE CLASSIFICATION .................................................................................................................................... 25
Landsat Imagery Data Sources........................................................................................................................... 25
Sources for GIS data layers ................................................................................................................................ 28
Land Use Classification ..................................................................................................................................... 34
Classification Accuracy Assessment .................................................................................................................. 38
Land Use Land Cover Change Detection ........................................................................................................... 44
Data Collection Methods.................................................................................................................................... 49
Statistical Analysis of Change in Program 120 Catches by Station ................................................................... 51
Results ......................................................................................................................................................................... 55
LAND USE CHANGE PATTERNS 1980-2000 .............................................................................................................. 55
CHANGES IN CATCHES OF TARGET SPECIES (1980-2004) ........................................................................................ 64
Spot .................................................................................................................................................................... 65
Southern flounder ............................................................................................................................................... 69
Atlantic croaker .................................................................................................................................................. 73
Atlantic menhaden ............................................................................................................................................. 77
Pinfish ................................................................................................................................................................ 81
Brown shrimp ..................................................................................................................................................... 85
LONG-TERM CHANGES IN SPECIES BY NCDMF STATIONS ....................................................................................... 89
ABIOTIC FACTORS AND ABUNDANCE ....................................................................................................................... 91
MULTIVARIATE CART ANALYSES OF LAND USE CHANGE ON TARGET SPECIES ........................................................ 97
Spot CART analysis ........................................................................................................................................... 98
Southern flounder CART analysis ................................................................................................................... 100
Atlantic croaker CART analysis ...................................................................................................................... 102
Atlantic menhaden CART analysis .................................................................................................................. 104
Pinfish CART analysis ..................................................................................................................................... 106
Brown shrimp CART analysis ......................................................................................................................... 108
Summary of the CART results ......................................................................................................................... 110
Discussion.................................................................................................................................................................. 111
Conclusion ................................................................................................................................................................. 115
References ................................................................................................................................................................. 116
2
List of Figures
Figure 1.The overall scheme used to relate land use and juvenile fish and shrimp catch statistics……………….... 24
Figure 2. Pamlico River NCDMF stations and stream networks displayed over the 14-digit USGS HU watershed
boundaries.................................................................................................................................................................... 31
Figure 3. Pamlico River NCDMF stations and the stream networks plotted over a USGS topographic quadrangle. . 32
Figure 4. Pamlico River NCDMF stations, stream network, USGS 14-digit HU watersheds, and new catchment
boundaries.................................................................................................................................................................... 32
Figure 5. Location of the 71 NCDMF core stations . .................................................................................................. 33
Figure 6. Three Landsat image scenes (a) for the North Carolina coastal area are combined into a single mosaic
image (b)...................................................................................................................................................................... 35
Figure 7. Classified land use from May 2000 Landsat imagery showing the 71 NCDMF stations. ............................ 37
Figure 8. Random point assignment ............................................................................................................................ 40
Figure 9. Schematic representation of the process used to relate land use change to changes in species catch between
1980, 1990, and 2000. ................................................................................................................................................. 53
Figure 10. Change in percent forested and wetland area compared to agriculture and developed area....................... 57
Figure 11. The mean percentage of each catchment surrounding the 71 NC DMF stations in each of the years
examined (1980, 1990, 2000) . .................................................................................................................................... 59
Figure 12. Spot, Leiostomus xanthurus. The geometric mean catch per trawl from 1978 through 2004. ................. 66
Figure 13. Spot, Leiostomus xanthurus. Long-term change in catch .......................................................................... 67
Figure 14. Spot, Leiostomus xanthurus. Change in trawl catch z-score (1980-2004) at each NCDMF station as a
function of percentage change in land use (1980-2000) within each catchment. ........................................................ 68
Figure 15. Southern flounder, Paralichthys lethostigma. The geometric mean of catch per trawl from 1978 through
2004.. ........................................................................................................................................................................... 70
Figure 16. Southern flounder, Paralichthys lethostigma. Long-term change in catch. ............................................... 71
Figure 17. Southern flounder, Paralichthys lethostigma. Change in trawl catch z-score (1980-2004) at each
NCDMF station as a function of percentage change in land use (1980-2000) within each catchment........................ 72
Figure 18. Atlantic croaker, Micropogonias undulatus. The geometric mean catch per trawl from 1978 through
2004. ............................................................................................................................................................................ 74
Figure 19. Atlantic croaker, Micropogonias undulatus. Long-term change in catch. ............................................... 75
Figure 20. Atlantic croaker, Micropogonias undulatus. Change in trawl catch Z-score (1980-2004) at each
NCDMF station as a function of percentage change in land use (1980-2000) within each catchment........................ 76
Figure 21. Atlantic menhaden, Brevoortia tyrannus. The geometric mean catch per trawl from 1978 through 2004. .
..................................................................................................................................................................................... 78
3
Figure 22. Atlantic menhaden, Brevoortia tyrannus. Long-term change in catch. ...................................................... 79
Figure 23. Atlantic menhaden, Brevoortia tyrannus. Change in trawl catch Z-score (1980-2004) at each NCDMF
station as a function of percentage change in land use (1980-2000) within each catchment. ...................................... 80
Figure 24. Pinfish, Lagodon rhomboides. The geometric mean of catch per trawl from 1978 through 2004.. ........ 82
Figure 25. Pinfish, Lagodon rhomboides. Long-term change in catch ........................................................................ 83
Figure 26. Pinfish, Lagodon rhomboides. Change in trawl catch Z-score (1980-2004) at each NCDMF station as a
function of percentage change in land use (1980-2000) within each catchment. ........................................................ 84
Figure 27. Brown shrimp, Farfantepenaeus aztecus. The geometric mean of catch per trawl from 1978 through
2004. ............................................................................................................................................................................ 86
Figure 28. Brown shrimp, Farfantepenaeus aztecus. Long-term change in catch. ..................................................... 87
Figure 29. Brown shrimp, Farfantepenaeus aztecus. Change in trawl catch Z-score (1980-2004) at each NCDMF
station as a function of percentage change in land use (1980-2000) within each catchment. ...................................... 88
Figure 30. The average catch of spot at stations with different salinities for 1980-84, 1990-94, and 2000-2004. ..... 93
Figure 31. The average catch of southern flounder at stations with different salinities for 1980-84, 1990-94, and
2000-2004. ................................................................................................................................................................... 94
Figure 32. The average catch of pinfish at stations with different salinities for 1980-84, 1990-94, and 2000-2004. .. 95
Figure 33. The average catch of brown shrimp at stations with different salinities for 1980-84, 1990-94, and 20002004. ............................................................................................................................................................................ 96
Figure 34. A Classification and Regression Tree (CART) analysis of the standardized z-score of change in catch of
spot between 1980 and 2000. ..................................................................................................................................... 99
Figure 35. A Classification and Regression Tree (CART) analysis of the standardized z-score of change in catch of
Southern flounder between 1980 and 2000. ............................................................................................................ 101
Figure 36. A Classification and Regression Tree (CART) analysis of the standardized Z-score of change in catch of
Atlantic croaker between 1980 and 2000. . ............................................................................................................... 103
Figure 37. A Classification and Regression Tree (CART) analysis of the standardized Z-score of change in catch of
Atlantic menhaden between 1980 and 2000. ............................................................................................................. 105
Figure 38. A Classification and Regression Tree (CART) analysis of the standardized z-score of change in catch of
pinfish between 1980 and 2000. .............................................................................................................................. 107
Figure 39. A Classification and Regression Tree (CART) analysis of the standardized z-score of change in catch of
Brown shrimp between 1980 and 2000. .................................................................................................................... 109
4
List of Tables
Table 1. North Carolina’s commercial fishery for 2007 including, catch, value, and rank in catch for the seven
selected estuarine-dependent species. .......................................................................................................................... 17
Table 2. Spectral and ground pixel resolution of the Multi-Spectral Scanner on Landsat 2 ....................................... 26
Table 3. Spectral and ground pixel resolution of sensors on the Thematic Mapper used on Landsat 5 and the
Enhanced Thematic Mapper used on Landsat 7 . ........................................................................................................ 27
Table 4. Satellite imagery, specific dates, and data sources. ....................................................................................... 28
Table 5. Sample field sheet showing site number, Easting, Northing and 1980, 1990, and 2000 land use . . ............. 41
Table 6. Error matrix showing the ground reference data versus the image classification for land use types in coastal
watersheds. .................................................................................................................................................................. 42
Table 7. Producer’s accuracy matrix for the image classification. .............................................................................. 43
Table 8. User’s accuracy matrix for the image classification ...................................................................................... 43
Table 9. Areal estimates of land use change from 1980 - 2000. .................................................................................. 45
Table 10. The number of hectares of forest and wetland area converted to developed and agriculture land uses
between 1980 and 2000. . ........................................................................................................................................... 49
Table 11. Sampling effort (number of stations visited each year and month) in the NC DMF Program 120.............. 50
Table 12. Summary statistics of changes in land use types between 1980 and 2000.. ................................................ 58
Table 13. Number of catchments showing dominant types of land use between 1980 and 2000. ............................... 59
Table 14. Loss of forested land by county ................................................................................................................... 62
Table 15. Gain of agricultural land by county. ........................................................................................................... 62
Table 16. Gain of developed land by county ............................................................................................................... 63
Table 17 Catchment Areas of change in catch per species in the 71 stations .............................................................. 90
Table 18. Mean catches of each of the target species and significant factors affecting abundance at all stations
identified in the CART analysis .................................................................................................................................. 92
Table 19. Summary of partitions of the CART tree for change in catch at 71 NCDMF stations for the six target
species. ...................................................................................................................................................................... 110
5
Acknowledgements
We thank the many individuals who contributed their time and effort in the field or
office, collecting or analyzing data, or sharing the knowledge of juvenile fish recruitment to
North Carolina’s estuarine waters. We thank especially Katy West and Sean McKenna,
biologists at the North Carolina Division of Marine Fisheries (NCDMF), Pamlico District Office
for their insight into program 120 as well as their efforts in data quality control. NCDMF’s field
technicians: Greg Judy, Lele Judy, and Mike Pulley were helpful in demonstrating field
procedures used in data collection. Jason Hassell and Leland Tetterton were helpful in sharing
their observations and knowledge about juvenile fish in the Pamlico Sound. Kevin Hart,
Biologist at the NCDMF and Chris Mascio, former East Carolina University student, provided
valuable office and field assistance.
6
Executive Summary
This is a study report on the relationship between land use change and recruitment of a
selected set of six commercially and ecologically important species of finfish and shrimp in
North Carolina sounds and coastal waters. The selected species were: Atlantic croaker
(Micropogonias undulatus), spot (Leiostomus xanthurus), pinfish (Lagodon rhomboides), brown
shrimp (Farfantepenaeus aztectus), Atlantic menhaden (Brevoortia tyrannus), and southern
flounder (Paralichthys lethostigma). This report follows a recent study (Luczkovich et al. 2007)
that analyzed land use changes effects on blue crab (Callinectes sapidus), and therefore uses
similar methodologies to determine the potential relationships between watershed land use
characteristics and aquatic life.
Estuarine-dependent fish and crustacean species rank as among the most important of
North Carolina’s fisheries, producing over $44 million and 50 million pounds in commercial
harvest annually. Commercial harvests are dependent on estuarine nursery habitats (shallow tidal
creeks and submerged aquatic vegetation areas) for the juvenile fish and crustaceans to avoid
predators, find food, and have access to high quality habitat conditions that promote good
growth. However, coastal development and agriculture have been identified as land uses that
may degrade the nursery function of estuaries, because runoff of water from such developed
landscapes can change the water quality and food availability of estuarine nursery habitats.
The goal of this report is to examine historical data collected by the NC Division of
Marine Fisheries’ juvenile fish trawl survey to see if the change in the abundances of the
estuarine-dependent species listed above were associated with land use change within
catchments adjoining North Carolina estuarine nursery areas over a period from 1980-2004. As
in Luczkovich et al. (2007), land use changes were determined by classifying and performing
7
change detection on satellite images of the coastal areas of North Carolina (images were obtained
from USGS Landsat program in 1980, 1990, and 2000). The Landsat visible wavelength bands
were classified into one of the following land use classes using a combination of unsupervised
and supervised classification routines: forested lands, wetlands, developed land, and agricultural
lands. Between 1980 and 2000, deforestation constituted the single greatest land use alteration,
with an average catchment loss of 28.7 ± 8.2 % (mean ± 1 standard deviation). No net wetland
loss occurred when averaged across all catchments; in fact there was a slight increase in wetland
acreage. Agricultural land showed a substantial increase, averaging 22.8 ± 12.7% per catchment.
Developed land showed no significant net change when averaged across all catchments. Most of
the land use change consisted of the conversion of forest to agriculture. Deforested land was
also a source of local gains in developed land, and occasionally to local increases in wetland
coverage. Loss of forest, and the conversion to agriculture was widely distributed among the
catchments within Beaufort, Carteret, Hyde, Onslow, and Pamlico counties. The only significant
(>10%) increases in developed land via deforestation occurred in catchments in Onslow County.
Overall, catchment basins became dominated by agricultural land, suggesting that conversion of
land from forest to agriculture would lead to water quality degradation, potentially reducing the
abundance of juvenile fishes and invertebrates in catchments with large changes in land use.
Abundance data on juvenile fish and crustacean were obtained from the North Carolina
Division of Marine Fisheries Program 120 data set (juvenile fish and invertebrate nursery
survey) and encompassed 71 sampling stations. Although these data have been collected since
1971 at some stations and during the months of May through October each year until 1989, we
limited the analysis to data collected in May and June since 1980 because these months have
been sampled consistently over the period for which we also had land use data. Data were
8
analyzed for temporal trends and spatial patterns of abundance during the same period as the
land use changes by computing 5-year averages during 1980-1984 and 2000-2004. By
examining time series plots for each station and computing a change in catch statistic using the
difference between the average catches in the early years (1980-1984) and in the later years
(2000-2004), we were able to examine long-term changes in the abundance of the selected
species and map them along with land use changes derived from the satellite images. Data were
smoothed using a time-series analysis approach in which a LOWESS (locally weighted) curve fit
was applied to the species data for each station and used to gauge trends in abundance over time.
Time series plots were generated for each station (shown in the Appendix) and a geometric mean
plot for all stations combined.
There was an overall increase in abundance of spot (Leiostomus xanthurus), southern
flounder (Paralichthys lethostigma), and pinfish (Lagodon rhomboides) during the study period,
with peak abundances occurring after 1996. The increase in pinfish was dramatic (increasing
from a geometric mean of 1.2 fish/trawl in 1980 to 36.3 fish/trawl in 2001), suggesting that these
increases may be associated with shifts in the food web of Pamlico Sound or coast-wide climate
change. Pinfish are sub-tropical fish that are normally abundant in warm water in estuaries of
the southeastern USA; North Carolina is near their northern range limit, but they appear to be
spreading further north. Brown shrimp (Farfantepenaeis aztectus), Atlantic menhaden
(Brevoortia tyrannus) and Atlantic croaker (Micropogonias undulatus) did not appear to change
when averaged over all stations.
Large differences in species abundances were noted for the 71 NCDMF Program 120
stations. Some stations had consistently high or low catches of individual species year after year.
The Pamlico River stations, in particular, are important nurseries for southern flounder, with the
9
largest catches recorded there year after year. Some stations showed increases and other showed
declines; most stations (66 or more) showed no significant change (i.e., catches were with 2
standard deviations of the long-term mean) over time. Some of the stations showing significant
declines for Atlantic croaker were in Pamlico (Aurora PAR11), Neuse (Oriental F3N) and Hyde
County (PUR5). Stations with significant declines for Atlantic menhaden were in Swan Quarter
Bay (SQB3) and the Neuse River at Oriental (F3N). Spot showed significant declines at Bath
Creek on the Pamlico River (PAR 7) and Atlantic in Carteret County (CC11).
Land use changes, other anthropogenic variables (the number of NPDES sites and human
population density in each catchment in 2000) and abiotic factors (mean bottom temperature,
mean bottom salinity and distance to nearest inlet for each NCDMF trawling station) were used
as predictors of change in catch of the six species 1980-2004 using a multivariate classification
and regression tree (CART) statistical program. The CART analysis indicated a significant
decline in catch occurred between1980-2004 at stations that had > 21 % change (for southern
flounder) and > 54 % (for Atlantic croaker) in land use in the surrounding catchment (conversion
of forested land use to agriculture and developed land use). Additionally, the CART suggested
that high densities of people in surrounding watershed catchments were associated with declines
in spot and pinfish catches. Some abiotic factors were associated with change in catches of the
species in the CART analyses. The CART analyses suggested that change in catch was also
influenced by bottom salinity for some species, with brown shrimp and southern flounder
increasing in abundance in salinities > 14 ppt. In the CART analyses, increased bottom
temperature was associated with increases in Atlantic menhaden and pinfish, and to a lesser
extent, brown shrimp and Atlantic croaker. Distance to an inlet influenced the change in catch of
spot, brown shrimp, and to a lesser extent, Atlantic menhaden and Atlantic croaker. Finally,
10
depth of water at the collection site significantly influenced the change in catch of southern
flounder and brown shrimp.
