Using enhanced vegetation index and land surface temperature to reconstruct the solar-induced chlorophyll fluorescence of forests and grasslands across latitude and phenology

Abstract

IntroductionForest and grassland are the two main carbon-collecting terrestrial ecosystems, and detecting their solar-induced chlorophyll fluorescence (SIF) enables evaluation of their photosynthetic intensity and carbon-collecting capacity. Since SIF that is retrieved directly from satellite observations suffers from low spatial resolution, discontinuity, or low temporal resolution, some vegetation indexes (VIs) and meteorological factors are used as predictors to reconstruct SIF products. Yet, unlike VIs, certain meteorological factors feature a relatively low space resolution and their observations are not always accessible. This study aimed to explore the potential of reconstructing SIF from fewer predictors whose high-resolution observations are easily accessible.MethodsA total of six forest and grassland regions across low, mid, and high latitudes were selected, and the commonly used predictors-normalized difference vegetation index (NDVI), enhanced vegetation index (EVI), and land surface temperature (LST)—were compared for their correlation with SIF. Results show that the combination of EVI and LST is more strongly correlated with SIF, but each contributed differently to SIF at differing growth stages of forest and grassland. Accordingly, we proposed the idea of a combined sampling approach that considers both location and phenological phase, to explore the extent to which time and space coverage samples' span could enlarge the disparity of EVI data in particular regions at specific growth stages. To do that, three kinds of sample combination methods were proposed: monthly regression at a global scale, seasonal regression at a regional scale, and monthly regression at a regional scale. Following this, Sentinel-3 EVI and MODIS LST data were used to reconstruct 500 m SIF in the six regions by implementing the proposed methodology.Results and discussionThese results showed that the R2 values were ≥0.90 between the reconstructed SIF and MODIS GPP (gross primary productivity), 0.70 with GOME-2 SIF and 0.77 with GOSIF, thus proving the proposed methodology could produce reliable results for reconstruction of 500 m SIF. This proposed approach, which bypasses dependence of traditional SIF reconstruction model on numerous predictors not easy to obtain, can serve as a better option for more efficient and accurate high-resolution SIF reconstructions in the future.