Detecting Drought-Prone Areas of Rice Agriculture Using a MODIS-Derived Soil Moisture Index

Detecting Drought-Prone Areas of Rice Agriculture Using a MODIS-Derived Soil Moisture Index B. R. Parida1 Geoinformatics Center, School of Engineering and Technology, Asian Institute of Technology (AIT), P.O. Box 4, Klong Luang, Pathumthani 12120, Thailand W. B. Collado Philippine Rice Research Institute, Central Experiment Station, Maligaya, Science City of Muñoz, Nueva Ecija, Philippines R. Borah, M. K. Hazarika, and L. Samarakoon Geoinformatics Center, School of Engineering and Technology, Asian Institute of Technology (AIT), P.O. Box 4, Klong Luang, Pathumthani 12120, Thailand Abstract: This study examined the use of remote sensing in detecting and assessing drought in Iloilo Province, Philippines. A remote sensing–based soil moisture index (SMI), rainfall anomaly data from the Tropical Rainfall Measuring Mission (TRMM), and rice production departure (Pd) data were used for drought detection and validation. The study was conducted using two drought years (2001, 2005) and one non-drought year (2002). According to SMI data, the drought distribution was classified into four major groups. SMI values > 0.3 were considered not to be drought and SMI values < 0.3 were classified as slight, moderate, and severe drought. Results based on SMI revealed that the study area experienced drought in 2001 and 2005, while 2002 exhibited no drought. On the other hand, TRMM-based rainfall anomaly data revealed negative values in 2001 and 2005 and positive values in 2002. Below-normal Pd values were observed in 2005 and above-normal values in 2002, whereas nearly normal values prevailed in 2001. Yield indicator data were crucial for the assessment of drought impacts on rice production. In most cases, the pattern of rice production and productivity revealed that the decline in the production or productivity of rice for a particular year coincided with lower SMI values and greater rainfall departure or negative anomaly. INTRODUCTION Drought is the occurrence of a protracted period of deficient precipitation resulting in extensive damage to crops and loss of yield. El Niño, a recurring period of 1 Email: bikashrp@gmail.com 109 GIScience & Remote Sensing, 2008, 45, No. 1, p. 109–129. DOI: 10.2747/1548-1603.45.1.109 Copyright © 2008 by Bellwether Publishing, Ltd. All rights reserved. 110 PARIDA ET AL. strong and prolonged warm weather, is considered to be the cause of dry spells and drought in the Philippines and other Southeast Asian countries. It produces serious impacts in many sectors of the economy and is very prominent in the agricultural sector. Drought generally leads to lower grain production, which is a function of the intensity and duration of drought. In the Philippines, approximately 27% of the land is vulnerable to drought. Provinces with Type 1 2 climates are more vulnerable to typhoons, drought, and El Niño episodes. The relationships between the Asian– Australian monsoons and El Niño–Southern Oscillation (ENSO) episodes in the tropical Pacific have long been the subject of considerable scientific and practical interest (Lau and Nath, 2000). The association of various ENSO events over the past century with monsoon fluctuations was documented in detail by Li (1990), Zhang et al. (1996), and Chang et al. (2000) for the eastern Asian sector. The Philippine Atmospheric, Geophysical, and Astronomical Services Administration (PAGASA) has recorded 13 serious El Niño events in the country over the period 1950–1998 (Concepcion, 2004). The period 1997–1998 is considered to be one of the worst El Niño episodes worldwide (Drexler and Ewel, 2001). The El Niño phenomenon that affected the Philippines in 1997–1998 was the most extensive in terms of areal coverage and adversely affected agricultural lands and crop production. Iloilo Province was one of the most heavily affected provinces in the country. An increasing number of provinces, particularly the small islands, have become vulnerable to drought and land degradation due to inadequate and inefficient irrigation systems, increasing population and rural poverty, poor land and watershed management, and the increasing incidence of El Niño and La Niña events (Concepcion, 2004). The agricultural crops most affected in the Philippines are rice and corn. For rice, water deficits affect many physiological processes associated with growth, and under severe deficits may result in death of the plants. The effect of a water deficit or drought stress may vary with the variety, degree, and duration of the stress and the growth stage of the rice crop. The seasonal water requirements of rice range from 1,200 to 1,600 mm for a growing period of 100–150 days (Jehangir et al., 2004). The optimum water requirement is about 200 mm/month, and if rainfall is < 100 mm/month, crop growth is seriously retarded, especially if the deficit occurs during the flowering and grainfilling stages (Syamisiah et al., 1993). In the past, climate and meteorological data have been the primary sources of drought information used to support decision-making. Various tools and indicators are currently in use for drought detection and monitoring. The preference for one or another of these methods depends on the characteristics of drought and the specific application (e.g., meteorological, hydrological, agricultural, and socio-economic), the size of the study area, and the availability of the basic input data. When studying drought events over a large area, the limited availability and spatial density of the necessary measurements may pose serious limitations to the spatial representation of the 2 The Philippines is divided into four