We conclude that there was not a major land use change effect on juvenile estuarine fish
populations that we studied in North Carolina between 1980 and 2004, with some notable
exceptions. Southern flounder and Atlantic croaker, based on the multivariate CART analyses,
declined where excessive land use change (i.e., > –21- 54% of the catchment converted from
forested land to other uses between 1980-2004) occurred. In addition, pinfish increased
dramatically in abundance between 1980 and 2004, but only where human population density
was low in a catchment. We note that these three species (and possibly spot, which declined as
the percentage of land use devoted to agriculture increased, although this species did not change
significantly in the CART analysis) are all dependent on benthic food resources. These species
may be declining when land use change is extensive because agricultural runoff may cause
reduced benthic food availability during the summer months. While benthic organisms in the
deep portions of the Neuse River, Pamlico River and Pamlico Sound are influenced by
summertime hypoxia and decline in summer, the relatively shallow stations sampled in NCDMF
Program 120 with extensive land use change may exhibit increased areas of hypoxia in the early
summer. Hypoxia or other stressors (agricultural runoff with pesticides and herbicides) may be
impacting in the nursery areas in early summer samples (May and June) examined in this study,
and could be reflected in these declines of benthos-consuming fishes. In contrast, Atlantic
menhaden, the one planktivorous species considered in our analysis, did not show any
observable trends in abundance as land use changed in the surrounding watershed.
To protect critical nursery habitats for the selected species, management steps should be
taken to conserve land use, limit agricultural run-off, and limit development in areas adjoining
11
the productive nursery areas surrounding the Pamlico and other North Carolina Sounds.
Continued monitoring of the stations surveyed in the Program 120 is recommended, but this
survey should be extended into the late summer months (July – Oct), when water temperatures
are highest and hypoxia is likely to reach the maximum extent. Late summer catches can be
compared to catches made during the same months in early years (pre-1989) of the Program 120.
Land use conservation and prevention of agricultural run-off entering the estuarine areas might
be encouraged through limiting agriculture and development along creek banks that drain
directly into the estuarine areas, greater protection of forested land as well as wetlands, and
reduction in sources of non-point water pollution.
12
List of abbreviations
AOI
Area of Interest
B-IBI
Benthic Index of Biotic Integrity
BASINPRO
A group of tools and layers for watershed visualization and analysis, developed
by the North Carolina Center for Geographic Information Analysis.
BSQ
Band Sequential (Format).
CART
Classification and Regression Tree
C-CAP
Coastal Change Analysis Program
CW
Carapace Width
DOQQ
Digital Ortho Quarter Quad
EROS
Earth Resources Observation Systems
ESRI
Environmental Systems Research Institute
ETM+
Enhanced Thematic Mapper Plus
GCLF
Global Land Cover Facility
GIS
Geographic Information Analysis
GLCF
Global Land Cover Facility
GPS
Global Positioning System
HU
Hydrologic Unit
LOWESS
Locally Weighted Smooth Surface
MSS
Multi Spectral Scanner
NC CGIA
North Carolina Center for Geographic Information Analysis
NC DMF
North Carolina Division of Marine Fisheries
NOAA
National Oceanic and Atmospheric Administration
13
NPDES
National Pollution Discharge Elimination System
NRCS
Natural Resources Conservation Service
PSU
Practical Salinity Unit
PRE
Proportional Reduction in Error
TM
Thematic Mapper
UMD
University of Maryland, College Park
USDA
US Department of Agriculture
USGS
United States Geological Survey
WAAS
Wide Area Augmentation System
WBD
Watershed Boundary Dataset
14
Introduction
Estuarine and coastal areas contain some of the nation’s most densely populated and
rapidly growing areas (Beach 2002, Crosset 2004, Crawford 2007). As population density
increases, so does the potential for degradation of the natural environment by human activities
(Cairns and Pratt 1992). In the year 2000, New Hanover County, North Carolina, had the highest
population density (803 persons / mi2) among the 20 CAMA counties in North Carolina, and
since the 1980’s, coastal counties have exceeded the statewide average growth rate by 3.6%
(Street et al. 2005). This population growth affects land use practices, which in turn can
influence aquatic species habitat quality by contributing excessive sediments, nutrients, and
pollutants through storm water and groundwater flows into adjacent water bodies. Accordingly, a
number of shellfish closures have resulted in the loss of harvests throughout the Pamlico Sound
region during the past several years (Street et al. 2005).
Alteration of watersheds by human activities like farming and coastal development could
threaten the functioning of the extensive system of North Carolina’s estuarine nursery areas.
Runoff from farmland, confined animal facilities, forests, wetlands and urban areas differ greatly
in the content of sediments, nutrients and pesticides (Kennish 2002). Farmland runoff and
industrial farming activities can promote phytoplankton growth (Boynton et al. 1985, Malone et
al. 1996), reduce the coverage of submerged aquatic vegetation (SAV) habitats due to
competition for light with phytoplankton (Livingston 2001), and decrease the abundance of
certain groups of fishes dependent on such SAV (Heck and Orth 1980), thus altering estuarine
food webs. Weisberg et al. (1997) and Bilkovic et al. (2006) demonstrated that low estuarine
15
macro-benthos community indices were associated with developed land uses in watersheds along
Chesapeake Bay.
The Benthic Index of Biotic Integrity, or B-IBI, which ranges from 1 (stressed) to 5
(unstressed), is a multimetric index that combines the abundance of various invertebrate species
(annelids, bivalves, and crustaceans) into a single metric. The lowest B-IBI values were
associated with commonly hypoxic aquatic habitats located near developed watersheds.
Estuarine benthic communities near areas with a forested watershed, by contrast, had higher BIBI values than those with developed watersheds. Bilkovic et al. (2006) concluded that if as
little as 10-12 % of the watershed area was deforested and developed, then the estuarine benthic
community would exhibit stress from hypoxia and land nutrient inputs.
In our investigation of the relationship between land use and nursery area function in
North Carolina, we focused on changes in the abundances of six estuarine-dependent species of
juvenile finfish and an invertebrate species. The species were: spot (Leiostomus xanthurus),
southern flounder (Paralichthys lethostigma), Atlantic croaker (Micropogonias undulatus),
Atlantic menhaden (Brevoortia tyrannus), pinfish (Lagodon rhomboides), and brown shrimp
(Farfantepenaeus aztectus). These species are common in the shallow estuaries in late spring
and early summer as juveniles, settling in the Pamlico Sound and coastal waters of North
Carolina after a planktonic larval dispersal stage. Once in the estuary, their feeding habits vary.
Several are benthic feeders (Atlantic croaker, spot, pinfish, brown shrimp), one is planktivore
(Atlantic menhaden), and one is a piscivore (southern flounder). We had planned to analyze
patterns for bay anchovies (Anchoa mitchilli); however, data gaps in the NCDMF Program 120
database precluded an analysis of that species.
16
This report follows a recent study (Luczkovich et al. 2007) that analyzed land use
changes effects on the abundance of the blue crab (Callinectes sapidus), and therefore uses
similar methodologies.
Estuarine-dependent fish and crustacean species rank as among the most important of
North Carolina’s fisheries, producing over $44 million and 50 million pounds in commercial
harvest annually (Table 1). Commercial harvests are dependent on the estuarine nursery habitats
(shallow tidal creeks and submerged aquatic vegetation areas) utilized by juvenile fishes and
crustaceans to avoid predators, find food and maintain water quality conditions that promote
good growth. However, coastal development and agriculture have been identified as land uses
that may threaten the nursery function of estuaries, because runoff of water from such developed
landscapes can change the water quality and food availability in estuarine nursery habitats.
Table 1. North Carolina’s commercial fishery for 2007 including, catch, value, and rank in catch
for the seven selected estuarine-dependent species (source: www.ncdmf.net, last accessed
February 23, 2009).
Common name
Scientific name
Catch (lbs)
Value ($)
Blue crab
Callinectes sapidus
21,420,719 $21,429,574
Penaeid shrimps
Farfantepenaeus aztecus,
9,548,371 $17,930,804
(brown, pink and
F. duorarum,
white)1
Litopenaeus setiferus
Atlantic croaker
Micropogonias undulatus
7,301,295 $ 2,726,029
Southern flounders
Paralichthys lethostigma
2,078,020 $ 4,958,657
Atlantic menhaden
Brevoortia tyrannus
1,134,167 $ 139,178
Spot
Leiostomus xanthurus
878,989 $ 612,608
Pinfish
Lagodon rhomboides
73,238 $
14,899
Totals
50,088,807 $44,212,810
1
These penaeid shrimp species are pooled by NCDMF in commercial catches
17
Rank
(lbs)
Rank
($)
1
2
1
2
3
7
10
12
51
5
4
37
26
65
Life History for the selected species
Based on the amount and completeness of information available for the first year of life
of Middle Atlantic Bight fish species, Able and Fahay (1998) synthesized information of
economically important species including six of the seven species studied. Able and Fahay
(1998) refer to estuarine-dependant species as transient species since they spend only a portion of
their lives in the estuaries. Estuarine-dependant species characteristically spawn in the ocean, but
where, when and how the larvae or juveniles return to the estuaries vary greatly according to
species. The small size of recently settled fish may make them experience a high mortality rate
due to predation or inhospitable abiotic factors such as extremely low temperatures.
Spot (Leiostomus xanthurus)
Spot are distributed between Massachusetts Bay to Campeche Bay, Mexico, and are most
abundant between the Chesapeake Bay and the North Carolina coast (Able and Fahay 1998). The
spot population is mostly euryhaline; their larvae and juveniles are more abundant in the Middle
Atlantic Bight estuaries between Cape Hatteras and the Hudson River. Spawning takes place in
the continental shelf from winter to spring and is more intense in warm waters in the outer shelf
and south of Cape Hatteras (Able and Fahay 1998). Larval development takes place on the
continental shelf before larvae enter estuaries and then migrate to lower salinity, colder water of
the upper estuary. Using field observations and laboratory experiments, Rakocinski et al. (2006)
found that temperature and salinity are the two most important abiotic factors affecting juvenile
spot migration.
18
Southern Flounder (Paralichthys lethostigma)
The geographic range of Southern flounder extends from Virginia to northern Mexico.
They are found in rivers, estuaries and coastal waters. The Southern flounder is part of the
Bothidae family (left eye flounders). They are closely related and morphologically similar to
summer flounder and Gulf flounder. The three species are found in North Carolina and along the
most of the Atlantic coast including the Gulf of Mexico waters (Able and Fahay 1998). Flounder
spend much of their life on the bottom, where camouflage helps reduce detection by prey and
predators.
According to Watterson and Monaghan (2001), adult southern flounder migrate out of the
rivers and estuaries in the late fall to spawn offshore in the warmer waters of the Gulf Stream
between November and February. However, juvenile flounder are believed to overwinter in the
low salinity waters of the rivers and bays for the first two years of their life rather than migrating
offshore (Powell and Schwartz 1977). After the spawning period, adult flounders return to the
estuaries, coastal waters and rivers through the inlets (NCDMF 2005). Newly hatched larvae are
transported back to estuaries and coastal waters by oceans currents (Powell and Henley 1995).
It is believed that these developing larval flounder remain in the offshore waters for 30 to 60
days before getting carried through the inlets into the estuaries during night time flood tides
(Warlen and Burke 1990, Burke et al. 1991, Burke et al. 1998). After metamorphosis, the
juvenile flounder settle on tidal flats towards the head of the estuaries and move upstream to
lower salinity habitats (Burke et al. 1991, Guidon and Miller (1995). Powell and Schwartz
(1977) have found that benthic substrate and salinity are the two most important factors
governing the distribution of southern and summer flounders; with southern being more
19
abundant in areas of low salinity and clayey silt or organic rich mud bottoms while summer
flounder are most abundant in areas of moderate to high salinities and sandy bottom.
Atlantic croaker (Micropogonias undulatus)
The Atlantic croaker is distributed from Massachusetts to the Gulf of Mexico (Able and
Fahay 1999). It is less common north of New Jersey but abundant farther south where it is a
bottom feeding fish of coastal waters and estuaries. During winter, adults move offshore and
head south. Spawning takes place in the Middle Atlantic Bight starting in September and peaks
in October and ending in December. Pelagic larvae enter the estuaries via ocean inlets and
ultimately end up in low salinity nursery areas. The Atlantic croaker’s important habitats include
the continental shelf for larvae and low salinity habitat such as tidal creeks and tributaries of
major bay systems for the early settlement stages.
Atlantic menhaden (Brevoortia tyrannus)
According to Able and Fahay (1998), Murky et al. (1997), and Epifanio and Carvine
(2001), menhaden occurs mainly along the Atlantic coast of North America, from Nova Scotia to
Florida. They migrate north during spring and return to the south during the fall where they
spend the winter, mostly south of Cape Hatteras, North Carolina. It has been determined that
there is limited spawning during the spring northward migration that extends as far north as Cape
Cod and that limited spawning also occurs during the summer. Spawning increases remarkably
during the fall southern migration. In Delaware estuaries, larvae are typically 10 to 20 mm total
length when they ingress from December through May. Larvae are pelagic and ride ocean
currents into estuaries where they transform into juveniles at about a length of 30 to 38 mm. In
New Jersey and North Carolina estuaries, Able and Fahay (1998) and Warlen (1994) observed
20
that periods of low catch are often associated with low water temperatures that are usually
followed by higher temperatures and great number of larvae from the South Atlantic. After
arrival in the estuary, juveniles move upstream in low salinity waters and in areas of maximum
phytoplankton (Friedland et al. 1996). They stay in estuarine habitat until temperatures start
dropping in September/October.
Pinfish (Lagodon rhomboides)
Pinfish spawn offshore and their larvae are transported shoreward where they migrate
into estuaries for continued development (Forward et al.1998). Adams et al. (2004) states that
juvenile pinfish, an abundant and trophically important species, are dependent on seagrass beds
because seagrass and associated habitats (macroalgae) provide not only shelter from predation,
but also a fertile food source. In a larval fish migration study in the Newport River estuary,
North Carolina, Warlen and Burke (1990) found that pinfish larvae constituted about 13% of the
immigrating larvae. The early season peak is in December, which is followed by a very low
recruitment from mid-January to mid-February. The main period of recruitment occurred from
late February to early April.
Brown Shrimp (Farfantepenaeus aztectus)
The range of brown shrimp extends from Massachusetts to Florida and the Gulf of
Mexico and extends south to the Yucatan Peninsula. They are most abundant in the Gulf of
Mexico. Adult brown shrimp spawn in deep ocean waters, and like most estuarine dependent
fish and invertebrates, their larvae are transported to estuaries and coastal waters by wind and
ocean currents. Brown shrimp larvae pass through several developmental stages before becoming
post-larvae and juveniles. Once post-larval shrimp enter the estuaries, growth is swift and
21
dependent on salinity and temperature (NCDMF 2006). Feeding occurs mostly at night;
however, they have also been observed feeding in turbid water during the day. Shrimp are
omnivorous; they feed on sediment, detritus, algae, and benthic organisms.
According to Wenner and Beatty (1993), wind driven currents transport shrimp larvae to
the upper reaches of estuaries beginning in February with peaks occurring in mid-March through
mid-April. Adult and sub adult shrimp seek higher and more stable salinities because of a
decrease in the ability for osmoregulation. Blanton et al. (2002) found that water temperature,
tidal phase and wind direction greatly influence the spatial and temporal variation in densities of
postlarval white and brown shrimp within two inlets in South Carolina and Georgia.
Shrimp feed primarily on sediment, detritus, algae and benthic organisms. Feeding occurs
mostly at night, although some daytime feeding will occur in turbid water. Growth and
production of penaeids in estuaries are related to temperature, salinity, and presence of
vegetation (Rulifson 1981, Wenner and Beatty 1993). Field observations and laboratory
experiments indicate that brown shrimp seem to prefer marshes and shallow water habitat with
sea-grass beds that provide food, substrate and protection for the young (Zimmerman et al. 1984,
Rulifson 1981).
Objectives
We first review the results of the land use classification and change analysis, based on
Landsat satellite imagery from 1980, 1990, and 2000, essentially identical to the analysis
reported in Luczkovich et al. (2007). Next, we analyze the data of the NC DMF Program 120
juvenile fish trawl survey from 1980-2004 with respect to the six species (spot, southern
flounder, Atlantic croaker, Atlantic menhaden, pinfish and brown shrimp). We examine their
22
long-term change in catch by plotting the time series of geometric means for each species in
catches for all stations. We also examine the change in average abundance of the target species
in three 5-year periods (1980-84, 1990-94, and 2000-04) with respect to abiotic factors.. Next,
we compute a change in catch statistic for each station by averaging the catches in trawls taken at
each station in May and June early in time series (1980-1984), at a point near the start of our
land-use analysis, and subtract them from the averages of trawl catches taken in May and June
2000-2004 at the same stations, near the end of the land-use analysis. Using normalized z-score
of this difference and a multivariate analysis (CART or Classification and Regression Trees), we
analyze the change in catch statistic with respect to anthropogenic (land use change, human
population density, number of pollution discharge points) and abiotic (temperature, salinity,
depth, distance to nearest inlet) factors.
Methods
To examine the long-term trends in land use and catches of selected species of juvenile estuarine
dependant fish and invertebrates, we used a combined approach of (a) satellite remote sensing,
(b) land use classification of a multi-date satellite imagery, (c) change analysis of the land-use
classes within defined catchment areas; and analysis of trawl catch data from NCDMF nursery
areas, and (d) analysis of trawl catch data (NCDMF Program 120). The imagery data sets were
classified into Anderson level 2 land use categories (Anderson et al. 1976) and analyzed using
ERDAS Imagine, an image processing software from Leica Geosystems and ARGIS 9.2, a GIS
software by Environmental Science Research Institute (ESRI, Redlands, California). The
resultant land use data was merged with summary statistics of the NCDMF Program 120 catch
23
data to allow a comparison of changes in land use and catches in each of the 71 associated
sampling stations. The overall approach is shown in a conceptual framework shown in Figure 1.
Figure 1.The overall scheme used to relate land use and juvenile fish and shrimp catch statistics.
