Cuina Li

In this study, a machine learning-based quality control algorithm for heavy rainfall was developed by integrating automatic weather station observations with remote sensing data, minute-level data, and metadata. Based on heavy rainfall samples from 1 June 2022 to 31 December 2024, the performances of four gradient boosting models—eXtreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM), Categorical Boosting (CatBoost), and Gradient Boosted Regression Trees (GBRT)—significantly outperformed precipitation-threshold-based conventional methods, including regional extreme value checks, temporal consistency checks, and others. Specifically, the XGBoost in particular achieves an increase in precision by 0.110 and recall by 0.162. This translates to a substantial reduction in both false alarms (higher precision) and missed detections (higher recall) of anomalous heavy rainfall events, thereby significantly enhancing the reliability of the quality-controlled data. The radar composite reflectivity, satellite cloud-top temperature, and minute-level precipitation were identified as dominant contributors to model predictions. The integration of multi-sensor observations effectively addressed limitations inherent in conventional threshold-based approaches. Through SHapley Additive exPlanations (SHAP)-based interpretability analysis, the model’s decision logic was shown to align with meteorological physical principles. Characteristic patterns such as combinations of low radar reflectivity and elevated cloud-top temperatures were flagged as anomalous rainfall events, typically corresponding to manual operational errors. Moreover, the model identified anomalous minute-level precipitation extremes to be critical signals for detecting instrument malfunctions, data encoding and transmission errors. The physical consistency of the model’s reasoning enhances its trustworthiness and supports its potential for operational implementation in heavy rainfall quality control.

In this study, a machine learning-based quality control algorithm for heavy rainfall was developed by integrating automatic weather station observations with remote sensing data, minute-level data, and metadata. Based on heavy rainfall samples from 1 June 2022 to 31 December 2024, the performances of four gradient boosting models (XGBoost, LightGBM, CatBoost, and GBRT) significantly outperformed conventional method, with XGBoost in particular achieving an increase in precision by 0.110, recall by 0.162, and F1-score by 0.140. This performance gain is attributed to the models’ ability to effectively learn nonlinear features from complex multi-source data, thereby reducing both false alarms and missed detections of anomalous rainfall events. The radar composite reflectivity, satellite cloud-top temperature, and minute-level precipitation were identified as dominant contributors to model predictions. The integration of multi-sensor observations effectively addressed limitations inherent in conventional threshold-based approaches. Through SHAP-based interpretability analysis, the model’s decision logic was shown to align with meteorological physical principles. Characteristic patterns such as combinations of low radar reflectivity and elevated cloud-top temperatures were flagged as anomalous rainfall events, typically corresponding to manual operational errors. Moreover, the model identified anomalous minute-level precipitation extremes to be critical signals for detecting instrument malfunctions, data encoding and transmission errors. The physical consistency of the model’s reasoning enhances its trustworthiness and supports its potential for operational implementation in heavy rainfall quality control.

The online coupling of aerosols and clouds and its effect on surface global horizontal irradiance (GHI) has not yet been thoroughly investigated in the Weather Research and Forecasting Model with Solar extensions (WRF-Solar), despite its potential significance for solar energy applications. This study addresses this critical gap by implementing a computationally efficient, coupled aerosol–cloud scheme and evaluating its impacts on GHI predictability. Simulations with online aerosol–cloud coupling are systematically compared to uncoupled simulations during March 2021, a period marked by two distinct pollution episodes over north China. The online coupling enhances aerosol optical depth (AOD) simulations, increasing the correlation coefficient from 0.19 to 0.51 while reducing the absolute bias from 0.54 to 0.48 and root mean square error from 0.82 to 0.72, compared to uncoupled simulations. Enhanced cloud microphysics (droplet concentration, water path) yields better cloud optical depth estimates, reducing all-sky GHI bias by 14.5% (63.5 W/m² for the uncoupled scenario and 54.3 W/m² for the coupled scenario) through improved aerosol–cloud–meteorology interactions. Notably, the simultaneous spatiotemporal improvement of both AOD and GHI suggests enhanced internal consistency in aerosol–cloud–radiation interactions, which is crucial for operational solar irradiance forecasting in pollution-prone regions. The results also highlight the practical value of incorporating online aerosol coupling in solar forecasting models.

Although the Weather Research and Forecasting model with solar extensions (WRF-Solar) is tailed for solar energy applications, its official version lacks the consideration of the online aerosol-radiation process. To overcome this limitation, we have coupled the aerosol module online with the radiation module, then assimilated the high-resolution aerosol optical depth (AOD) from the Himawari-8 next-generation geostationary satellite using a three-dimensional variational (3DVAR) AOD data assimilation system to optimize the irradiance predictions with the better aerosol–radiation interaction. The results show that data assimilation can significantly eliminate the AOD underestimations and reasonably reproduce the AOD temporal distributions, improving 51.63% for biases and 61.29% for correlation coefficients. Compared with the original WRF-Solar version, coupled online with an advanced aerosol module minifies the bias value of global horizontal irradiance (GHI) up to 44.52%, and AOD data assimilation contributes to a further reduction of 17.43%.