Canopy color indices for estimating gross primary production: insights from a savanna ecosystem in the Dry-Hot Valley of Southwest China

Zhou, Ruiwu

FRONTIERS IN FORESTS AND GLOBAL CHANGE

2026

9

Zhou, RW;Song, QH; Zhang, YP

Introduction The deployment of digital cameras on flux towers offers a practical approach to exploring how seasonal canopy photosynthesis is related to canopy color indices (CCIs). However, the pronounced spatial and temporal heterogeneity of savanna ecosystems remains a major challenge.Methods We compared EC-derived gross primary production (GPP) with six LUE model variants, including two direct-FPAR models, three CCI-proxy models, and one MODIS NDVI-proxy model. This framework allowed us to distinguish direct FPAR inputs from vegetation-index proxies for the FPAR term in LUE-based GPP estimation.Results During the extreme high-temperature drought of 2019, the peak values of Sgreen, Hue, and GEI decreased by approximately 50%. Annual GPP declined from 850 to 567 g C m-2, leaf area index (LAI) decreased from 0.989 to 0.670 m2 m-2, and LUE decreased from 0.40 to 0.25 g C MJ-1. The seasonal pattern of GPP remained broadly consistent, although the length of the growing season (LOS) varied substantially, ranging from 213 to 111 d in 2016 and 2019, respectively. During the growing season (GS), the CCIs were closely synchronized with daily GPP and LAI; during the non-growing season (NGS), a lag of approximately 20 days was observed. Among the six LUE variants, the MODIS NDVI-proxy model performed best (R2 = 0.77, slope = 1.19), whereas the MODIS FPAR model showed moderate performance and underestimated GPP (R2 = 0.61, slope = 0.66). The tower-derived FPAR model overestimated GPP, and the CCI-proxy models performed moderately, especially with lower accuracy during the non-growing season.Discussion Overall, digital camera-derived CCIs can be used as empirical FPAR-like proxies for estimating seasonal GPP dynamics in savanna ecosystems, but they should not be interpreted as physical FPAR measurements. The comparison with tower-derived FPAR and MODIS FPAR highlights the importance of explicitly distinguishing direct FPAR inputs from vegetation-index proxies in LUE-based GPP modeling.