climatic types, depending on how rainfall is distributed throughout the year. Type 1 has two pronounced seasons, wet and dry, with the maximum rain period occurring from June to September and a dry season lasting from 3 to 6 months. Type II lacks a dry season, and is characterized by a very pronounced precipitation maximum in December and January. Type III does not have a pronounced precipitation maximum, with a short dry season lasting from 1 to 3 months. In Type IV climates, precipitation is distributed more or less evenly throughout the year (Moog, 2005). DETECTING DROUGHT-PRONE AREAS OF RICE AGRICULTURE 111 derived information (Rossi et al., 1992). Thus, use of satellite data in conjunction with relevant field data has been studied for mapping and monitoring droughts at the regional scale. In addition, satellite observations have proven to be a valuable source of timely, spatially continuous data with improved detail for monitoring vegetation dynamics over large areas. Many previous studies of vegetation conditions have been based on analyses of numerical transforms known as vegetation indices (VI). Remote sensing technology is an economical and promising tool for obtaining land surface parameters for drought analysis. Remote sensing technology used to assess or monitor regional drought is mainly based on an index that is a function of spectral vegetation and/or land surface temperature. Wang et al. (2004) concluded that drought information is not closely related to NDVI data, and that a drought index based on NDVI is insensitive to soil water status. A drought index based on land surface temperature (LST or Ts) should be more efficient than those based on NDVI only. A drought index based on NDVI falls short in monitoring drought because NDVI is a rather conservative indicator of water stress, which means that vegetation remains green after initial water stress (Sandholt et al., 2002). In contrast, LST is more sensitive to water stress (Goetz, 1997). The combination of NDVI and LST provides information on both vegetation and moisture status. The scatter plot of remotely sensed temperature and the spectral vegetation index often exhibits a triangular (Carlson et al., 1994) or trapezoidal (Moran et al., 1994) shape and is called the NDVI-Ts space if a full range of fractional vegetation cover and soil moisture content is represented. The vegetation index and surface temperature are important parameters for describing the dry/wet condition of the land surface. The Vegetation Temperature Condition Index (VTCI), Water Deficit Index (WDI), Drought Severity Index (DEVNDVI), and Vegetation Condition Index (VCI) are remote sensing–derived indices found to be sensitive indicators of drought condition (Thenkabail et al., 2004). These indices are radiometric measures of vegetation condition and dynamics, exploiting the unique spectral signatures of canopy elements, particularly in the red and near-infrared portions of the spectrum (Huete et al., 1997) and are sensitive to vegetation type, growth stage, canopy cover, and structure (Clevers et al., 1993; Thenkabail, 2003). Surface soil moisture is also an important variable that could significantly improve drought monitoring (Zhan et al., 2004). This research examined the use of a surface moisture indicator—i.e., the Soil Moisture Index (SMI)—derived from the eight-day composite MODIS satellite images, specifically land surface temperature (LST) and bands 1 (red) and 2 (near infrared) reflectance data. It was developed for drought detection in drought-prone areas of the Philippines. The method was used for the detection of drought-prone/affected areas in Iloilo Province for two prominent drought years (2001 and 2005) and one non-drought year (2002). Rice crop production and productivity data were used for quantifying and assessing the drought in terms of crop loss in the study area. This made it possible to assess drought in regions with scarce ground meteorological observations (Thenkabail, 2003). The objectives of the study were to: (1) use a remote sensing–derived index for drought detection and monitoring; and (2) study the impacts of drought on rice production and productivity. 112 PARIDA ET AL. Fig. 1. The study area—Iloilo Province, Philippines. STUDY AREA AND CAUSES OF DROUGHT Iloilo is the largest and most developed province on Panay Island, Philippines (Fig. 1). It borders on Capiz Province in the north and Antique Province in the west. The waters of Panay Gulf, and the Iloilo and Guimaras straits surround the province on the south and east. The province extends from 10°27'36'' to 11°36'00'' N. Lat. and 122°1'48'' to 123°9'00'' E. Long., has a total land area of 466,342 ha, and contains 113,995 ha of timberland (Iloilo Provincial Planning and Development Office, 2006). The province has 43 municipalities and a total population of 1,559,182 in 2000 (National Statistics Office, 2000). Level plains in the southeast, mountains in the west, and hills and rolling terrain in the northeast characterize the topography of the province. The mountain ranges lie along the border between Iloilo and Antique and Capiz provinces, and roll down to flat plains toward the coastal towns. Almost onethird of the entire province has rather level terrain. There are 17 soil types in the province, with the loam type being predominant and conducive to farming. Generally, the soil is fertile and suitable for almost all types of agricultural crops (The Official Home of the Province of Iloilo, 2006). In the Philippines, El Niño events are associated with drier than normal conditions that cause dry spells and drought. The effects of El Niño can be felt in various sectors, such as the environment, public