Assessing gross primary production dynamics in differently managed tallgrass prairies using Eddy covariance and satellite-derived estimates.

Accurately quantifying gross primary production (GPP) is crucial for understanding grassland carbon dynamics. We examined spatiotemporal variations in eddy covariance-derived GPP (GPPEC) across six native tallgrass prairie pastures (∼32 to 64 ha each) in central Oklahoma and compared them against GPP estimates from the global simulation of the Vegetation Photosynthesis Model (GPPVPM_global) and the site-level simulation of the Vegetation Photosynthesis and Ecosystem Respiration Model (GPPVPERM_site). These sites encompass varied burning frequencies (annual vs. 4-5-year rotation), grazing intensities (light rotational vs. intensive), and a hayed pasture, as well as diverse growing conditions, enabling us to thoroughly evaluate the models' capability to estimate GPP in tallgrass prairie. We observed a strong seasonal pattern in GPPEC, though peak timing and magnitude varied greatly across pastures and years due to differences in management and weather. Variations in GPPEC-enhanced vegetation index (EVI) linear regression slopes (16.78 to 25.82) among sites reflected management impacts on carbon uptake. Linear regression, stepwise ordinary least squares, and linear mixed-effects models consistently identified EVI as the dominant predictor of GPPEC, while air temperature, solar radiation, and rainfall added little explanatory power. Narrow marginal (0.74-0.79) and conditional (0.76-0.79) R2 ranges indicated minor spatial and interannual effects, suggesting a robust GPPEC-EVI relationship across sites and years. Both GPPVPM_global and GPPVPERM_site closely tracked GPPEC dynamics (R2 = 0.72-0.85, P < 0.01). The seasonal peaks of GPPVPM_global and GPPVPERM_site matched those of GPPEC in lightly grazed pastures (0-2% annual overestimation at Pasture 14), but models substantially overpredicted annual cumulative fluxes (15-30%) in intensively grazed and hayed pastures (RMSE = 2.08-2.66 g C m-2 d-1). These results underscore the general robustness of light use efficiency models while highlighting the need to incorporate dynamic C₃:C₄ functional composition and management disturbances to improve model accuracy in mixed species grasslands.