Exploring denoising diffusion probabilistic model for daily streamflow gap filling in Central Asia typical watersheds

Abstract

Study area: The Manas River Basin (84°01′−86°32′E, 42°27′−45°21′N) on the northern slopes of the Tianshan Mountains and the Kaidu River Basin (82°18′−85°57′E, 40°54′−44°30′N) on the southern slopes. Study focus: Traditional gap filling models struggle to accurately simulate daily streamflow in data-scarce regions. The underutilize of multi-source data leads to insufficient simulation ability for extremely high values of streamflow. Currently, a quantitative relationship between daily streamflow and hydro-meteorological factors has not been systematically established, despite the potential for improving accuracy of hydrological models through such relationships. To address this, a Denoising Diffusion Probabilistic Model were employed to investigate the driving factors of daily streamflow data. This study tested and compared the gap filling efficiency of mainstream models in the Tianshan Mountains region and provided a quantitative gap filling model for data-scarce areas. New hydrological insights for the region: This study has revealed a strong correlation between the missing rate and the efficiency of gap filling models. It is revealed that the DDPM model accurately simulates streamflow in the climatic conditions of the region and outperformed other models in terms of all missing rates, which holds significant implications for gaining a deeper understanding of the water balance process and enhancing hydrological and meteorological simulations in data-scarce regions.

Keywords

• Random forest
• Denoising diffusion probabilistic model
• Gap filling
• Hydrological data reconstruction
• Budyko equation