health, agriculture, water resources, and energy production and consumption. The agricultural sector is the most vulnerable to drought, and particularly rice, the dominant and most important agricultural crop in DETECTING DROUGHT-PRONE AREAS OF RICE AGRICULTURE 113 Table 1. Satellite Imagery Used for Drought Detection and Land Use Mapping in Iloilo Province Satellite Product name MOD09A1 Terra/ MODIS Terra/ MODIS Images Date Surface reflectance (SR) Spatial resolution 500 m MOD11A2 Land surface temperature (LST) Eight-day composite, Aug–October, 2001–2002, 2005 MOD13Q1 Vegetation indices 16-day composite, 2001 1 km 250 m the province. Climatological records reveal that major drought events in the Philippines are associated with El Niño occurrences or warm episodes in the central and eastern equatorial Pacific. Provinces in the western portions of the country experiencing a Type 1 climate are characterized by two pronounced seasons, dry and wet, with the precipitation maximum from June to September due to prevalence of a southwest monsoon. The climate is dry from December to June and wet from July to November in Iloilo Province. Annual precipitation is about 1,789 mm. Seasonal aridity is exacerbated by the increasing incidence of El Niño, which now occurs on a two- to threeyear cycle, as compared to previous five-year intervals (Concepcion, 2004). DATA MODIS land surface temperature (LST) and surface reflectance (SR) data were used for the detection and assessment of drought in Iloilo for 2001, 2002, and 2005. A total of 15 LST and SR images were downloaded from the Earth Observation Systems website (Table 1). Only cloud-free and partially contaminated data were used for this study. Another MODIS product, MOD13Q1, with 250 × 250 m spatial resolution on August 29, 2001 was used for land use/land cover mapping in the province. Rainfall data from the Visayas Experiment Station Agro meteorological station, Jaro, and Iloilo City from 1990 to 2005 were obtained from the International Rice Research Institute (IRRI). Rice production data from 1994 to 2005 were obtained from the Provincial Agriculture Office of Iloilo. GPS points of different land cover and information obtained by interviewing farmers were used for land use mapping and to integrate the remote sensing data with those on rice production and productivity. Land Surface Temperature (LST) Land surface temperature is generally defined as the skin temperature of the ground. For bare soil surfaces, LST is the soil surface temperature. For densely vegetated ground, LST can be viewed as the canopy surface temperature of the vegetation, and for sparsely vegetated ground LST is determined by the temperature of the vegetation canopy, vegetation body, and soil surface (Qin and Karnieli, 1999). LST is 114 PARIDA ET AL. one of the key parameters in the physics of land-surface processes at regional and global scales, combining the results of all surface-atmosphere interactions and energy fluxes between the atmosphere and the ground (Mannstein, 1987). The satellite data used in this study were the MODIS land surface temperature product (MOD11A2), consisting of 1 × 1 km MODIS LST data averaged over eight days. LSTs in MOD11A1 were retrieved by the generalized split-window algorithm (Wan and Dozier, 1996). Emissivities in bands 31 and 32 were estimated by the classificationbased emissivity method (Snyder et al., 1998) according to land cover types in the pixel determined by the input data in quarterly land cover (MOD12Q1) and daily snow cover (MOD10_L2). Normalized Difference Vegetation Index (NDVI) The NDVI is a measure of the amount and vigor of vegetation at the surface. The magnitude of NDVI is related to the level of photosynthetic activity in the observed vegetation. In general, higher values of NDVI indicate greater vigor and amounts of vegetation. The reason NDVI is related to vegetation is that healthy vegetation reflects very well in the near-infrared part of the spectrum and absorbs much of the incident energy in the red portion of the spectrum. It is defined as: NDVI = (band 2 – band 1)/(band 2 + band 1) or (NIR – RED)/(NIR + RED), where band 2 = 858 nm and band 1 = 645 nm. NDVI is a good indicator of green biomass, leaf area index, and patterns of production (Thenkabail et al., 2004). NDVI is the most commonly used vegetation index and it varies from –1 to + 1. In this study NDVI was computed using two bands of MODIS surface reflectance. METHODOLOGY MODIS products (i.e., LST and SR) were first converted to a geotiff format. The converted images were then subjected to preprocessing including reprojection, resizing, and subsetting. Scaling was also performed to obtain the real values for LST and SR images. The SR images were used to produce NDVI data. The value for each point that represents the warm edge (the upper limit to surface temperature for a given vegetation cover) and the cold edge (well-watered condition) were derived from a scatter plot of NDVI and LST. The warm- and cold-edge values were obtained from linear regression equations and then used to generate the SMI images. SMI values ranged from 0 (severe drought) to 1 (no drought). In general, a lower SMI value indicated heavier drought occurrence. The land use map was prepared from the MODIS image using the RuleGen decision tree supervised classification technique. Decision tree classifiers have several advantages over traditional supervised classification procedures. They have significant intuitive appeal because the classification structure is explicit and therefore easily interpretable (Fayyad and Irani, 1992). A decision tree is a classification procedure that recursively partitions a data set into