Adapted from Mohsen A. Environmental land use Change detection and assessment using with
multi-temporal satellite imagery (visited 12/15/2009)
http://www.gisdevelopment.net/application/environment/overview/frpf0004.htm
24
Land Use Classification
Landsat Imagery Data Sources
Land use information was derived by classification of satellite imagery of the North
Carolina coast: 1980 Landsat 2 Multispectral Scanner (MSS), 1990 Landsat 5 Thematic Mapper
(TM), and 2000 Landsat 7 Enhanced Thematic Mapper (ETM+). Table 2 shows the spectral and
spatial resolution of the Landsat MSS data, which are the oldest and thus the best in terms of the
time series that matches the start of our NCDMF data. However, the MSS data have lower
spectral resolution (only 4 spectral bands) and lower ground pixel resolution (57m x 79m pixels)
than the Landsat TM and ETM+ data (Table 3) which has 8 spectral bands and 30m x 30m
ground-pixel resolution. Band 4 in MSS (visible green) is similar to Band 2 in TM and ETM+;
Band 5 in MSS (visible red) is similar to Band 3 in TM and ETM+; Band 6 (near infrared) in
MSS is similar to Band 4 in TM and ETM+; Band 7 (also near infrared) in MSS is similar to
Bands 4 and 5 in TM and ETM+. These differences in spectral and ground pixel resolution
presented us with problems in comparison of the early Landsat MSS and later Landsat TM and
ETM+ data. We chose to project the MSS, TM, ETM+ data in the same geo-referenced data
frame in ERDAS Imagine using the smaller ground pixel resolution of the later Landsat TM and
ETM+ satellite data by resampling the earlier MSS data at 30m X 30 m resolution to obtain a
correct registration of the earlier imagery.
25
Table 2. Spectral and ground pixel resolution of the Multi-Spectral Scanner (MSS) on Landsat 2
which was operational from January 22, 1975 - February 25, 1982. (Source : US Geological
Survey: http://landsat.usgs.gov/about_landsat2.php, last visited 12/15/2009).
Sensor
Description
Wavelength
(µm)
Band 4
Visible Green
0.5 - 0.6
Band 5
Visible Red
0.6 - 0.7
Band 6
Near Infrared
0.7 - 0.8
Band 7
Near Infrared
0.8 - 1.1
26
Common use
Emphasizes sedimentladen water and
delineates areas of
shallow water.
Emphasizes cultural
features.
Emphasizes vegetation
boundary between land
and water.
Penetrates atmospheric
haze best, emphasizes
vegetation, bou ndary
between land and water,
and landforms.
Ground resolution
(m )
57 x 79 m
57 x 79 m
57 x 79 m
57 x 79 m
Table 3. Spectral and ground pixel resolution of sensors on the Thematic Mapper (TM) used on
Landsat 5 (March 1, 1984 – present) and the Enhanced Thematic Mapper (ETM+ ) used on
Landsat 7 (April 15, 1999 - present). (Source: US Geological Survey:
http://landsat.usgs.gov/about_landsat5.php, last visited 12/15/2009)
Electromagnetic
Spectrum
Wavelength
(µm)
Band 1
Visible blue-green,
reflected
0.450 - 0.515
Sediments in water
30
Band 2
Visible green,
reflected
0.525 - 0.605
Vegetation
30
Band 3
Visible red,
reflected
0.630 - 0.690
Vegetation
30
Band 4
Near infrared,
reflected
0.75 - 0.90
Vegetation, Chlorophyll
30
Band 5
Mid infrared,
reflected
1.55 - 1.75
Moisture in plants/soils, clouds
vs snow, mineral content
30
Band 6*
Thermal infrared,
reflected
10.4 -12.5
Geothermal activity, moisture
120, 60
2.09 - 2.35
Moisture in vegetation /soils,
rock formations
30
0.52 - 0.90
High resolution used to
enhance details in other bands
15
Sensor #
Band 7
Band 8**
Shortwave infrared,
reflected
Panchromatic,
visible light,
reflected
*The thermal Band 6 has a 120-m resolution for TM and 60-m resolution for ETM
**Landsat TM does not have Band 8.
27
Resolution
(m)
Common use
+
Table 4. Satellite imagery, specific dates, and data sources. [Sources: Earth Resources
Observation Systems (EROS) data center at The USGS; The Global Land Cover Facility (GLCF)
at the University of Maryland (UMD), College Park.]
Dataset
Date
Source
1980 Landsat MSS
p14r035
p14r036
p15r036
1990 Landsat TM
p14r035
p14r036
p15r036
+
2000 Landsat ETM
08/02/1980
08/02/1980
08/03/1980
USGS
USGS
USGS
06/12/1988
09/06/1990
05/08/1990
UMD
UMD
UMD
p14r035
p14r036
p15r036
09/23/1999
09/23/1999
05/11/2000
USGS
USGS
USGS
Sources for GIS data layers
Physiographic and topographic features of landscapes associated with primary nursery areas
were derived from The North Carolina Center for Geographic Information Analysis (CGIA),
which serves as a hub for Geographic information Systems (GIS) data used and or produced by
State agencies. We used GIS data layers contained in BASINPRO 8.0, a statewide data
compilation produced by CGIA. Among the layers of interest, we used hydrography, watershed
boundaries, confined animal operations, and water bodies. We also used data layers available to
the public at The North Carolina Department of Transportation. These include detailed county
roads and 1998 Digital Orthographic Quarter Quadrangle (DOQQ) photos, both of which were
used as reference to locate some satellite imagery features.
28
After the basic datasets were acquired and archived, the first task was to bring together all
the GIS data layers that we judged useful in computing landscape variables. We built a multilayer GIS database in ARCGIS 9.2 (ESRI Inc.) that brought together many watershed descriptive
layers. These included hydrography, watershed boundaries, confined animal operations,
National Pollution Discharge Elimination System (NPDES), water bodies, and detailed county
road layers. We added the NCDMF sampling stations, watershed boundaries, shorelines, parks
and preserves, inlets, and streams to the GIS database. 1998 DOQQ photos were used as
references to locate several satellite imagery features that are not easily identifiable on a 30
meter resolution.
These variables were merged within the NRCS watershed boundaries and were used to
relate to catches with respect to stations and years. Because these NRCS watersheds were much
larger than the extent of the drainage area impacting tidal creeks where NCDMF took samples,
we delineated our own catchment boundaries for the analysis. These are sub-watershed or
catchments, each of which encompasses all the streams and ditches draining to a nursery area as
denoted by the location of specific NCDMF sampling stations. Catchment boundaries were
determined using the Federal Standards for delineation of hydrologic unit boundaries (USGS and
USDA, 2009) that establishes standards and guidelines for creating and delineating hydrologic
unit boundaries, modifying existing hydrologic units, and establishing a national Watershed
Boundary Dataset (WBD). Additional guidelines for coastal watersheds (Ferguson and Mew
2000) were also followed.
Catchments were delineated in 3 distinct steps using ArcGIS 9.2 for display and analysis:
(1) Overview of the NRCS 14-digit hydrologic unit boundaries. We displayed major stream
networks (from NC Geographic Information Analysis data) and NCDMF sampling
29
stations over the NRCS 14-digit watershed boundaries (Hydrologic Units or HU’s) to
generate a general view of the watershed area as displayed in Figure 2.
(2) Topographic overlays: We displayed the 14-digit HU’s over the USGS topographic
quadrangles to note the topographic setting of the watershed with respect to sampling
station (Figure 3).
(3) Taking into account the stream network and topographic setting in steps 1 and 2, we
performed on-screen digitizing for the catchments draining to each of the NCDMF
sampling stations. Figure 4 shows sample catchments in the lower Pamlico. Catchments
boundaries are displayed in yellow while the USGS watershed lines are in grey. The
resulting catchment layer is then used to compute land-use parameters related to the
NCDMF sampling stations.
Depending on the location of a given NCDMF sampling station, catchment size can be
smaller or equal to the source watershed. Using this technique, we found that the entire 105
NCDMF Program 120’s core sampling stations were located within just 99 catchment areas
because some catchments contained more than one sampling station (Figure 5). Figure 5 also
shows the final catchments and NCDMF locations for the entire coast. Note however, that we
limited our analysis to the 71 stations out of 105 that had been sampled consistently from 1980 to
2004; the remaining 34 stations (yellow stations on the map) were established more recently than
1980. Images were clipped to exactly the study area with a 300 feet buffer. A buffer zone was
added to the catchment boundary to account for possible spatial inaccuracies in layer boundaries.
Each clipped image was visually inspected and classified using ERDAS Imagine 9.0 (remote
sensing software from Leica Geosystems GIS & Mapping, LLC) to denote type of land use.
30
Figure 2. Pamlico River NCDMF stations and stream networks displayed over the 14-digit
USGS HU watershed boundaries (grey lines).
31
Figure 3. Pamlico River NCDMF stations and the stream networks plotted over a USGS
topographic quadrangle.
Figure 4. Pamlico River NCDMF stations, stream network, USGS 14-digit HU watersheds (grey
lines), and new catchment boundaries (yellow lines) displayed on the USGS watershed layer.
Catchment boundaries were drawn within each USGS watershed area.
32
Figure 5. Location of the 71 NCDMF core stations (red circles) and, the other NCDMF core
stations not analyzed (yellow circles).
33
Land Use Classification
Several techniques can be used for land use land cover classification. The most common
techniques are supervised classification, unsupervised classification, or a combination of both.
Supervised classification of land use in a remote sensing image is done interactively while the
analyst is looking at a display of an image, delimiting areas of similar land use, either by having
visited the areas on the ground or by using other imagery of higher resolution to determine land
use. In contrast, unsupervised classification of an image is done using a computer algorithm that
simply recodes each pixel in the image into classes of land use based on the spectral signatures in
the digital data (Lillesand and Kiefer 1994). Given that the images we have chosen vary in
resolution (1980 Landsat has 60-m resolution while 1990 and 2000 have a 30-m resolution), we
used a combination of both supervised and unsupervised to improve the quality of the resulting
thematic map.
Landsat satellite imagery was processed using ERDAS IMAGINE 9.0 software (Leica
Geosystems, Inc. 2003). The dataset obtained from USGS in generic binary format (BSQ) was
imported into .img format, which is the main format for ERDAS IMAGINE software. This
import procedure brings each scene’s images in separate wavelength spectral bands. For Landsat
MSS, we used all four spectral bands for classification (Table 1), while for Landsat TM and
ETM+ we used six of the seven and 8 bands, respectively, for analysis (bands 1, 2, 3, 4, 5, and 7;
Table 2). Band 6 or the thermal band, has a much coarser resolution than the other bands and
was not used. After the bands were imported for each imagery type, they were stacked and
placed together as a mosaic in a geo-referenced grid (Figure 6b). The geo-referenced imagery
was then imported to ArcGIS 9.2. Areas of interest were extracted from a mosaic of the images
34
using an ArcGIS layer of the catchment areas surrounding the 71 NCDMF stations as shown in
Figure (6c).
Figure 6. Three Landsat image scenes (a) for the North Carolina coastal area are combined into a
single mosaic image (b). The coastal watersheds that were studied are outlined in red, and
NCDMF program 120 sampling stations are shown in blue (c).
35
Images were clipped to the study area with a three mile buffer. Each clipped image was
visually inspected / analyzed before classification in ERDAS Imagine 9.0 to generate land use
categories.
Image classification was conducted using a combination of unsupervised and supervised
classification scheme (Jensen 2007). Unsupervised classification was first used because its
unbiased mathematical algorithm can help generate any desired number of land use categories
simply by analyzing images’ spectral signatures. We followed methods outlined by the NOAA
Coastal Services Center guidelines (Dobson et al. 1995) and NOAA C-CAP (2002) to create 21
classes of level 2 Anderson land use classification categories (Anderson et al. 1976). The
unsupervised classification was supplemented with a supervised classification, whereby an
analyst inspected the land-use classes automatically generated by the ERDAS software and
reclassified them using ancillary and ground-truth data to reduce possible classification errors.
The ERDAS “RECLASS” algorithm was then trained to identify the pixels in all the imagery
with similar characteristics based on the ground-truth data. By setting class identifier values for
these pixels, and using the trained reclassification algorithm to assign the same class value to
each similar pixel in the original data set, each resulting class corresponds to a land-use pattern
that was originally identified in the ground-truth (Lillesand and Keifer 1994).
This process
helped cluster the 21 land-use classes (Anderson Level 2) into 5 coarse categories (Anderson
Level 1, Anderson et al. 1976) for ease of comparison of large scale maps. We collapsed all
wetlands subcategories into “wetlands” (Anderson classes 10 through 15), all forested
subcategories to “forested” (Anderson classes 6 through 9), all development subcategories to
“developed” (Anderson classes 2 and 3), and all agricultural subcategories to “agriculture”
(Anderson classes 4 and 5); the “water” class remaining the same. The reconstituted and
36
renumbered classes are: 1 = Water, 2 = Developed,3 = Forested, 4 = Wetland, 6 = Agriculture.
A map showing these classifications based on the 2000 imagery is shown in Figure 7.
Figure 7. Classified land use from May 2000 Landsat imagery showing the 71 NCDMF stations
sampled consistently since 1980 and the catchment boundaries.
37
Classification Accuracy Assessment
After the 1980, 1990, and 2000 Landsat images (ETM+, TM and MSS) were classified,
an accuracy assessment protocol was developed. This consisted of conducting field verifications
on a set of sample points randomly distributed over the study area and determining the frequency
of congruence with the corresponding satellite imagery. Ground-truthing was conducted in the
summer of 2006 using a combination of aerial photographs, USGS’s Digital Ortho Quadrangles
(DOQQ), and site visits.
Accuracy assessment also requires that an adequate number of points be sampled for each
map category in order to yield valid statistical analyses. To determine how many sample points
are required for a given size of area of interest, researchers have used equations based on the
binomial distribution or normal approximation, which performs well in determining the overall
mapping accuracy. Congalton and Green (1999) suggested that a multinomial approximation
tends to provide a good balance between statistical validity and field practicality. Also, the
number of samples for each category can be adjusted based on the relative importance of that
category within the objectives of the mapping project or by the inherent variability within one of
the categories. Categories with less variability such as water or even-aged forest plantations can
be sampled at a lower density, while more variable categories such as uneven-aged forests and
agriculture should be sampled at higher densities to capture the wide spectrum of those classes.
The procedure for generating the appropriate sample size using a multinomial
distribution, originally presented by Tortora (1978) and described by Congalton and Green
(1999), can be summarized by the following equation
38
where n = sample size, i = 1,…k are the number of land use classes, пi is the proportion of the
population in the ith category, B is a constant determined from a Chi square table with 1 degree
of freedom, and bi is the desired precision (for example, if 95% confidence level is desired, bi=
0.05). In our study, we computed that we needed to visit at least 239 points to achieve a 95%
confidence interval, based on the proportion of the “wetlands” category (π = 0.10 for wetlands,
bi= 0.05, B(1,0.01) = 6.63). Similar calculations were made for the other classes.
In choosing ground-truth points, we used both simple random sampling and stratified
random sampling. In simple random sampling, points are randomly chosen within land use strata
and each sample unit in the study area has an equal chance of being selected. This has an
advantage over stratified random sampling because it does not assume prior knowledge of the
study area. One problem with simple random sampling is that it tends to under-sample rare map
categories. However, one can increase the number of sample points to compensate for rarity. To
balance statistical validity and practical application, we used a simple random sampling scheme
early in the project and then complemented it with stratified random sampling later because the
“developed” land use category had been initially under-sampled. Thus, on later trips, we visited
additional random ground-truth points within the “developed” category.
The random point generator function of Excel was utilized to generate a random
distribution of sampling points. At first, we started with 104 points spanning between (xmin, ymin)
and (xmax, ymax) coordinates, which are respectively the lower left and upper right coordinates of
our study area (Figure 8 b). This is relatively a high number of points, but it is done in this
manner knowing that many of the points would be eliminated for being outside the study area or
39
would be inaccessible because they are located on a water body, at gated properties, or on
inaccessible private property. After elimination of unusable points, we only retained 256 points
as our final number of sample sites to visit (Figure 8b).
(a)
(b)
Figure 8. Random point assignment. (a) 104 random points overprinted on the image; (b) 256
random points that intersected with catchment areas and were accessible to ground-truthing are
highlighted in red; random points outside of catchments were removed
The coordinates from these 256 points were transferred to a Global Positioning System
(GPS)-GARMIN GPSmap 76S (Garmin International Inc. 2005). We used the “send to GPS”
function of ExpertGPS software (TopoGrafix Inc.) to transfer waypoints/random sample sites
from ARCGIS to GARMIN. This GPS receiver has an accuracy of ± 3 meters using the Wide
Area Augmentation System (WAAS) capability. Table 5 is an example field sheet.
40
Table 5. Sample field sheet showing site number, Easting, Northing and 1980, 1990, 2000 land
use and comments. The easting is the projected distance of the position from the central
meridian, while the northing is the projected distance of the point from the Equator.
Site #
GIS ID
EASTING
NORTHING
Land use 1980 Land use 1990 Landuse 2000
COMMENT
1
46
335835
3931190
2
131
347702
3894787
3
290
361023
3868910
4
292
337602
3926415
5
323
231927
3781800
6
366
339385
3901330
7
509
409064
3926463
8
511
337485
3861975
9
528
284808
3844072
10
603
390878
3921448
11
632
365832
3855203
12
663
278146
3846343
W
W
water body west of Croatan Sound
13
667
417821
3944121 F
F
F
spoke to owner, forest about 50 years old
14
683
426151
3941618
15
685
347407
3880977
Ag
Ag
O.G.F cotton fields
16
686
330086
3932700
17
741
290618
3836513
F
18
750
382527
3920965
Ag
19
757
189615
3768903
20
845
225759
3800104
Site unaccessible
21
871
361594
3869576
Site unaccessible
22
879
351025
3922017
23
943
352472
3919829
Ag
Agriculture - currently soy field, behind a small
church
24
958
184467
3763836
Ag
Agriculture
25
1009
446714
3969988
F
26
1073
348465
3895600
F
was forested-recently cut about 2 years old
Forested close to Pamlico beach, between
Waders and Dave Moore pt.
