Biological Conservation 198 (2016) 60–69

Contents lists available at ScienceDirect

Biological Conservation

journal homepage: www.elsevier.com/locate/bioc

Seeing the forest from drones: Testing the potential of lightweight drones
as a tool for long-term forest monitoring
Jian Zhang a,b,c,⁎,1, Jianbo Hu d,1, Juyu Lian e, Zongji Fan e, Xuejun Ouyang e, Wanhui Ye e
a
School of Ecological and Environmental Sciences, East China Normal University, Shanghai 200241, China
b
Tiantong National Forest Ecosystem Observation and Research Station, Ningbo 315114, Zhejiang, China
c
Section for Ecoinformatics and Biodiversity, Department of Bioscience, Aarhus University, Ny Munkegade 114, Aarhus C, DK-8000, Denmark
d
Tianjin Research Institute of Water Transport Engineering, Ministry of Transport, Tianjin 300456, China
e
Key Laboratory of Vegetation Restoration and Management of Degraded Ecosystems, South China Botanical Garden, Chinese Academy of Sciences, Guangzhou 510650, China

a r t i c l e i n f o a b s t r a c t

Article history: Long-term ecological monitoring has contributed substantially towards advancements in theoretical and applied
Received 9 November 2015 ecology. However, the costs to maintain a long-term monitoring site are enormous. Lightweight unmanned aerial
Received in revised form 22 January 2016 vehicles (UAVs or drones) have been rapidly emerging as a new tool for local-scale monitoring. To evaluate the
Accepted 21 March 2016
value of drone applications in long-term ecological studies, we combined drone-derived canopy variables,
Available online 18 April 2016
detailed ground-based stem-mapping data and topographic and edaphic variables from a 20-ha forest dynamics
Keywords:
plot in a species-rich subtropical forest. Specifically, we evaluated the relative importance of these variables in
Drone ecology explaining local-scale variation in forest stand and species measures. We found that drone-derived canopy var-
Forest canopy iables contributed substantially towards explaining local patterns of biodiversity and more specifically in
Forest dynamics supporting a gap dynamics hypothesis in structuring observed forest biodiversity. Stand basal area was positively
Remote sensing related with canopy closure, indicating the importance of protecting old-growth forests as carbon sinks. The
Species coexistence importance of topographic and edaphic variables was also demonstrated, supporting a niche differentiation hy-
Unmanned Aerial Vehicle (UAV) pothesis in structuring patterns in biodiversity. Species-level analyses illustrated that light-demanding species
were more strongly correlated with canopy variables than shade-tolerant species. We provide convincing
evidence that drones can add substantial value to long-term ecological monitoring by providing low cost, high
resolution data. Drones should be included in the ecologist's toolbox to complement traditional field surveys.
© 2016 Elsevier Ltd. All rights reserved.

1. Introduction frequent as required for analysis and monitoring of short-term

change. Therefore, a key challenge remains on how to collect forest
High-quality biodiversity data on species' distributions and its attribute data in a timely and cost effective manner.
integration with environmental variables are critical for addressing Remote sensing techniques are increasingly being used to assess
basic research questions in ecology, tracking biodiversity changes, and changes in forest cover (Hansen et al., 2013; Nijland et al., 2015), tree
developing effective conservation actions. Although we gain a wealth density (Crowther et al., 2015), species distributions (Cord et al.,
of knowledge by spending an enormous amount of time and energy 2013), canopy height (Simard et al., 2011; Nijland et al., 2015; Zhang
in the field, traditional field surveys can be exhausting and costly et al., 2016), and carbon stocks (Saatchi et al., 2011; Zhang et al.,
(Lawton et al., 1998; Gardner et al., 2008). For example, a field team of 2014). However, satellite and airborne sensors can be expensive and
12 to 14 individuals took ~3 years to complete the first tree census for inaccessible for most researchers, requiring trade-offs between resolu-
a 50 ha forest dynamics plot in Barro Colorado Island, Panama. tion, scale, and frequency (Anderson and Gaston, 2013). For ecological
Costs to establish similar plots are estimated at about US$100,000 studies at local and regional scales, satellite and airborne data are not
to US$500,000 (Condit, 1998). Additional measurements and moni- often well-suited to the scale of the study (Wulder et al., 2004). Small
toring of tree height, canopy openness, forest disturbance and other unmanned aircraft systems, also known as lightweight unmanned aerial
forest parameters are limited by available human labor and financial vehicles (UAVs) or drones, provide “a promising route to responsive,
resources. Due to these limitations, ground-based surveys are not as timely, and cost-effective monitoring of environmental phenomena”
(Anderson and Gaston, 2013). Although drones in military applications
have a relatively long history, civilian applications have only recently
⁎ Corresponding author at: School of Ecological and Environmental Sciences, East China
Normal University, Shanghai 200241, China.
emerged (Koh and Wich, 2012; Anderson and Gaston, 2013). Pioneering
E-mail address: jzhang1982@gmail.com (J. Zhang). ecologists and conservation biologists have recently been using drones to
1
These authors contributed equally to this work. monitor wildlife and plant populations (Jones et al., 2006; Chabot and

http://dx.doi.org/10.1016/j.biocon.2016.03.027
0006-3207/© 2016 Elsevier Ltd. All rights reserved.
 J. Zhang et al. / Biological Conservation 198 (2016) 60–69 61

