Spatially Explicit Estimation of Optimal Light Use Efficiency for Improved Satellite Data Driven Ecosystem Productivity Modeling

Abstract:

Remote sensing based light use efficiency (LUE) models, including the MODIS (MODerate resolution Imaging Spectroradiometer) MOD17 algorithm are commonly used for regional estimation and monitoring of vegetation gross primary production (GPP) and photosynthetic carbon (CO2) uptake. A common model assumption is that plants in a biome matrix operate at their photosynthetic capacity under optimal climatic conditions. A prescribed biome maximum light use efficiency parameter defines the maximum photosynthetic carbon conversion rate under prevailing climate conditions and is a large source of model uncertainty. Here, we used tower (FLUXNET) eddy covariance measurement based carbon flux data for estimating optimal LUE (LUEopt) over a North American domain. LUE_opt was first estimated using tower observed daily carbon fluxes, meteorology and satellite (MODIS) observed fraction of photosynthetically active radiation (FPAR). LUE_opt was then spatially interpolated over the domain using empirical models derived from independent geospatial data including global plant traits, surface soil moisture, terrain aspect, land cover type and percent tree cover. The derived LUE_opt maps were then used as primary inputs to the MOD17 LUE algorithm for regional GPP estimation; these results were evaluated against tower observations and alternate MOD17 GPP estimates determined using Biome-specific LUE_opt constants. Estimated LUE_opt shows large spatial variability within and among different land cover classes indicated from a sparse North American tower network. Leaf nitrogen content and soil moisture are two important factors explaining LUE_opt spatial variability. GPP estimated from spatially explicit LUE_opt inputs shows significantly improved model accuracy against independent tower observations (R² = 0.76; Mean RMSE < 257 g C m^-2 yr^-1) relative to GPP modeled using biome-specific LUE_opt constants (R² = 34; RMSE = 439 g C m^-2 yr^-1). We show that general landscape and plant trait information can explain spatial heterogeneity in LUEopt, leading to improved GPP estimates from satellite based LUE models.