Stage‐Specific Associations Between Extreme Climate Events and Crop Failures Across China

Abstract

Crop failures represent severe lower‐tail risks to food security. However, their climate associations remain insufficiently quantified because localized severe losses are difficult to link to large‐scale climate variability. Using station‐based crop yield, phenology and daily meteorological observations across China from 1981 to 2018, we developed an event‐level diagnostic framework to detect and cluster crop failures and evaluate their statistical associations with temperature‐precipitation extremes during the vegetative growth period (VGP) and reproductive growth period (RGP). Crop failures were defined as LOWESS‐detrended yield anomalies below crop‐specific 10th percentiles and clustered in a longitude‐latitude‐time matrix. Detected crop failures affected 70.0%–100.0% of monitored crop sites across crop systems and contributed 48.8%–54.4% of cumulative negative yield anomalies. Although most events were spatially localized, crop‐site failures showed a clear synchronization peak in the early 1990s. After 2000, background occurrence rates of single and compound extremes during VGP each increased by 1.4 percentage points, whereas corresponding rates during RGP decreased by 3.1 and 1.2 percentage points, respectively. Under the nominal p  < 0.10 screening criterion, 22.7%–56.0% of VGP failures and 14.8%–56.0% of RGP failures showed candidate associations with at least one temperature–precipitation extreme. Candidate associations with single extremes were generally more prevalent than those with compound extremes. Sensitivity analyses further showed that low temperature and low precipitation produced the most stable high‐loss signals across crops, stages, FDR subsets and alternative processing choices. These findings support event‐level, phenology‐aware diagnostics for resolving crop‐specific climate risks, identifying critical growth‐stage windows, and improving agricultural risk assessment and adaptation strategies.