Bird, 2012; Vermeulen et al., 2013), wildlife poachers (Schiffman, 2014), windfall and drought, may not be captured by satellite remote sensors
vegetation structure (Getzin et al., 2012; Dandois and Ellis, 2013; Puliti (Wulder et al., 2004). Therefore, drone-based ecosystem monitoring
et al., 2015), and mapping of land cover change (Rango et al., 2009). that can be applied at the temporal and spatial scales relevant to ground
Compared with satellite and airborne remote sensing techniques, drones measures will greatly benefit long-term studies of ecological properties,
can fly at low altitudes and at slow speeds, allowing them to take ultra- potentially “revolutionizing spatial ecology” (Anderson and Gaston,
high spatial resolution (1–20 cm) imagery and thereby collect near- 2013).
earth data of plant and wildlife populations and biophysical variables In this study, we explore the utility of using lightweight drones as a
(Rango et al., 2009; Koh and Wich, 2012; Whitehead and Hugenholtz, flexible, cost-effective, and accurate method for mapping forest stand
2014). Using drones also avoids many limitations associated with satellite characteristics in a 20-ha CTFS-ForestGEO subtropical forest plot in
data, including the lack of sufficient spatial resolution to detect and mea- China. By combining aerial photographs collected by the drone with
sure certain critical biophysical properties (e.g., forest canopy gaps and photogrammetry and using detailed ground survey data on species
single-tree identification), the lack of sufficient temporal resolution data distribution, topography and edaphic variables, we mapped forest canopy
to detect changes in phenology and stand structure by disturbance events, structure, and analysed the relative contribution of drone-derived canopy
and long-duration cloud contamination over many types of tropical and attributes, topography and edaphic variables to observed patterns of
subtropical forests (Paneque-Gálvez et al., 2014; Whitehead et al., biodiversity and tree regeneration. Specially, we assess: (1) the feasibility
2014). Additionally, the cost of one camera-equipped drone is relatively of using drone technology to collect high resolution aerial photographs for
low (Koh and Wich, 2012). Despite these advantages, current cost- mapping three-dimensional (3D) forest canopy structure; (2) to what
effective drones have relatively limited spatial extent per flight, small extent drone-derived canopy attributes contribute to our understanding
payloads and low spectral resolution (Paneque-Gálvez et al., 2014; of local-scale patterns in biodiversity and biomass storage; and (3) how
Whitehead et al., 2014), and therefore this technology has yet to receive species with different life history strategies (i.e., light-demanding vs.
much attention by field ecologists, especially for long-term ecological shade-tolerant species) respond to drone-derived canopy attributes and
studies. other environmental variables.
Long-term ecosystem monitoring is the keystone of ecological re-
search and management (Callahan, 1984; Likens, 1989; Condit, 1995; 2. Materials and methods
Lindenmayer et al., 2012) because these data provide important insights
to complex ecological systems. There are a number of well-known long- 2.1. Ground inventory data
term ground-based monitoring programs, such as International Long
Term Ecological Research Network (ILTER, http://www.ilternet.edu), This study was conducted in a 20-ha (500 m × 400 m) forest dynam-
North American Breeding Bird Survey (BBS, https://www.pwrc.usgs. ic plot in the Dinghushan (DHS) National Nature Reserve (23°09′–
gov/bbs), United States Forest Inventory and Analysis National Program 23°11′N, 112°30′–112°33′E) in Southern China (Fig. 1). The DHS
(FIA, http://www.fia.fs.fed.us), Amazon Forest Inventory Network reserve, which was established in 1956 as China's first nature reserve,
(RAINFOR, http://www.rainfor.org), and the Center for Tropical Forest encompasses approximately 1155 ha of forests with elevations ranging
Science and Forest Global Earth Observatory (CTFS-ForestGEO, http:// from 14.1 m to 1000.3 m above sea level. This reserve joined the In-
www.forestgeo.si.edu). Using the CTFS-ForestGEO network as an exam- ternational Man and Biosphere Reserve Network (MAB) as a global
ple, this network comprises over 60 plots from 24 countries across the conservation hotspot in 1979. The region is characterized by a south sub-
Americas, Africa, Asia, and Europe, on over 10,000 woody plant species, tropical monsoon climate, with mean annual temperature of 20.8 °C, and
and more than 6 million living individuals (Anderson-Teixeira et al., monthly mean temperatures varying between 12.6 °C in January to 28 °C
2015). Since the first 50-ha forest dynamics plot was established at in July (Huang et al., 1998). Mean annual precipitation is 1929 mm, with
Barro Colorado Island in Panama in 1980, over 200 researchers have most precipitation occurring between April and September. Mean annual
published over 1000 scientific articles using CTFS-ForestGEO data. evaporation is 1115 mm with relative humidity averaging 80% (Huang
These publications have had a substantial impact across a large variety et al., 1998). The vegetation is mainly covered by well-protected monsoon
of science and policy issues (e.g., Hubbell et al., 1999; Harms et al., evergreen broadleaved forest. In contrast to the surrounding disturbed
2000; Hubbell, 2001; Condit et al., 2006; John et al., 2007; He and forests, the reserve contains rare primary forests of at least 400 years of
Hubbell, 2011; Stephenson et al., 2014). Likewise, remote-sensing- age that were conserved by monks at the Buddhist temple near the plot
based long-term monitoring, such as the National Ecological Observato- (Fig. 1c).
ry Network (NEON, http://www.neoninc.org) and NASA's Moderate The 20 ha DHS plot was established in 2004–2005 (Fig. 1). Following
Resolution Imaging Spectroradiometer (MODIS) platform, have contrib- the protocols from the CTFS-ForestGEO network (Condit, 1998), all
uted substantially to our understanding of a variety of ecosystems at re- stems with ≥ 1 cm diameter at breast height (DBH) were tagged,
gional and global scales (Kerr and Ostrovsky, 2003). georeferenced, and identified to species. The first inventory of this plot
Despite the tremendous value of these long-term data sets, many was completed in 2005, and has been re-censused at five-year intervals
data gaps remain. First, ground-based monitoring sites only cover a ever since. Data for the 2010 census were used for the current analysis
small fraction of the Earth's surface and are not representative with sev- which includes 60,015 living individual stems with ≥ 1 cm DBH
eral geographic biases. Martin et al. (2012) analysed the distributions of representing 177 woody plant species. Among these species, 47% (84 spe-
2573 terrestrial ecological sites, and found that these sites overrepre- cies) are considered rare as defined as being less than one individual per
sented protected areas and wealthy countries, and were rarely distrib- hectare (Ye et al., 2008). Based on importance values, Chinese chestnut
uted in the 75% of the terrestrial world where humans live. One main (Castanopsis chinensis, Fagaceae), Chinese gugertree (Schima
reason for the paucity of these ground-based sites is their high monitor- superba, Theaceae), and yellow basket-willow (Engelhardtia roxburghiana,
ing and maintenance costs. Second, challenges remain between linking Juglandaceae) are the three most dominant species.
broad-scale remote sensing data with local-scale ground data (Kerr and
Ostrovsky, 2003; Wulder et al., 2004). The problem of mismatch of spa- 2.2. Topographic and edaphic variables
tial scales results in limitations to monitoring and predictions of species
distributions and dynamics (e.g., Saveraid et al., 2001). Likewise, mis- Topography of the study plot was measured by four variables:
matches in temporal scales also occur. Broad-scale remote sensing elevation, slope, aspect, and convexity. Elevation was surveyed on
data are not generally as frequent as required to address a number of a 20 m × 20 m grid within the 20-ha plot using an Electronic Total
pressing ecological questions. For instance, human and natural distur- Station with elevation values averaged from the four corner of
bance events at local and/or regional scales, such as forest harvesting, each 20 m × 20 m quadrat (Ye et al., 2008). Slope was defined as
62 J. Zhang et al. / Biological Conservation 198 (2016) 60–69

Fig. 1. Geographic location of the study area in Dinghushan National Nature Reserve, Southern China. (a) Aerial photography of our study area. The dotted blue lines represent the bound-
ary of the 20-ha (400 m × 500 m) forest dynamics plot. (b) Point cloud map for DSM of the region covered by the aerial survey using the drone. (c) This image shows the Buddhist temple
used for takeoff and landing. (d) This image illustrates a small area with close canopy occupied by multiple species, while (e) illustrates an area with open canopy. The two red circles in
(e) highlight the locations of two ground-level seed traps.

the mean angular deviation from the horizontal of each of the 2.3. Aerial drone survey
four triangular planes formed by connecting three of the corners of
each 20 m × 20 m quadrat. Aspect was defined as the compass direc- The Microdrones MD4-1000 small drone (https://www.microdrones.
tion in which a slope faces. Convexity of each 20 m × 20 m quadrat com) was used for aerial survey of the 20-ha plot (Appendix A). This UAV
was calculated as the elevation of the focal quadrat minus the mean weighed 2.65 kg, had a cruising speed of 12 m/s, maximum flight duration
elevation of eight surrounding quadrats. For the edge quadrat, convexity of 88 min without payload under optimal weather conditions, and maxi-
was the elevation of the center point minus the mean of its four corners mum payload mass of 1.2 kg. This quadcopter (four brushless battery
(Harms et al., 2001; Wang et al., 2009). The landform of this plot is com- powered motors) drone can fly by remote control or autonomously
plex, with a 230 m range in elevation (240–470 m), resulting in numerous with the aid of its GPS receiver and its waypoint navigation system. A
extremely steep slopes ranging from 30° to 50° (Wang et al., 2009). The Sony NEX-5 still-photograph camera was mounted to the bottom of the
elevation of each 1 × 1 m2 point was interpolated by ordinary kriging drone to acquire aerial imagery. Our team member (J.H.) has used this
from the 20 m × 20 m base grid. The ordinary kriging interpolation tech- system for over 200 successful missions investigating land cover changes
nique has been used in many CTFS-ForestGEO plots (e.g., Harms et al., and vegetation dynamics.
2001; John et al., 2007) and recent drone study by Dandois and Ellis The flight mission was planned with the mdCockpit software that
(2013). The accuracy of this technique was judged by the mean error came with the drone. The flight route was planned using orthoimages
(ME) and the root mean square error (RMSE) between estimations and a digital elevation model (DEM) of the flight area (Appendix A).
and observations. The values of ME and RMSE were 0.04 m and 4.28 m To keep the pixel size of the orthoimages relatively constant, the flight
respectively. could not be horizontal (as with traditional fixed-wing planes) because
Edaphic variables were collected using a standard protocol de- of the large elevational difference (230 m) in this study site. The vertical
veloped by the CTFS-ForestGEO (John et al., 2007). Specifically, we distance between the drone and the ground surface was kept at about
collected topsoil samples (0–10 cm depth) by using a regular grid 240 m. Total flight time was nearly 28 min with the speed of 5 m per
of points at each 30 m interval within the 20-ha plot. Each sampling second to cover the whole study plot (Fig. 1 & Appendix A). In total,
point was paired with two additional sampling locations at 2, 5, or we collected 312 images with about 70% average overlap and the aver-
15 m in a random compass bearing from the point to capture fine- age pixel size was 4.3 cm.
scale variation in soil properties (John et al., 2007). In total, 710 Aerial photographs were processed into georeferenced orthoimages,
samples were collected with eight edaphic variables measured for a digital surface model (DSM), and point clouds using the photogram-
each sample. These variables included soil organic carbon (SOC), metric software “Pix4dmapper” (http://pix4d.com), which has been
available potassium (AK), total potassium (TK), available phosphorus widely used for UAV photogrammetry. Based on the aerial images
(AP), total phosphorus (TP), available nitrogen (AN), total nitrogen collected from the drone, we selected 9 ground control points (GCPs)
(TN), and soil pH (Lin et al., 2013). Soil values for each 20 m × 20 m quad- to geocorrect the point cloud. The XYZ locations of each GCP were
rat were calculated using ordinary kriging, and the accuracy of these esti- measured using a Trimble RTK (Real-Time Kinematic) GPS within 1 m
mations was reported in Lin et al. (2013). accuracy (UTM Zone 49 N, WGS84 horizontal datum). We followed a
 J. Zhang et al. / Biological Conservation 198 (2016) 60–69 63

