Resolving sub-daily variability in the GPP–SIF relationship and its implications for GPP estimation at a temperate savanna site

Savannas cover approximately 20% of the global land surface and contribute nearly 30% of terrestrial net primary production, making them an important component of the global carbon cycle. Solar-induced chlorophyll fluorescence (SIF), which is closely linked to photosynthetic activity, has shown considerable potential for estimating ecosystem gross primary productivity (GPP). However, temporal variability in the GPP–SIF relationship, particularly at sub-daily timescales, remains insufficiently understood, limiting the accurate estimation of high-frequency GPP dynamics. Here, we used synchronous observations of GPP, SIF, and meteorological variables collected between 2020 and 2024 at a temperate elm savanna site in northern China to characterize the temporal variability of the GPP–SIF relationship across sub-daily and seasonal scales, quantify its environmental dependence, and evaluate an environmentally informed two-stage stacking approach for GPP estimation. The results revealed significant sub-daily variability in the GPP–SIF relationship, with regression slopes ranging from 2.32 to 5.32 and correlation coefficients exhibiting two daily peaks at 09:00 (r = 0.64) and 15:00 (r = 0.62). Meteorological drivers exerted different temporal effects on GPP and SIF across both sub-daily and seasonal timescales. Incorporating these environmental dependencies improved GPP estimation, with the proposed two-stage stacking model explaining 72% of the half-hourly GPP variability (R 2 = 0.72, RMSE = 2.53, MAE = 1.91, RPD = 1.82). Including SIF further improved the Stacked model, increasing R 2 by 0.05. Overall, this study provides a site-level evaluation of temporal GPP–SIF coupling and demonstrates the value of accounting for time-varying environmental dependence, with SIF providing an additional contribution to high-frequency GPP estimation in temperate savannas.