Agricultural landscape spatial heterogeneity analysis and optimal scale selection: An example applied to Sanjiang plain

Agricultural monitoring requires high temporal frequency data which are currently provided only by moderate spatial resolution sensors. At such moderate spatial resolutions, farmland that is heterogeneous within a pixel will be averaged and hence obscured. This would bias any non-linear estimation of crop growing processes (e. g., net primary productivity (NPP), leaf area index (LAI)). To modify this bias, a first approach is used to explicitly take into account the intra-pixel spatial heterogeneity in the retrieval algorithm. A second approach is to use the surface heterogeneity to disaggregate moderate spatial resolution estimates of land surface variable at a proper scale of spatial variation. Both approaches are required to quantify spatial heterogeneity, and a proper scale selection should be necessary for agricultural monitoring. To this ends, four typical landscape pattern sites in the Jiansanjiang Reclamation Area which is an important basin of commercial grain production in China, were selected and Landsat/TM NDVI image data were analyzed in this study. Based on the variogram analysis, some conclusions can be drawn. (1) Directional experiment variograms analysis can make clear how the human activates and natural factors affect the agricultural spatial heterogeneity qualitatively. For example, dry lands (including the landscape only with dry land and the landscape which is mosaic of dry land and paddy fields in this study) have the largest heterogeneity in North-South direction, while the landscape pattern which only have paddy fields have the largest heterogeneity in East-West direction. Based on this, we can demonstrate that spatial heterogeneity caused by human and natural factors can be examined deeply through variogram analysis. (2) The fitted variograms can present how different landscape patterns have their own spatial heterogeneity quantificationally. In this study, for example, the same type of land use can have lower heterogeneity as different types of land use landscape patterns have larger heterogeneity. (3) Through the variogram analysis of heterogeneity, a method used to select a proper scale (pixel size) for agricultural remote sensing monitoring is discussed.