standard workflow in photogrammetry to extract the DSM and ortho- correlations (coefficient N 0.7). After collinearity assessments, we kept
mosaic from images (McGlone, 2013). Then we evaluated the spatial ac- four drone-derived canopy variables (canopy height, mean height, stan-
curacy using the RMSE by comparing digitized and known coordinates dard deviation of heights, and canopy closure), three topographic vari-
from ground. Finally, we generated the DSM with a resolution of 5 cm. ables (elevation, convexity and slope), and four edaphic variables (soil
The RMSEs were 32 cm in X (east), 44 cm in Y (north), and 69 cm in Z. organic carbon, available potassium, total potassium, and total phos-
phorus). Considering the spatial accuracy of topographic and edaphic
2.4. Canopy height model (CHM) generation and canopy height metrics predictor variables, we primarily limited our statistical analysis to the
20-m scale (grain resolution). This spatial scale is also widely used for
We generated a canopy height model (CHM) with a pixel size of 1 m field sampling in forest communities. It is worth mentioning that the
by subtracting the DEM from the DSM. We then developed a suite of for- further analyses at finer scales can shed light on many important eco-
est canopy metrics across the 20-ha plot at three spatial scales (5 m, logical questions, such as forest regeneration and phenology, and there-
10 m, and 20 m) by aggregating the CHM data with 1 m resolution to fore should explore deeply depending on research questions and overall
three raster grids consisting of cells of 5 m × 5 m, 10 m × 10 m and data quality of available attributes.
20 m × 20 m, respectively. In total, there were 8000, 2000 and 500 Spatial simultaneous autoregressive error models (SARs), which
cells at the scales of 5 m, 10 m and 20 m, respectively. These metrics allow the inclusion of residual spatial autocorrelation in data (Kissling
included canopy height (the average of three highest values), mean et al., 2008), were used to evaluate the relative importance of each var-
height, skewness of the heights, standard deviation of heights, vertical iable to forest stand and species patterns. For each analysis, all possible
distribution ratio (VDR; Goetz et al. (2007)), and canopy closure. We combinations of 11 predictor variables were used to fit the models.
calculated VDR using the equation: VDR ¼ HT max HT med
, where HTmax and Among all the 2047 (211-1) combinations, the best combination of var-
HT max
iables was selected by comparing all model subsets using Akaike's Infor-
HTmed were maximum and median values of canopy height at each
mation Criterion (AIC) (Burnham and Anderson, 2002). Then, for the
grid cell. We quantified canopy closure by the percentage of 1 × 1 m
‘best’ model, we calculated the relative importance of predictor vari-
pixels with N10 m height. Values of canopy closure ranged from 0 to
ables by using the standardized partial regression coefficients of all pre-
100 with higher values indicating close canopy and lower values indi-
dictor variables (cf. Kissling et al., 2008; Zhang et al., 2013). The Akaike
cating open canopy.
weight (w) for each variable based on all possible combinations of pre-
dictor variables was also calculated.
2.5. Variables generated from ground inventory data
All statistical analyses were carried out using R 3.1.2 software
(R Core Team, 2014). Pearson correlation coefficients after accounting
Since we had the detailed georeferenced locations of each individual
for spatial autocorrelation were calculated with the R package
tree with ≥1 cm DBH in the entire 20-ha plot from the ground inventory,
‘SpatialPack 0.2–3′ (Osorio and Vallejos, 2014), and SARs were calculat-
we assigned each of individuals to three raster grids (5 m × 5 m,
ed using the R package ‘spdep 0.5–88′ (Bivand et al., 2015). The spatial
10 m × 10 m, and 20 m × 20 m) to match with drone-driven variables.
weight matrices of the SARs were calculated with the nearest neighbour
Both community- and species-level biodiversity metrics were calculat-
and a row-standardized coding style (Kissling and Carl, 2008).
ed at these three spatial scales.
3. Results
2.5.1. Community-level metrics
At each of three spatial scales, we calculated species richness (S), 3.1. Stand-level analysis
Shannon diversity index (H), species evenness, and stand basal area
for each subplot for two tree size groups: all stems (≥ 1 cm DBH) and Drone-derived canopy variables, stand-level attributes, and topo-
saplings (1 cm ≤ DBH b 5 cm). Shannon diversity index (H) accounts graphic and edaphic variables varied greatly at the three spatial scales
for both abundance and evenness of the species present, and is calculat- (Fig. 2). At the 20-m scale, species richness ranged from 12 to 49 species
ed as follows: H = - ∑si=1(pi) ln (pi), where pi is the proportion of total with stand basal area varying from 5.5 m2 ha−1 to 64.9 m2 ha−1
individuals represented by species i. Species evenness was calculated (Table 1). For the drone-derived canopy variables, canopy height ranged
as the Shannon diversity index divided by the natural logarithm of spe- from 7.0 m to 44.3 m with canopy closure varying from open canopy
cies richness (Pielou, 1975). (0%) to completely close canopy (100%) and averaging overall 73.5%.
Comparing spatial patterns of these variables at three spatial scales
2.5.2. Species-level metrics (Fig. 2), we detected similar patterns between canopy height and cano-
Considering possible species-specific requirements for growth and py closure: tall and close canopies were found in the northwest corner
reproduction, we selected two light-demanding species and two of the plot, while low and open canopies occurred in the southeast cor-
shade-tolerant species to compare how these species respond to their ner. Patterns in abundance and species richness differed (Fig. 2) with
environment measured by the drone-derived canopy metrics, topogra- higher abundance and species richness in low to open canopies with
phy, and soils. The two light-demanding species were fissure chestnut abundance and richness lower in tall and close canopies.
(Castanopsis fissa; Fagaceae; CAFI) and panicled Mallotus (Mallotus After accounting for spatial autocorrelation, negative correlations
paniculatus; Euphorbiaceae; MAPA), and two shade-tolerant species with canopy closure and skewness of the height (ht_skewness) were
were willow-leaf Acmena (Acmena acuminatissima; Myrtaceae; ACAC) found for all stems and only the saplings, species richness, Shannon
and fleshy nut tree (Sarcosperma laurinum; Sapotaceae; SALA). For index and evenness, while weak correlations with ground measures
each species, the abundance for each subplot and for different tree were found for canopy height (ht_canopy) and standard deviation
size groups was calculated. of the heights (ht_sd) (Tables 2 and 3). Among three topographic
variables, slope and convexity were significantly related with species
2.6. Statistical analysis richness and the Shannon index, but not for evenness. Among four se-
lected edaphic variables, soil organic carbon (SOC) was positively corre-
We first used Pearson correlations to analyse pair-wise relationships lated with species richness for all stems and the saplings and Shannon
among different variables to identify multicollinearity and we used index for the saplings, while total potassium (TK) was negatively corre-
Dutilleul's (1993) modified t-test to calculate statistical significance lated with species richness for all stems and the saplings and positively
accounting for spatial autocorrelation. To avoid multicollinearity related with the evenness for the saplings. For stand basal area (BA),
problems (Dormann et al., 2013), we excluded variables with high canopy closure was the most important explanatory variable at the
64 J. Zhang et al. / Biological Conservation 198 (2016) 60–69

Fig. 2. Maps for drone-derived canopy height and closure, and woody plant abundance and species richness generated from ground inventory data at three spatial scales, 5 m, 10 m, and
20 m, respectively. Canopy height at each scale was calculated by averaging the three tallest canopies of 1 m × 1 m pixels. Canopy closure was measured by the percentage of 1 m × 1 m
pixels with heights N10 m as estimated by the drone. Canopy closure values range from 0 to 100; higher values indicate more close canopy, while lower values indicate more open canopy.
Note that the maps at 5-m and 10-m scales are presented to show the potential applications of drones as a tool for forest monitoring but were not used for analyses in this study.

20-m scale, following by convexity, total potassium (TK) and total phos- demanding species, MAPA, the abundance had significantly negative rela-
phorus (TP) (Table 3). tionships with canopy closure, ht_sd, and ht_skewness, and positive rela-
tionships with canopy height, but weak relationships with all selected
3.2. Species-level analysis topographic and edaphic variables. Abundance of MAPA saplings also
showed a negative relationship to canopy closure and a positive relation-
Species abundance distributions for the four selected species ship to canopy height.
showed individual patterns, while similar patterns were detected for For two shade-tolerance species, ACAC and SALA, the correlations
all stems and only the saplings of the same species (Fig. 3). For all living between abundance of all stems and drone-derived canopy variables
stems of each species, the abundance of CAFI, one light-demanding were quite weak (Table 4). In contrast, elevation and several edaphic
species, was positively related to elevation and convexity, and negatively variables showed relatively strong relationships with abundance. Simi-
related to canopy closure (Tables 4 and 5). For CAFI saplings, abundance lar trends were observed when analyzing only saplings of these two
was only related to elevation and convexity. However, for another light- species (Table 4).

Table 1
Descriptive statistics of drone-derived canopy variables, stand-level attributes, and topographic and edaphic variables at the 20-m scale.