27
1096
340044
3924262
28
1169
363832
3863034
Ag
South of Lake Mattamuskeet
29
1187
445719
3900300
Ag
Young forest,less than 5 years old, 2000 clear cut?
30
1194
275346
3825469
31
1277
443823
3966758
F
F
pine forest dying off, near DMF station
No access
No access same as 51/A40
Ag
young mixed forest, possible transition from pine
to hardwood, unknown age
off Jackson Swamp Rd/Jackson Hungting club
across
No access at Beaufort/Pamlico county line
41
Land classified using ground-truthing data was compared to that classified in the office,
and based on satellite data using a confusion matrix (Congalton and Green 1999). A confusion
matrix contains information about actual and predicted classifications done by a classification
scheme (Goodchild 1994). Performance of such systems is commonly evaluated using the data
in the matrix. Observations made in the field were tabulated and compared to maps made in the
laboratory. Table 6 shows an error matrix displaying the number of sample units assigned to a
particular category in one computer based classification relative to the field based reference data,
assumed to be correct.
Table 6. Error matrix showing the ground reference data versus the image classification for land
use types in coastal watersheds.
Ground Reference
Image Classification
Developed
Developed
11
Forest
2
Wetland
Forest
Wetland
Agriculture
Total
1
12
63
1
1
67
1
13
1
15
48
54
56
148
Agriculture
5
1
Total
18
65
14
Using this matrix, we computed producer’s, user’s and overall accuracy for the classified
land use maps. Producer’s accuracy or commission error tells how often the map correctly
predicts known features. It is calculated by dividing the number of correct pixels for a class by
the actual number of ground truth pixels for that class (Congalton and Green 1991, Jensen 2007).
User’s accuracy or omission error shows how often the map successfully leads the user to
42
unknown features. It is a measure of the reliability of an output map generated from a
classification scheme, and a statistic that can tell the map user what percentage of a class
corresponds to the ground-truthed class. User's accuracy is calculated by dividing the number of
correct pixels for a class by the total pixels assigned to that class (Congalton and Green 1998,
Jensen 2007). The overall accuracy is computed by dividing the total correct (i.e., the sum of the
major diagonal) by the total number of pixels in the error matrix (Congalton 1991). The
computed overall classification accuracy was 91% (Table 6), the producer’s accuracy was 84%
(Table 7), while user accuracy was 90% (Table 8).
Table 7. Producer’s accuracy matrix for the image classification.
Producer’s Accuracy
Developed
11/18
Forest
63/65
Wetland
13/14
Agriculture
48/56
Overall producer’s
accuracy
135/148
Percentage
61
97
93
86
84
Table 8. User’s accuracy matrix for the image classification
User Accuracy
Developed
Forest
Wetland
Agriculture
Overall user’s
accuracy
Percentage
11/12
63/67
13/15
48/54
92
94
87
89
90
43
Land Use Land Cover Change Detection
Change analysis was performed on the 1980, 1990, and 2000 satellite imagery. This was
done by aligning the images and matching pixel to pixel, the five classified categories
(developed, forest, wetland, agriculture, water). Change detection usually involves two steps: (1)
comparison between land cover maps which are independently produced using selected
statistical algorithms and (2) change enhancement by editing the resulting layer into readily
interpretable “from to” classes. Change detection was performed using the ERDAS post
classification image differencing algorithm. In this algorithm, if two pixels have the same class
of wetland (class 4) in 1980 and 2000, the “from to” class is 44 and this is coded as “no change”
in the final image. A gray pixel is shown for this change class. If, however, the pixel changed
from forest (class 3) in 1980 to developed (class 2) land use in 2000, a non-zero change was
detected, so a “from to” class value of 32 was assigned to that pixel. By examining all pixels in
this way, changes of all classes were detected and mapped as an image of the watersheds. For
each time series (1980, 1990, 2000) classified areas of interest (AOI) were made into a mosaic
using the histogram stretch and minimum value options.
The catchments layer was overlaid on thematic maps and zonal statistics generated using
the spatial analysis module of ARCGIS 9.2 to generate land uses within each catchment area.
The zonal statistics function was used to calculate statistics on values of an image (raster) within
the zones defined in another dataset that can be either raster or vector. We used “Tabulate Area,”
an algorithm that cross-tabulates areas between two datasets (ESRI, ARCGIS 9.2, 2006). For this
study, we used the catchment's boundary layer as a zone layer to extract land use changes from
the changed map the whole study area by catchment. Table 9 shows the resulting land use
change areas tabulated for each catchment. Land changes are summarized in Table 10.
44
Table 9. Areal estimates of land use change from 1980 - 2000. Shown for each catchment are the identification numbers (ID#),
NCDMF station codes (Station), longitude and latitude of the stations, county name, USGS topographic quadrangle name, total area,
water area, and land area of catchment (in ha), change in land use between 1980 and 2000 for the following change classes (in ha):
forested land to developed land; forested land to agricultural land; wetland to developed land; wetland to agricultural land; agricultural
land to developed land, total area of change (in ha), and % land changed.
Catch
ment
ID
Station
Long.
Lat
4
CFR11
-77.922
33.98
5
CFR4
-77.929
34.12
5
CFR5
-77.95
34.13
6
CFR1
-77.968
34.25
8
CFR2
-77.947
9
VC1
10
County
NEW
HANOVER
NEW
HANOVER
NEW
HANOVER
USGS
Quadrangle
Kure Beach
Total
Area
(ha)
Water
Area
(ha)
Land
Area (ha)
Forest
to
Develo
ped
(ha)
Forest
to
Agriculture (ha)
Wetland
to
Develop
ed (ha)
Wetland
to
Agriculture (ha)
Agriculture to
Develop
ed (ha)
Total
Area of
Change
(ha)
% of
Land
Use
Change
6410.41
2913.39
3497.02
254.88
270.8
36.06
51.58
29.48
642.81
18.38
Carolina
3801.6
945.99
2855.61
301.75
311.99
37.12
48
30.22
729.08
25.53
Wilmington
3801.6
945.99
2855.61
301.75
311.99
37.12
48
30.22
729.08
25.53
Castle Hayne
738.24
9.72
728.52
6.01
33.55
1.71
5.04
5.77
Castle Hayne
7170.9
169.65
7001.25
334.32
869.27
19.66
17.3
379.24
52.07
1619.7
9
7.15
34.27
BRUNSWICK
NEW
HANOVER
23.14
-77.606
34.43
PENDER
Holly Ridge
1662.99
27.72
1635.27
180.48
422.78
1.71
4.47
17.95
627.38
38.37
SSO1
-77.475
34.48
ONSLOW
Spicer Bay
1300.56
438.48
862.08
48.25
50.2
33.79
27.54
10.88
170.65
19.8
12
SSI1
-77.424
34.52
ONSLOW
Sneads Ferry
1662.44
1.98
1660.46
164.07
396.62
1.3
3.57
23.47
589.04
35.47
13
NR10
-77.398
34.59
ONSLOW
Sneads Ferry
315.24
78.66
236.58
8.29
24.04
0.57
1.14
0.08
34.11
14.42
14
NR13
-77.43
34.63
ONSLOW
Jacksonville
1105.91
2.88
1103.03
48.98
378.83
0
3.09
4.47
39.47
15
NR6
-77.331
34.64
ONSLOW
Camp Lejeune
3395.14
39.24
3355.9
163.59
1211.55
2.84
15.03
3.17
435.37
1396.1
8
16
NR2
-77.426
34.74
ONSLOW
Jacksonville
1900.74
114.03
1786.71
198.6
205.74
2.44
6.01
101.61
514.4
28.79
17
NR1
-77.436
34.76
ONSLOW
Jacksonville
2324.75
23.22
2301.53
306.3
249.04
11.78
3.17
209.07
33.86
18
NR4
-77.383
34.72
ONSLOW
Jacksonville
5316.28
283.5
5032.78
748.49
502.3
7.47
6.09
493.36
779.35
1757.7
1
21
CC3
-76.751
34.75
CARTERET
Newport
2952.6
182.16
2770.44
101.77
299.96
2.03
45.73
81.79
531.29
19.18
22
CC5
-76.623
34.82
CARTERET
Williston
2401.67
18.09
2383.58
29.81
986.8
2.44
64.09
5.36
1088.5
45.67
23
CC7
-76.458
34.83
CARTERET
Davis
1456.17
101.07
1355.1
4.71
235.55
2.68
6.82
0.57
250.34
18.47
25
CC6
-76.481
34.81
CARTERET
Davis
3275.94
343.35
2932.59
10.15
304.68
2.03
8.53
3.09
328.47
11.2
45
41.6
34.93
Table 9 (continued). Areal estimates of land use change from 1980 – 2000.
Catchment
ID
Station
Longitude
Latitude
County
USGS
Quadrangle
Total
Area
(ha)
Water
Area
(ha)
Land
Area
(ha)
Forest to
Developed (ha)
Forest to
Agriculture (ha)
Wetland
to
Developed (ha)
Wetland
to
Agriculture (ha)
Agriculture to
Developed
(ha)
Total
Area of
Change
(ha)
%
Land
Use
Change
26
H2
-76.762
34.87
CRAVEN
Newport
6199.66
352.35
5847.31
225.07
760.59
8.29
28.1
80.01
1102.06
18.85
27
CC10
-76.359
34.92
CARTERET
Atlantic
2985.57
932.31
2053.26
157.74
50.12
13
9.26
0.81
230.92
11.25
28
J2
-76.464
34.94
CARTERET
Long Bay
1461.94
272.61
1189.33
16.33
864.15
0.16
14.62
0.41
895.67
75.31
29
G19
-76.637
34.94
CARTERET
Merrimon
1461.94
72
1389.94
11.13
371.28
0.32
3.49
0.49
386.71
27.82
30
J10
-76.406
34.94
CARTERET
Long Bay
2690.79
426.51
2264.28
64.9
64.17
0.65
5.04
0
134.75
5.95
31
CC11
-76.289
34.95
CARTERET
Atlantic
754.29
24.93
729.36
2.19
17.87
0.16
1.54
0
21.77
2.98
33
G3
-76.507
34.98
CARTERET
South River
4823.01
809.01
4014
10.88
557.85
2.68
12.1
0
583.52
14.54
35
F3N
-76.666
35.04
PAMLICO
Oriental
786.96
149.49
637.47
11.94
90.32
1.62
0.41
18.44
122.73
19.25
36
F12
-76.724
35.04
PAMLICO
Oriental
2188.41
12.69
2175.72
8.77
283.39
0
1.06
6.09
299.31
13.76
37
F1
-76.645
35.06
PAMLICO
Oriental
942.51
176.85
765.66
5.69
174.8
0
0.57
0.73
181.78
23.74
38
E10
-76.658
35.08
PAMLICO
Oriental
1292.01
38.61
1253.4
6.25
393.21
0.16
0.49
1.14
401.25
32.01
39
E15
-76.586
35.11
PAMLICO
Broad Cre
582.45
100.35
482.1
0.41
60.51
0.08
0.89
0
61.89
12.84
40
D5
-76.594
35.14
PAMLICO
Jones Bay
1029.51
86.31
943.2
0.16
142.63
0
0.08
0
142.87
15.15
41
CS13
-76.667
35.15
PAMLICO
Vandemere
340.75
6.21
334.54
1.87
79.03
0.08
1.3
0.57
82.85
24.77
42
CS2
-76.636
35.15
PAMLICO
Vandemere
1045.15
81.9
963.25
12.51
248.14
0
0.08
1.79
262.52
27.25
43
D8
-76.569
35.16
PAMLICO
Jones Bay
988.44
279.45
708.99
0
0.81
0
0
0
0.81
0.11
44
CN1
-76.675
35.18
PAMLICO
Vandemere
894.33
31.95
862.38
7.55
310.77
0
12.27
6.09
336.68
39.04
45
CN6
-76.558
35.2
PAMLICO
Jones Bay
286.57
35.11
251.46
0
6.82
0
0.08
0
6.9
2.75
46
CN14
-76.714
35.15
PAMLICO
Vandemere
2315.59
47.34
2268.25
10.64
792.92
1.95
50.36
5.69
861.55
37.98
47
B43
-76.572
35.22
PAMLICO
Jones Bay
560.04
69.3
490.74
0.49
16.33
0
0.08
0
16.89
3.44
49
CN3
-76.621
35.2
PAMLICO
Jones Bay
1668.33
261.81
1406.52
5.2
532.67
0
3.74
0.89
542.5
38.57
50
B40
-76.591
35.24
PAMLICO
Jones Bay
1165.89
197.19
968.7
10.07
122.41
2.76
9.67
1.62
146.53
15.13
51
B20
-76.502
35.26
PAMLICO
Lowland
883.68
281.16
602.52
0
45.24
0
14.95
0
60.19
9.99
46
Table 9 (continued). Areal estimates of land use change from 1980 – 2000.
Catchment
ID
Station
Longitud
e
Latitud
e
County
USGS
Quadrangle
Total
Area
(ha)
Water
Area
(ha)
Land
Area
(ha)
Forest
to
Develop
- ed (ha)
Forest
to
Agricul
-ture
(ha)
Wetlan
d to
Developed (ha)
Wetlan
d to
Agriculture
(ha)
Agricul
-ture to
Developed (ha)
Total
Area
Change
(ha)
%
Land
Change
53
A12
-76.601
35.3
PAMLICO
Lowland
1316.1
66.51
1249.59
30.22
121.27
2.84
8.61
4.87
167.81
13.43
54
A58
-76.511
35.3
Lowland
475.63
235.8
239.83
0
1.46
0
34.44
0
35.9
14.97
55
PAR11
-76.783
35.31
PAMLICO
BEAUFOR
T
9988.57
37.08
9951.49
183.41
3745.69
0.49
79.76
45.24
4054.59
40.74
56
B10
-76.501
35.32
103.36
38.3
65.06
0
1.46
0
6.5
0
7.96
12.24
58
A2
-76.637
35.32
South Creek
1011.13
64.35
946.78
33.79
162.61
0
15.68
2.76
214.84
22.69
59
PAR16
-76.644
35.34
South Creek
215.57
7.47
208.1
0
20.06
0
7.47
0
27.54
13.23
60
PAR13
-76.685
35.33
South Creek
794.09
21.42
772.67
27.7
152.05
0.97
11.78
28.67
221.18
28.62
63
PAR7
-76.817
35.38
PAMLICO
BEAUFOR
T
BEAUFOR
T
BEAUFOR
T
BEAUFOR
T
1967.35
23.22
1944.13
39.48
329.85
0
9.99
0.08
379.4
19.52
65
AB1
-76.508
35.41
HYDE
Pamlico B
739.74
130.23
609.51
0.16
33.55
0
64.66
0
98.36
16.14
66
OC1
-76.13
35.36
Bluff Point
2453.53
190.53
2263
0
44.59
0.16
0
0
44.76
1.98
67
PUR3
-76.596
35.4
HYDE
BEAUFOR
T
Pamlico B
662.75
32.13
630.62
0.49
118.59
0.32
23.23
0
142.63
22.62
68
RB3
-76.434
35.43
HYDE
455.27
61.02
394.25
0
64.82
0
12.91
0
77.73
19.72
69
SQB3
-76.312
35.39
1805.09
34.11
1770.98
2.92
815.5
0
4.14
0
822.57
46.45
70
PAR9
-76.761
35.43
HYDE
BEAUFOR
T
Scranton
Swanquarte
r
838.03
26.37
811.66
10.97
305.41
0.57
18.03
0.16
335.13
41.29
72
PUR5
-76.528
35.45
HYDE
Pamlico B
1480.57
95.04
1385.53
0.08
450.96
0
38.26
0
489.3
35.32
73
RB1
-76.431
35.44
HYDE
494.04
16.92
477.12
0
164.56
0
8.2
0
172.77
36.21
74
SQB1
-76.357
35.41
HYDE
1811.78
59.76
1752.02
0.49
200.95
0
20.23
0
221.66
12.65
76
JB1
-76.255
35.39
HYDE
Scranton
Swanquarte
r
Swanquarte
r
5445.91
100.44
5345.47
1.79
380.86
0
27.54
0
410.19
7.67
77
WB3
-76.065
35.41
HYDE
Middletown
1996.05
77.4
1918.65
69.93
239.9
0
21.12
2.36
333.31
17.37
79
WB1
-76.064
35.43
HYDE
Middletown
3331.92
205.83
3126.09
39.96
429.76
0
4.63
0.89
475.25
15.2
Aurora
Lowland
Bath
Bath
47
Table 9 (continued). Areal estimates of land use change from 1980 – 2000.
