A research team led by Prof. WANG Shuo, Member of Research Institute for Land and Space (RILS), Otto Poon Charitable Foundation Smart Cities Research Institute (SCRI) and Research Institute for Sustainable Urban Development (RISUD), and Associate Professor of Department of Land Surveying and Geo-Informatics, has utilised innovative explainable machine learning to uncover the persistent and significant negative impact of snow droughts on winter wheat yields, highlighting that global food security is facing unprecedented challenges. This breakthrough study provides key scientific evidence for building climate-resilient agriculture, ensuring food security and promoting sustainable development. The study was conducted in collaboration with scholars from The University of Hong Kong and the University of California, Irvine. The findings have been published in the international journal Nature Food.
Over the past 60 years, the frequency of snow droughts across the Northern Hemisphere’s winter wheat croplands has increased significantly. The proportion of croplands affected by snow drought surged from 46–54% in 1960–1970 to 70–99% in 2010–2020, reflecting the fact that snow droughts have evolved from a localised risk into a widespread phenomenon.
The research team developed the XGB-SHAP model framework to accurately identify the direct impact of snow drought on crop yields. The framework combines Extreme Gradient Boosting (XGBoost) with Shapley Additive exPlanations (SHAP) to effectively exclude interference from other climate factors such as high temperature and rainfall, enabling a quantitative analysis of the direct impact of snow drought on winter wheat yields. This study also systematically analysed snow–crop–water interactions, providing a robust empirical basis for developing climate-resilient agricultural systems.
Using the XGB-SHAP framework, the study found that approximately 45% of croplands in the Northern Hemisphere experienced significant adverse impacts from snow droughts, with Europe, Central Asia and the United States being the most severely affected regions. Meanwhile, in East Asia, the yield benefits of longer growing seasons due to warmer winters are gradually diminishing and the regional hydrothermal balance is becoming increasingly unstable. The study also identifies increased fertiliser use, intensified freezing stress and reduced precipitation as the three primary factors contributing to winter wheat’s increased sensitivity to snow drought. While enhanced soil nutrient levels can promote crop growth, they also make crops more dependent on the water supply and thermal insulation provided by snow cover. When snow drought occurs, these factors can amplify the risk of yield loss.
Looking ahead, the research team recommends strengthening the climate resilience of agricultural systems by enhancing crop varieties, improving agricultural management and monitoring risks. Alongside intensifying efforts to develop crop varieties with both cold and drought tolerance, agricultural management models have to shift from traditional, input-intensive production to more precise and sustainable nutrient management systems. Furthermore, snow cover monitoring should be incorporated into agricultural risk assessment and early warning systems. Identifying snow drought risks at an early stage and enhancing the resilience of agricultural systems will ensure that stable food production is maintained in the face of climate change, steering global agriculture towards a more sustainable future.
Press release: https://polyu.me/4nOZhRH
Online coverage:
The National Tribune - https://polyu.me/4dNTZ4t
Mirage - https://polyu.me/42YLSx2
Ta Kung Pao - https://polyu.me/4e8weVZ
Wen Wei Po - https://polyu.me/3RKk4dd
由理大土地及空間研究院、潘樂陶慈善基金智慧城市研究院及可持續城市發展研究院成員兼土地測量及地理資訊學系副教授王碩教授帶領的研究團隊，利用創新的可解釋機器學習技術，揭示雪旱對冬小麥產量帶來顯著而持續的負面影響，反映全球糧食安全正面對前所未有的挑戰。這項突破性研究為建構氣候韌性農業、保障糧食安全及推動可持續發展，提供關鍵的科學依據。此研究與香港大學及美國加州大學爾灣分校學者合作進行，研究結果已刊登於國際權威期刊《Nature Food》。
過去60年間，北半球冬小麥種植區的雪旱發生頻率顯著上升。受雪旱影響的農地比例，已由1960至1970年間的46%至54%，大幅增至2010至2020年間的70%至99%，反映雪旱已由局部風險演變為廣泛性影響。
為準確識別雪旱對作物產量的直接影響，研究團隊開發了XGB-SHAP模型框架，結合極端梯度提升（XGBoost）與夏普利加性解釋（SHAP）方法，有效排除高溫、降雨等其他氣候因素的干擾，定量分析雪旱對冬小麥產量的直接影響。研究亦系統性分析積雪、作物生長與水分條件之間的交互關係，為推動氣候韌性農業系統建設提供堅實的實證基礎。
透過XGB-SHAP模型分析，研究發現北半球約45%的農地受到雪旱顯著衝擊，其中以歐洲、中亞及美國情況最為嚴重。至於東亞地區，過往因暖冬而延長作物生長期所帶來的增產效益正逐步消退，區域水熱平衡亦愈見不穩。研究進一步指出，施肥量增加、凍害加劇及降水減少，是提高冬小麥對雪旱敏感度的三大主因。雖然提升土壤養分有助促進作物生長，但同時亦會增加作物對水分供應及積雪保溫的依賴；一旦發生雪旱，減產風險反而會被進一步放大。
展望未來，研究團隊建議從作物品種改良、農業管理及風險監測三方面提升農業系統的氣候韌性。除了應加強培育兼具抗寒及抗旱能力的作物品種，農業管理模式亦需由傳統投入密集型生產，逐步轉向更精準且可持續的養分管理系統，而積雪監測亦應納入農業風險評估與預警系統。透過及早識別雪旱風險並提升農業系統適應能力，將有助在氣候變化下維持糧食生產穩定，推動全球農業邁向更可持續的未來。
網上報導：
The National Tribune - https://polyu.me/4dNTZ4t
Mirage - https://polyu.me/42YLSx2
大公報 - https://polyu.me/4e8weVZ
文匯報 - https://polyu.me/3RKk4dd