Variables Abbreviation Mean ± SD Min Max

Stand-level attributes Species richness Richness 26.41 ± 6.45 12 49

Shannon diversity index Shannon 2.66 ± 0.29 1.38 3.23
Species evenness Evenness 0.82 ± 0.06 0.50 0.94
Stand basal area (m2 ha−1) BA 26.98 ± 0.35 5.50 64.87
Species richness for saplings Richness_S 19.12 ± 5.91 5 40
Shannon diversity index for saplings Shannon_S 2.32 ± 0.38 0.80 3.09
Species evenness for saplings Evenness_S 0.80 ± 0.09 0.38 0.97
Drone-derived canopy variables Canopy height (m) ht_canopy 23.56 ± 4.21 7.02 44.26
Mean height (m) ht_mean 13.90 ± 5.01 0 25.97
Standard deviation of heights (m) ht_sd 4.60 ± 1.60 1.61 10.62
Canopy closure (%) Closure 73.45 ± 28.02 0 100
Topographic variables Elevation (m) elev 339.23 ± 50.83 237.12 466.16
Slope (°C) slope 32.59 ± 7.96 8.54 57.91
Convexity (m) Conv 0.47 ± 6.51 −13.98 17.67
Edaphic variables Soil organic carbon (g kg−1) SOC 60.98 ± 11.21 42.61 98.05
Available potassium (g kg−1) AK 18.15 ± 3.41 8.45 31.22
Total potassium (g kg−1) TK 54.99 ± 19.58 30.11 121.18
Total phosphorus (g kg−1) TP 1.80 ± 1.43 0.41 4.87
 J. Zhang et al. / Biological Conservation 198 (2016) 60–69 65

Table 2
Pearson correlation coefficients between stand plant measures (species richness, Shannon diversity index, species evenness, and stand basal area) and drone-derived canopy measures,
terrain variables, and edaphic variables. p values were calculated after accounting for spatial autocorrelation using Dutilleul's (1993) method: *** p b 0.001; ** p b 0.01; * p b 0.05. The cor-
relation coefficients with p b 0.05 are marked in bold. Abbreviations of predictors are explained in Table 1.

All stems Saplings

Richness Shannon Evenness BA Richness Shannon Evenness

ht_canopy −0.154 −0.134 −0.027 0.133 −0.154 −0.189 −0.102

ht_sd 0.152 0.086 −0.032 −0.231** 0.168 0.170 0.085
ht_skewness 0.151 0.125 0.045 −0.230*** 0.162 0.244* 0.220*
Closure −0.248 −0.191 −0.029 0.409*** −0.263 −0.338* −0.238
elev 0.364 0.193 −0.075 0.131 0.357 0.143 −0.153
conv 0.268*** 0.166* −0.036 0.315*** 0.261*** 0.161* −0.042
slope 0.450** 0.289** −0.027 −0.145 0.422** 0.377** 0.120
SOC 0.397 0.204 −0.094 −0.165 0.441 0.420* 0.170
AK 0.099 0.079 0.004 −0.145 0.138 0.230 0.197
TK 0.019 −0.017 −0.045 −0.362*** 0.086 0.226 0.235
TP −0.317 −0.159 0.077 0.15 −0.371 −0.316 −0.097

4. Discussion Stand basal area, which has frequently been used as a surrogate for
forest biomass and carbon stocks (Houghton et al., 2009), was signifi-
4.1. Effects of drone-derived canopy variables on stand attributes cantly related to canopy closure in our study forest. This positive rela-
tionship indicates that forests with undisturbed close canopy store
By combining data from lightweight drones and ground-based sur- higher aboveground biomass and carbon than disturbed open forests,
veys on species distribution, topography and edaphic variables, we supported by Liu et al.'s (2007) finding on biomass allocation in five
found that drone-derived canopy variables contributed substantially forest types in Dinghushan Nature Reserve. It also highlights the impor-
to explaining patterns of biodiversity in this species-rich subtropical for- tance of protecting old-growth forests as carbon sinks (Luyssaert et al.,
est plot. For both all living stems and saplings, forest canopy closure and 2008), especially in this highly fragmented region packed with intense
skewness of canopy heights, which are directly related to canopy light human activities.
heterogeneity, had much stronger relationships (negative) with species
richness and Shannon diversity than the average and standard deviation 4.2. Effects of topographic and edaphic variables on stand attributes
of canopy heights. This suggests that local light availability plays a crit-
ical role on determining biodiversity patterns (Grubb, 1977; Denslow, After accounting for the effects of edaphic variables and drone-
1987). Local disturbances, such as gaps formed by treefall and standing derived canopy variables, topographic variations in this plot were
dead alders, increase heterogeneity in light conditions and may pro- among the most important factors relating to biodiversity patterns for
mote the coexistence of species by providing opportunities for niche both all living stems and saplings. With a 230-m elevation range in
differentiation (Denslow, 1987). Rüger et al. (2009) evaluated the influ- this 20-ha plot, the DHS plot has high microhabitat heterogeneity lead-
ence of light gap disturbances on tree recruitment in a 50-ha tropical ing to topographic niche differentiation among species (Brown et al.,
forest plot in Barro Colorado Island, which was the first established 2013). The importance of habitat heterogeneity is supported by previ-
plot of the CTFS-ForestGEO network, concluding that nearly all species ous studies in DHS where habitat conditions affected species abundance
increased in recruitment with increasing light. These findings by distributions for 83% of all woody species (Wang et al., 2009) and signif-
Rüger et al. (2009) were contrary to the early work by Hubbell et al. icantly influencing on intraspecific variation of one common species
(1999). Due to the practical difficulties in measuring canopy structure C. chinensis (Wang et al., 2012). In contrast, there was little evidence
directly, both of the studies were based on a rough estimation of the for habitat heterogeneity affecting species in a 50-ha tropical forest plot
light environment in the forest, which may affect their results (Rüger in Barro Colorado Island with only 40 m in elevation range (Harms
et al., 2009). Compared with traditional ground-based surveys, light- et al., 2001). Our finding confirms the niche differentiation hypoth-
weight drones provide a high-quality measure of forest vertical structure, esis in this species-rich forest (Brown et al., 2013).
and could serve as a cost-effective and time-saving tool for monitoring Among the four selected edaphic variables, soil organic carbon ex-
canopy dynamics. plained additional variation in species richness and Shannon diversity

Table 3
Spatial simultaneous autoregressive models of response variables (species richness, Shannon diversity index, species evenness, and stand basal area) against combinations of predictor
variables. Standardized coefficients (Coef) for the model with the highest Akaike weight (w) for a given variable group are given, as well as the Akaike weight (w) for each variable based
on all possible combinations of predictor variables. Pseudo r2 of each model was in bold. *** p b 0.001; ** p b 0.01; * p b 0.05. Abbreviations of predictors are explained in Table 1.

All stems Saplings (1 cm ≤ DBH b 5 cm)

Richness Shannon Evenness BA Richness Shannon Evenness

Coef w Coef w Coef w Coef w Coef w Coef w Coef w

ht_canopy 0.319 0.477 0.374 0.285 0.341 0.498 0.502

ht_sd 0.293 0.419 0.362 0.276 0.328 0.391 0.351
ht_skewness −0.110** 0.911 −0.065 0.422 0.365 0.270 −0.108** 0.959 −0.074 0.464 0.422
Closure −0.111* 0.619 −0.144* 0.579 −0.097 0.495 0.421*** 1.000 0.431 −0.121* 0.509 −0.117 0.550
elev 0.202* 0.777 0.183 0.526 0.347 0.351 0.192 0.795 0.353 0.469
conv 0.090* 0.789 0.120** 0.879 0.441 0.135** 0.953 0.087* 0.806 0.106* 0.830 0.401
slope 0.273*** 1.000 0.192*** 0.995 0.317 0.277 0.230*** 1.000 0.178*** 0.992 0.316
SOC 0.336** 0.935 0.506 0.294 0.384 0.338** 0.952 0.294*** 0.947 0.381
AK 0.369 0.350 0.288 0.285 0.310 0.443 0.413
TK −0.265* 0.887 0.490 0.289 −0.173* 0.688 −0.219 0.764 0.328 0.217* 0.635
TP 0.439 0.286 0.307 −0.159* 0.592 −0.154 0.503 0.336 0.286
Pseudo r2 0.537 0.430 0.357 0.334 0.538 0.435 0.401
66 J. Zhang et al. / Biological Conservation 198 (2016) 60–69

Fig. 3. Spatial distribution at the 20-m scale of species abundance for two light-demanding species (Mallotus paniculatus (MAPA) and Castanopsis fissa (CAFI)) and two shade-tolerant
species (Acmena acuminatissima (ACAC) and Sarcosperma laurinum (SALA)). Patterns for all individuals (a–d) and saplings (e–f) are shown here.

