【摘要】 降水融合技术通过将多种降水数据整合在一起，以获取更准确和更可靠的降水信息。然而，这些降水数据源通常本身存在系统偏差，同时降水表现出显著的时空异质性。为应对这些问题，本文提出了一种偏差校正和降水时空融合相结合的两阶段降水融合方法。在偏差校正阶段，采用经验累积分布函数匹配法（ECDF）对降水产品进行偏差校正；在时空融合阶段，先通过动态约束线性回归模型（DCLR）获取时空权重，随后对偏差校正后的降水产品进行加权平均。将本文提出的降水融合方法称为ECDF＿DCLR。此外，在时空融合阶段分别使用动态贝叶斯模型平均（DBMA）和简单模型平均（SMA）与偏差校正阶段的ECDF结合，形成对照降水融合方法 ECDF＿DBMA和ECDF＿SMA，以验证ECDF＿DCLR的有效性。将ECDF＿DCLR、ECDF＿DBMA和ECDF＿SMA分别应用于西南地区2005—2017年卫星降水产品IMERG和再分析降水产品ERA5-Land的融合，以地面气象站降水数据作为基准评估融合结果。结果表明：(1) ECDF能有效降低IMERG和ERA5-Land的系统偏差，同时提高它们的精度，两者RB绝对值减小的比例分别为95.5%和99.6%,KGE增加的比例分别为12.7%和41.5%。ECDF还可以增强ERA5-Land的探测降水事件能力（ERA5-Land的CSI增加了7.8%），但对IMERG的影响不大（IMERG的CSI保持0.53不变）。此外，在融合前进行偏差校正是必要的，偏差校正与时空融合相结合生成的降水产品的KGE和CSI分别平均高出只时空融合生成的降水产品的KGE和CSI的11.5%和3.1%。(2) DCLR、SMA和SMA均能有效融合降水产品，3种方法中DCLR降水融合的精度最好，它们提高降水事件探测能力的差异不大。在不同时间尺度、空间尺度和不同海拔等级上，3种融合降水产品的KGE和CSI大都大于或接近最佳数据源的KGE和CSI，它们当中通过DCLR融合的降水产品具有最高的KGE，但它们CSI的差距基本不超过0.01。与地理加权回归相比和带漂移项的克里金相比，ECDF＿DCLR的大部分指标表现更好，其KGE和CSI分别至少高出前者的4.3%和1.8%。总之，ECDF与DCLR相结合的降水融合方法能够为西南地区提供更准确的降水数据，给多源降水数据融合领域提供研究思路。更多还原

【Abstract】 Precipitation merging technology integrates muliple precipitation datasets to obtain more accurate and reliable precipitation information. However, these data sources have inherent systematic biases and precipitation exhibits spatiotemporal heterogeneity. To address these issues, this paper proposed a two-stage precipitation merging method combining bias correction and precipitation spatiotemporal fusion. In the first stage, the biases in precipitation products are corrected by the Experience Cumulative Distribution Function matching method(ECDF). In the second stage, a Dynamic Constrained Linear Regression model(DCLR) is used to determine spatiotemporal weights, followed by weighted averaging of the the bias-corrected precipitation products. The proposed method is termed as ECDF＿DCLR. In addition, the Dynamic Bayesian Model Average(DBMA) and Simple Model Average(SMA) are used in the second stage along with ECDF, forming the contrasting methods ECDF＿DBMA and ECDF＿SMA, to verify the effectiveness of ECDF＿DCLR. ECDF＿DCLR, ECDF＿DBMA and ECDF＿SMA were applied to integrate satellite precipitation product IMERG and reanalysis precipitation product ERA5-Land in Southwest China from 2005 to 2017, using precipitation data from ground meteorological stations as the reference for evaluation. Results show that:(1) ECDF can effectively reduce the systematic bias in IMERG and ERA5-Land while improving their accuracy, with the absolute values of RB decreasing by 95.5%and 99.6%, and KGE increasing by 12.7% and 41.5%, respectively. ECDF also enhances the precipitation event detection capability of ERA5-Land(The CSI of ERA5-Land increases by 7.8%), but has a minimal impact on IMERG(The CSI of IMERG remains unchanged at 0.53). Additionally, it is necessary to perform bias correction before fusion, as the KGEs and CSIs of precipitation products generated by combining bias correction and spatiotemporal fusion are, on average, 11.5% and 3.1% higher than those generated by spatiotemporal fusion alone, respectively.(2) DCLR, SMA, and DBMA all effectively integrate precipitation products. Among the three methods, DCLR has the best accuracy of precipitation fusion. There is little difference between them in improving the detection ability of precipitation events. At different time scales, spatial scale, and different altitude grades,the KGEs and CSIs of three fusion precipitation products are mostly greater than or close to the KGE and CSI of the best data source. Among fusion precipitation products, the precipitation product fused by DCLR has the highest KGE. While the differences in CSIs between fusion precipitation products do no exceed 0.01. Compared to Geographically Weighted Regression and Kriging with External Drift, the most metrics of ECDF＿DCLR perform better, with KGE and CSI at least 4.3% and 1.8% higher than the former, respectively. In short, the precipitation merging method combined with ECDF and DCLR can provide more accurate precipitation data for Southwest China and offer new insights into multi-source precipitation data merging research.更多还原