Remote sensing proxies underestimate fire-induced gross primary productivity loss and overestimate recovery in forests

Abstract

Wildfires significantly alter terrestrial carbon cycling by reducing vegetation productivity and reshaping ecosystem functioning, yet satellite-based estimates of gross primary productivity (GPP) remain highly uncertain under fire disturbance. Here, we evaluated five global GPP products—BESS GPP (process-based), FLUXCOM and FluxSat GPP (machine learning-based), GOSIF GPP (derived from reconstructed solar-induced chlorophyll fluorescence, SIF), MODIS GPP (light-use efficiency–based)—together with three complementary proxies: GOSIF (reconstructed SIF), the near-infrared reflectance of vegetation (NIRv), and leaf area index (LAI). These products were benchmarked against eddy covariance (EC) tower GPP measurements from ten fire-affected sites (five forest sites, five grass/shrub sites) with multi-year pre- and post-fire records. Results show that satellite proxies generally underestimated fire-induced GPP loss, with forest sites showing the largest discrepancy: EC GPP declined by ∼94%, compared to 47–88% from satellites. During recovery, most satellite products overestimated post-fire carbon gain and underestimated recovery time, often signaling premature recovery in forests. In contrast, grass and shrub ecosystems showed faster rebound and closer agreement with satellite estimates. Among these products, BESS GPP and GOSIF better reproduced immediate loss and recovery time, though still underestimated persistent suppression and overestimated cumulative uptake. Moreover, EC data further revealed reduced post-fire GPP sensitivity to light, temperature, and vapor pressure deficit in forests, which satellite products failed to capture. These findings highlight systematic biases in current satellite proxies, emphasize the challenges in monitoring forest recovery, and underscore the need for disturbance-responsive models and expanded flux benchmarks to improve post-fire carbon cycle assessments.