DBN-LightGBM hybrid spatial downscaling of GSMaP based on the adaptive lag effect of vegetation

Résumé fourni par la source

The spatial resolution limitations of satellite-based precipitation products, such as GSMaP, hinder their application in localized disaster warnings and extreme weather monitoring within basin regions. Traditional studies employ downscaling methods to improve precipitation data resolution, the lagged effect of vegetation to precipitation remains a critical constraint. Currently, there is still a lack of systematic analysis on the stable hysteresis characteristics of vegetation in response to precipitation. Therefore, this paper proposes an Adaptive Lag Enhanced Vegetation Index (ALEVI) that incorporates stable response characteristics, enabling a more precise depiction of the lagged response of vegetation to precipitation. With ALEVI, a more accurate and reasonable influencing factor set (ALEVI-IFS) is established to explore the impact of lag effects on downscaling performance. Building on this, we develop a novel downscaling model that links precipitation with ALEVI-IFS through a combined Deep Belief Network (DBN) and LightGBM approach, validated with ground observation data. Case analyses demonstrate that ALEVI significantly improves the downscaling model’s precision by effectively characterizing vegetation-precipitation interactions. Moreover, the DBN-LightGBM model shows substantial enhancements over traditional methods, yielding a more accurate spatial representation of local precipitation patterns. This framework offers robust data support for hydrological simulations and climate change analyses, advancing localized predictive capability.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.