Catchment
ID
Station
Longitude
Latitude
County
USGS
Quadrangle
81
FC3
-76.008
35.47
HYDE
Middletown
83
FC1
-75.986
35.51
HYDE
84
PAR8
-76.818
35.45
86
LSR3
-75.903
87
LSR5
88
89
Total
Area
(ha)
Water
Area
(ha)
Land
Area
(ha)
Forest
to
Developed (ha)
Forest
to
Agriculture
(ha)
Wetland
to
Developed (ha)
Wetland
to
Agriculture
(ha)
Agriculture to
Developed (ha)
Total
Area
Change
(ha)
%
Land
Change
866.19
38.61
827.58
18.93
125.31
0
60.68
0.16
205.08
24.78
Engelhard
2462.95
287.1
2175.85
0
391.75
0
6.99
0
398.73
18.33
BEAUFORT
Bath
9897.73
348.84
9548.89
38.26
3779.4
1.87
48.41
5.28
3873.21
40.56
35.6
HYDE
Engelhard
2605.55
22.59
2582.96
0
0
0
0
0
0
0
-75.818
35.6
DARE
Long Shoal
2082.72
471.51
1611.21
0.16
0.08
0
0.24
0.02
0.5
0.03
LSR1
-75.864
35.62
DARE
Long Shoal
3979.51
317.43
3662.08
0.16
3.66
0
3.82
0.1
7.74
0.21
SPB1
-75.771
35.7
DARE
Engelhard
4119.5
812.34
3307.16
0.16
15.43
0
15.6
0.47
31.66
0.96
48
Table 10. The number of hectares of forest and wetland area converted to developed and
agriculture land uses between 1980 and 2000. These land use change categories were used in
later multivariate analyses of land use change.
Land Use Changes
Forest to Developed
Forest to Agriculture
Wetland to Developed
Wetland to Agriculture
Agriculture to Developed
1980 - 1990
1990 - 2000
(hectares)
(hectares)
5,341
3,145
33,047
21,408
887
444
1,357
3,685
2,067
4,481
NC DMF Program 120 Trawl Data
Data Collection Methods
The North Carolina Division of Marine Fisheries has conducted juvenile fish and
invertebrate trawl surveys (“Program 120”) since 1972 to assess population trends in North
Carolina nursery areas. The NCDMF stations were initially selected on the basis of depth,
salinity, character of the bottom sediments, and the abundance of commercially important
estuarine-dependent fish and invertebrates. Monthly surveys were carried out March through
November each from 1972 until 1987; subsequent data collections were restricted to sampling in
May and June because of budget limitations. Consequently, we only analyzed the May and June
data for the entire 25-year period. Table 11 lists the 71 stations of focus, which were sampled in
May and June from 1978 to 2004.
49
Table 11. Sampling effort (number of stations visited each year and month) in the NC DMF
Program 120 data set.
Year
Total number
Stations sampled
Number of Stations
sampled in May
Number of Stations
sampled in June
1978
87
32
55
1979
136
67
69
1980
141
70
71
1981
146
73
73
1982
154
77
77
1983
183
91
92
1984
186
94
92
1985
195
97
98
1986
204
104
100
1987
206
105
101
1988
209
104
105
1989
206
103
103
1990
206
102
104
1991
207
104
103
1992
208
105
103
1993
204
101
103
1994
202
99
103
1995
204
102
102
1996
205
103
102
1997
206
103
103
1998
208
104
104
1999
206
103
103
2000
206
103
103
2001
208
104
104
2002
208
104
104
2003
208
105
103
2004
208
104
104
50
Stations were sampled using an otter trawl with a 10.5-foot-headrope and 3.2 mm mesh
cod-end that was towed for 1 minute (~ 75 m distance) in the middle of a sampling day.
Sampling of the 71 stations was not done on the same day in each month, because of NCDMF
manpower limitations. Although sampling days were spread throughout the months of May and
June across regions, all stations within a region (~10stations/region) were sampled on the same
day. Thus, it took approximately 1 week to ten days of sampling each month to sample all
stations. The NCDMF field crew recorded the total numbers of each of the selected species
collected per trawl. In addition, surface and bottom water temperature and salinity were
measured at each station using water quality meters, thermometers, and refractometers. Depth
was measured using a measurement stick or weight and line at the time of trawling. We used the
average of depths measured over 25 years. Dissolved oxygen was not recorded over the entire
sequence of sampling (only after 1987), so was not considered in the analysis here.
Statistical Analysis of Change in Program 120 Catches by Station
We plotted the 25-year time series of the catches of juvenile fishes and shrimp averaged
over all the stations (using a geometric mean) to examine the overall temporal trends in
abundance and assess how physical factors might have affected catches. Each of the selected
species abundances in trawls at each station were correlated with surface and bottom temperature
(◦C), surface and bottom salinity (psu, practical salinity unit or parts per thousand), and bottom
depth (m).
We created time series plots of mean (May and June trawls averaged) of each species’
abundance for all 71 stations from 1980 through 2004. Because of the extreme variability
inherent in these trawl data, we used a geometric mean to compute the averages by years, which
51
is a log-transformed mean number per trawl after adding 1 to each data point (which makes all
zeros = 1, but after log-transformation, they are 0 for computing the mean), then backtransforming using the exponential to get geometric mean number/trawl. This method of
averaging smoothed the data, and a LOWESS line (a locally weighted curve fit to the data, with
a smoother span f = 0.667, which is the proportion of the data used to make the smoothed fit at
each year) was superimposed over each plot to further identify the long-term trends over all these
stations. To analyze for trends in abundance of a species at a station or within a county, we used
a time series plot of the May and June catches per trawl with DWLS (Distance Weighted Least
Squares with a tension value = 0.2) line plots (SYSTAT 12.0). Distance weighted least squares
smoothing is similar to a moving average; it fits a line through a set of points by least-squares
method, but it weights the points near the mean being computed more than the points farther
away. Thus, the 25-year time-series line for each station and species is allowed to flex at any
given year on the series to fit the data. The amount of flex is controlled by the tension parameter.
This method produces a true, locally weighted curve running through the points (SYSTAT 12.0
user manual) using an algorithm attributed to McLain (1974).
Associated with the time series plots, are maps of each catchment in which we examine
the spatial and temporal patterns in the trawl catch data as a function of land use. We averaged
catches from each station into three 5-year periods (1980-1984, 1990-1994, and 2000-2004)
connected with each of the Landsat images(1980, 1990, and 2000). Intervening years were
omitted. This allowed us to use the classified Landsat imagery and catch data together in the
classification and regression tree analysis. In this way, the trawl catches of selected fishes and
brown shrimp and the associated land use change in each catchment can be visualized and
integrated. Figure 9 offers a schematic display of how these variables are related.
52
Figure 9. Schematic representation of the process used to relate land use change to changes in
species catch between 1980, 1990, and 2000.
For each of the species, we computed for every station a z-score of the normalized
change in mean catch during May and June using the following formulae:
( x( 2000−2004 ) - x(1980−1984 ) ) for station s
and
z-score of change in catch for station s =
(∆ s ) − x (∆ s )
σ ( ∆s )
where Δs is the change in catch at each station during the interval 1980-2004, x(1980−1984 ) is the
average catch at each station during the 5-year period 1980-1984 at the start of the time series
considered in this analysis, x( 2000−2004 ) is the 5-year average catch at each station at the end of the
53
time series period, and σ ( ∆ s ) is the standard deviation of the change in catch at each station.
These Z-scores of change in catch were computed for each of the target species in this report. A
z-score indicates how many standard deviations an observation is above or below the mean
change in catch for all stations. It allows comparison of observations of trawl catches with
different ranges, i.e., it normalizes the changes in catch for comparison across species and with
land-use. A z-score can be negative, positive, or zero. If z is negative, the corresponding Δs
value is below the mean of all stations. If z is positive, the corresponding Δs value is above the
mean of all stations. And if z = 0, the corresponding Δs value is equal to the mean of all stations.
For example, a z-score of 1.5 means that the Δs is 1.5 standard deviations above the mean. A
negative z-scores of 1.5 means that the species in question has declined by 1.5 standard
deviations below the long-term mean change in catch for that species at that station.
The z-score approach allowed us to assess, using a normalized index of catches for each
station, whether or not a station had exhibited a significant increase or decrease of each species
over the 25-year period between 1980 and 2004. A z-score ≤ -1.9 represents a statistically
significant decline (two standard deviations below the mean) for that station. A z-score between 1.96 and 1.96 represents no statistically significant change (within two standard deviations of the
overall mean) for that station. Finally, a z-score ≥ 1.96 represents a statistically significant
increase (two standard deviations above the mean) for that station. These symbols were plotted
on maps of land use change in the position where the sampling station is located in the catchment
(see Appendix).
We used the Classification and Regression Tree (CART) analysis (De’Ath and Fabricus
2000, King et al. 2005) in SYSTAT 11.0 to determine the most significant factors that affected
the 25-year average species abundances during 1980-2004 at each station (dependent variable).
54
For the CART analysis, the response variable was the z-score of catch change at each station (see
section above for computation), where the difference in average abundance at each station
between trawls taken in 1980-1984 and 2000-2004 was converted to a z-score (relative to the
differences in catch observed at all stations) to correct for heteroscedasticity and non-normality
of the data. The land use change predictor variables for the CART analysis were the total
percentage of land use change in each catchment; that is, the sum of (1) forested land changed to
developed land, (2) forested land that changed to agriculture, (3) agricultural land that changed to
development use, (4) wetland that changed to development use, and (5) wetland to developed, all
divided by the catchment land areas (see Table 9, last column). Additional landscape parameters
considered as predictor variables included (a) distance of each station to closest inlet in km, (b)
the number of National Pollution Discharge Elimination System (NPDES) point sources
(including animal feeding operations) within the 14-digit watershed containing the catchment
area, (c) human population within the census track containing the catchment area during the
2000 US Census , (d) average station depth in m, (e) 24-year average bottom temperature in °C
at time of capture, and (f) 24-year average bottom salinity in psu measured at time of capture.
Results
Land Use Change Patterns 1980-2000
Forested areas along the coast declined between 1980 and 2000; much of this land was
converted to agriculture. The percent of catchment area surrounding the NC DMF Program 120
stations represented by forest declined between 1980 and 2000, from a mean of 71.8 ± 2.4%
(standard error of the mean or SEM) in 1980, to 55.3 ± 2.4% in 1990, and to 43.1 ± 2.3% in 2000
55
(Figure 10). These values are similar when forested land use classes were summed across all
catchments in the study, then divided by total catchment area (Table 12). Over the same time
period, areal percentage of agriculture within these watersheds increased from 6.2 ± 0.8% in
1980 to 18.2 ± 2.0 % in 1990, and to 28.4 ± 2.0 % in 2000 (Figure 10). Thus the incidence of
forest as a dominant land form (defined as covering at least 55% of the land) decreased by nearly
half (from 57 catchments in 1980 to 28 catchments in 2000) over the course of these two
decades. Fifty-eight catchments showed a net loss of at least 10% of forest between 1980 and
2000; the average loss was 28.7% ± 8.2. The total amount of wetlands expressed comparatively
little net change in size, and actually showed a small increase in average percent cover of 4.5 ±
1.8 (Figure 10). Overall, there were no significant changes in percentage developed land (urban)
or wetlands in the catchments, although increases percentage of catchments covered by wetlands
and urban development within certain watersheds were evident (Figure 10 and Table 12).
56
100
Wetland
Forest
Agriculture
Urban
90
Percent land use
80
70
60
50
40
30
20
10
0
1975
1980
1985
1990
1995
2000
2005
Year
Figure 10. Change in percent forested and wetland area compared to agriculture and developed
area.
57
Table 12. Summary statistics of changes in land use types between 1980 and 2000. Values were
computed by summing total areas of catchments and land use class totals for all watersheds
surrounding the NC DMF stations.
Land use
type
1980
1990
2000
Total
Total
Total
area
area
area
Percent
Percent (ha)
(ha)
Percent (ha)
Agriculture
10335
7.5 31743.1
23.3 42621.2
31.6
Wetland
15146
11.0 18667.7
13.7 20913.1
15.5
Forest
104510
76.1 76698.5
56.3 61560.5
45.6
Developed
7371
5.4
9122.2
6.7
9834.1
7.3
Percent change
Percent Percent Percent
change change change
1980 to 1990 to 1980 to
2000
2000
1990
15.8
8.3
24.1
2.7
1.8
4.5
-19.8
-10.7
-30.5
1.3
0.6
1.9
Combining forest and wetland areas to serve as a proxy for “unaltered” land, and
combining agriculture and developed land area as a proxy for “altered land”, revealed that
unaltered land declined as a percentage of the catchment areas, while “altered” land increased
(Table 13). In 1980, forest and wetland covered 89.7 ± 1.6 % (SEM) of the total area of the 67
catchments; this percentage declined to 76.1 ± 2.4 % by 1990 and to 66.0 % ± 2.3 % by 2000. In
the same time period, agriculture and developed areas increased from 10.3 ± 1.6 % to 23.9 ± 2.4
% in 1990, and to 34.0 ± 2.3 % in 2000. Sixty-five of the 67 catchments were dominated forest
and wetland in 1980 (Table 13). By 1990, forest and wetland cover was dominant in only 53 of
the 67 catchments. In 2000, dominance of combined forest and wetland cover declined to 43
catchments.
58
Table 13. Number of catchments showing dominant types of land use between 1980 and 2000.
F+W = combined forest and wetland acreage, a proxy for “unaltered” land; A+D = combined
agriculture and developed, a proxy for “altered” land. A dominant land use type is here defined
as a type that comprises 55% or more of the catchment.
Watersheds Watersheds
Percent Percent dominated dominated
by A+D
by F+W
F+W
A+D
87.1
12.9
65
2
70.0
30.0
53
14
61.1
38.9
43
34
Year
1980
1990
2000
Reductions in the dominance of the combined forested and wetland categories were primarily
due to the loss of forested land (Figure 11).
100
Forest and Wetland
Agriculture and urban
Percent land use
80
60
40
20
0
1975
1980
1985
1990
1995
2000
2005
Year
Figure 11. The mean percentage of each catchment surrounding the 71 NC DMF stations that
was covered by “unaltered” land (forest and wetland) and “altered” land (agriculture and
development) in each of the years examined (1980, 1990, 2000) for land use on Landsat imagery.
Error bars are 1 SEM.
59
Number of Watersheds
67
54
40
27
13
0
1980
1985
1990
1995
2000
Year
Watersheds dominated by F+W
Watersheds dominated by A+D
Figure 12. Change in number of catchments dominated by “unaltered” land (forest and wetland)
versus the number of watershed dominated by “altered” (agriculture and developed) land.
Land used for agriculture showed the largest increases during the study period (Table 12).
Agriculture was a minor land category in 1980, but had become the dominant land category in 10
of the 67 catchments by 2000. The average increase in land used for agriculture between 1980
and 2000 was 22.1% ± 6.5. Increases in developed land were localized; thus at the level of all
of the catchments, developed land cover increased by an average of only 1.5% ± 5.9. Nearly all
of the gain in agricultural land was a consequence of conversion of forested land (Table 12).
Deforestation was also the primary source of new developed land. Some wetlands were also
converted to agriculture; six catchments showed a conversion of 10-20% of original wetland area
to agricultural use.
Loss of forested land was generally widely distributed among the counties containing the
nursery area catchments. Thus Beaufort, Hyde, Onslow, and Pamlico counties each contained
catchments that spanned the entire range of deforestation, from less than 10% to over 50% loss
60
(Table 14). These counties are also the best represented in the catchments sampled by NC DMF,
with each of them containing at least 8 stations. Beaufort County had catchments that exhibited
a wide range of increases in agriculture, although considerable gains were also evident in Hyde,
Onslow and Pamlico counties (Table 15). In contrast, the most extensive accumulation of
developed lands (> 20% of land area) was confined to 4 catchments in Onslow County (Table
16).
61
Table 14. Loss of forested land by county. The numbers within the table represent the number of
catchments within each county showing a particular percentage range of land lost or gained. The
numbers in parentheses indicate the total number of catchment per county.
County (catchment #) > 50 %
Beaufort (8)
2
Brunswick (1)
Carteret (10)
Craven (1)
1
Dare (3)
Hanover (3)
1
Hyde (15)
1
Onslow (8)
1
Pamlico (17)
2
Pender (1)
Totals
50-41 %
2
3
1
1
4
1
8
40-31%
1
3
1
1
2
1
4
-
12
30-21%
1
1
1
1
3
1
-
13
20-11%
1
2
1
7
2
3
-
8
< 10%
1
1
1
1
3
3
-
16
Total
8
1
10
1
3
3
15
8
17
1
10
67
Table 15. Gain of agricultural land by county. The numbers within the table represent the
number of catchments within each county showing a particular percentage range of land lost or
gained
County (cathment #) > 50%
Beaufort (8)
Brunswick (1)
Carteret (10)
Craven (1)
Dare (3)
Hanover (3)
Hyde (15)
Onslow (8)
Pamlico (17)
Pender (1)
50-41%
1
40-31%
3
30-21%
1
4
1
2
3
12
62
1
1
2
< 10%
Total
1
1
5
1
1
1
4
6
2
1
3
4
2
1
3
5
5
17
4
13
18
1
67
2
1
1
1
1
>20-11%
16
8
1
10
1
3
3
15
8
Table 16. Gain of developed land by county. The numbers within the table represent the number
of catchments within each county showing a particular percentage range of land lost or gained
County
Beaufort (8)
Brunswick (1)
Carteret (10)
Craven (1)
Dare (3)
Hanover (3)
Hyde (15)
Onslow (8)
Pamlico (17)
Pender (1)
> 50%
50-41%
40-31%
30-21%
20-11%
1
1
63
< 10% Total
8
8
1
1
10
10
1
1
3
3
3
3
15
15
3
4
8
17
17
1
1
3
63
67
Changes in Catches of Target Species (1980-2004)
We evaluated long-term trends in the NC DMF Program 120 trawls at 71 stations for
selected species over the period 1980-2004. Catches were recorded as number of individual fish
or shrimp per trawl during May and June each year at each station. We plotted the overall time
series for each species, computing a geometric mean (computed over all 71 stations and from
trawls in both months) for each year and fitting a LOWESS curve to the data. Next we
examined a z-score of change in catch (the difference between the 1980-1984 average and the
2000-2004 average catch), which was computed for each station and can be interpreted as the
change in normalized abundance (relative to all stations) in the trawls for each species (in units
of standard deviation). This normalized index allows for the comparison of relative changes at
stations and species with very different baseline abundances. Thus, a z-score of +1.0 indicates
and increase in catch by 1 standard deviation beyond the mean for a particular species and station
between 1980-1984 and 2000-2004. The z-scores for each species were first plotted to examine
the relative variation in catch among stations within each of the North Carolina coastal counties
(some counties have many stations, some only one station), and thereby relate trends in catch to
pattern in land use. Subsequently, z-scores for each species were plotted against the percentage
of land use that had changed in the catchments surrounding each station (between 1980 and
2000), with a LOWESS smoothing function applied to the data. These plots allow the reader to
make a visual inspection of the change in catch in NCDMF program 120 data for each species as
a function of land use change over the same 24-year period.
