Quantitative precipitation forecasting in numerical weather prediction (NWP) models rely on physical parameterization schemes. However, these schemes involve considerable uncertainties due to limited knowledge of the mechanisms involved in the precipitating process, ultimately leading to degraded precipitation forecasting skills. To address this issue, our study proposes using a Swin-Transformer based deep learning (DL) model to quantitatively map fundamental variables solved by NWP models to precipitation maps. Our results show that the DL model effectively extracts features over meteorological variables, leading to improved precipitation skill scores of 21.7%, 60.5%, and 45.5% for light rain, moderate rain, and heavy rain, respectively, on an hourly basis. We also evaluate two case studies under different driven synoptic conditions and show promising results in estimating heavy precipitation during strong convective precipitation events. Overall, the proposed DL model can provide a vital reference for capturing precipitation-triggering mechanisms and enhancing precipitation forecasting skills. Additionally, we discuss the sensitivities of the fundamental meteorological variables used in this study, training strategies, and performance limitations.

Ziping Zuoa, Jimmy C.H. Funga,b, Zhenning Lia,*, Yiyi Huangd, Mau Fung, Wonga, Alexis K.H. Laua,c, Xingcheng Luea Division of Environment and Sustainability, The Hong Kong University of Science and Technology, Clear Water Bay, Hong Kong, Chinab Department of Mathematics, The Hong Kong University of Science and Technology, Clear Water Bay, Hong Kong, Chinac Department of Civil and Environmental Engineering, The Hong Kong University of Science and Technology, Clear Water Bay, Hong Kong, Chinad Department of Hydrology and Atmospheric Sciences, University of Arizona, Tucson, AZ, USAe Department of Geography and Resource Management, The Chinese University of Hong Kong, Shatin, Hong Kong, ChinaCorresponding author: Zhenning Li,[email protected] worldwide heatwaves have shattered temperature records in many regions. In this study, we applied a dynamical downscaling method on the high-resolution version of the Max Planck Institute Earth System Model (MPI-ESM-1-2-HR) to obtain projections of the summer thermal environments and heatwaves in the Pearl River Delta (PRD) considering three shared socioeconomic pathways (SSP1-2.6, SSP2-4.5, and SSP5-8.5) in the middle and late 21st century. Results indicated that relative to the temperatures in the 2010s, the mean increases in the summer (June–September) daytime and nighttime temperatures in the 2040s will be 0.7–0.8 °C and 0.9–1.1 °C, respectively. In the 2090s, the mean difference will be 0.5–3.1 °C and 0.7–3.4 °C, respectively. SSP1-2.6 is the only scenario in which the temperatures in the 2090s are expected to be lower than those in the 2040s. Compared with those in the 2010s, hot extremes are expected to be more frequent, intense, extensive, and longer-lasting in the future in the SSP2-4.5 and SSP5-8.5 scenarios. In the 2010s, a heatwave occurred in the PRD lasted for 6 days on average, with a mean daily maximum temperature of 34.4 °C. In the 2040s, the heatwave duration and intensity are expected to increase by 2–3 days and 0.2–0.4 °C in all three scenarios. In the 2090s, the increase in these values will be 23 days and 36.0 °C in SSP5-8.5. Moreover, a 10-year extreme high temperature in the 2010s is expected to occur at a monthly frequency from June to September in the 2090s.SIGNIFICANCE STATEMENTPearl River Delta (PRD) has been experiencing record-shattering heatwaves in recent years. This study aims to investigate the future trends of summer heatwaves in the PRD by modeling three future scenarios including a sustainable scenario, an intermediate scenario, and a worst-case scenario. Except the sustainable scenario, summer temperatures in the intermediate and worst-case scenarios will keep increasing, and heatwaves will become more frequent, intense, extensive, and longer-lasting. In the worst-case scenario, extreme heat events that occurred once in 10 years in the 2010s will shorten to once a month in the 2090s. A better understanding of heatwave trends will benefit implementing climate mitigation methods, urban planning, and improving social infrastructure.

Recent worldwide heatwaves have shattered temperature records in many regions. In this study, we applied a dynamical downscaling method on the high-resolution version of the Max Planck Institute Earth System Model (MPI-ESM-1-2-HR) to obtain projections of the summer thermal environments and heatwaves in the Pearl River Delta (PRD) considering three Shared Socioeconomic Pathways (SSP1-2.6, SSP2-4.5, and SSP5-8.5) in the middle and late 21st century. Results indicated that relative to the temperatures in the 2010s, the mean increases in the summer daytime and nighttime temperatures in the 2040s will be 0.7–0.8 °C and 0.9–1.1 °C, respectively. In the 2090s, they will be 0.5–3.1 °C and 0.7–3.4 °C, respectively. SSP1-2.6 is the only scenario in which the temperatures in the 2090s are expected to be lower than those in the 2040s. Compared with those in the 2010s, hot extremes are expected to be more frequent, more intense, more extensive, and longer-lasting in the future in the SSP2-4.5 and SSP5-8.5 scenarios. In the 2010s, a heatwave occurred in the PRD lasted for 6 days on average, with a mean daily maximum temperature of 34.4 °C. In the 2040s, the heatwave duration and intensity are expected to increase by 2–3 days and 0.2–0.4 °C in all three scenarios. In the 2090s, the increase in these values will be 23 days and 36.0 °C in SSP5-8.5. Moreover, a 10-year extreme high temperature in the 2010s is expected to occur at a monthly frequency from June to September.