Analysis of time-series MODIS 250 m vegetation index data for crop classification in the U.S. Central Great Plains

https://doi.org/10.1016/j.rse.2006.11.021Get rights and content

Abstract

The global environmental change research community requires improved and up-to-date land use/land cover (LULC) datasets at regional to global scales to support a variety of science and policy applications. Considerable strides have been made to improve large-area LULC datasets, but little emphasis has been placed on thematically detailed crop mapping, despite the considerable influence of management activities in the cropland sector on various environmental processes and the economy. Time-series MODIS 250 m Vegetation Index (VI) datasets hold considerable promise for large-area crop mapping in an agriculturally intensive region such as the U.S. Central Great Plains, given their global coverage, intermediate spatial resolution, high temporal resolution (16-day composite period), and cost-free status. However, the specific spectral–temporal information contained in these data has yet to be thoroughly explored and their applicability for large-area crop-related LULC classification is relatively unknown. The objective of this research was to investigate the general applicability of the time-series MODIS 250 m Enhanced Vegetation Index (EVI) and Normalized Difference Vegetation Index (NDVI) datasets for crop-related LULC classification in this region. A combination of graphical and statistical analyses were performed on a 12-month time-series of MODIS EVI and NDVI data from more than 2000 cropped field sites across the U.S. state of Kansas. Both MODIS VI datasets were found to have sufficient spatial, spectral, and temporal resolutions to detect unique multi-temporal signatures for each of the region's major crop types (alfalfa, corn, sorghum, soybeans, and winter wheat) and management practices (double crop, fallow, and irrigation). Each crop's multi-temporal VI signature was consistent with its general phenological characteristics and most crop classes were spectrally separable at some point during the growing season. Regional intra-class VI signature variations were found for some crops across Kansas that reflected the state's climate and planting time differences. The multi-temporal EVI and NDVI data tracked similar seasonal responses for all crops and were highly correlated across the growing season. However, differences between EVI and NDVI responses were most pronounced during the senescence phase of the growing season.

Introduction

Land use/land cover (LULC) data are among the most important and universally used terrestrial datasets (IGBP, 1990) and represent key environmental information for many science and policy applications (Cihlar, 2000, DeFries and Belward, 2000). The emergence of environmental change issues has generated critical new requirements for LULC information at regional to global scales. More accurate, detailed, and timely LULC datasets are needed at these scales to support the demands of a diverse and emerging user community (Cihlar, 2000, DeFries and Belward, 2000).

The environmental, economic, and social implications of LULC change have led to the recognition that LULC patterns must be mapped on a repetitive basis for large geographic areas in order to provide ‘up-to-date’ LULC information and to characterize major human-environment interactions (National Aeronautics and Space Administration (NASA), 2002, National Research Council (NRC), 2001, Turner et al., 1995). As a result, the remote sensing community has been challenged to develop regional to global scale LULC products that characterize ‘current’ LULC patterns, document major LULC changes, and include a stronger land use component. Several major research programs and documents, which include NASA's Land Cover–Land Use Change (LCLUC) program (NASA, 2002), the International Geosphere–Biosphere Program (IGBP)/International Human Dimensions Program (IHDP) Land Use/Land Cover Change (LUCC) Program (Turner et al., 1995), the National Research Council's (NRC) “Grand Challenges in Environmental Sciences” (NRC, 2001), and the U.S. Carbon Cycle Science Plan (Sarmiento & Wofsy, 1999) have identified the development of such LULC products as a research priority.

Improved and up-to-date LULC datasets are particularly needed for regions dominated by agricultural land cover such as the U.S. Central Great Plains. The cropland component of the agricultural landscape is of specific interest because it is intensively managed and continually modified, which can rapidly alter land cover patterns and influence biogeochemical and hydrologic cycles, climate, ecological processes, groundwater quality and quantity, and the economy. At the regional scale, cropland areas are characterized by a diverse mosaic of LULC types that change over various spatial and temporal scales in response to different management practices. As a result, detailed regional-scale cropping patterns need to be mapped on a repetitive basis to characterize ‘current’ LULC patterns and monitor common agricultural LULC changes. Such information is necessary to better understand the role and response of regional cropping practices in relation to various environmental issues (e.g., climate change, groundwater depletion) that potentially threaten the long-term sustainability of major agricultural producing areas such as the U.S. Central Great Plains.

