2025-08-07 中国科学院(CAS)
＜関連情報＞
- https://english.cas.cn/newsroom/cas_media/202508/t20250807_1049806.shtml
- https://www.sciencedirect.com/science/article/pii/S016980952500479X
チリアン山脈のアルプス寒冷地域における高精度格子降水データセットの再構築：ダウンスケーリングから校正までの知能型技術フレームワーク Reconstruction of high-precision gridded precipitation dataset in the alpine cold regions of the Qilian Mountains: An intelligent technological framework from downscaling to calibration
Renjun Wang, Xiang Qin, Yushuo Liu, Jianyu Zhao, Rui Zhang, Zizhen Jin, Yanzhao Li, Wentao Du, Jizu Chen, Weijun Sun
Atmospheric Research Available online: 22 July 2025
DOI:https://doi.org/10.1016/j.atmosres.2025.108387
Highlights
- A novel framework (TLIDC) was developed for high-accuracy precipitation products in alpine cold regions.
- TLIDC-generated data outperforms three widely used precipitation datasets in accuracy.
- TLIDC reconstructed daily precipitation for Qilian Mountains (1950-2024) at 0.01° resolution.
Abstract
Accurate precipitation data play a vital role in hydrological and climate studies, with their significance being especially pronounced in alpine cold regions where in-situ observational data are limited. However, existing gridded precipitation datasets often suffer from low resolution and significant biases, making them inadequate for addressing the strong spatiotemporal heterogeneity of alpine areas. To address these challenges, this study developed a novel Three-Layer Intelligent Downscaling and Calibration (TLIDC) framework, integrating Geographically Weighted Regression (GWR) and Convolutional Neural Network–Bidirectional Long Short-Term Memory (CNN-BiLSTM) model, to generate high-precision gridded precipitation data tailored for alpine regions.
The TLIDC framework was quantitatively evaluated using data from 100 rain gauge stations in the Qilian Mountains and applied to reconstruct daily precipitation data at a 0.01° × 0.01° spatial resolution for the Qilian Mountains from 1950 to 2024. The results demonstrate that: (1) The TLIDC framework effectively downscales the coarse spatial resolution ERA5-Land precipitation data, producing high spatial resolution outputs that preserve the temporal periodicity and overall spatial distribution, while markedly enhancing spatial detail and visual clarity. (2) The calibration module of the TLIDC framework effectively corrected the bias in the raw precipitation data, significantly improving data performance, particularly in areas with scarce ground observation data. Compared to CHM_PRE, CN05.1, and TRMM, the generated data showed a 15.95 % ∼ 25.20 % improvement in precipitation event identification accuracy. Furthermore, the Pearson correlation coefficient (CC) for precipitation simulation accuracy increased by 0.30–0.55, while the root mean square error (RMSE) and mean absolute error (MAE) decreased by 3.33–4.58 mm/day and 1.42–2.27 mm/day, respectively. (3) The high-precision precipitation dataset for the Qilian Mountains, reconstructed using the TLIDC framework, has a multi-year average of 296.84 mm/year for the period 1999–2019. This value is close to the multi-year averages of three other precipitation products, which range from 296.43 to 352.47 mm/year. Additionally, the spatial distribution pattern of this dataset aligns with those of the other products. (4) From 1950 to 2024, precipitation in the Qilian Mountains has increased at a linear rate of 2.49 mm per decade (p < 0.05), exhibiting a spatial pattern of decreasing precipitation from southeast to northwest. Our findings offer a viable solution for generating high-precision precipitation data in alpine cold regions with complex topography and sparse observational networks, addressing a critical gap in current climate and hydrological research.