for both all living stems and saplings, but explained less for the evenness of all species across different tree life stages will provide important
and stand basal area. Zhou et al. (2006) analyzed soil carbon dynamics evidence for how and why species respond to their environment in
from 1979 to 2003 in the same forest and found that old-growth forests different ways.
accumulated atmospheric carbon at an unexpectedly high rate, but they
did not give a clear explanation for the potential mechanisms and pos-
4.4. Drone ecology and long-term ecosystem monitoring
sible consequences for biodiversity and ecosystem functioning. The re-
lationships we found between soil organic carbon and biodiversity
We demonstrate an example of how drone-derived variables can
and stand basal area give some insight to this question.
contribute to our understanding of biodiversity maintenance and spe-
cies coexistence for a diverse subtropical forest plot. Clearly, drones
4.3. Species-specific responses to their environment hold great potential for providing advancements in mapping and mon-
itoring of forest dynamics. Compared with satellite and airborne remote
The results of the species-specific analyses of abundance dis- sensing techniques, drones collect the data with ultra-high spatial reso-
tributions support the hypothesis that different habitat requirements lution (e.g., nearly 5 cm in our study) in a cost-effective manner, which
among species result in the coexistence of diverse species in one com- can be used to measure some key stand attributes that have been
munity (Levine and HilleRisLambers, 2009). The abundance distribution demonstrated by a few recent drone studies in forest ecosystems
of the light-demanding species M. paniculatus (MAPA) was markedly af- (Getzin et al., 2012; Dandois and Ellis, 2013; Puliti et al., 2015; Zahawi
fected by canopy closure when considering all stems and only saplings, et al., 2015; this study). Our analyses also showed that the drone imag-
supporting traditional gap dynamics theory (Oliver and Larson, 1996). ery of our study plot matched very well with ground reference points,
In contrast, shade-tolerant species had only weak relationships with with the precision of 32–44 cm horizontal RMSE and 69 cm vertical
canopy variables. Species-specific responses to topography and edaphic RMSE. High accuracy between drone imagery and ground data was
variables found in this study have been documented previously (e.g., Li also reported in boreal (Puliti et al., 2015) and temperate forests
et al., 2009; Lin et al., 2013; Wang et al., 2009). By combining drone- (Getzin et al., 2012; Dandois and Ellis, 2013). The mismatch between
derived variables with other biotic and abiotic variables, future studies most remote sensing data sources and ground inventory data has

Table 4
Pearson correlation coefficients of four selected species' abundances and drone-derived canopy measures, terrain variables, and edaphic variables, after accounting for spatial autocorre-
lation. CAFI (Castanopsis fissa; Fagaceae) and MAPA (Mallotus paniculatus; Euphorbiaceae) are light-demanding species, while ACAC (Acmena acuminatissima; Myrtaceae) and SALA
(Sarcosperma laurinum; Sapotaceae) are shade-tolerant species. The results that are statistically significant are typed in bold. *** p b 0.001; ** p b 0.01; * p b 0.05. Other symbols and
explanations are as in Table 1.

All stems Saplings

CAFI MAPA ACAC SALA CAFI MAPA ACAC SALA

ht_canopy −0.105 −0.077 −0.138 −0.110 −0.086 0.009 −0.026 −0.021

ht_sd −0.047 0.187 0.270 0.098 −0.048 0.192* 0.127 0.041
ht_skewness 0.063 0.241* 0.244 0.043 0.045 0.152 0.088 −0.013
Closure −0.065 −0.358** −0.399* −0.151 −0.033 −0.225* −0.153 −0.016
elev 0.387* −0.060 −0.296 −0.310 0.394* −0.062 −0.184 −0.293
conv 0.226** −0.095 −0.162* −0.055 0.203** −0.071 −0.084 0.005
slope 0.193 0.206 0.286 0.084 0.169 0.114 0.157 0.012
SOC 0.206 0.268 0.329 0.113 0.171 0.168 0.169 0.029
AK −0.099 0.183 0.277 0.246 −0.096 0.138 0.135 0.205
TK −0.091 0.408** 0.474* 0.167 −0.105 0.304** 0.222 0.069
TP −0.104 −0.237 −0.362 −0.121 −0.086 −0.179 −0.186 −0.042
 J. Zhang et al. / Biological Conservation 198 (2016) 60–69 67

Table 5
Spatial simultaneous autoregressive models of response variables (the abundance of each of four selected species) against combinations of predictor variables. *** p b 0.001; ** p b 0.01; * p b 0.05.
Other symbols and explanations are in Tables 3 and 4.

Groups Variables CAFI MAPA ACAC SALA

Coef w Coef w Coef w Coef w

All stems ht_canopy 0.304 0.260*** 0.950 0.381 0.425

ht_sd 0.281 −0.233** 0.950 0.334 0.296
ht_skewness 0.282 −0.094 0.592 0.275 −0.074 0.642
Closure −0.098 0.550 −0.475*** 0.999 0.337 0.301
elev 0.390*** 0.988 0.412 −0.274** 0.937 −0.348** 0.954
conv 0.106* 0.760 0.386 0.480 0.284
slope 0.321 0.296 0.096* 0.744 0.316
SOC 0.554 0.326 0.382 0.193 0.566
AK 0.438 0.287 0.338 0.136 0.656
TK 0.430 0.305*** 1.000 0.273** 0.871 0.348
TP 0.282 0.308 −0.185* 0.693 0.435
Pseudo r2 0.334 0.192 0.490 0.493
Saplings (1 cm ≤ DBH b 5 cm) ht_canopy 0.299 0.119* 0.691 0.287 0.284
ht_sd 0.310 0.362 0.296 0.301
ht_skewness 0.274 0.321 0.298 −0.070 0.565
Closure 0.367 −0.152* 0.703 0.293 0.349
elev 0.384*** 0.995 0.363 −0.211** 0.857 −0.383*** 0.952
conv 0.075 0.570 0.300 −0.069 0.527 0.286
slope 0.298 0.295 0.447 0.311
SOC 0.437 0.375 0.443 0.251* 0.548
AK 0.338 0.271 0.298 0.152 0.648
TK 0.362 0.229*** 0.998 0.365 −0.234 0.532
TP 0.277 0.284 −0.196* 0.580 0.395
Pseudo r2 0.315 0.109 0.225 0.384

been documented and much debated in recent years (Kerr and affected by the relatively coarse ground-based DEM data (4.28 m in
Ostrovsky, 2003; Turner et al., 2003). For example, the debates on the RMSE when interpolating to 1-m scale). This concern has also been
accuracy of biomass estimation in Amazon forests (Mitchard et al., raised by recent drone studies in temperate deciduous forests
2014; Saatchi et al., 2015) and the quality of global forest cover change (Dandois and Ellis, 2013) and tropical forests (Zahawi et al., 2015). De-
maps (Hansen et al., 2013; Tropek et al., 2014). Considering their high spite this limitation, our analyses at the 20-m scale were still robust to
accuracy, drone surveys provide an effective solution for this issue at explain local-scale variations in biodiversity and other stand attributes.
local or regional scales. This spatial scale has been widely used for many studies in community
Although our research has shown several applications of drone tech- ecology and forest management (e.g., Dandois and Ellis, 2013; Zahawi
nology, there are an increasing number of potential ecological applica- et al., 2015). We detected some inconsistencies when comparing spatial
tions that we have not discussed (Koh and Wich, 2012; Floreano and variations of drone-derived canopy variables at three spatial scales
Wood, 2015). Here we list several potential applications, especially for (Fig. 2), suggesting the importance of spatial scale in ecological studies
use in long-term monitoring networks. First, drones can be used to (Levin, 1992). Further analyses at finer scales (even the individual
monitor long-term ecosystem dynamics. Long-term ecological studies level) could be used to address questions involving forest regeneration,
are critical for understanding how biodiversity and ecosystem function phenology, and others, and therefore it is worthwhile to continue
responds to natural and anthropogenic disturbances (e.g., Condit et al., exploring when high-quality data on topographic and edaphic vari-
2006; Hubbell et al., 1999; Zhang et al., 2015). Similar with several ables are available. To improve the quality of ground-based terrain
other pioneering studies (e.g., Getzin et al., 2012; Zahawi et al., 2015), data, one ongoing development is to mount light detection and
our study had only a one-time snapshot of the forest canopy structure. ranging (LiDAR) systems onto drones (Lin et al., 2011; Wallace
With increases in time of drone-derived variables, we will be able to et al., 2012), but the cost of aerial LiDAR sensors for drones is cur-
quantify how different ecosystems evolve under the changes in climate rently very high and specific technical expertise for expensive pro-
and land use (Whitehead et al., 2014). Second, using drones to collect cessing softwares is required.
multispectral and hyperspectral images will provide a biophysical and The second challenge is to accurately identify species and individuals
biochemical approach for mapping ecosystems. These images can pro- using drones, especially in subtropical and tropical forests with high
vide detailed information on plant chemical and structural properties, species richness and dense canopies. The application of hyperspectral
such as canopy water content and leaf nitrogen concentration (Asner remote sensing may provide insight into this problem (Asner et al.,
et al., 2015; Malenovský et al., 2015) and Normalized Difference Vegeta- 2015). Third, although drones are becoming much easier to operate
tion Index (NDVI) (Zarco-Tejada et al., 2013). These data can be used for than other types of remote sensing equipment, training is needed for
single individual tree detection, phenology monitoring, biomass map- designing flight paths, operating drones, and image post-processing
ping and so on (Whitehead and Hugenholtz, 2014). (Dandois and Ellis, 2013; Paneque-Gálvez et al., 2014). Previous experi-
Our use of drones in this study identified several caveats and practi- ence on over 200 successful missions by our team members ensured our
cal challenges that need to be considered for further studies. First, ter- current achievement. Collaborations between ecologists and experts on
rain data collected from the field may not spatially match the digital remote sensing will advance the application of drones in ecology. Addi-
surface model (DSM) generated from the drone. Following the standard tionally, drones with high performance and flexibility, such as the
protocols of the CTFS-ForestGEO network (Condit, 1998), nearly all of equipment used here, can be costly, although it is cheaper than the
over 60 plots had topographic variables measured at the 20-m scale. cost of acquiring imagery from very high spatial resolution satellites
Some errors of this approach exist, especially for the plots with large (e.g., Quickbird, WorldView, IKONOS, and RapidEye) or piloted aircraft
ranges of topography. Although we selected 9 ground control points missions in most cases. Ongoing efforts to design inexpensive equip-
to geocorrect the point cloud and the results showed high precision ment (e.g., http://conservationdrones.org) and open-source computer
for these control points, the estimation of canopy height data was still vision software (e.g., http://ecosynth.org; http://ardupilot.com) will
68 J. Zhang et al. / Biological Conservation 198 (2016) 60–69

