Accurate estimation of terrestrial gross primary production (GPP) in ecologically and topographically complex
regions remains a major challenge. This study evaluated four global GPP products—BESS, GOSIF, MOD17, and
VPM—against eddy covariance (EC) observations from 11 flux towers in Southwest China. The region spans
diverse vegetation types and climatic gradients, making it an ideal testbed for satellite-based GPP assessment.
Site-level evaluation showed that GOSIF had the strongest correlation with GPP-EC (R2 > 0.61) but consistently
overestimated GPP, especially during summer. BESS achieved the lowest RMSE (2.02 g C/m2/d) and better
captured interannual variation, including drought impacts. MOD17 and VPM tended to underestimate summer
peaks, particularly in evergreen broadleaf forests. Seasonal analysis revealed that all four products performed
better in the non-growing season, with BESS showing the highest growing-season accuracy (R2 = 0.39, RMSE =
2.44 g C/m2/d). GOSIF notably overestimated seasonal totals, especially in shrubland and grassland ecosystems.
Spatially, mean annual GPP estimates varied by over 30 % among products, with GOSIF being the highest
(1497.32 g C/m2) and BESS being the lowest (1109.73 g C/m2). The vegetation-type analysis revealed the
highest level of agreement in deciduous broadleaf forests, while the greatest discrepancies were observed in open
shrublands. Our findings highlight the strengths and limitations of current GPP products in a mountainous,
heterogeneous region. Incorporating physiological indicators such as solar-induced chlorophyll fluorescence
(SIF), improving model parameterization of phenology, and accounting for local environmental stressors are
essential for enhancing GPP estimation in Southwest China and similar landscapes.

Gross primary production (GPP) Breathing earth system simulator (BESS) Solar-induced chlorophyll fluorescence (SIF) Moderate resolution imaging spectroradiometer (MODIS) Vegetation photosynthesis model (VPM)