Over the past decade, remotely sensed data from satellite-based sensors have proven useful for large-area LULC characterization due to their synoptic and repeat coverage. Considerable progress has been made classifying LULC patterns at the state (Eve & Merchant, 1998) and national (Craig, 2001, Homer et al., 2004, Vogelmann et al., 2001) levels using multispectral, medium resolution data from the Landsat Thematic Mapper (TM) and Enhanced Thematic Mapper (ETM+) as a primary input. Similar advances in LULC classification have also been made at national (Loveland et al., 1991, Lu et al., 2003) to global (DeFries et al., 1998, DeFries and Townshend, 1994, Hansen et al., 2000, Loveland and Belward, 1997, Loveland et al., 2000) scales using multi-temporal, coarse resolution data (1 and 8 km) from the Advanced Very High Resolution Radiometer (AVHRR). However, few of these mapping efforts have classified detailed, crop-related LULC patterns (Craig, 2001), particularly at the annual time step required to reflect common agricultural LULC changes. The development of a regional-scale crop mapping and monitoring protocol is challenging because it requires remotely sensed data that have wide geographic coverage, high temporal resolution, adequate spatial resolution relative to the grain of the landscape (i.e., typical field size), and minimal cost. Remotely sensed data from traditional sources such as Landsat and AVHRR have some of these characteristics, but are limited for such a protocol due to their spatial resolution, temporal resolution, availability, and/or cost.

Landsat TM/ETM+ data are appropriate for detailed crop mapping given the instruments' multiple spectral bands, which cover the visible through middle infrared wavelength regions, and 30 m spatial resolution. However, most crop classification using Landsat data has been limited to local scales (i.e., sub-scene level) (Mosiman, 2003, Price et al., 1997, Van Niel and McVicar, 2004, Van Niel et al., 2005). Most state/regional-scale LULC maps derived from Landsat TM and ETM+ data, such as the United States Geological Survey's (USGS) National Land Cover Dataset (NLCD) (Homer et al., 2004, Vogelmann et al., 2001) and the Gap Analysis Program (GAP) datasets (Eve & Merchant, 1998), have classified cropland areas into a single or limited number of thematic classes and are infrequently updated. The exception is the United States Department of Agriculture (USDA) National Agricultural Statistics Service (NASS) 30 m cropland data layer (CDL), which is a detailed, state-level crop classification that is annually updated (Craig, 2001). However, the CDL is only produced for a variable and limited number of states (10 total states in 2004). The production of LULC datasets comparable to the CDL in other countries with large broad-scale farming systems is also lacking. The use of Landsat data (and data from similar sensors such as SPOT) for repetitive, large-area mapping has been limited primarily by the considerable costs and time associated with the acquisition and processing of the large number of scenes that are required. Data availability/quality issues (e.g., cloud cover) associated with acquiring imagery at optimal times during the year are also a factor (DeFries & Belward, 2000).

The value of coarse resolution, time-series AVHRR normalized difference vegetation index (NDVI) data for land cover classification at national (Loveland et al., 1991, Loveland et al., 1995) to global (DeFries et al., 1998, DeFries and Townshend, 1994, Hansen et al., 2000, Loveland and Belward, 1997, Loveland et al., 2000) scales has clearly been demonstrated. The high temporal resolution (e.g., 10 to 14-day composite periods, with near-daily image acquisition) of the time-series data coupled with the NDVI's correlation with biophysical parameters (e.g., leaf area index (LAI) and green biomass) (Asrar et al., 1989, Baret and Guyot, 1991) allows land cover types to be discriminated based on their unique phenological (seasonal) characteristics. The spectral–temporal information in time-series NDVI data also has been used to monitor vegetation conditions (Jakubauskas et al., 2002, Reed et al., 1996) and major phenological events (Reed et al., 1994, Zhang et al., 2003). However, the 1-km resolution limits the spatial and thematic detail of LULC information that can be extracted from AVHRR data. Most AVHRR pixels have an integrated spectral–temporal response from multiple land cover types contained within the 1 km footprint (Townshend and Justice, 1988, Zhan et al., 2002). As a result, coarse resolution sensors are appropriate for mapping ‘natural’ systems, but the high spatial variability and complexity of agricultural systems requires higher resolution data than AVHRR provides (Turner et al., 1995). Most LULC classifications derived from 1 km AVHRR data emphasize broad scale natural vegetation classes and/or are comprised of ‘mixed’ classes representing multiple LULC types. Cropland areas are typically represented as a generalized crop class or as a mixed crop/natural vegetation class.

The Moderate Resolution Imaging Spectroradiometer (MODIS) offers an opportunity for detailed, large-area LULC characterization by providing global coverage of science quality data with high temporal resolution (1–2 days) and intermediate spatial resolution (250 m) (Justice & Townshend, 2002). An ‘AVHRR-like’ 250 m dataset is available at no cost, which includes a time series of visible red (620–670 nm) and near infrared (841–876 nm) surface reflectance, NDVI, and enhanced vegetation index (EVI) composited at 16-day intervals. The spatial, spectral, and temporal components of the MODIS 250 m VI data construct are appropriate for crop mapping and monitoring activities in the U.S. Central Great Plains. However, few studies have evaluated the potential of these data for detailed LULC characterization (Hansen et al., 2002, Wessels et al., 2004), particularly in an agricultural setting (Lobell and Asner, 2004, Wardlow et al., 2006).

