An improved approach for remotely sensing water stress impacts on forest C uptake.

Given that forests represent the primary terrestrial sink for atmospheric CO2 , projections of future carbon (C) storage hinge on forest responses to climate variation. Models of gross primary production (GPP) responses to water stress are commonly based on remotely sensed changes in canopy 'greenness' (e.g., normalized difference vegetation index; NDVI). However, many forests have low spectral sensitivity to water stress (SSWS) - defined here as drought-induced decline in GPP without a change in greenness. Current satellite-derived estimates of GPP use a vapor pressure deficit (VPD) scalar to account for the low SWSS of forests, but fail to capture their responses to water stress. Our objectives were to characterize differences in SSWS among forested and nonforested ecosystems, and to develop an improved framework for predicting the impacts of water stress on GPP in forests with low SSWS. First, we paired two independent drought indices with NDVI data for the conterminous US from 2000 to 2011, and examined the relationship between water stress and NDVI. We found that forests had lower SSWS than nonforests regardless of drought index or duration. We then compared satellite-derived estimates of GPP with eddy-covariance observations of GPP in two deciduous broadleaf forests with low SSWS: the Missouri Ozark (MO) and Morgan Monroe State Forest (MMSF) AmeriFlux sites. Model estimates of GPP that used VPD scalars were poorly correlated with observations of GPP at MO (r(2) = 0.09) and MMSF (r(2) = 0.38). When we included the NDVI responses to water stress of adjacent ecosystems with high SSWS into a model based solely on temperature and greenness, we substantially improved predictions of GPP at MO (r(2) = 0.83) and for a severe drought year at the MMSF (r(2) = 0.82). Collectively, our results suggest that large-scale estimates of GPP that capture variation in SSWS among ecosystems could improve predictions of C uptake by forests under drought.