Improved net ecosystem exchange (NEE) modelling under drought conditions in the Argentine Pampas

The terrestrial biosphere is responsible for most CO₂ exchanges between land surfaces and the atmosphere, with ecosystems functioning as either carbon sinks or sources. These exchanges are primarily controlled by photosynthesis and ecosystem respiration, which depend on vegetation traits, environmental drivers, and soil water availability. Under drought conditions, plants tend to reduce stomatal conductance to conserve water, decreasing photosynthetic efficiency and limiting atmospheric CO₂ uptake.

In Argentina, observational studies of ecosystem CO₂ fluxes remain scarce, partly due to the high costs of instrumentation. Models such as the Vegetation Photosynthesis and Respiration Model (VPRM) provide an alternative approach to estimate ecosystem–atmosphere carbon exchange using meteorological forcing and satellite-derived vegetation indices. Recent developments include a modified VPRM formulation that explicitly accounts for water availability effects on respiration (Gourdji et al., 2022), which may improve model skill during drought. Additionally, high-resolution satellite observations have been demonstrated to more accurately represent heterogeneous agricultural landscapes, such as the crop mosaics characteristic of central Argentina.

In this work, we assess the ability of a modified VPRM driven by high-resolution satellite data to reproduce net ecosystem exchange (NEE) under contrasting hydroclimatic conditions. We use eddy-covariance observations from three sites (two grasslands and one under crop rotation cycles) in the Argentine Pampas. For each site, information on vegetation conditions in the vicinity of the flux tower was extracted from MODIS Terra and Sentinel-2 images. Time series of the Enhanced Vegetation Index (EVI) and the Land Surface Water Index (LSWI) were derived. NEE was simulated using both the original and the modified VPRM forced by each satellite data source, evaluating all model–satellite combinations. Drought conditions were characterized using the Standardized Precipitation–Evapotranspiration Index (SPEI) computed from CRU TS v4 gridded data at the nearest grid cells. Based on SPEI thresholds, the observational period was classified into “normal”, “mild drought”, and “moderate-to-severe drought”. Also, model performance statistics were computed for each regime.

Across sites, the configuration combining the modified VPRM with Sentinel-2 inputs (VPRMnew_S2) achieved improved skill (R² = 0.49, 0.24, 0.65) compared with the original VPRM driven by MODIS imagery (R² = 0.43, 0.23, 0.52). For the grassland sites, VPRMnew_S2 consistently outperformed the other configurations across all moisture regimes (higher R², lower RMSE, and near-zero bias). At the cropland site, VPRMnew_S2 showed similar skill to the original model in terms of R² and RMSE, but substantially reduced bias under water-limited conditions. These findings suggest that high-resolution satellite indices, coupled with drought-sensitive parameterizations, better capture NEE responses to water stress in the Argentine Pampas. Improved modelling of drought impacts on CO₂ exchange is essential to reduce uncertainty in regional carbon budgets and to assess ecosystem vulnerability under increasing drought frequency.

 

Keywords: Net Ecosystem Exchange, Eddy Covariance, MODIS, Sentinel-2, VPRM (Vegetation Photosynthesis and Respiration Model)

Acknowledgements

This research was financed by the UBACyT 2020–2025 program (N° 20020190100128BA) and by PIP-CONICET (N° 11220200100794CO) grants. Rodrigo Merino is supported by a scholarship granted by CONICET.