Land cover characterization of Temperate East Asia using multi-temporal VEGETATION sensor data

Abstract

Temperate East Asia (TEA) is characterized by diverse land cover types, including forest and agricultural lands, one of the world's largest temperate grasslands, and extensive desert and barren landscapes. In this paper, we explored the potential of SPOT-4 VEGETATION (VGT) data for the classification of land cover types in TEA. An unsupervised classification was performed using multi-temporal (March–November 2000) VGT-derived spectral indices (Land Surface Water Index [LSWI] and Enhanced Vegetation Index [EVI]) to generate a land cover map of TEA (called VGT-TEA). Land cover classes from VGT-TEA were aggregated to broad, general class types, and then compared and validated with classifications derived from fine-resolution (Landsat) data. VGT-TEA produced reasonable results when compared to the Landsat products. Analysis of the seasonal dynamics of LSWI and EVI allows for the identification of distinct growth patterns between different vegetation types. We suggest that LSWI seasonal curves can be used to define the growing season for temperate deciduous vegetation, including grassland types. Seasonal curves of EVI tend to have a slightly greater dynamic range than LSWI during the peak growing season and can be useful in discriminating between vegetation types. By using these two complementary spectral indices, VGT data can be used to produce timely and detailed land cover and phenology maps with limited ancillary data needed.

Introduction

Information on land cover status at the regional scale is needed for natural resource management, carbon cycle studies, and modeling of biogeochemistry, hydrology, and climate. Satellite-based remote sensing products can meet these data needs in a timely and consistent manner. Numerous studies of large-scale mapping of land cover have used data from the Advanced Very High Resolution Radiometer (AVHRR; Defries & Townshend, 1994, Loveland et al., 2000). However, the AVHRR sensors, originally designed for meteorological applications, have only two spectral bands (red and near-infrared) that can be used to generate spectral indices of vegetation vigor. Recently, a new generation of optical sensors designed for terrestrial applications has been launched. These include the VEGETATION (VGT) sensor onboard the SPOT-4 satellite and the Moderate Resolution Imaging Spectroradiometer (MODIS) onboard the Terra and Aqua satellites. VGT and MODIS have a number of advantages over AVHRR, including more spectral bands that can be used for vegetation analyses (see Section 2.1 for more details about the VGT sensor). Multi-temporal VGT data have been used to characterize forests in northeastern China (Xiao et al., 2002b) and cropland in southern China (Xiao et al., 2002a). The Global Land Cover program (GLC2000) is an on-going effort to provide a harmonized global land cover product from VGT data using a hierarchical classification scheme. Preliminary results from northern Eurasia are promising (Bartalev et al., 2003); however, only the Russian forest cover has been tested for accuracy, with an R² of 0.93 existing between GLC2000 forest areas and official Russian forest statistics (Bartalev et al., 2003).

A number of vegetation indices have been developed and used for monitoring vegetation structure and function, as well as land cover classification at large spatial scales. The Normalized Difference Vegetation Index (NDVI, Eq. (1)), which uses spectral information from the red and near infrared bands, is the most widely used. NDVI has served as the input data for various satellite-based land cover mapping activities Defries & Townshend, 1994, Loveland et al., 2000. The shortwave infrared (SWIR) band is sensitive to vegetation cover, leaf moisture and soil moisture (Tucker, 1980), and a combination of the NIR and SWIR bands has the potential for retrieving canopy water content Ceccato et al., 2002a, Ceccato et al., 2002b. The Land Surface Water Index (LSWI, Eq. (2)) is calculated using NIR and SWIR reflectance values Jürgens, 1997, Xiao et al., 2002b. Recently, LSWI has been used together with NDVI as input to land cover mapping efforts (Xiao et al., 2002b), with the expectation that the increased amount of spectral information provided from LSWI would improve the discrimination of vegetation types. It is known that NDVI has several limitations, including sensitivity to both atmospheric conditions (Xiao et al., 2003) and the soil background, and a tendency to saturate at closed vegetation canopies with large leaf area index values. To account for these limitations, the Enhanced Vegetation Index (EVI, Eq. (3)) was proposed, which directly adjusts the reflectance in the red spectral band as a function of the reflectance in the blue band Huete et al., 1997, Liu & Huete, 1995:

$$\text{NDVI}\text{=(ρ}{}_{\mathrm{<mtext>nir</mtext>}}\text{−ρ}{}_{\mathrm{<mtext>red</mtext>}}\text{)/(ρ}{}_{\mathrm{<mtext>nir</mtext>}}\text{+ρ}{}_{\mathrm{<mtext>red</mtext>}}\text{)}$$

$$\text{LSWI}\text{=(ρ}{}_{\mathrm{<mtext>nir</mtext>}}\text{−ρ}{}_{\mathrm{<mtext>swir</mtext>}}\text{)/(ρ}{}_{\mathrm{<mtext>nir</mtext>}}\text{+ρ}{}_{\mathrm{<mtext>swir</mtext>}}\text{)}$$

$$\text{EVI}\text{=2.5×}\text{ρ}{}_{\mathrm{<mtext>nir</mtext>}}\text{−ρ}{}_{\mathrm{<mtext>red</mtext>}}\text{ρ}{}_{\mathrm{<mtext>nir</mtext>}}\text{+6×ρ}{}_{\mathrm{<mtext>red</mtext>}}\text{−7.5×ρ}{}_{\mathrm{<mtext>blue</mtext>}}\text{+1}$$

where ρ_blue, ρ_red, ρ_nir, and ρ_swir represent the surface reflectance values of blue, red, NIR, and SWIR bands, respectively. In the evolution of land cover product generation, spectral input data have become increasingly tailored for land cover analyses. There is a need to assess the potential of using both EVI and LSWI for generating improved land cover classifications that take advantage of a much greater portion of the electromagnetic spectrum than previous NDVI-based products.

In this study, we explore the utility of multi-temporal VGT data for the mapping of land cover in Temperate East Asia (TEA). Our objectives were threefold: (1) to document the land cover of TEA using VGT data acquired in 2000; (2) to perform a validation of the land cover map using products derived from fine-resolution imagery; and (3) to analyze the seasonal dynamics of the various land cover types. This study outlines the potential for using VGT data to monitor the seasonal and inter-annual ecosystem dynamics in TEA. Such a regional level product can greatly improve the estimation of carbon and greenhouse gas fluxes in very dynamic and changing landscapes.

TEA is characterized by diverse land cover types, ranging from productive agricultural lands (e.g., North China Plain) to some of the most barren landscapes on Earth (e.g., Taklimakan and Gobi Deserts). The northern portion of TEA is covered by boreal forest and taiga, primarily dominated by larch and pine species. One of the most extensive temperate steppe grasslands in the world exists in the middle of the region. Land cover variability in TEA is strongly influenced by two factors: the monsoon climate system and elevation. The monsoon circulation creates a strong seasonality in precipitation availability, where the majority of annual precipitation falls in the critical growing season months. The amount and spatial distribution of the annual monsoon precipitation is controlled by several moisture sources and prevailing winds (Xue, 1996). Elevation exerts a large influence on the land cover distribution of TEA because several sizable mountain ranges are located within the region, and much of the region is situated on plateaus with an average elevation well over 1000 m (Fig. 1).

While much of TEA remains remote and inaccessible, the area has experienced increased human and natural activities that have significantly altered the structure and function of the constituent ecosystems. Forests have become increasingly fragmented due to harvesting and agricultural encroachment (Wang et al., 2001), while natural wildfires cause landscape-level changes on an annual basis (Kasischke & Bruhwiler, 2002). Grasslands in TEA have been subjected to a range of anthropogenic activities, including cropland conversion and livestock grazing, which leads to overgrazing and desertification. Such changes can significantly alter the carbon dynamics of terrestrial ecosystems. It has been suggested that a large carbon sink is located in northern Asia (Bousquet et al., 1999). While the role of forests in this Northern Hemisphere carbon sink has been documented (Schimel et al., 2001), grasslands play a significant but poorly recognized role in the global carbon cycle (Scurlock & Hall, 1998). Grassland soil carbon stocks (where the vast majority of grassland carbon is stored) have been estimated at 10–30% of global soil carbon Anderson, 1991, Eswaran et al., 1993.