An appendix is provided at the end of this report that contains detailed maps of catches of
each species at each NCDMF station, and the surrounding catchment land use classifications that
were measured and used in the analysis. We provide notes in this section on the maps, indicting
64
which stations showed changes in catches. Changes in catch are represented by color codes
where red circles = decline, green circles = no change, and pink circles = increase. Red circles
indicate that catch means declined z-score values greater than 1.96; pink circles indicate that
mean catch increased by a z-score greater than 1.96. Green circles indicate that changes in catch
remained within the upper (positive) and lower (negative) z-score boundaries; accordingly, we
interpret this to mean that no significant change in catch occurred.
Spot
The catch for spot was highest overall, with a geometric mean catch of 114.7 fish/trawl
Over all stations, spot increased in mean abundance from a geometric mean of 60.7 per trawl in
1980 to178.4 per trawl in 2004, but this increase was not significant (Figure 12, correlation r =
0.25, p = 0.23). There were significant declines of spot in Carteret and Dare counties (Figure
13). Finally, there was a decline in catch z-score as land use change increased (Figure 14),
although a correlation analysis suggested this decline was not significant (r = -0.08, p = 0.51).
65
Figure 12. Spot, Leiostomus xanthurus. The geometric mean catch per trawl (May and June
averages from the 71 NC DMF stations) from 1978 through 2004. The curve is a LOWESS fit
(f=0.67). The correlation between year and geometric mean catch was not significant (r = 0.25, p
= 0.23).
66
Figure 13. Spot, Leiostomus xanthurus. Long-term change in catch (reported as a z-score of
change in trawl catches between 1980-2004; see Methods for computation of this index) for each
NCDMF station in different coastal counties. A z-score of 0 indicates no change; a negative zscore indicates a decline in the catch, and a positive z-score indicates an increase in catch
between 1980 and 2004.
67
Figure 14. Spot, Leiostomus xanthurus. Change in trawl catch z-score (1980-2004) at each
NCDMF station as a function of percentage change in land use (1980-2000) within each
catchment.
Notes on the station maps in Appendix: Out of 71 stations, three have had an increase (A2, A12,
and B10) in catch of juvenile spot, while two (CC11 and PAR7) have had a decline (Appendix
Map 1). Of the two stations showing declines, CC11 is located on the southwest prong of Lewis
Creek in Carteret County, and drains into Core Sound. The watershed around station CC11 has
had a negligible amount of land use change (~3%) during the study period, probably because it is
mostly covered by estuarine and riverine marsh, and therefore is not suitable to agriculture or
68
urban development. Station PAR7 is found on Porter Creek in Beaufort County, downstream
from the Potash Corporation mining site. Several large farming operations are located in its
wider watershed (Map 4). During the course of the study, the watershed upstream has lost around
33% of its forested area to farming and mining. Stations A2, A12 and B10, which showed
increases, are located in the lower Pamlico River area. They have had differing amount of forest
and wetland loss, which are respectively ~21%, 2.3% and ~0%, in the 20 years covered by
imagery.
Southern flounder
The average catch for Southern flounder was among the lowest of the species examined, with a
geometric mean = 3.7 per trawl, but their abundance has increased significantly from a geometric
mean of 2.8 fish per trawl 1980 to 4.2 per trawl in 2004 (Figure 15, correlation between year and
geometric mean catch, r = 0.44, p=0.03). Increases in z-score were most noticeable in Pamlico,
Hyde, Dare, Craven, and Brunswick counties (Figure 16). The z-score declined as percentage of
land use change increased (Figure 17), which is a significant linear decrease (linear regression:
southern flounder z-score in catch change = -1.47 * (% land use change) + 0.41, p=0.03).
69
Figure 15. Southern flounder, Paralichthys lethostigma. The geometric mean catch per trawl
(May and June averages from the 71 NC DMF stations) from 1978 through 2004. The curve is a
LOWESS fit (f=0.67).
70
Figure 16. Southern flounder, Paralichthys lethostigma. Long-term change in catch (reported as
a z-score of change in trawl catches between 1980-2004; see Methods for computation of this
index) for each NCDMF station in different coastal counties. A z-score of 0 indicates no change;
a negative z-score indicates a decline in the catch, and a positive z-score indicates an increase in
catch between 1980 and 2004.
71
Figure 17. Southern flounder, Paralichthys lethostigma. Change in trawl catch z-score (19802004) at each NCDMF station as a function of percentage change in land use (1980-2000) within
each catchment. Line is a Lowess fit.
Notes on the station maps in Appendix: A map of the changes in southern flounder catches
between 1980 and 2004 using standardized scores (z-scores) for the entire coastal region is
shown in Appendix Map 10. Appendix Maps 11 through 18 show time series plots at each
station of changes in southern flounder’s catch 1980-2004. Some stations have very high
average catches (FC3 and LSR1, Map 11; A2, A58, B10, B20, B40, B43 on Map13) and some
72
have low catches (SPB1, Map 11; CN1, CN14, CN3, CS13 and F1, Map 14; all stations on the
Carteret County region map, Map 16; the New river stations, Map 17; and the Cape Fear River
stations, Map 18). Some stations show an increase (AB1, SB3, Map 12 ; B20, B43, Map 13; D8,
Map 14; E10, E15,F3N, and J2, Map 15) and some display a slight decline in catch of juvenile
southern flounder ( F1, Map 15; NR13, Map 17).
The number of juvenile flounder caught in the 71 NCDMF stations between 1980 and
2004 remained unchanged at 68 stations, and significantly increased at three stations. These are
CFR1, Cape Fear station 1, located on the lower reaches of the Cape Fear River (Map 18);
station H2, Club foot creek in Craven County (Map 15), and D8 located on Dipping creek,
tributary of Long Creek in Pamlico County (Map 14). These stations have had respectively 14%,
~26% and ~0% land conversions from forest/wetland to agriculture urban areas. The stations
with increasing catch of juvenile southern flounder have no clear common denominator,
suggesting that multiple factors might be at play to cause the observed results.
Atlantic croaker
Atlantic croaker catches exhibited large fluctuations in abundance, with a low of 6.1
fish/trawl in 1991, highs of 58.4 fish/trawl in 1983 and an overall geometric mean abundance of
21.6 fish/trawl. Hyde and Dare County stations exhibited declines, while New Hanover and
Brunswick County stations exhibited increases (Figure 19). There was a decline in the z-score
of change in catch for Atlantic croaker with increased percentage of land use change (Figure 20),
although this was not a significant effect in a linear regression analysis (P = 0.493)
73
Figure 18. Atlantic croaker, Micropogonias undulatus. The geometric mean catch per trawl (May
and June averages from the 71 NC DMF stations) from 1978 through 2004. The curve is a
LOWESS fit (f=0.67).
74
Figure 19. Atlantic croaker, Micropogonias undulatus. Long-term change in catch (reported as
a Z-score of change in trawl catches between 1980-2004; see Methods for computation of this
index) for each NCDMF station in different coastal counties. A Z-score of 0 indicates no
change; a negative Z-score indicates a decline in the catch, and a positive Z-score indicates an
increase in catch between 1980 and 2004.
75
Figure 20. Atlantic croaker, Micropogonias undulatus. Change in trawl catch Z-score (19802004) at each NCDMF station as a function of percentage change in land use (1980-2000) within
each catchment.
Notes on the station maps in Appendix: Appendix Map 19 shows Atlantic croaker catches at 71
stations and land use changes in adjacent catchments. Two of the stations registered an increase
while three have had a decline. Stations A2, located on Betty Creek, tributary of Spring Creek in
Pamlico County and PAR16, Cypress Branch, secondary tributary of the East Prong Creek
76
showed an increase in juvenile Atlantic croaker catch between 1980 and 2004 (Map 22). Stations
F1, PAR11 and PUR 5 have experienced a decline. Station F1 is found in the upper reaches of
Orchard creek in Pamlico County and is a tributary of the Neuse River (Map 23). F1 catchment
has had about 21.5% of its forested land developed. Station PAR11 is found on South Creek in
Beaufort County, downstream from a heavy agricultural watershed, where about 51% of the
forested land has been converted to agriculture (Map 22); PUR5 is located on Warner Creek, a
tributary of Fortescue Creek in Hyde County and which drains to the Pamlico River (Map 21).
The upstream watershed has known a forest land conversion of about 38%.
Atlantic menhaden
The catches of Atlantic menhaden declined significantly over all stations from a geometric mean
of 8.6 fish/trawl during 1980 to 3.2 fish/trawl in 2002 with an overall geometric mean of 7.2
individuals per trawl (Figure 21, correlation between year and geometric mean catch r = -0.39, p
= 0.05). However, the Z-scores of average change in catch at the various counties’ stations was
remarkably consistent, with few long-term changes noted (Figure 22). There was no relationship
between the Z-score and the percentage of land use change within the catchments (linear
regression, P = 0.55; Figure 23).
77
Figure 21. Atlantic menhaden, Brevoortia tyrannus. The geometric mean catch per trawl (May
and June averages from the 71 NC DMF stations) from 1978 through 2004. The curve is a
LOWESS fit (f=.67).
78
Figure 22. Atlantic menhaden, Brevoortia tyrannus. Long-term change in catch (reported as a Zscore of change in trawl catches between 1980-2004; see Methods for computation of this index)
for each NCDMF station in different coastal counties. A Z-score of 0 indicates no change; a
negative Z-score indicates a decline in the catch, and a positive Z-score indicates an increase in
catch between 1980 and 2004.
79
Figure 23. Atlantic menhaden, Brevoortia tyrannus. Change in trawl catch Z-score (1980-2004)
at each NCDMF station as a function of percentage change in land use (1980-2000) within each
catchment.
Notes on Maps in the Appendix: Appendix Map 28 shows Atlantic menhaden catches at 71
stations on a color-coded map to show long-term changes. Sixty nine stations showed no
significant change and two registered a decline. The stations that show a decline in menhaden
80
catch are F3N, located on Pierce Creek in Oriental, NC (Map 33) and SQB3, on Oyster Creek off
Swanquarter Bay (Map 29).
Pinfish
The catches of pinfish increased significantly during 1980 through 2004, with a
geometric mean of 1.2 fish per trawl in 1980, peaking at 36.3 per trawl during 2001 (Figure 24,
correlation between year and geometric mean catch r = 0.65, p = 0.0002). Their dramatic
increase suggests that some external factor (land-based nutrient inputs, climate changes leading
to increased survival, increases in food availability, or the release from predation by a decline in
a predator in the food web) is causing them to increase at most stations. The differences in
mean catch between 1980-84 and 2000-2004 showed mostly positive values at the 71 stations.
The Z-scores of change in catch was variable among counties, but increased more in Pamlico,
Onslow, and Hyde counties (Figure 25). There was no decline in the Z-score of change in catch
of pinfish as the percentage of land use change increased (Figure 26, correlation between Z-score
and land use change r = -0.07, p = 0.59).
81
Figure 24. Pinfish, Lagodon rhomboides. The geometric mean catch per trawl (May and June
averages from the 71 NC DMF stations) from 1978 through 2004. The curve is a LOWESS fit
(f=0.67).
82
Figure 25. Pinfish, Lagodon rhomboides. Long-term change in catch (reported as a Z-score of
change in trawl catches between 1980-2004; see Methods for computation of this index) for each
NCDMF station in different coastal counties. A Z-score of 0 indicates no change; a negative Zscore indicates a decline in the catch, and a positive Z-score indicates an increase in catch
between 1980 and 2004.
83
Figure 26. Pinfish, Lagodon rhomboides. Change in trawl catch Z-score (1980-2004) at each
NCDMF station as a function of percentage change in land use (1980-2000) within each
catchment.
Notes on the maps in the Appendix: Appendix Map 37 displays the overall picture of changes in
pinfish catch at 71 stations, color-coded to show long-term changes. Sixty eight stations showed
84
no significant change while three showed a significant increase. The stations that showed an
increase in pinfish were CFR1 (lower Cape Fear, Map 45), H2 (Clubfoot Creek, tributary of the
lower Neuse River, and close to Cherry Point Air Force Base, Map 42), and D8 (Long Creek,
tributary Bonner Bay, which drains in Bay River, Map 40), an area that has been extensively
drained. Although three stations out of 71 are not a significant number, we find that they are all
associated with increased land use change.
Brown shrimp
Finally, brown shrimp were relatively abundant (geometric mean = 16.7 individuals/trawl)
during the study period, and increased from 20.3 individuals/trawl during 1980 to 37.0
individuals/trawl in 2002 (there was an extreme peak of 47.8 individuals/trawl in 1985 and 5.1
individuals/trawl an extreme low in 1998), however, this increase was not statistically significant
(Figure 27, correlation between year and geometric mean catch r = 0.09, p = 0.65). There were
long-term increases in the Z-score change in catch in Pamlico, Dare, Hyde and Craven counties
(Figure 28). However, there was no significant association between Z-score of change in catch
and percent land use change (linear regression, P = 0.77; Figure 29).
85
Figure 27. Brown shrimp, Farfantepenaeus aztecus. The geometric mean catch per trawl (May
and June averages from the 71 NC DMF stations) from 1978 through 2004. The curve is a
LOWESS fit (f=0.67).
86
Figure 28. Brown shrimp, Farfantepenaeus aztecus. Long-term change in catch (reported as a Zscore of change in trawl catches between 1980-2004; see Methods for computation of this index)
for each NCDMF station in different coastal counties. A Z-score of 0 indicates no change; a
negative Z-score indicates a decline in the catch, and a positive Z-score indicates an increase in
catch between 1980 and 2004.
87
Figure 29. Brown shrimp, Farfantepenaeus aztecus. Change in trawl catch Z-score (1980-2004)
at each NCDMF station as a function of percentage change in land use (1980-2000) within each
catchment.
Notes on maps in Appendix: Appendix Map 46 displays a map of overall changes in brown
shrimp catch along the North Carolina coast. Three of the stations showed a significant increase,
one station showed a decrease. The three that increased are FC3 (Middletown Creek, Map 47),
WB1 (Douglas Bay, Map 47) and WB3 (Wysocking Bay, Map 47), all in Hyde County. Overall
Brown shrimp catch declined slightly at station CC6 (Jarrett Bay) off Core Sound (Map 52).
88
Long-term changes in species by NCDMF stations
Most species did not show any long-term declines or increases in abundance in the May
and June trawls taken at individual NCDMF stations 1980-2004. Table 17 provides a summary
list of stations where the target species declined or increased. Using a Z-score of the change in
abundance at each station (equivalent to a confidence interval of 95%; Z-scores were computed
for each station of less 1.96, or greater than -1.96, see Z-score analysis in methods section), we
find that all species have remained unchanged at > 90% of the 71 stations (Table 17). Five
species have shown increases at a few stations: southern flounder (3 stations), Atlantic croaker (2
stations), spot (3 stations), pinfish (5 stations) and brown shrimp (3 stations). At a few stations,
Atlantic croaker (3 stations), spot (2 stations), Atlantic menhaden (3 stations) and brown shrimp
(1 station) have gotten less abundant over time. Interestingly, there seems to be a concentration
of the stations showing these declines in the lower Pamlico River for spot and croaker. Note that
some stations would be expected to show increases or declines by chance alone (5 % of the time,
or 3-4 stations), so these results are unsurprising.
89
Table 17 Catchment Areas of change in catch per species in the 71 stations
Species
Southern flounder,
Paralichtys
lethostigma
Atlantic croaker,
Micropogonias
undulatus
Spot,
Leiostomous
xanthurus
Pinfish,
Lagodon rhomboides
Atlantic menhaden,
Brevoortia tyrannus
Brown shrimp,
Farfantepenaeus
aztecus
Number of
stations without
a significant
change
(percentage)
Number of
stations with
catch increase
(Station IDs)
Number of
stations with
catch decline
(Station IDs)
68 (96%)
3
(D8,H2,CFR1)
0
66 (93%)
2
(PAR13,A2)
both stations with
increase are located
3
(PAR11,F3N,PUR5) in Lower Pamlico
66 (93%)
3 (A2,A12,B10)
2 (PAR7, CC11)
66 (93%)
5 (CFR1,H2,D8)
0
69 (97%)
0
2 (SQB3,F3N)
67 (94%)
3
(FC3,WB1,WB3) 1(CC6)
90
Notes
Decline at Potash
Corp., Cedar Island
Abiotic factors and abundance
Abiotic factors (bottom temperature and salinity measured by NC DMF personnel at the
time of the trawl collection) were associated with changes in abundance of the target species.
Note that dissolved oxygen was not measured by NCDFM until 1989, and no attempt was made
to include those data in analyses reported here.
