A data-model fusion approach for upscaling gross ecosystem productivity to the landscape scale based on remote sensing and flux footprint modelling

摘要

In order to use the global available eddy-covariance (EC) fluxdataset and remote-sensing measurements to provide estimates ofgross primary productivity (GPP) at landscape(101–102 km2), regional(103–106 km2) and global land surface scales, wedeveloped a satellite-based GPP algorithm using LANDSAT data and anupscaling framework. The satellite-based GPP algorithm uses twoimproved vegetation indices (Enhanced Vegetation Index – EVI, LandSurface Water Index – LSWI). The upscalling framework involves fluxfootprint climatology modelling and data-model fusion. This approachwas first applied to an evergreen coniferous stand in thesubtropical monsoon climatic zone of south China. The ECmeasurements at Qian Yan Zhou tower site (26°44′48" N,115°04′13" E), which belongs to the China flux networkand the LANDSAT and MODIS imagery data for this region in 2004 wereused in this study. A consecutive series of LANDSAT-like images ofthe surface reflectance at an 8-day interval were predicted byblending the LANDSAT and MODIS images using an existing algorithm(ESTARFM: Enhanced Spatial and Temporal Adaptive Reflectance FusionModel). The seasonal dynamics of GPP were then predicted by thesatellite-based algorithm. MODIS products explained 60% of observedvariations of GPP and underestimated the measured annual GPP(= 1879 g C m?2) by 25–30%; while the satellite-basedalgorithm with default static parameters explained 88% of observedvariations of GPP but overestimated GPP during the growing seasonalby about 20–25%. The optimization of the satellite-based algorithmusing a data-model fusion technique with the assistance of EC fluxtower footprint modelling reduced the biases in daily GPPestimations from about 2.24 g C m?2 day?1(non-optimized, ~43.5% of mean measured daily value) to1.18 g C m?2 day?1 (optimized, ~22.9% of meanmeasured daily value). The remotely sensed GPP using the optimizedalgorithm can explain 92% of the seasonal variations of EC observedGPP. These results demonstrated the potential combination of thesatellite-based algorithm, flux footprint modelling and data-modelfusion for improving the accuracy of landscape/regional GPPestimation, a key component for the study of the carbon cycle.