smaller subdivisions based on a set of tests defined at each node in the tree (Friedl and Brodley, 1997). In this framework, a data set was classified by sequentially subdividing it DETECTING DROUGHT-PRONE AREAS OF RICE AGRICULTURE 115 according to the decision defined by the tree, and class labels were assigned to each observation according to the leaf node into which the observation fell. The generalized land use classification was done using a hierarchical rule-based decision tree supervised classification method. This classification was done by obtaining training samples from the satellite image. The rice production and productivity were used for validation of results of remote sensing–based indices by computing production departure. Soil Moisture Index In this study, the scatter plot of remotely sensed LST and NDVI used in the LSTNDVI space resulted in a trapezoidal shape (Fig. 2), as described by Lambin and Ehrlich (1996). All types of land cover fell within the trapezoid of the LST-NDVI space. The upper envelope of the trapezoid, A–C, representing the dry condition, was called the “warm edge”—the upper limit of surface temperature for a given vegetation cover. The lower limit of the trapezoid, B–D, corresponding to the well-watered condition, was called the “cold edge.” Following the concept illustrated in Figure 2, the LST and NDVI data were combined to form an index for identifying soil moisture. A soil moisture index (SMI) based on the location of a pixel in the LST-NDVI space was defined in a scatter plot. Further, SMI having values of 0 at the “warm edge” and 1 at the “cold edge” were established. The SMI for a given pixel E (LST, NDVI) in the LST-NDVI space was estimated as the proportion between line M–E and M–N (Zhan et al., 2004): LST max – LST SMI = ------------------------------------------LST max – LST min (1) where Tmax and Tmin are the maximum and minimum land surface temperatures for a given NDVI. LST is the observed surface temperature at the given pixel for a given NDVI where LSTmax = (a1) NDVI + b1 and LSTmin = (a2) NDVI + b2, and, ai and bi (i = 1, 2) are parameters defining the warm and cold edges modeled as a linear fit to data (Zhan et al., 2004). TRMM Product, Rice Production (Pd) and Yield Departure (Yd) Rainfall data from the Tropical Rainfall Measuring Mission (TRMM) were compared with seasonal rainfall and remote sensing results. The TRMM product 3B43 (monthly 0.25° × 0.25°–TRMM and other sources of rainfall; i.e., merge rain rate from TRMM, geosynchronous IR, SSM/I, and rain gauges) was used for generating rainfall anomalies based on NASA’s Glovani rainfall analysis tool. Based on these data, the rainfall anomalies for 1998, 2001, 2002, and 2005 were generated. The rice production and yield data from 1994 to 2005 were used in the analysis of rice production departure because the long-term production data indicate the variation of production over time. Rice production and yield departure were calculated to observe the relationships with crop loss due to drought. The rice yield trend in Iloilo Province was plotted against time for yield analysis using historical rice yield data for 1994–2004 (Fig. 3). There is an increasing trend in yield over this period due to 116 Fig. 2. (right) Scatter plot in LST-NDVI space and the definition of SMI. PARIDA ET AL. Fig. 3. (below) Linear yield trend for rain-fed rice in Pototan and San Enrique, Iloilo Province. DETECTING DROUGHT-PRONE AREAS OF RICE AGRICULTURE 117 technological advances (assuming farmers were using high-yielding rice varieties and more fertilizer). In the graph, the year 1994 was considered as the base year (0) for plotting the yield trend, whereas the year 2004 is indicated as 10. The two municipalities of Pototan (central region) and San Enrique (northern part) were plotted for rice yield trend analysis as a representative test case in the province, where Pototan and San Enrique yielded regression coefficients of 0.7 and 0.52, respectively. The production departure of rice was computed using the following equation to study the impacts of drought on crop performance: Actual production ( P a ) Production departure ( Pd ) = ---------------------------------------------------------------------Average production ( P m ) (2) where Pa is the actual production in a particular year and Pm is the average production of 12 years (1994–2005). Similarly, the yield departure was computed using the rice yield based on: Actual yield ( Y a ) Yield departure ( Yd ) = -------------------------------------------------------Average yield ( Y m ) (3) Field Survey A field survey was conducted to validate the remote identification of droughtprone areas in Iloilo Province. Land use maps, topographic maps, and soil moisture index (SMI) images were used as references for determining the locations of GPS points. The areas that consistently exhibited drought in the SMI images together with areas that exhibited lush vegetation or showed no signs of drought were given high priority as target areas for field visits and GPS point collection. Sixty-five GPS points were collected, which represented different land covers such as rice, grass, trees, sugarcane, water, and built-up areas. Interviews with farmers were carried out simultaneously with GPS point collection. Questions during the survey included the farmer’s name, crop calendar, cropping pattern, rice ecosystem, average yield, rainfall pattern, soil type, drought occurrence, and losses or problems encountered due to drought. RESULTS AND DISCUSSION Land Use Classification of Iloilo Province The land use map of Iloilo Province was prepared using the RuleGen decision tree supervised classification