advance the development of this field (Koh and Wich, 2012; Dandois S., Valencia, R., Vallejo, M., Villa, G., Zillio, T., 2006. The importance of demographic
niches to tree diversity. Science 313, 98–101.
and Ellis, 2013). Cord, A.F., Meentemeyer, R.K., Leitao, P.J., Vaclavik, T., 2013. Modelling species distribu-
In conclusion, our research provides one of the first tests of how light- tions with remote sensing data: bridging disciplinary perspectives. J. Biogeogr. 40,
weight drones can be used for long-term ecological research programs. 2226–2227.
Crowther, T.W., Glick, H.B., Covey, K.R., Bettigole, C., Maynard, D.S., Thomas, S.M., Smith,
We believe that drone technology, as a complementary tool of traditional J.R., Hintler, G., Duguid, M.C., Amatulli, G., Tuanmu, M.N., Jetz, W., Salas, C., Stam, C.,
field surveys, offers the potential to collect a large volume of data with Piotto, D., Tavani, R., Green, S., Bruce, G., Williams, S.J., Wiser, S.K., Huber, M.O.,
high spatiotemporal resolution and at low costs (Anderson and Gaston, Hengeveld, G.M., Nabuurs, G.J., Tikhonova, E., Borchardt, P., Li, C.F., Powrie, L.W.,
Fischer, M., Hemp, A., Homeier, J., Cho, P., Vibrans, A.C., Umunay, P.M., Piao, S.L.,
2013). Further, the combining drone-derived variables and detailed Rowe, C.W., Ashton, M.S., Crane, P.R., Bradford, M.A., 2015. Mapping tree density at
ground inventory data allows us to advance our understanding of key a global scale. Nature 525, 201–205.
processes in complex ecosystems in a changing world. Dandois, J.P., Ellis, E.C., 2013. High spatial resolution three-dimensional mapping of vege-
tation spectral dynamics using computer vision. Remote Sens. Environ. 136, 259–276.
Supplementary data to this article can be found online at http://dx.
Denslow, J.S., 1987. Tropical rainforest gaps and tree species diversity. Annu. Rev. Ecol.
doi.org/10.1016/j.biocon.2016.03.027. Syst. 18, 431–451.
Dormann, C.F., Elith, J., Bacher, S., Buchmann, C., Carl, G., Carré, G., Marquéz, J.R.G., Gruber,
B., Lafourcade, B., Leitão, P.J., Münkemüller, T., McClean, C., Osborne, P.E., Reineking,
Acknowledgements B., Schröder, B., Skidmore, A.K., Zurell, D., Lautenbach, S., 2013. Collinearity: a review
of methods to deal with it and a simulation study evaluating their performance.
Ecography 36, 27–46.
We greatly acknowledge the financial support from CTFS-ForestGEO Dutilleul, P., 1993. Modifying the t test for assessing the correlation between two spatial
Research Grant to J.Z. and J.H., the National Natural Science Foundation processes. Biometrics 49, 305–314.
Floreano, D., Wood, R.J., 2015. Science, technology and the future of small autonomous
of China (No. 41371078 to J.L.) and Chinese Forest Biodiversity Monitor- drones. Nature 521, 460–466.
ing Network (http://www.cfbiodiv.org/). We thank Scott E. Nielsen and Gardner, T.A., Barlow, J., Araujo, I.S., Avila-Pires, T.C., Bonaldo, A.B., Costa, J.E., Esposito,
Robert Muscarella for their helpful comments and discussions on the M.C., Ferreira, L.V., Hawes, J., Hernandez, M.I.M., Hoogmoed, M.S., Leite, R.N.,
Lo-Man-Hung, N.F., Malcolm, J.R., Martins, M.B., Mestre, L.A.M., Miranda-Santos, R.,
manuscript, and Jian Han, Honglin Cao, Guojun Lin, Gang Feng and Overal, W.L., Parry, L., Peters, S.L., Ribeiro, M.A., da Silva, M.N.F., Motta, C.D.S., Peres,
Zhongliang Huang for their support on drone data collection and/or C.A., 2008. The cost-effectiveness of biodiversity surveys in tropical forests. Ecol.
data analysis. We also thank all the field crews involved in ground Lett. 11, 139–150.
Getzin, S., Wiegand, K., Schöning, I., 2012. Assessing biodiversity in forests using very
data collection in the 20-ha Dinghushan forest dynamic plot.
high-resolution images and unmanned aerial vehicles. Methods Ecol. Evol. 3,
397–404.
Goetz, S., Steinberg, D., Dubayah, R., Blair, B., 2007. Laser remote sensing of canopy habitat
References heterogeneity as a predictor of bird species richness in an eastern temperate forest,
USA. Remote Sens. Environ. 108, 254–263.
Anderson, K., Gaston, K.J., 2013. Lightweight unmanned aerial vehicles will revolutionize Grubb, P.J., 1977. The maintenance of species-richness in plant communities: the impor-
spatial ecology. Front. Ecol. Environ. 11, 138–146. tance of the regeneration niche. Biol. Rev. 52, 107–145.
Anderson-Teixeira, K.J., Davies, S.J., Bennett, A.C., Gonzalez-Akre, E.B., Muller-Landau, H.C., Hansen, M.C., Potapov, P.V., Moore, R., Hancher, M., Turubanova, S.A., Tyukavina, A., Thau,
Wright, S.J., Abu Salim, K., Zambrano, A.M.A., Alonso, A., Baltzer, J.L., Basset, Y., Bourg, D., Stehman, S.V., Goetz, S.J., Loveland, T.R., Kommareddy, A., Egorov, A., Chini, L.,
N.A., Broadbent, E.N., Brockelman, W.Y., Bunyavejchewin, S., Burslem, D.F.R.P., Butt, Justice, C.O., Townshend, J.R.G., 2013. High-resolution global maps of 21st-century
N., Cao, M., Cardenas, D., Chuyong, G.B., Clay, K., Cordell, S., Dattaraja, H.S., Deng, forest cover change. Science 342, 850–853.
X.B., Detto, M., Du, X.J., Duque, A., Erikson, D.L., Ewango, C.E.N., Fischer, G.A., Harms, K.E., Wright, S.J., Calderon, O., Hernandez, A., Herre, E.A., 2000. Pervasive density-
Fletcher, C., Foster, R.B., Giardina, C.P., Gilbert, G.S., Gunatilleke, N., Gunatilleke, S., dependent recruitment enhances seedling diversity in a tropical forest. Nature 404,
Hao, Z.Q., Hargrove, W.W., Hart, T.B., Hau, B.C.H., He, F.L., Hoffman, F.M., Howe, 493–495.
R.W., Hubbell, S.P., Inman-Narahari, F.M., Jansen, P.A., Jiang, M.X., Johnson, D.J., Harms, K.E., Condit, R., Hubbell, S.P., Foster, R.B., 2001. Habitat associations of trees and
Kanzaki, M., Kassim, A.R., Kenfack, D., Kibet, S., Kinnaird, M.F., Korte, L., Kral, K., shrubs in a 50-ha neotropical forest plot. J. Ecol. 89, 947–959.