The specific LULC information that can be extracted at the 250 m resolution is still relatively unexplored (Zhan et al., 2000). The 250 m bands were included in the MODIS instrument to detect anthropogenic-driven land cover changes that commonly occur at or near this spatial scale (Townshend & Justice, 1988). Land cover changes associated with anthropogenic and natural causes have been detected in the MODIS 250 m imagery (Hansen et al., 2002, Morton et al., 2006, Zhan et al., 2002). Wessels et al. (2004) found that general land cover patterns (e.g., agricultural, deciduous/evergreen forest, and grassland) could be successfully mapped with MODIS 250 m data. These results suggest that the MODIS 250 m data would be appropriate for crop mapping in the U.S. Central Great Plains given the region's relatively large field sizes. Fields are frequently 32.4 ha or larger, with such sites corresponding areally with approximately five or more 250-m MODIS pixels.

Two VIs, the NDVI and the EVI, are produced at 250-m resolution from MODIS. The NDVI is a normalized difference measure comparing the near infrared and visible red bands defined by the formulaNDVI=(ρNIRρred)/(ρNIR+ρred),where ρNIR (846–885 nm) and ρred (600–680 nm) are the bidirectional surface reflectance for the respective MODIS bands. It serves as a ‘continuity index’ to the existing AVHRR NDVI record. The EVI takes the formEVI=G((ρNIRρred)/(ρNIR+C1×ρredC2×ρblue+L)),where the ρ values are partially atmospherically corrected (Rayleigh and ozone absorption) surface reflectances, L is the canopy background adjustment (L = 1), C1 and C2 are coefficients of the aerosol resistance term that uses the 500 m blue band (458–479 nm) of MODIS (Huete et al., 1999) to correct for aerosol influences in the red band (C1 = 6 and C2 = 7.5), and G is a gain factor (G = 2.5) (Huete et al., 1994, Huete et al., 1997). The EVI is designed to minimize the effects of the atmosphere and canopy background that contaminate the NDVI (Huete et al., 1997) and to enhance the green vegetation signal (Huete et al., 2002). The MODIS VIs provide a consistent spatial and temporal coverage of vegetation conditions and complement each other for vegetation studies (Huete et al., 2002). Gao et al. (2000) found that the NDVI was more chlorophyll sensitive and saturated at high biomass levels, whereas the EVI was more responsive to canopy structure variations (e.g., LAI, plant physiognomy, and canopy type) and had improved sensitivity over high biomass areas.

Huete et al. (2002) evaluated the time-series MODIS 500 m and 1 km VI data products over several biome types (e.g., forest, grassland, and shrubland) and found that the multi-temporal signatures (or profiles) of both VIs well represented the phenology of each biome. However, the general EVI–NDVI relationship varied among the biomes and reflected differences in both their canopy structures and climate regimes. The two VIs were more strongly correlated for grassland and shrubland than for forests, and their dynamic ranges varied according to climate regime. Huete et al. (2002) also found that the EVI was more sensitive to variations over high biomass areas (e.g., tropical forest), whereas the NDVI tended to saturate. The response of the EVI and NDVI at the 250 m resolution and over cropped areas has yet to be evaluated and is a necessary first step in determining the suitability of these MODIS datasets for detailed crop characterization.

The objective of this study was to investigate the general applicability of the time-series MODIS 250 m EVI and NDVI datasets for crop-related LULC classification in the U.S. Central Great Plains. Initial results from other LULC characterization work using the MODIS 250 m VI data suggest that the spectral–temporal information in these data holds considerable potential for discriminating detailed crop classes based on their crop calendars (phenology). In this study, three primary research questions were addressed regarding the data's applicability for crop classification. First, do the time-series MODIS 250 m VI data have sufficient spatial, spectral, and temporal resolution to discriminate the region's major crop types (alfalfa, corn, sorghum, soybeans, and winter wheat) and crop-related land use practices (double crop, fallow, and irrigation)? Second, are the regional variations in climate and management practices (e.g., planting times) that occur across the study area detected in the time-series MODIS 250 m VI data for the crop classes? Third, how do the EVI and NDVI respond over the various crop cover types and how informationally distinct are the VIs in this setting? To address these questions, a combination of graphical and statistical analyses was performed on a 12-month time series of MODIS 250 m EVI and NDVI data (January to December) from 2179 cropped field sites across the state of Kansas.