technique. MOD13Q12001241, a MODIS product with a 250 m spatial resolution, was used in the classification using red and NIR reflectance. The 16-day composite reflectance image was classified into four major classes: rice, other crops, FSG,3 and others (Fig. 4). The categories for rice and other crops encompassed the prominent agricultural fields in the province, whereas the “others” class included built-up urban area, roads, barren land, fish ponds, etc. The land use map indicates that the rice crop is dominant mainly in the central part of the province, PARIDA ET AL. 118 Fig. 4. Land use map of Iloilo Province prepared from MOD13Q12001241 (250 m data). with forest cover being found along the borders of the province, including the eastern part. Soil Moisture Index The Soil Moisture Index (SMI) was computed based on the NDVI-Ts space using MODIS-based LST and NDVI images. SMI was computed for 2001–2002 and 2005, which comprised 233 to 289 days of the year (DOY). This index was calculated from August to October in each year for regional drought detection and assessment in Iloilo Province. DOY 265 and 273 were used for drought identification and mapping because they represent the critical growth phases of the rice crop, which are more sensitive to thermal stress. Goetz (1997) reported that the negative correlation between LST and NDVI observed at several scales is largely due to changes in vegetation cover and soil moisture, and indicated that the surface temperature can rise rapidly 3 FSG stands for forest land, shrubs, and grass land, and primarily includes dense forest, open forest, trees, shrubs, plantations, etc. DETECTING DROUGHT-PRONE AREAS OF RICE AGRICULTURE 119 Table 2. Warm- and Cold-Edge Linear Equations Used for SMIa DOY Warm edge R2 Cold edge R2 2001 249 LSTmax = –6.8837x + 310.57 0.90 LSTmin = 10.478x + 286.14 0.97 257 LSTmax = –7.7875x + 312.76 0.91 LSTmin = 6.5268x + 289.38 0.94 273 LSTmax = –9.3998x + 313.61 0.94 LSTmin = 5.4319x + 286.65 0.91 289 LSTmax = –5.7279x + 307.46 0.92 LSTmin = 5.4620x + 289.51 0.97 241 LSTmax = –7.3533x + 310.97 0.92 LSTmin = 4.1443x + 290.78 0.9 265 LSTmax = –5.9890x + 309.64 0.91 LSTmin = 6.8369x + 288.74 0.9 273 LSTmax = –4.7784x + 308.22 0.96 LSTmin = 11.321x + 284.53 0.91 289 LSTmax = –3.8092x + 307.61 0.93 LSTmin = 9.5103x + 285.87 0.94 2002 2005 233 LSTmax = –4.5920x + 308.43 0.9 LSTmin = 10.835x + 285.76 0.98 265 LSTmax = –7.4628x + 310.66 0.91 LSTmin = 4.6937x + 289.59 0.95 273 LSTmax = –6.5322x + 309.77 0.9 LSTmin = 7.8194x + 281.85 0.95 a x indicates NDVI. with water stress. The selected linear regression equations derived from NDVI-Ts space for the SMI index are given in Table 2. Spatial and Temporal Distribution of Land Surface Temperature. The spatial and temporal patterns of LST at different periods are shown in Figure 5. LST is considered to be one of the important parameters for drought detection and deriving possible indices by integrating it with other vegetation parameters. It is seen from Figure 5 that the distribution of temperature varies mainly with time and the type of vegetation cover. The temperature ranged from 285 to 307 K and was higher in 2001 compared to 2002 and 2005. The year 2002 exhibits the lowest temperature. In 2005, most of the areas experienced temperatures in the range of 300 to 304 K. The surface temperature in forest areas exhibited the lowest temperature in all the years compared to the agricultural lands. The central part of the province exhibited higher surface temperatures compared to the other parts of the province. SMI Distribution and Drought Severity Mapping. Drought occurrence in the study area was classified into four major groups based on SMI values (Table 3). SMI values > 0.3 were classified as no drought or favorable soil moisture conditions, whereas SMI values < 0.3 were classified into three categories of drought, namely severe, moderate, and slight drought. The distribution of drought was also explained in a temporal pattern using series of SMI-based satellite images. The SMI images PARIDA ET AL. 120 Fig. 5. Spatio-temporal LST distribution in Iloilo Province, 2001–2002 and 2005. Table 3. The Drought Category Based on SMI Values SMI range Drought classification 0.0 – 0.1 Severe drought 0.1 – 0.2 Moderate drought 0.2 – 0.3 Slight drought > 0.3 No drought were subjected to visual interpretation. It was observed that in 249 DOY in 2001, drought conditions were concentrated mainly on the central part of the province. The municipalities that exhibited drought were Santa Barbara, Cabatuan, Mina, San Miguel, Oton, New Lucena, Iloilo city, Calinog, Bingawan, part of Lambunao, Igbaras, Maasin, Alimodian Tabungan, Zarraga, and Leon. In succeeding stages, the spatial extent of drought was reduced in 273 DOY. In 289 DOY, there was not much change in the spatial distribution of drought as compared to 273 DOY except for a DETECTING DROUGHT-PRONE AREAS OF RICE AGRICULTURE 121 Table 4. Area Experiencing Different Levels of Drought in Iloilo Province Areas in different drought classes, km2 Year Severe Moderate Slight No drought Total area 2001 168 545 1164 2594 4471 2002 120 336 480 3599 4535 2005 178 602 1182 2593 4555 few municipalities including Dumalag, Duenas, and Iloilo City where the severity of drought increased. On the other hand, the northwestern part of the province exhibited favorable moisture conditions or no sign of drought due to forest cover. Similarly, many parts of the province with mountainous topography exhibited SMI values of > 0.3, indicating no drought due to trees/shrubs covers. In 2002, the drought effect was found to be much less when compared to 2001. The spatial pattern of SMI is similar for both 241 and 265 