Salinity and temperature affected the change in abundance of several species, but effects
were apparent only after multivariate analysis. Table 18 provides a summary of the geometric
mean catches per trawl and the significant abiotic factors influencing the change in abundance at
each station as analyzed using a Classification and Regression Tree (CART) for each species.
The complete results of the CART analyses will be discussed in the next section.
Bottom temperature was a significant factor associated with the change in abundance of
Atlantic croaker, Atlantic menhaden, and pinfish. Atlantic croaker increased more between 19802005 at stations where temperature was > 25 °C, and similarly, Atlantic menhaden increased
more where temperatures were greater > 24 °C. Pinfish increased more between 1980 and 2004
at stations where temperature was < 25 °C or over 26 °C, depending on other factors (see CART
results in the next section).
Overall, the most important abiotic factor for spot, flounder, and brown shrimp was
salinity (Table 18). Peak catches of juvenile spot in all years were associated with low salinity
stations (< 15 ppt, Figure 30), and peak catches of southern flounder occurred at stations with
intermediate salinity (10-15 ppt, Figure 31). Pinfish (Figure 32) peak catches were bimodal in
2000-04 (when they had increased to significant levels though the study area; no trends were
apparent in earlier years), with peaks both at low salinity (oligohaline, 5-10 ppt) and mesohaline
91
(20-25 ppt) stations. Brown shrimp abundance was highest at stations with high salinity in
1980-84, 1990-94, and 2000-04 (> 25 ppt, Figure 33).
The greatest increases of southern flounder and brown shrimp occurred in shallow depth,
and this was another significant abiotic factor for these species in the CART analysis (Table 18).
Table 18. Mean catches (number per trawl) of each of the target species and significant factors
affecting abundance at all stations identified in the Classification and Regression Tree (CART)
analysis, discussed in the next section.
SPECIES
Spot
Southern flounder
Atlantic croaker
Atlantic menhaden
Pinfish
Brown shrimp
Geometric
mean catch
per trawl
Bottom
temperature
Bottom
salinity
Depth
114.7
x
3.7
21.6
7.2
7.3
16.7
x
x
x
x
92
x
x
x
300
400
2000-04
1990-94
1980-84
0
100
200
Number/trawl
500
Spot
0
5
10
15
20
25
30
35
Salinity
Figure 30. The average catch of spot at stations with different salinities; LOWESS curves
(f=0.67) for 1980-84, 1990-94, and 2000-2004.
93
10
2000-04
1990-94
1980-84
0
5
Number/trawl
15
Southern flounder
0
5
10
15
20
25
30
35
Salinity
Figure 31. The average catch of southern flounder at stations with different salinities; LOWESS
curves (f=0.67) for 1980-84, 1990-94, and 2000-2004.
94
80
2000-04
1990-94
60
1980-84
0
20
40
Number/trawl
100
Pinfish
5
10
15
20
25
30
Salinity
Figure 32. The average catch of pinfish at stations with different salinities; LOWESS curves
(f=0.67) for 1980-84, 1990-94, and 2000-2004.
95
200
150
2000-04
1990-94
1980-84
0
50
100
Number/trawl
250
Brown shrimp
0
5
10
15
20
25
30
35
Salinity
Figure 33. The average catch of brown shrimp at stations with different salinities; LOWESS
curves (f=0.67) for 1980-84, 1990-94, and 2000-2004.
96
Multivariate CART analyses of land use change on target species
We used a multivariate approach called Classification and Regression Trees (CART) to
test our hypothesis that land use changes, along with abiotic factors or other anthropogenic
variables, were associated with changes in the long-term abundance of the juvenile stages of the
target species. In CART, the Z-scores of change in catch at each of station is partitioned into two
groups in a hierarchical way, so that the greatest amount of variation occurs between groups, and
the least amount within groups. Each predictor variable, including the percentage of land use
change within each catchment around the 71 NCDMF stations, is used to split the response
variable data though a recursive, iterative procedure to assess variation in Z-score of catch
change at each step. The predictor variable that explains the greatest amount of variation (has
the greatest PRE or Proportional Reduction in Error) is used to create the first partition of the
stations into two groups. The variable with next greatest PRE creates the second partition of each
of those groups, and so on. A threshold for dividing the groups on the basis of that variable is
also reported with each split. Splitting continues until the groups are too small to be split (PRE
is not significant). We assessed the influence of the following predictor variables on Z-score of
change in catch (1980-2004) for each station: (1) percentage of land use that changed between
1980-2000 within the surrounding catchment (PLU_CHAN), (2) bottom depth in m
(DEPTH_M), (3) bottom temperature in °C (BTEMP), (4) bottom salinity in ppt (BSALIN), (5)
the distance to the nearest inlet in km (DIS_INLE), (6) number of National Pollution Discharge
Elimination System discharge points in the 14 digit hydrologic unit area (NPDES), and (7) the
human population (people/km2) in the census tract containing the station’s catchment in 2000
(POP_2000).
97
Spot CART analysis
Spot showed positive catch change Z-scores (mean = 0.5 SD units) at 29 stations that
were greater than 42 km from the nearest inlet (first partition). At those stations, even greater Zscores (mean = 2 SD units) occurred at five stations where the human population density was
greater than 883 persons/km2 (second partition; Figure 34). A third partition was also significant
within the 24 stations where the human population did not exceed that density: if the human
population was less than 883 persons/km2, then positive catch Z-scores were observed at 18
stations located in less than 64 km from an inlet.
98
Figure 34. A Classification and Regression Tree (CART) analysis of the standardized z-score of
change in catch of spot between 1980 and 2000. Predictor variables included in CART are depth
in meters (DEPTH_M), distance to nearest inlet in km (DIS_INLE), bottom salinity (BSALIN),
bottom temperature (BTEMP), number of NPDES sites in the watershed (NPDES), total human
population in year 2000 (POP_2000) and overall change in land use as a percentage of catchment
area (PLU_CHAN).
99
Southern flounder CART analysis
For the southern flounder, the first CART partition involved 50 stations where the bottom
salinity was < 14.0 ppt; at these stations, there were positive Z-scores of catch change (mean =
0.3 SD units, Figure 35). In the second partition of these 50 stations, there were 20 stations that
had positive Z-scores (mean = 0.8 SD units) if the percentage land use change was less than 21.3
%; if the land use change exceeded that threshold, the Z-scores became slightly negative (mean =
0.1 SD units). Within the group of 30 stations that had land use change exceeding 21%, if station
depths < 0.9 m (N= 6 stations), then the Z-score was once again positive (mean = 0.8 SD units,
Figure 31). In our study, CART shows juvenile southern flounder catch increasing in lower
salinity water (<14 ppt) and low percentage of land use change (< 21%).
100
Figure 35. A Classification and Regression Tree (CART) analysis of the standardized z-score of
change in catch of Southern flounder between 1980 and 2000. Predictor variables included in
CART are depth in meters (DEPTH_M), distance to nearest inlet in km (DIS_INLE), bottom
salinity (BSALIN), bottom temperature (BTEMP), number of National Pollution Discharge
Elimination Sites (NPDES) in the watershed, total human population in the year 2000
(POP_2000) and overall change in land use as a percentage of catchment area (PLU_CHAN).
101
Atlantic croaker CART analysis
The first partition of the CART indicated that Atlantic croaker’s catches increased
slightly (mean = 0.1 SD units) at the 65 stations that had percentage land use change less than
54%; if there were greater change in land use, the Z-score of catch change became negative
(mean = -1.1 SD units at 6 stations, Figure 36) . The second partition of the 65 stations indicated
that the Z-score of change in catch increased at 21 stations (mean = 0.6 SD units) where the
distance to an inlet was greater than 47 km; at the 44 stations that were closer than 47 km to an
inlet, catches declined slightly (Z-score of catch change mean = -0.1 SD units) . The third
partition suggested that croaker increased more (mean = 0.9 SD units) at 14 stations (of 21
stations > 47 km from an inlet) that were less than 62 km from an inlet (Figure 32). Another
partition suggested that increased Z-scores occurred at 22 stations closer than 23 km from an
inlet, among the 44 stations closer than 47 km to an inlet. Finally, of the 14 stations between 47
and 62 km from an inlet, seven stations that had bottom temperatures greater than 25 °C and also
had high Z-scores of change in catch (mean = 1.7 SD units). Thus, for Atlantic croaker, the
CART analysis suggests than land use changes exceeding 54% are associated with declines in
catches (low Z-scores), but that stations between 47 and 62 km from inlets with temperatures
greater than 25 °C had increasing catches (greatest Z-scores) between 1980 and 2004.
102
Figure 36. A Classification and Regression Tree (CART) analysis of the standardized Z-score of
change in catch of Atlantic croaker between 1980 and 2000. Predictor variables included in
CART are depth in meters (DEPTH_M), distance to nearest inlet in km (DIS_INLE), bottom
salinity (BSALIN), bottom temperature (BTEMP), number of NPDES sites in the watershed
(NPDES), total human population in year 2000 (POP_2000) and overall change in land use as a
percentage of catchment area (PLU_CHAN).
103
Atlantic menhaden CART analysis
The first partition of the CART indicated that Atlantic menhaden’s catches increased
slightly (mean = 0.1 SD units) at the 66 stations that had bottom water temperatures > 24 °C
(Figure 37). The second partition of the 66 stations indicated that the Z-score of change in catch
increased at 20 stations (mean = 0.4 SD units) where the distance to an inlet was less than 24 km;
at the 46 stations that were greater than 47 km to an inlet, catches declined slightly (Z-score of
catch change mean = -0.1 SD units) . Percentage land use change was not a significant predictor
variable in the CART analysis for menhaden.
104
Figure 37. A Classification and Regression Tree (CART) analysis of the standardized Z-score of
change in catch of Atlantic menhaden between 1980 and 2000. Predictor variables included in
CART are depth in meters (DEPTH_M), distance to nearest inlet in km (DIS_INLE), bottom
salinity (BSALIN), bottom temperature(BTEMP), number of NPDES sites in the watershed
(NPDES), total human population in year 2000 (POP_2000) and overall change in land use as a
percentage of catchment area (PLU_CHAN).
105
Pinfish CART analysis
The first partition of the CART indicated that pinfish’s catches increased (mean = 0.43 zscore units) at the 27 stations that had temperature < 24.8 °C (Figure 38). If temperatures were
greater than this, the Z-score of catch change became negative (mean = -0.28 Z-score units at 44
stations). The second partition of these 27 stations indicated that the Z-score of change in catch
increased at 7 stations (mean = 1.5 Z-score units) where the population in 2000 was less than 2
persons/km2 and the temperature was less than 24.8 °C; if population exceeded this threshold,
catches remained close to the 1980-1984 level (catch change mean = 0.09 Z-score units at 20
stations). The third partition suggested that if the 2000 US Census population exceeded 2
people per km2, pinfish increased (mean = 1.08 Z-score units) at 5 stations with temperatures >
24. 7 °C (Figure 38). Thus, the CART analysis shows that pinfish increases 1980-2004 were
associated with stations with temperatures < 24.8 °C and low human populations in the
surrounding watershed during that period, or at temperatures > 24.7 C if the human populations
were higher than 2 persons/km2.
106
Figure 38. A Classification and Regression Tree (CART) analysis of the standardized z-score of
change in catch of pinfish between 1980 and 2000. Predictor variables included in CART are
depth in meters (DEPTH_M), distance to nearest inlet in km (DIS_INLE), bottom salinity
(BSALIN), bottom temperature(BTEMP), number of NPDES sites in the watershed, total human
population in year 2000 (POP_2000) and overall change in land use as a percentage of catchment
area (PLU_CHAN).
107
Brown shrimp CART analysis
The first partition of the CART indicated that brown shrimp catches increased (mean =
0.3 SD units) at the 50 stations that were > 21 km to the nearest inlet (Figure 39). If a station
were closer to an inlet than this, the z-score of catch change became negative (mean =
-0.7 SD units at 21 stations, Figure 39). In addition, a second partition within this group of 21
stations close to inlets suggested that there were relatively more shrimp (but still a declining
number, mean z-score= -0.3 SD) at 16 stations where the water temperature was not greater than
25 °C at the time of capture. The second partition of the 50 stations that were distant from inlets
indicated that the z-score of change in catch increased at 6 stations (mean = 1.7 SD units) where
the bottom water salinity >14 ppt; if the salinity was less than this threshold, catches declined (zscore of catch change mean = 0.1 SD units at 44 stations). The third partition within the 44
stations with salinity < 14 ppt suggested that brown shrimp increased (mean = 0.6 SD units) at
14 stations that had depths less than 0.9 m. The percentage land use change in the surrounding
watershed during 1980-2000 was not a significant factor for brown shrimp catches. Thus, brown
shrimp increases were associated with great distances from inlets, shallow water, and bottom
salinity greater than 14 ppt.
108
Brown shrimp, Farfantepenaeus aztectus
Decision Tree
MEAN = 0
SD = 1
N = 71
DIS_INLE < 20.91
MEAN = -0.7
SD = 1.1
N = 21
MEAN = 0.3
SD = 0.8
N = 50
PLU_8090 < 0.51
MEAN = -1
SD = 1.1
N = 16
BSALIN < 13.60
MEAN = 0.1
SD = 0.9
N=5
MEAN = 0.1
SD = 0.5
N = 44
BTEMP < 25.01
MEAN = -1.7 MEAN = -0.6
SD = 0.3
SD = 1.6
N= 6
N = 10
MEAN = 1.7
SD = 1.1
N=6
DEPTH < 0.96
MEAN = -0.1
SD = 0.3
MEAN = 0.6
SD = 0.5
N = 31
N = 13
Figure 39. A Classification and Regression Tree (CART) analysis of the standardized z-score of
change in catch of Brown shrimp between 1980 and 2000. Predictor variables included in CART
are depth in meters (DEPTH_M), distance to nearest inlet in km (DIS_INLE), bottom salinity
(BSALIN), bottom temperature (BTEMP), number of NPDES sites in the watershed (NPDES),
total human population in year 2000 (POP_2000) and overall change in land use as a percentage
of catchment area (PLU_CHAN).
109
Summary of the CART results
According to CART results (Table 19), distance from an inlet, bottom salinity, bottom
water temperature, percent land use change within a catchment surrounding the station, and
station depth are the factors that most influence these species.
Table 19. Summary of partitions of the CART tree for change in catch at 71 NCDMF stations
for the six target species (1= first CART tree split, based on the highest amount of variation
explained in z-score of change in catch 1989-2004, 2 = second CART tree split, etc.).
Species
Spot
Southern flounder
Atlantic croaker
Atlantic menhaden
Pinfish
Brown shrimp
Bottom
temperature
( oC )
Bottom
salinity
(ppt)
NPDES
(#
permits)
Distance
from an
inlet
Depth
(km)
(m)
1,3
1
2
Popula- % Land
use
tion in
change
2000
2
3
4
1
1,3
2
2
1
2,3
2
2
1
3
The number of NPDES permits in the watershed were not associated with changes in catch for
any species. Each species responded in somewhat different ways to the changes in land use and
other factors in the study period (1980-2004). It appears that southern flounder, Atlantic croaker,
and pinfish are the species significantly affected by an increased amount of land use change in
the catchment surrounding certain stations, which appeared to cause a decline in abundance
relative to other stations. Spot, Atlantic menhaden and brown shrimp did not show an association
with increased percentage of land use change in the surrounding watershed and declining catches
in the DMF program 120 trawls.
110
Discussion
Land use changed significantly between 1980 and 2000 in North Carolina coastal areas
near nursery area monitoring stations surveyed by the North Carolina Division of Marine
Fisheries (NCDMF Program 120), with most of the change being forested land converted to
agriculture, and to a lesser extent, suburban and urban development. In contrast, the abundance
of the juvenile fishes and brown shrimp in the estuarine areas immediately adjacent to the
changing land area stayed largely unchanged at more than 90% of the stations surveyed. Where
changes in abundance happened, however, they were associated with change in land use or
human population density. Declines in abundance of Atlantic croaker and southern flounder
were associated with increased land use associated with agriculture. Pinfish increased
significantly between 1980 and 2004 in watersheds where population density (based in the 2000
Census data) was low. Spot, brown shrimp, and Atlantic menhaden did not show any changes
associated with change in land use in the surrounding watersheds. Spot actually showed
increased abundance in watersheds where the US Census was > 883 persons/km2 in 2000.
This lack of agreement (significant land use changes and the lack of significant fish and
shrimp changes) may be due to four possible factors: 1) the change in land use or human
population did not exceed a threshold that would cause a detectable change in the water quality,
ecosystem quality and function or estuarine species and food webs (Holland et al. 2004); 2) the
change in juvenile fish abundance in estuarine areas was not detectable by the NCDMF Program
120 sampling because of inadequate statistical power of the methods used to detect change (1
trawl per station per month in May and June, which means that no replication was available for
statistical testing); 3) the impact of land use changes occurred at times other than the NCDMF
sampling period of May and June ( greater stress to aquatic organisms would be expected in July
111
through September when hypoxia associated with high water temperatures and benthic
respiration would be more likely (Christian et al. 2009)); or 4) there may be many more factors at
play such as varying substrate type (West et al.2000), ocean currents, as well as presence or
absence of aquatic vegetation as reported in Ross and Epperly (1985), or changes related to
global warming and sea level rise.