Kumar, J., Larson, A.J., Li, Y.D., Li, X.K., Liu, S.R., Lum, S.K.Y., Lutz, J.A., Ma, K.P., He, F.L., Hubbell, S.P., 2011. Species-area relationships always overestimate extinction
Maddalena, D.M., Makana, J.R., Malhi, Y., Marthews, T., Serudin, R.M., McMahon, rates from habitat loss. Nature 473, 368–371.
S.M., McShea, W.J., Memiaghe, H.R., Mi, X.C., Mizuno, T., Morecroft, M., Myers, J.A., Houghton, R.A., Hall, F., Goetz, S.J., 2009. Importance of biomass in the global carbon cycle.
Novotny, V., de Oliveira, A.A., Ong, P.S., Orwig, D.A., Ostertag, R., den Ouden, J., J. Geophys. Res. 114, G00E03.
Parker, G.G., Phillips, R.P., Sack, L., Sainge, M.N., Sang, W.G., Sri-ngernyuang, K., Huang, Z.L., Meng, M.L., Zhang, Y.C., 1998. Climate of Dinghushan Biosphere Reserve.
Sukumar, R., Sun, I.F., Sungpalee, W., Suresh, H.S., Tan, S., Thomas, S.C., Thomas, Tropical and subtropical forest ecosystems VIII. China Meteorological Press, Beijing,
D.W., Thompson, J., Turner, B.L., Uriarte, M., Valencia, R., Vallejo, M.I., Vicentini, A., pp. 134–139.
et al., 2015. CTFS-ForestGEO: a worldwide network monitoring forests in an era of Hubbell, S.P., 2001. The Unified Neutral Theory of Biodiversity and Biogeography.
global change. Glob. Chang. Biol. 21, 528–549. Princeton University Press, Princeton, New Jersey, USA.
Asner, G.P., Martin, R.E., Anderson, C.B., Knapp, D.E., 2015. Quantifying forest canopy Hubbell, S.P., Foster, R.B., O'Brien, S.T., Harms, K.E., Condit, R., Wechsler, B., Wright, S.J., de
traits: imaging spectroscopy versus field survey. Remote Sens. Environ. 158, 15–27. Lao, S.L., 1999. Light-gap disturbances, recruitment limitation, and tree diversity in a
Bivand, R., Altman, M., Anselin, L., Assunção, R., Berke, O., Bernat, A., Blanchet, G., neotropical forest. Science 283, 554–557.
Blankmeyer, E., Carvalho, M., Christensen, B., Chun, Y., Dormann, C., Dray, S., John, R., Dalling, J.W., Harms, K.E., Yavitt, J.B., Stallard, R.F., Mirabello, M., Hubbell,
Gómez-Rubio, V., Gubri, M., Halbersma, R., Krainski, E., Legendre, P., Lewin-Koh, N., S.P., Valencia, R., Navarrete, H., Vallejo, M., Foster, R.B., 2007. Soil nutrients influ-
Li, H., Ma, J., Millo, G., Mueller, W., Ono, H., Peres-Neto, P., Piras, G., Reder, M., ence spatial distributions of tropical tree species. Proc. Natl. Acad. Sci. U. S. A.
Tiefelsdorf, M., Yu, D., 2015. spdep: Spatial Dependence: Weighting Schemes, Statis- 104, 864–869.
tics and Models. R package version 05-88 (https://cran.r-project.org/web/packages/ Jones, G.P., Pearlstine, L.G., Percival, H.F., 2006. An assessment of small unmanned aerial
spdep/index.html). vehicles for wildlife research. Wildl. Soc. Bull. 34, 750–758.
Brown, C., Burslem, D.F.R.P., Illian, J.B., Bao, L., Brockelman, W., Cao, M., Chang, L.W., Kerr, J.T., Ostrovsky, M., 2003. From space to species: ecological applications for remote
Dattaraja, H.S., Davies, S., Gunatilleke, C.V.S., Gunatilleke, I.A.U.N., Huang, J., Kassim, sensing. Trends Ecol. Evol. 18, 299–305.
A.R., LaFrankie, J.V., Lian, J., Lin, L., Ma, K., Mi, X., Nathalang, A., Noor, S., Ong, P., Kissling, W.D., Carl, G., 2008. Spatial autocorrelation and the selection of simultaneous
Sukumar, R., Su, S.H., Sun, I.F., Suresh, H.S., Tan, S., Thompson, J., Uriarte, M., autoregressive models. Glob. Ecol. Biogeogr. 17, 59–71.
Valencia, R., Yap, S.L., Ye, W., Law, R., 2013. Multispecies coexistence of trees in trop- Kissling, W.D., Field, R., Böhning-Gaese, K., 2008. Spatial patterns of woody plant and bird
ical forests: spatial signals of topographic niche differentiation increase with environ- diversity: functional relationships or environmental effects? Glob. Ecol. Biogeogr. 17,
mental heterogeneity. Proc. R. Soc. B Biol. Sci. 280, 20130502. 327–339.
Burnham, K.P., Anderson, D.R., 2002. Model Selection and Multimodel Inference: A Koh, L.P., Wich, S.A., 2012. Dawn of drone ecology: low-cost autonomous aerial vehicles
Practical Information-Theoretic Approach. second ed. Springer-Verlag, New York. for conservation. Trop. Conserv. Sci. 5, 121–132.
Callahan, J.T., 1984. Long-term ecological research. Bioscience 34, 363–367. Lawton, J.H., Bignell, D.E., Bolton, B., Bloemers, G.F., Eggleton, P., Hammond, P.M., Hodda,
Chabot, D., Bird, D.M., 2012. Evaluation of an off-the-shelf Unmanned Aircraft System for M., Holt, R.D., Larsen, T.B., Mawdsley, N.A., Stork, N.E., Srivastava, D.S., Watt, A.D.,
surveying flocks of geese. Waterbirds 35, 170–174. 1998. Biodiversity inventories, indicator taxa and effects of habitat modification in
Condit, R., 1995. Research in large, long-term tropical forest plots. Trends Ecol. Evol. 10, tropical forest. Nature 391, 72–76.
18–22. Levin, S.A., 1992. The problem of pattern and scale in ecology. Ecology 73, 1943–1967.
Condit, R., 1998. Tropical Forest Census Plots: Methods and Results from Barro Colorado Levine, J.M., HilleRisLambers, J., 2009. The importance of niches for the maintenance of
Island, Panama and a Comparison with Other Plots. Springer, Berlin. species diversity. Nature 461, 254–257.
Condit, R., Ashton, P., Bunyavejchewin, S., Dattaraja, H.S., Davies, S., Esufali, S., Ewango, C., Li, L., Huang, Z., Ye, W., Cao, H., Wei, S., Wang, Z., Lian, J., Sun, I.F., Ma, K., He, F., 2009. Spa-
Foster, R., Gunatilleke, I.A., Gunatilleke, C.V., Hall, P., Harms, K.E., Hart, T., Hernandez, tial distributions of tree species in a subtropical forest of China. Oikos 118, 495–502.
C., Hubbell, S., Itoh, A., Kiratiprayoon, S., Lafrankie, J., de Lao, S.L., Makana, J.R., Noor, Likens, G.E., 1989. Long-term Studies in Ecology: Approaches and Alternatives. Springer-
M.N., Kassim, A.R., Russo, S., Sukumar, R., Samper, C., Suresh, H.S., Tan, S., Thomas, Verlag, New York.
 J. Zhang et al. / Biological Conservation 198 (2016) 60–69 69