DOY. Only Santa Barbara, Iloilo City, Calinog, and the City of Passi experienced slight to moderate drought, while the other municipalities showed favorable moisture conditions or no drought during these periods. In 2005, the spatio-temporal pattern of SMI revealed that drought was limited to the municipalities of Janiuay and Badiangan in 233 DOY. The majority of the municipalities showed slight drought conditions. In 265 DOY, the severe drought extended to the municipalities of Duenas, San Enrique, Santa Barbara, Leganes, San Miguel, and Oton, while the rest of the municipalities showed moderate to slight drought. Drought severity mapping was carried out based on SMI threshold values (Table 3) using 265 DOY satellite images for each year because this period represented one of the critical growth stages of the rice crop. In 2001, 273 DOY was used instead of 265 because of the lack of availability of cloud-free satellite data for the latter. The rice crop was the most dominant agricultural crop in Iloilo Province. The rice area distribution is shown in the land use map (Fig. 4) for comparison with the remotely sensed index and the land use and land cover. Thus, SMI values can be easily compared with respect to land cover by knowing the drought-affected areas with respect to their geographical locations and the particular land use types. The drought distribution in 2001–2002 and 2005 varied from year to year (Fig. 6). In 2001, drought conditions covered mainly the central part of the province with severe to slight drought. It was observed that the total area under drought was less in 2002 versus 2001 and 2005, while the area under slight drought was greater in 2001 and 2005 compared to 2002 (Table 4). In 2002, drought conditions were very low compared to the other years, providing a favorable soil moisture condition for the rice crop. In 2005, drought was confined to the central part of the province with different degrees of severity. The area under moderate to slight drought was greater in 2005. The degree of severity in 2005 was quite similar to that in 2001. Thus, it can be concluded that the year 2002 was not affected much by drought, whereas the other years were affected by drought with varying degrees of severity. In general, drought was confined to the rice areas and 122 PARIDA ET AL. Fig. 6. Drought intensity map based on SMI for 2001–2002 and 2005. those of other agricultural crops, whereas forest showed a higher SMI with no sign of drought. In Figure 7, a greater percentage of the province was affected by slight to moderate drought in 2001 and 2005. Nearly 26% of the area experienced slight drought in 2001 and 2005. Rice Production, Productivity and Yield Departure in Drought Analysis The importance of rice is well known in this region, and hence the trend of rice production was studied. Rice was the most widely planted agricultural crop in Iloilo Province, is produced in surplus, and is exported. The rice statistics for the entire province from 1994 to 2005 are given in Table 5 and were analyzed to examine the impacts of drought on crop performance. The results revealed that the linear trend of rice production in rain-fed rice is increasing (Fig. 8). However, total production was lower in 1997 and 1998 due to the adverse El Niño event. The total area harvested under rain-fed rice, as seen in Figure 8, was drastically reduced in 1998, 2000, 2001, and 2005. Rice production in 2000, 2001, and 2005 was much lower compared to 2002 to 2004. The period 2002–2004 showed the highest production in the last 12 DETECTING DROUGHT-PRONE AREAS OF RICE AGRICULTURE 123 Fig. 7. Total percentage of area under drought based on above SMI in each year during DOY 265. Table 5. Rice Statistics for Rain-Fed Ecosystems in Iloilo Province from 1994–2005a Year Area Production Productivity 1994 110,882 306,902 2.85 1995 104,598 293,423 2.87 1996 100,960 281,277 2.87 1997 95,062 246,717 2.66 1998 67,553 136,404 1.96 1999 117,284 318,145 2.68 2000 97,451 300,909 3.32 2001 112,678 370,095 3.53 2002 127,245 426,423 3.44 2003 119,432 444,139 3.75 2004 118,271 433,685 3.7 2005 106,266 290,883 2.91 320,750.17 3.05 Average aArea is in ha, production in mt, and productivity in t/ha). years. The area during 1994 was similar to the area during 2003 and 2004 but there was a big difference in production, owing to agricultural development as discussed in yield trends. The maximum output was obtained in 2003, with a total of 444,139 metric tons (mt), while the minimum production of 136,404 mt was in 1998. The 124 PARIDA ET AL. Fig. 8. Total rain-fed rice production and area in Iloilo Province, 1994–2005. drought (El Niño event) caused a substantial crop loss in 1998 compared to other years. Crop loss or failure also was observed in 2001 and 2005 with the occurrence of drought, which resulted in a decrease in rice production compared to 2002 to 2004. Rice productivity results revealed that productivity was quite low (2.91) in 2005 compared to 2000–2004. Values of Pd or Yd < 1 indicate the loss of production due to drought or other causes, whereas values >1 indicate more than normal, non-drought production (Table 6). Results show high Pd values (>1.3) prevailed during the period 2002–2004, and lower values <1.0 Pd during 1998, 2000, and 2005. In 2001, the Pd value was a little more than 1, or just about normal production. The base year 1994 and subsequent years up to 1997 consistently showed Pd values <1; however, it could not be concluded that these were drought years. The likely reason for this is the technological gaps between 1994 and 2000. Results also indicate that the lower the Pd value, the lower the yield or the higher the yield departure from