Effects of Deep Learning Bias Correction and Carbon Neutrality on Projections of Future Population Exposure to Extreme Precipitation

Abstract

Spatiotemporal variations in extreme precipitation and their impacts on population exposure remain poorly understood due to biases in climate model projections and assumptions inherent in emission scenarios. In this study, we evaluate the bias‐correction performance of three deep learning techniques—Convolutional Neural Networks (CNN), Long Short‐Term Memory networks (LSTM), and hybrid CNN‐LSTM—using extreme precipitation indices derived from 10 Coupled Model Intercomparison Project Phase 6 (CMIP6) models. Building on this, we assess exposure for China's nine river basins under Shared Socioeconomic Pathway (SSP)1–2.6 and SSP3–7.0 in the near future (2031–2060) and far future (2071–2100) and quantify the potential benefits of achieving carbon neutrality on reducing population exposure to extreme precipitation. Our findings reveal that CNN outperforms both LSTM and CNN‐LSTM in correcting biases. Compared to the historical period (1985–2014), both R95p and Rx1day exhibit increasing trends across China at the scenario and temporal levels. The SSP3–7.0 scenario shows more pronounced increases than SSP1–2.6, and the long‐term trends (2071–2100) are greater than those observed in the short‐term period (2031–2060). Spatially, the strongest responses are observed in the southwest and southeast coastal regions, while changes in inland areas are more moderate. On the exposure side, SSP3–7.0 results in a substantial national increase, with far‐future exposure projected to reach approximately 2.7 times the near‐future levels for Rx1day and 1.2 times for R95p. Hotspot areas of exposure are found along the Yangtze–Pearl River Delta–Southeast coastal belt, the southwestern uplands, and the Beijing–Tianjin–Hebei–Bohai corridor. In contrast, the carbon‐neutral pathway (SSP1–2.6) leads to approximately 80% reductions in national exposure in the far future, with reductions greater than 70% in all nine basins. We suggest that climate extreme changes, rather than population dynamics or extreme‐population interactions, are anticipated to dominate these reductions in the future. These results provide important scientific support for ongoing efforts aimed at achieving carbon neutrality by the 2060s to reduce the potential risk of extreme precipitation in China and its nine major river basins.

Bibliographic Information