How well do recently reconstructed solar-induced fluorescence datasets model gross primary productivity?

Abstract

The collection of various long-term reconstructed solar-induced fluorescence (SIF) datasets derived at a range of spatio-temporal scales provides new opportunities for modelling vegetation dynamics, in particular, gross primary productivity (GPP). Information about the proximity of the reconstructed SIF (SIFr) datasets to GPP across land cover types and climatic conditions provides important support for a better application of these products for modelling applications. We conducted a multiscale analysis of four different long-term (12 years, 2007–2018) high-resolution global SIFr datasets (0.05° × 0.05°), namely – CSIF (Contiguous SIF), GOSIF (Global OCO-2 SIF), LUE-SIF (Light Use Efficiency SIF), and HSIF (Harmonized SIF) - at 4-day, 8-day, and monthly time scales and found that for the majority of sites, the SIFr is linearly related to ground-based GPP measurements with the eddy covariance method. While the relationship between SIFr and GPP (i.e., the slope - GPP/SIFr) varied significantly across the SIFr datasets, sites, and land cover types, all four SIFr datasets were unequivocally a better predictor of GPP compared to remotely sensed vegetation indices – NDVI (normalized difference vegetation index) and EVI (enhanced vegetation index), sensed by the MODIS satellite. Furthermore, we also analyzed SIF-GPP relationships during drought vs non-drought conditions and found that for about 30% of the sites, comprising mostly non-forests site, the SIF-GPP relationship became weaker (decreased R2) with a lower slope during drought conditions compared to non-drought conditions. Among the four different products, the CSIF (at 4-day timescale) and GOSIF (at 8-day timescale) predicted GPP better compared to LUE-SIF and HSIF across all land cover types. Owing to their long-term availability (since 2000 for CSIF and GOSIF), these SIFr datasets combined with proxies of ecosystem properties can be used to appropriately capture vegetation dynamics and the interannual variabilities across a wide range of climatic conditions.