normal. Recent years show greater rice production compared to 1994, which can be attributed to recent agricultural development, although it is obvious that output in 2001 and 2005 was influenced by drought, as evidenced by relatively low production departure values. Similarly, Yd indicated that yield was influenced by drought in 2005 relative to 2002, whereas in 2001 it exhibited a nearly normal departure. Overall, the production departure exhibited more accurate results than the yield departure. Rainfall Anomaly Based on TRMM and Seasonal Rainfall Climatic data from the Agro meteorological station in Iloilo City were used to analyze seasonal rainfall. These results were compared with the TRMM-based rainfall anomaly and SMI results. The seasonal rainfall pattern from June to October DETECTING DROUGHT-PRONE AREAS OF RICE AGRICULTURE 125 Table 6. Rice Production and Yield Departure for Rain-Fed Ecosystems in Iloilo Province, 1994–2005 Year Pd Yd 1994 0.96 0.94 1995 0.91 0.94 1996 0.88 0.94 1997 0.77 0.87 1998 0.43 0.64 1999 0.99 0.88 2000 0.94 1.09 2001 1.15 1.16 2002 1.33 1.13 2003 1.38 1.23 2004 1.35 1.21 2005 0.91 0.95 Fig. 9. Total monthly rainfall from June to October taken from the Visayas Experiment Station Agro meteorological Station, Iloilo City (Source: IRRI). (2001–2002 and 2005) in Iloilo City is shown in Figure 9. The trend shows that rainfall was highest in July and August and decreased over time. However, it can be seen that in 2001 there was a decrease in the amount of rainfall received during the month 126 PARIDA ET AL. Fig. 10. Rainfall anomalies generated from TRMM rainfall data for 1998, 2001, 2002, and 2005. of July (193 mm) from June (351 mm). Although rainfall increased again in August, the rainfall in September was only about 102 mm, 50% below the normal average monthly rainfall during the three-year study period. In 2001 and 2005, total rainfall was lower than in 2002, which may have caused the water shortage to the rice crop and resulted in reduced production. Furthermore, although the total rainfall received in 2005 was more than in 2001, output was much lower in 2005. This was due to the limited meteorological observation data available (for only only one station), representing only part of the province rather than its entirety. Rainfall anomalies generated from the rainfall data (near the surface) from TRMM were negative in 1998, 2001, and 2005 (Fig. 10), and positive in 2002. This indicates that in 1998, 2001 and 2005 precipitation was much lower than normal in the area, leading to drought. The above results were similar to those obtained from the remote sensing–based SMI. Hence, the decrease in rice production in 1998, 2001, and 2005 can be attributed to the effects of drought in the study area. It can be observed from the spatial and DETECTING DROUGHT-PRONE AREAS OF RICE AGRICULTURE 127 temporal distribution of LST that the temperature was higher in 2001 and 2005 than in 2002. This is an important parameter in determining vegetation stress and influences the performance of the crop as a result of drought. Moreover, the TRMM- and SMI-based results indicate the influence of drought on crop performance, which was validated using long-term rice production and yield data. Conversely, the groundbased rainfall anomaly information fell short in comparison with the satellite data due its limited spatial distribution and lack of availability. Thus, there is a need to integrate both climatological as well as satellite data for drought detection, assessment, and validation in drought-prone areas. SUMMARY AND CONCLUSIONS The study provided a high level of spatial and temporal information for drought assessment using remote sensing methods. The remote sensing approach (SMI) and other techniques showed good agreement in identification of the occurrence, severity, and distribution of drought in the study area. The SMI results revealed that drought was mainly concentrated in the central part of Iloilo Province, a predominantly agricultural area. The areas covered by dense and open forest and shrubs, with mountainous topography, exhibit SMI values with little sign of drought. According to the SMI data, it can be concluded that 2002 was not much affected by drought, whereas the other years were affected by drought of differing degrees of severity. The yield indicator also provided valuable information in the assessment of the impacts of drought on rice production. The SMI, rainfall anomalies from TRMM, and Pd gave results that described the occurrence, severity, and impacts of drought on crop production. In most cases, the pattern of rice production and productivity revealed that the decline in the production or productivity of rice for a particular year coincided with lower SMI values and greater rainfall departure from the normal. Hence, the remote sensing approaches have the capability of detecting, mapping, and monitoring drought at the regional level. ACKNOWLEDGMENTS The authors would like to extend their most sincere gratitude to the Japanese Aerospace Exploration Agency (JAXA) for financial assistance during the study. Special thanks are also due to the Iloilo Provincial Agriculture Office (PAO) and the International Rice Research Institute (IRRI) for providing the rice yield and climatic data, respectively. The authors express their thanks to personnel from the Philippine Rice Research Institute for assistance and encouragement during the field work. REFERENCES Carlson, T. N., Gillies, R. R., and E. M. Perry, 1994, “A Method to Make Use of Thermal Infrared Temperature and NDVI Measurements to Infer Soil Water Content and Fractional Vegetation Cover,” Remote Sensing Reviews, 9:16–173. Chang, C. P., Zhang, Y., and T. Li, 2000, “Interannual and Interdecadal Variations of the East Asian Summer Monsoon and Tropical Pacific SSTs. Part I: Roles of the Subtropical Ridge,” Journal of Climate, 13:4310–4325. 