What question did this study set out to answer?

This research focuses on developing a forecasting system for extreme weather events in Northeast China, driven by climate change.

June 6, 2026Open Access

Using remote sensing radiation and meteorological data to assess climate change: prediction of extreme weather events in Northeast China

Key Points

This research focuses on developing a forecasting system for extreme weather events in Northeast China, driven by climate change.
Integrated multi-source remote sensing data, meteorological observations, and topographic parameters.
Used gradient-boosting machine learning model (XGBoost) on stratified samples for prediction.
Applied SHAP analysis for variable contribution quantification.
Achieved AUC values of 0.91 for annual maximum daily temperature at 37 °C threshold, 0.89 for maximum 1-day precipitation, and 0.84 for consecutive dry days.
Probability of Detection (POD) was 76.3% for temperature, 72.1% for precipitation, and 69.4% for consecutive dry days.
The prediction error rate in the black soil belt was 10.2%, highlighting effective multi-source data integration.

Abstract

Global warming is driving a significant increase in the frequency and intensity of extreme weather events, posing serious challenges to regional ecological security and sustainable agricultural development. Northeast China, situated at the climatic transition between temperate monsoon and cold temperate zones, is particularly sensitive to extreme heatwaves, heavy rainfall, and prolonged drought. This study aims to develop a spatially explicit extreme climate event forecasting system for Northeast China for the period 2000–2024 by integrating multi-source remote sensing data (NDVI, LST, albedo), MODIS surface products, topographic parameters, and ground-based meteorological observations from the China Meteorological Administration (CMA). Three ETCCDI indices—annual maximum daily temperature (TXx), maximum 1-day precipitation (RX1day), and consecutive dry days (CDD)—are used as prediction targets. A gradient-boosting machine learning model (XGBoost) was trained on an 80/20 stratified split of station-grid matched samples and validated using a leave-one-city-out cross-validation strategy. SHAP analysis was applied to quantify variable contributions. The model achieves AUC values of 0.91 (TXx at 37 °C threshold), 0.89 (RX1day), and 0.84 (CDD), with Probability of Detection (POD) of 76.3%, 72.1%, and 69.4%, respectively. For continuous prediction, the root mean square error (RMSE) is 1.3 °C for TXx and 6.8 mm for RX1day. The prediction error rate in the black soil belt is 10.2%, demonstrating the effectiveness of multi-source data integration for high-precision extreme weather forecasting. These results provide technical support for climate risk assessment and agricultural disaster warning in Northeast China.

Ask AI

Mark Helpful

Bookmark

Relay

View Full Paper