All-sky hourly estimation over East Asia using Himawari-8 AHI and multi-source data: investigating the main climatic drivers of afternoon depression and intraday variability in gross primary productivity

Abstract

Intraday observations from geostationary satellites provide key information for estimating terrestrial productivity and analyzing environmental drivers, but cloud cover often hinders continuous monitoring. In this study, we addressed this limitation by combining multi-source data and a data-driven approach to develop hourly, all-sky, regional-scale gross primary productivity (GPP). Our all-sky GPP showed strong consistency with ground measurements in East Asia (coefficient of determination (R2) = 0.86, root mean squared error = 2.4 mu mol CO2/m(2)/s) and outperformed conventional hourly GPP products derived from the observations of International Space Station (ISS) sensors and polar-orbiting satellites. To investigate how the importance of input variables differs between clear-sky and cloudy-sky conditions, we applied Shapely Additive exPlanation (SHAP) analysis. Notably, the Himawari-8-derived features contributed most in both clear- and cloudy-sky models. Under heat stress in clear-sky conditions, water-content features exhibited increasing impact compared to normal conditions, while latent heat flux demonstrated high contribution beneath the clouds. By capturing these regime shifts in feature importance, our all-sky GPP effectively captured the widespread afternoon depression in summer across East Asia. Regional analysis by land cover, using the Pearson correlation coefficient (r), revealed that vapor pressure deficit drove afternoon depression under clear skies (r = -0.519), whereas surface latent heat flux was the primary driver under cloudy conditions (r = -0.785). These findings highlight the synergistic use of high-frequency geostationary observations and the detailed spatial information from polar-orbiting satellites, which enhances our understanding of the shifting environmental drivers of GPP across regimes.