128 PARIDA ET AL. Clevers, J. G. P. W. and W. Verhoef, 1993, “LAI Estimation by Means of the WDVI: A Sensitivity Analysis with a Combined PROSPECT-SAIL Model,” Remote Sensing of Environment, 7:43–64. Concepcion, R. N., 2004, Summary Report: The Philippine National Action Plan to Combat Desertification, Land Degradation, Drought and Poverty, DA-BSWM, DENR, DOST, DAR, Manila, Philippines: National Country Report on the UNCCD Implementation (Philippines), [http://www.unccd.int/actionprogrammes/asia/national/2004/philippines-eng.pdf]. Drexler, J. Z. and K. C. Ewel, 2001, “Effect of the 1997–1998 ENSO-Related Drought on Hydrology and Salinity in a Micronesian Wetland Complex,” Estuaries, 24:347–356. Fayyad, U. M. and K. B. Irani, 1992, “On the Handling of Continuous-Valued Attributes in Decision Tree Generation,” Machine Learning, 8:87–102. Friedl, M. A. and C. E. Brodley, 1997, “Decision Tree Classification of Land Cover from Remotely Sensed Data,” Remote Sensing of Environment, 61:399–409. Goetz, S. J., 1997, “Multi-sensor Analysis of NDVI, Surface Temperature, and Biophysical Variables at a Mixed Grassland Site,” International Journal of Remote Sensing, 18:71–94. Huete, A. R., Liu, H. Q., Batchily, K. and W.V. Leeuwen, 1997, “A Comparison of Vegetation Indices over a Global Set of TM Images for EOS-MODIS,” Remote Sensing of Environment, 59:440–451. Iloilo Provincial Planning and Development Office, 2006, website [http:// www.iolilo.gov.ph]. Jehangir, W. A., Turral, H., and I. Masih, 2004, “Water Productivity of Rice Crops in Irrigated Areas,” paper presented at the 4th International Crop Science Congress, Brisbane, Queensland, Australia, 27 September to 1 October 2004. Lambin, E. F. and D. Ehrlich, 1996, “The Surface Temperature–Vegetation Index Space for Land Cover and Land-Cover Change Analysis,” International Journal of Remote Sensing, 17:463–487. Lau, N. C. and M. J. Nath, 2000, “Impact of ENSO on the Variability of the AsianAustralian Monsoons as Simulated in GCM Experiments,” Journal of Climate, 13:4287–4309. Li, C., 1990, “Interaction between Anomalous Winter Monsoon in East Asia and El Niño Events,” Advances in Atmospheric Sciences, 7:36–46. Mannstein, H., 1987, “Surface Energy Budget, Surface Temperature, and Thermal Inertia,” in Vaughan, R. A. (Ed.), Remote Sensing Applications in Meteorology and Climatology, Dordrecht, The Netherlands: D. Reidel, NATO Advanced Study Institutes Series, Series C, Mathematical and Physical Sciences, vol. 201. Moog, F. A., 2005, “Philippines,” Rome, Italy: U.N. Food and Agricultural Organization, Agriculture Department, Crop and Grassland Service (AGPC) Country Pasture/Forage Resource Profiles [http://www.fao.org/ag/AGP/AGPC/doc/ Counprof/Philippines/Philipp.htm]. Moran, M. S., Clarke, T. R., Inoue, U., and A. Vidal, A., 1994, “Estimating Crop Water Deficit Using the Relation between Surface Air Temperature and Spectral Vegetation Index,” Remote Sensing of Environment, 49:246–263. National Statistics Office, 2000, Philippines census report [http://www.census.gov.ph/ data/pressrelease/2002/pr0286tx.html DETECTING DROUGHT-PRONE AREAS OF RICE AGRICULTURE 129 Qin, Z. and A. Karnieli, 1999, “Progress in the Remote Sensing of Land Surface Temperature and Ground Emissivity using NOAA-AVHRR Data,” International Journal of Remote Sensing, 20:2367–2393. Rossi, G., Benedini, M., Tsakiris, G., and S. Giakoumis, 1992, “On Regional Drought Estimation and Analysis,” Water Resources Management, 6: 249–277. Syamisiah, I., Suprapto, D., and S. Bhuiyan, 1993, “Rainwater Collection and Conservation for Alleviation of Drought and Increasing Productivity of Rain-Fed Ricelands in Indonesia,” paper presented at the International Symposium on Rainfed Lowland, Semarang, Central Java, Indonesia. February 8–13, 1993. Sandholt, I., Rasmussen, K., and J. Anderson, 2002, “A Simple Interpretation of the Surface Temperature/Vegetation Index Space for Assessment of the Surface Moisture Status,” Remote Sensing of Environment, 79:213–224. Snyder, W. C., Wan, Z., Zhang, Y., and Y. Z. Feng, 1998, “Classification-Based Emissivity for Land Surface Temperature Measurement from Space,” International Journal of Remote Sensing, 19(14):2753–2774. The Official Home of the Province of Iloilo, 2006, website [http://iloilo.gov.ph]. Thenkabail, P. S., 2003, “Biophysical and Yield Information for Precision Farming from Near-Real-Time and Historical Landsat TM Images,” International Journal of Remote Sensing, 24:839–877. Thenkabail, P. S., Gamage, M. S. D. N., and V. U. Smakhtin, 2004, The Use of Remote Sensing Data for Drought Assessment and Monitoring in Southwest Asia, Colombo, Sri Lanka: International Water Management Institute. Wan, Z. and J. Dozier, 1996, “A Generalized Split-Window Algorithm for Retrieving Land-Surface Temperature from Space,” IEEE Transactions on Geoscience and Remote Sensing, 34(4):892–905. Wang, C., Qi, S., Niu, Z., and J. Wang, 2004, “Evaluating Soil Moisture Status in China Using the Temperature-Vegetation Dryness Index (TVDI),” Canadian Journal of Remote Sensing, 30(5):671–679. Zhan, Z., Qin, Q., and X. Wang, 2004, “The Application of LST/NDVI Index for Monitoring Land Surface Moisture in Semiarid Area,”. Geoscience and Remote Sensing, 2004, IGARSS ‘04. Proceedings 2004 IEEE International, 3:1551– 1555. Zhang, R., Sumi, A., and M. Kimoto, 1996, “Impact of El Niño on the East Asian Monsoon: A Diagnostic Study of the 86/87 and 91/92 Events,” Journal of the Meteorological Society of Japan, 74:49–62.