Lin, Y., Hyyppa, J., Jaakkola, A., 2011. Mini-UAV-Borne LIDAR for fine-scale mapping. IEEE Schiffman, R., 2014. Drones flying high as new tool for field biologists. Science 344, 459-
Geosci. Remote Sens. Lett. 8, 426–430. 459.
Lin, G.J., Stralberg, D., Gong, G.Q., Huang, Z.L., Ye, W.H., Wu, L.F., 2013. Separating the Simard, M., Pinto, N., Fisher, J.B., Baccini, A., 2011. Mapping forest canopy height globally
effects of environment and space on tree species distribution: from population to with spaceborne lidar. J. Geophys. Res. 116.
community. PLoS One 8, e56171. Stephenson, N.L., Das, A.J., Condit, R., Russo, S.E., Baker, P.J., Beckman, N.G., Coomes, D.A.,
Lindenmayer, D.B., Likens, G.E., Andersen, A., Bowman, D., Bull, C.M., Burns, E., Dickman, Lines, E.R., Morris, W.K., Ruger, N., Alvarez, E., Blundo, C., Bunyavejchewin, S.,
C.R., Hoffmann, A.A., Keith, D.A., Liddell, M.J., Lowe, A.J., Metcalfe, D.J., Phinn, S.R., Chuyong, G., Davies, S.J., Duque, A., Ewango, C.N., Flores, O., Franklin, J.F., Grau, H.R.,
Russell-Smith, J., Thurgate, N., Wardle, G.M., 2012. Value of long-term ecological Hao, Z., Harmon, M.E., Hubbell, S.P., Kenfack, D., Lin, Y., Makana, J.R., Malizia, A.,
studies. Austral Ecol. 37, 745–757. Malizia, L.R., Pabst, R.J., Pongpattananurak, N., Su, S.H., Sun, I.F., Tan, S., Thomas, D.,
Liu, S., Luo, Y., Huang, Y.H., Zhou, G.Y., 2007. Studies on the community biomass and its van Mantgem, P.J., Wang, X., Wiser, S.K., Zavala, M.A., 2014. Rate of tree carbon accu-
allocations of five forest types in Dinghushan Nature Reserve. Ecol. Sci. 26, 387–393. mulation increases continuously with tree size. Nature 507, 90–93.
Luyssaert, S., Schulze, E.D., Borner, A., Knohl, A., Hessenmoller, D., Law, B.E., Ciais, P., Grace, Tropek, R., Sedlacek, O., Beck, J., Keil, P., Musilova, Z., Simova, I., Storch, D., 2014.
J., 2008. Old-growth forests as global carbon sinks. Nature 455, 213–215. Comment on “High-resolution global maps of 21st-century forest cover change”.
Malenovský, Z., Turnbull, J.D., Lucieer, A., Robinson, S.A., 2015. Antarctic moss stress Science 344, 981.
assessment based on chlorophyll content and leaf density retrieved from imaging Turner, W., Spector, S., Gardiner, N., Fladeland, M., Sterling, E., Steininger, M., 2003.
spectroscopy data. New Phytol. 208, 608–624. Remote sensing for biodiversity science and conservation. Trends Ecol. Evol. 18,
Martin, L.J., Blossey, B., Ellis, E., 2012. Mapping where ecologists work: biases in the global 306–314.
distribution of terrestrial ecological observations. Front. Ecol. Environ. 10, 195–201. Vermeulen, C., Lejeune, P., Lisein, J., Sawadogo, P., Bouche, P., 2013. Unmanned aerial
McGlone, J.C., 2013. Manual of Photogrammetry. sixth ed. American Society for Photo- survey of elephants. PLoS One 8, e54700.
grammetry and Remote Sensing (ASPRS), Bethesda. Wallace, L., Lucieer, A., Watson, C., Turner, D., 2012. Development of a UAV-LiDAR system
Mitchard, E.T.A., Feldpausch, T.R., Brienen, R.J.W., Lopez-Gonzalez, G., Monteagudo, A., with application to forest inventory. Remote Sens. 4, 1519–1543.
Baker, T.R., Lewis, S.L., Lloyd, J., Quesada, C.A., Gloor, M., ter Steege, H., Meir, P., Wang, Z., Ye, W.H., Cao, H.L., Huang, Z.L., Lian, J.Y., Li, L., Wei, S.G., Sun, I.F., 2009. Species-
Alvarez, E., Araujo-Murakami, A., Aragao, L.E.O.C., Arroyo, L., Aymard, G., Banki, O., topography association in a species-rich subtropical forest of China. Basic Appl. Ecol.
Bonal, D., Brown, S., Brown, F.I., Ceron, C.E., Moscoso, V.C., Chave, J., Comiskey, J.A., 10, 648–655.
Cornejo, F., Medina, M.C., Da Costa, L., Costa, F.R.C., Di Fiore, A., Domingues, T.F., Wang, Z.F., Lian, J.Y., Huang, G.M., Ye, W.H., Cao, H.L., Wang, Z.M., 2012. Genetic groups in
Erwin, T.L., Frederickson, T., Higuchi, N., Coronado, E.N.H., Killeen, T.J., Laurance, the common plant species Castanopsis chinensis and their associations with topo-
W.F., Levis, C., Magnusson, W.E., Marimon, B.S., Marimon, B.H., Polo, I.M., Mishra, P., graphic habitats. Oikos 121, 2044–2051.
Nascimento, M.T., Neill, D., Vargas, M.P.N., Palacios, W.A., Parada, A., Molina, G.P., Whitehead, K., Hugenholtz, C.H., 2014. Remote sensing of the environment with small
Pena-Claros, M., Pitman, N., Peres, C.A., Poorter, L., Prieto, A., Ramirez-Angulo, H., unmanned aircraft systems (UASs), part 1: a review of progress and challenges.
Correa, Z.R., Roopsind, A., Roucoux, K.H., Rudas, A., Salomao, R.P., Schietti, J., Silveira, J. Unmanned Veh. Syst. 02, 69–85.
M., de Souza, P.F., Steininger, M.K., Stropp, J., Terborgh, J., Thomas, R., Toledo, M., Whitehead, K., Hugenholtz, C.H., Myshak, S., Brown, O., LeClair, A., Tamminga, A., Barchyn,
Torres-Lezama, A., van Andel, T.R., van der Heijden, G.M.F., Vieira, I.C.G., Vieira, S., T.E., Moorman, B., Eaton, B., 2014. Remote sensing of the environment with small un-
Vilanova-Torre, E., Vos, V.A., Wang, O., Zartman, C.E., Malhi, Y., Phillips, O.L., 2014. manned aircraft systems (UASs), part 2: scientific and commercial applications.
Markedly divergent estimates of Amazon forest carbon density from ground plots J. Unmanned Veh. Syst. 02, 86–102.
and satellites. Glob. Ecol. Biogeogr. 23, 935–946. Wulder, M.A., Hall, R.J., Coops, N.C., Franklin, S.E., 2004. High spatial resolution remotely
Nijland, W., Coops, N.C., Macdonald, E.S., Nielsen, S.E., Bater, C.W., John Stadt, J., 2015. sensed data for ecosystem characterization. Bioscience 54, 511–521.
Comparing patterns in forest stand structure following variable harvests using Ye, W., Cao, H., Huang, Z., Lian, J., Wang, Z., Li, L., Wei, S., Wang, Z., 2008. Community struc-
airborne laser scanning data. For. Ecol. Manag. 354, 272–280. ture of a 20 hm2 lower subtropical evergreen broadleaved forest plot in Dinghushan.
Oliver, C.D., Larson, B.C., 1996. Forest Stand Dynamics. Updated ed. Wiley, New York. Chin. J. Plant Ecol. 32, 274–286.
Osorio, F., Vallejos, R., 2014. SpatialPack: Package for analysis of spatial data. R package Zahawi, R.A., Dandois, J.P., Holl, K.D., Nadwodny, D., Reid, J.L., Ellis, E.C., 2015. Using light-
version 0.2-3 (http://cran.r-project.org/package=SpatialPack). weight unmanned aerial vehicles to monitor tropical forest recovery. Biol. Conserv.
Paneque-Gálvez, J., McCall, M., Napoletano, B., Wich, S., Koh, L., 2014. Small drones for 186, 287–295.
community-based forest monitoring: an assessment of their feasibility and potential Zarco-Tejada, P.J., Guillen-Climent, M.L., Hernandez-Clemente, R., Catalina, A., Gonzalez,
in tropical areas. Forests 5, 1481–1507. M.R., Martin, P., 2013. Estimating leaf carotenoid content in vineyards using high res-
Pielou, E.C., 1975. Ecological Diversity. John Wiley & Sons, New York. olution hyperspectral imagery acquired from an unmanned aerial vehicle (UAV).
Puliti, S., Orka, H.O., Gobakken, T., Naesset, E., 2015. Inventory of small forest areas using Agric. For. Meteorol. 171, 281–294.
an Unmanned Aerial System. Remote Sens. 7, 9632–9654. Zhang, J., Kissling, W.D., He, F., 2013. Local forest structure, climate and human distur-
R Core Team, 2014. R: A language and environment for statistical computing. R Foundation bance determine regional distribution of boreal bird species richness in Alberta,
for Statistical Computing, Vienna, Austria. Canada. J. Biogeogr. 40, 1131–1142.
Rango, A., Laliberte, A., Herrick, J.E., Winters, C., Havstad, K., Steele, C., Browning, D., 2009. Zhang, J., Huang, S., Hogg, E.H., Lieffers, V., Qin, Y., He, F., 2014. Estimating spatial variation
Unmanned aerial vehicle-based remote sensing for rangeland assessment, monitor- in Alberta forest biomass from a combination of forest inventory and remote sensing
ing, and management. J. Appl. Remote. Sens. 3, 033542. data. Biogeosciences 11, 2793–2808.
Rüger, N., Huth, A., Hubbell, S.P., Condit, R., 2009. Response of recruitment to light avail- Zhang, J., Huang, S.M., He, F.L., 2015. Half-century evidence from western Canada shows
ability across a tropical lowland rain forest community. J. Ecol. 97, 1360–1368. forest dynamics are primarily driven by competition followed by climate. Proc.
Saatchi, S.S., Harris, N.L., Brown, S., Lefsky, M., Mitchard, E.T., Salas, W., Zutta, B.R., Natl. Acad. Sci. U. S. A. 112, 4009–4014.
Buermann, W., Lewis, S.L., Hagen, S., Petrova, S., White, L., Silman, M., Morel, A., Zhang, J., Nielsen, S.E., Mao, L.F., Chen, S.B., Svenning, J.-C., 2016. Regional and historical
2011. Benchmark map of forest carbon stocks in tropical regions across three conti- factors supplement current climate in shaping global forest canopy height. J. Ecol.
nents. Proc. Natl. Acad. Sci. U. S. A. 108, 9899–9904. 104, 469–478.
Saatchi, S., Mascaro, J., Xu, L., Keller, M., Yang, Y., Duffy, P., Espírito-Santo, F., Baccini, A., Zhou, G.Y., Liu, S.G., Li, Z., Zhang, D.Q., Tang, X.L., Zhou, C.Y., Yan, J.H., Mo, J.M., 2006.
Chambers, J., Schimel, D., 2015. Seeing the forest beyond the trees. Glob. Ecol. Old-growth forests can accumulate carbon in soils. Science 314, 1417-1417.
Biogeogr. 24, 606–610.
Saveraid, E.H., Debinski, D.M., Kindscher, K., Jakubauskas, M.E., 2001. A comparison of
satellite data and landscape variables in predicting bird species occurrences in the
Greater Yellowstone Ecosystem, USA. Landsc. Ecol. 16, 71–83.