Algorithms for spatial scaling of net primary productivity using subpixel information

摘要

Spatial scaling is of particular importance in remote sensing applications to terrestrial ecosystems where spatial heterogeneity is the norm. Surface parameters derived at different resolutions can be considerably different even though they are derived using the same algorithms or models. This article addresses issues related to spatial scaling of net primary productivity (NPP). The main objective is to develop algorithms for spatial scaling of NPP using subpixel information. NPP calculations at 30 m and 1km resolutions were performed using the Boreal Ecosystem Productivity Simulator (BEPS). The area of interest is near Fraserdale, Ontario. It is found from this investigation that lumped (coarse resolution) calculations can be considerably biased (up to 64 %) from distributed (fine resolution) case, suggesting that global and regional NPP maps can be biased by the same amount if surface heterogeneity within the mapping resolution is ignored. The bias is negative when conifer-labeled pixels contain considerable deciduous forests. Due to relatively high and variable NPP values of open land areas with growing grasses, the bias is negative when deciduous-labeled pixels are mixed with open land. There is no trend between the biasness and open land fractions within conifer-labeled pixels. Based on these results, algorithms for removing these biases in lumped NPP are developed using subpixel land cover information.