We tentatively conclude that that there was not a major land use effect on juvenile
estuarine fish populations in North Carolina between 1980 and 2004 (cause 1), with some
notable exceptions. The exceptions are southern flounder and Atlantic croaker, based on the
multivariate CART analyses, which suggest that where excessive land use change (i.e., > 20%
of the catchment converted from forested land to other uses between 1980-2004) occurred,
estuarine juvenile stages of southern flounder and Atlantic croaker declined. In addition, pinfish
increased only where human population density was low in a catchment. We note that these three
species (and possibly spot, which declined as the percentage of land use devoted to agriculture
increased, although this was change not significant in a linear regression) are all dependent on
benthic food resources. These species may be stressed by reduced benthic food availability that
persists at certain stations due to summertime hypoxia. In contrast, Atlantic menhaden, the one
planktivorous species considered in our analysis, did not show any observable trends in
abundance as land use changed in the surrounding watershed.
Given the lack of statistical power to such detect long-term changes in estuarine species
(Cause 2 above) and the lack of later summer sampling in Program 120 (Cause 3), our analyses
may have under-estimated the influence of land use change in North Carolina estuaries. We note
that in other estuarine systems [which were focused on blue crabs and mollusks in King et al.
(2005), macrobenthic taxa in Holland et al. 2004, and a benthic community Index of Biological
112
Integrity in Bilkovic et al. (2006) in Chesapeake Bay], land use changes have been associated
with degradation of benthic estuarine conditions. Runoff from agricultural fields has been
demonstrated to increase phytoplankton growth, depress SAV, degrade benthic habitats, and
lowering dissolved oxygen in receiving estuarine waters. We expected to see similar effects here
in North Carolina coastal areas, especially for Atlantic croaker and Southern flounder, which
depend directly or indirectly on benthos. The levels of change in previous studies in Chesapeake
Bay suggested that developed (urban) land use would have to exceed 10% of the land area, and
North Carolina coastal areas never reached such high levels (except in Onslow County, which
had consistently low catches and z-scores of catch change in our analyses). We conclude that
our analyses, because it focused on fishes and invertebrates that are mobile, highly variable
spatially and temporally, and operating at a higher trophic level than organisms measured in
these other studies, may have been insensitive to the level of land use changes that North
Carolina experienced between 1980-2004. It is possible that all three causes contributed to this
insensitivity.
In order to address the concern raised with possible cause 2 above, we recommend that a
power analysis of the existing Program 120 data sets (i.e., what amount of change in abundance
could be detected, given the past sampling effort and variability in the Program 120 data?). Our
initial analysis indicates that changes in land use must be large (> 20%) to show a detectable
influence on the Program 120 dataset, and that a 10% change in land use may not register a
strong signal, given the extreme variability in the juvenile stages of fishes and invertebrates.
Increased sampling each month at selected stations in the future (more trawls, perhaps 7-10 per
station, not simply a single trawl) may improve the statistical power to detect small changes in
fish and invertebrate abundance.
113
To address the third possible cause of the insensitivity, and to improve the monitoring of
coastal areas in the future, NCDMF should consider adding additional trawling surveys in the
late summer (July through September) at selected stations. This increased sampling effort
should be applied to selected watershed areas, but not at all 105 Program 120 stations, or even at
the subset of 71 stations considered in this study. The rationale for increased temporal sampling
of nursery areas is that the stress induced by summer temperatures, due to climate warming , and
increased urbanization of the coast, due to current development trends that emphasize waterfront
residences along the estuarine shoreline, will continue. A good pre-1990 late-summer baseline
Program 120 dataset for comparison with the increased sampling exists for all 71 stations
examined here. Thus, in the future, late summer trawling surveys at selected Program 120
stations can be compared to data taken in July – October in 1980-1989 (and data which we did
not use in this analysis, because we did not have post-1990 late summer data to compare it with).
This sampling may provide additional insight into cause 3 above (late summer stress due to
increase benthic community respiration) as a possible consequence of land use change. An
additional benefit of this approach would be an opportunity to monitor growth and survival of
these species after the peak in recruitment in May and June in watersheds with differing amounts
of land use. We would also recommend monitoring of benthic community respiration and
ecosystem function (benthic and planktonic communities, similar to work done in the
Chesapeake Bay noted earlier) at these times and at different stations with varying levels of landuse alteration.
114
Conclusion
Analysis of long-term land use change in coastal watersheds revealed an increase in the
amount of land devoted to agricultural use and a decline of forested areas between 1980 and
2000. There were negligible changes in wetlands and a slight increase in developed land for
urban and suburban residential uses. The long-term analyses of the Program 120 data indicated
that 90% of the stations were unaffected by land use change in the surrounding watershed, and
the abundance of Atlantic menhaden, and brown shrimp did not change significantly between
1980 and 2004. Atlantic croaker exhibited a long-term decline in abundance. At some stations,
spot, Southern flounder, and pinfish have increased, and a coastwide increase was noted for these
three species (based on geometric means and smoothed data plots). Most stations did not show
an increase in these species, however, and a few showed declines that can be attributed to land
use changes or human population increases. It appears that climate-related or other unexplained
factors may be causing an increase of these three species coast-wide. Stations with large scale
land use modifications (> 20% area changed, mostly from conversion of forest to agriculture) in
the surrounding watershed were associated with long-term declines (negative z-scores of catch
change) of some species (Atlantic croaker and southern flounder). Pinfish, while increasing, did
not increase if the human population was dense in the surrounding watershed. It appears that
these direct or indirect benthos-dependent species did not increase at these stations with
extensive land use change or human populations. The remaining species (spot, Atlantic
menhaden, and brown shrimp) did not show any land use associated declines in abundance, but
varied greatly due to abiotic factors.
115
References
Able, K. W., and M. P. Fahay 1998. The first year in the life of estuarine fishes in the Middle
Atlantic Bight. Rutgers University Press, New Brunswick, N.J. 342pp.
Adams, A. J., J.V. Locascio, and B.D. Robbins. 2004. Microhabitat use by a post-settlement
stage estuarine fish: evidence from relative abundance and predation among habitats. Journal of
Experimental Marine Biology and Ecology 299:17–33.
Anderson, J. R., E.E. Hardy, J.R. Roach, and R.E. Witmer. 1976. A land cover and land use
classification system for use with remote sensor data. U.S. Geological Survey Professional Paper
964. 36 pp.
Beach, D. 2002. Coastal sprawl: the effects of urban design on aquatic ecosystems in the United
States. Pew Oceans Commission, Arlington, VA, 32 pp.
Bilkovic, D. M., M. Roggero, C. H. Hershner, and K. H. Havens. 2006. Influence of land use on
macrobenthic communities in near shore estuarine habitats. Estuaries and Coasts 29:1185–1195.
Blanton, J. O., P. G. Verity, J. Amft, E. L. Wenner, C. A. Barans, D.M. Knott, B.W. Stender, and
S. B. Wilde. 2002. Key factors influencing transport of white shrimp postlarvae in southeastern
U.S. estuaries. Report to Georgia Sea Grant Program and South Carolina Sea Grant Consortium.
Boynton, W. R., J. H. Garber, R. Summers, and W. M. Kemp. 1995. Inputs, transformations and
transport of nitrogen and phosphorus in Chesapeake Bay and selected tributaries. Estuaries
18:285-314.
Burke, J. S., J. M. Miller, and D. E. Hoss. 1991. Immigration and habitat selection of
Paralichthys dentatus and Paralichthys lethostigma larvae and juveniles in an estuarine nursery
ground, North Carolina, USA. Netherlands Journal of Sea Research 27:393–405.
Burke, J. S., M. Ueno, Y. Tanaka, H. Walsh, T. Maeda, I. Kinoshita, T. Seikai, D.E. Hoss, and
M. Tanaka. 1998. The influence of environmental factors on early life history patterns of
flounders. Journal of Sea Research 40:19–32.
Cairns, Jr. J., and J.R. Pratt. 1992. Restoring ecosystem health and integrity during a human
population increase to ten billion. Journal of Aquatic Ecosystem Health 1:59-68.
Christian R. R., M. M. Brinson, J. K. Dame, G.J. Johnson, C. H. Peterson, and D. Baird. 2009.
Ecological network analyses and their use for establishing reference domain in functional
assessment of an estuary. Ecological Modeling. 220:3113-3122.
116
Congalton R. M. 1991. A Review of Assessing the Accuracy of Classifications of Remotely
Sensed Data. Remote Sensing of the Environment. 37:35-46.
Congalton, R. G., and K. Green. 1999. Assessing the Accuracy of Remotely Sensed Data:
Principles and Practices. Lewis Publishers, New York, 137 pp.
Cowardin, L.M., V. Carter, F.C. Golet, and E.T. LaRoe. 1979. Classification of wetlands and
deepwater habitats of the United States. U. S. Department of the Interior, Fish and Wildlife
Service, Washington, D.C. 79pp.
Crawford, W. T. 2007. Where does the coast sprawl the most? Trajectories of residential
development and sprawl in coastal North Carolina, 1971-2000. Landscape and Urban Planning
83:294-307.
Crossett, K.M. T. J. Culliton, P. C. Wiley, and T.R. Goodspeed. 2004. Population trends along
the coastal United States: 1980-2008. National Oceanic and Atmospheric Administration Coastal
Trends Report Series. 51pp.
Dauer D. M., J. A. Ranasinghe, and S. B. Weisberg. 2000. Relationships between Benthic
Community Condition, Water Quality, Sediment Quality, Nutrient Loads, and Land Use Patterns
in Chesapeake Bay. Estuaries 23:80–96.
De’ Ath, G., and K.E. Fabricius. 2000. Classification and regression trees: a powerful yet simple
technique for ecological data analysis. Ecology. 81:3178–3192.
Dobson, J. E., E. A. Bright, R. L. Ferguson, D. W. Field, L. L. Wood, K. D. Haddad, J. R.
Jensen, H. Iredale, V. V. Klemas, R. J. Orth, and J. P. Thomas. 1995. NOAA Coastal Change
Analysis Program (C-CAP): Guidance for Regional Implementation. NOAA Technical Report #
123.Washington, DC. 140pp.
Epifanio, C. E., and R.W. Carvine .2001. Larval transport on the Atlantic Continental Shelf of
North America: a review. Estuarine, Coastal and Shelf Science 52:51-77.
Ferguson R. L., and H. E. Mew. 2000. Proposal to Change Hydrologic Unit Delineations over
the Estuarine Waters of North Carolina. White Paper. Coastal Subcommittee, North Carolina
River Basin Coordinating Committee, Natural Resources Conservation Service.13pp.
Forward R. B., R.A. Tankersley, and K. A. Reinsel. 1998. Selective tidal stream transport of spot
(Leistomus xanthurus Lacepede) and pinfish Lagodon rhomboides (Linnaeus) larvae:
Contribution of circatidal rhythms in activity. Journal of Experimental Marine Biology and
Ecology 226:19–32.
Friedland K. D., D. W. Ahrenholz, and J. F. Githrie. 1996. Formation and seasonal evolution of
Atlantic menhaden juvenile nurseries in coastal estuaries. Estuaries 19:105–114.
117
Garmin International Inc. 2005. GARMIN GPSmap 76S User Manual. Olathe, Kansas
Goodchild M.F. 1994. Integrating GIS and Remote Sensing for Vegetation Analysis and
Modeling: Methodological Issues. Journal of Vegetation Science. 5:615-626.
Guidon K.Y., and J. M. Miller. 1995. Growth of juvenile southern flounder, paralichthys
lethostigma, in low salinity nursery areas of Pamlico Sound, North Carolina, USA. Netherlands
Journal of Sea Research. 34:89-100.
Hill J., and D. Aifadopoulou. 1990. Comparative analysis of Landsat-5TM and SPOT data for
use in multiple sensor approaches. Remote Sensing of The Environment 34:55-70.
Heck K.L. Jr., and R.J. Orth. 1980. Structural components of eelgrass (Zostera marina)
Meadows in the Lower Chesapeake Bay-Decapod Crustacea. Estuaries. 3:289-295.
Holland, A. F., D. M. Sanger, C. P. Gawle, S. B. Lerberg, M. S. Santiago, G. H. M. Riekerk, L.
E. Zimmerman, and G. I. Scott.2004. Linkages between tidal creek ecosystems and the landscape
and demographic attributes of their watersheds. Journal of Experimental Marine Biology and
Ecology. 298:151–178.
Jensen, J.R. 2007. Remote sensing of the environment: an earth resource perspective. Second
edition. Prentice Hall Series in geographic information science. Upper Saddle River, New
Jersey.592pp.
Kennish, M. J. 2002. Environmental threats and environmental future of estuaries.
Environmental Conservation. 29: 78–107.
King R. S., A.H. Hines, F. D. Craige, S. Grap. 2005. Regional, watershed and local correlates of
blue crab and bivalve abundances in sub-estuaries of Chesapeake Bay, USA. Journal of
Experimental Marine Biology and Ecology. 319:101–116
Leica Geosystems Geospatial Imaging, LLC.2005. ERDAS Field Guide. Software reference
document. Atlanta, GA.
Lillesand, T. M., R.W. Kiefer, and J.W. Chipman. 2004. Remote sensing and image
interpretation. John Wiley & Sons Limited. New York. 763pp.
Livingston R.J.2001. Eutrophication processes in coastal systems: Origin and succession of
plankton blooms and effects on secondary production in the Gulf Coast estuaries. 327p. CRC
Press Inc.
Luczkovich J.J, M.M. Brinson, T.L. West, G.F.R. Meyer 2007. Analysis of potential effects of
land cover change on Blue crab (Callinectes sapidus) populations in primary nursery areas.
Final Report, (05-POP-01), Blue Crab Resource Grant Program, NC Sea Grant, Raleigh, NC.
118
Malone, T. C., D. J. Conley, T. R. Fisher, P. M. Glibert, L.W. Harding, and K. G. Sellner. 1996.
Scales of nutrient-limited phytoplankton productivity in Chesapeake Bay. Estuaries and Coasts
19: 371-385.
Mclain D.H. 1974. Two dimensional interpolation from random data. Computer Journal. 19:178181.
NCDMF 2006. North Carolina fishery management plan: Shrimp. North Carolina Department of
Environment and Natural Resources. Division of Marine Fisheries. Morehead City, NC. 390pp.
NOAA C-CAP. 2002. Internal procedures for the implementation of land cover and change
analysis projects. Version 2.0. Charleston, SC. 71pp.
Powell A.B., and F. J. Schartz. 1977. Distribution of Paralichthid Flounders (Bothidae:
Paralichthys) in North Carolina estuaries. Chasapeake Bay Science. 18:334-339.
Powell, A. N., and T. Henley. 1995. Egg and larval development of laboratory-reared gulf
flounder, Paralichthys albigutta and southern flounder, P. lethostigma (Pisces, Paralichthydae).
Fishery Bulletin. 93: 504–515.
Rakocinski C.F., M. S. Peterson, B.H. Comynsa, G. A. Zapfe, and G. L. Fulling. 2006. Do
abiotic factors drive the early growth of juvenile spot (Leiostomus xanthurus)? Fisheries
Research. 82:186-193.
Ross S.W., and S.P. Epperly.1985. Utilization of shallow estuarine nursery areas by fishes in
Pamlico Sound and adjacent tributaries, North Carolina. In: Yanez-Arancibia A. (ed)
Fish community ecology in estuaries and coastal lagoons: Towards an ecosystem integration.
UNAM Press, Mexico. p 207–232.
Rulifson, R. A. 1981. Substrate preferences of juvenile penaeid shrimps in estuarine habitats.
Contributions in Marine Sciences. 24:35-52.
Street, M.W., A.S. Deaton, W.S. Chappell, and P.D. Mooreside. 2005. North Carolina Coastal
Habitat Protection Plan. North Carolina Department of Environment and Natural Resources,
Division of Marine Fisheries, Morehead City, NC. 656 pp.
Tortora, R. 1978. A note on sample size estimation for multinomial populations. The American
Statistician 32:100-102.
U.S. Geological Survey and U.S. Department of Agriculture, Natural Resources Conservation
Service.2009. Federal guidelines, requirements, and procedures for the national Watershed
Boundary Dataset: U.S. Geological Survey Techniques and Methods 11–A3, 55 p.
119
Warlen, S. M., J.S. Burke. 1990. Immigration of larvae of fall/winter spawning marine fishes
into a North Carolina Estuary. Estuaries. 13: 453-462.
Warlen S.M.1994. Spawning time and recruitment dynamics of larval Atlantic menhaden,
Brevoortia tyrannus,into a North Carolina estuary. Fishery Bulletin. 92:420-433.
Watterson, J. C., and J. P. Monaghan. 2001. Flounder pound net fishery. In: Assessment of North
Carolina commercial finfisheries, 1997–2000.
North Carolina Department of Environment and Natural Resources, Division of Marine
Fisheries, Completion Report NA 76 FI 0286. Morehead City, NC.
Weisberg S. B., J.A. Ranasinghe, D.M. Dauer, L.C. Schaffner, R.J. Diaz, and J.B. Frithsen.
1997. An estuarine benthic index of biotic integrity (B-IBI) for Chesapeake Bay. Estuaries.
20:149-158.
Wenner E. L., and H. R. Beatty. 1993. Utilization of Shallow Estuarine Habitats in South
Carolina, U.S.A. by postlarval and juvenile stages of Penaeus spp. (Decapoda: Penaeidae).
Journal of Crustacean biology. 13: 280-295.
West, T.L., L.M. Clough, and W.G. Jr Ambrose. 2000. Assessment of function in an oligohaline
environment: Lessons learned by comparing crested and natural habitats. Ecological
Engineering. 15:303–321.
Zimmerman, R. J., T. J. Minello, and G. Zamora. 1984. Selection of vegetated habitat by
Penaeus aztecus in a Galveston Bay salt marsh. United States National Marine Fisheries Service
Fishery Bulletin. 82:325-336.
120