Section snippets

Study area

Kansas (Fig. 1), which is situated approximately between 37° and 40°N latitude and 94° and 102°W longitude, is an agriculturally dominated state that occupies 21.3 million ha of the U.S. Central Great Plains. A cropland/grassland mosaic comprises most of the state, with 46.9% (10.0 million ha) of its total area dedicated to intensive crop production. Cropland areas primarily consist of a mosaic of relatively large fields (∼ 32.4 ha or larger) with diverse crop types and management practices that

Time-series MODIS VI data

A 12-month time series of 16-day composite MODIS 250 m EVI and NDVI data (MOD13Q1 V004) spanning one growing season (January–December 2001) was created for Kansas. The time series consisted of 23 16-day composite periods, and three tiles (h09v05, h10v05, and h10v04) of the MODIS data were required for statewide coverage. For each composite period, the EVI and NDVI data were extracted by tile, mosaicked, and reprojected from the Sinusoidal to the Lambert Azimuthal Equal Area projection.

Methods

Several graphical and statistical analyses were performed to evaluate the applicability of the time-series MODIS 250 m VI datasets for crop discrimination. First, MODIS 250 m and Landsat ETM+ 30 m imagery were visually compared to examine the spatial cropping patterns that could be resolved at the 250 m resolution in the U.S. Central Great Plains.

Second, the field sites were aggregated by crop type and management practice, and average, state-level multi-temporal VI profiles were calculated for

MODIS 250 m imagery and agricultural LULC patterns

Fig. 3 illustrates the potential of MODIS 250 m data for detecting crop-related LULC patterns in the U.S. Central Great Plains. Similar LULC patterns were detected in the multi-temporal MODIS 250 m and single date, multi-spectral Landsat ETM+ imagery at the landscape level (Fig. 3a and b). General land cover types (e.g., grassland and cropland), specific crop types (e.g., winter wheat and summer crops), and cropland under center pivot irrigation were visually evident in both images. Fig. 3a

Conclusions

The objective of this research was to evaluate the applicability of time-series MODIS 250 m VI data for large-area crop-related LULC classification in the U.S. Central Great Plains region. From this work, we drew several conclusions regarding the suitability of the data for this specific application.

First, we concluded that a time-series of the 16-day composite MODIS 250 m VI data had sufficient spectral, temporal, and radiometric resolutions to discriminate the region's major crop types and

Acknowledgements

This research was supported by NASA Headquarters under the Earth System Science Fellowship Grant NGT5‐50421 and the USGS AmericaView program. The work was conducted at the Kansas Applied Remote Sensing (KARS) Program (Edward A. Martinko, Director; Kevin P. Price, Associate Director). The authors thank the Kansas Farm Service Agency and 48 of its county offices for providing the annotated aerial photographs used for field site selection. The authors also thank the three anonymous reviewers for

References (55)

  • T.G. Van Niel et al.

    Determining temporal windows of crop discrimination with remote sensing: A case study in south-eastern Australia

    Computers and Electronics in Agriculture

    (2004)
  • T.G. Van Niel et al.

    On the relationship between training sample size and data dimensionality: Monte Carlo analysis of broadband multi-temporal classification

    Remote Sensing of Environment

    (2005)
  • K.J. Wessels et al.

    Mapping regional land cover with MODIS data for biological conservation: Examples from the Great Yellowstone Ecosystem, USA and Para State, Brazil

    Remote Sensing of Environment

    (2004)
  • R.E. Wolfe et al.

    Achieving sub-pixel geolocation accuracy in support of MODIS land science

    Remote Sensing of Environment

    (2002)
  • X. Zhan et al.

    Detection of land cover changes using MODIS 250 m data

    Remote Sensing of Environment

    (2002)
  • X. Zhang et al.

    Monitoring vegetation phenology using MODIS

    Remote Sensing of Environment

    (2003)
  • G. Asrar et al.

    Estimation of plant canopy attributes from spectral reflectance measurements: Chapter 7

  • J. Cihlar

    Land cover mapping of large areas from satellites: Status and research priorities

    International Journal of Remote Sensing

    (2000)
  • M.E. Craig

    The NASS cropland data layer program

  • R.S. DeFries et al.

    Global and regional land cover characterization from satellite data: An introduction to the special issue

    International Journal of Remote Sensing

    (2000)
  • R.S. DeFries et al.

    Global land cover classifications at 8 km spatial resolution: The use of training data derived from Landsat imagery in decision tree classifiers

    International Journal of Remote Sensing

    (1998)
  • R.S. DeFries et al.

    NDVI-derived land cover classifications at a global scale

    International Journal of Remote Sensing

    (1994)
  • Eve, M. D., & Merchant, J. W. (1998). National Survey of Land Cover Mapping Protocols Used in the GAP Analysis Program....
  • M.C. Hansen et al.

    Global land cover classification at 1 km spatial resolution using a classification tree approach

    International Journal of Remote Sensing

    (2000)
  • J. Havlin et al.
  • C. Homer et al.

    Development of a 2001 national land-cover database for the United States

    Photogrammetric Engineering and Remote Sensing

    (2004)
  • A. Huete et al.
  • Cited by (0)

    View full text