Abstract: Abstract: Historical disaster documents can greatly contribute to machine learning from disaster experiences, particularly in understanding the interaction mechanism of regional disaster systems. Historical, agricultural, and meteorological disaster characteristics can be explored by the representation and reanalysis of disaster processes and events, integrating meteorological data and agricultural information. However, compared with field crops, the historical disaster records of cash crops, such as forests and fruits, are relatively scarce, which can make the disaster representation and reanalysis with low accuracy. Therefore, it is of great merit to develop a method for identifying forest and fruit disasters, based on the limited historical disaster data, as well as long series meteorological and fruit growth data, for the cash crop meteorological disasters researches which are lack of historical disaster records. In this study, taking the drought disaster of Fuji Apple in Shaanxi Province as an example, the meteorological data, historical drought disaster, and phenological data were integrated to identify the trigger threshold of apple drought. According to the phenological data of apple collected in this study, tree growth stages, that is, tree germinating to flower budding, flower budding to full bloom, and full bloom to mature, were analyzed as target growth stages for apple. Referring to the drought index construction method for crops, such as wheat and corn, pre- and current water demand and precipitation supply were fully considered in the construction of the apple Drought Index (DI). The probability analysis, K-means clustering, and Euclidean distance were used to comprehensively analyze the distribution and classification characteristics of DI between 35 stations from 1981 to 2018 and historical disaster samples. According to the Euclidean distance between the DI in historical disaster sample and the center point of the cluster sample, the drought trigger threshold of Fuji Apple in tree germinating to flower budding, flower budding to full bloom and full bloom to mature were identified by the corresponding minimum Euclidean distance. Afterwards, the trigger threshold was verified by comparing the sequence of disaster-causing factors and reserving samples. The results showed that: 1) The trigger thresholds of apple drought in tree germinating to flower budding, flower budding to full bloom and full bloom to mature were 0.87, 0.84, and 0.73, respectively. 2) The DI sequences that extracted based on the threshold value in tree germinating to flower budding, flower budding to full bloom and full bloom to mature stages were detected the same characters with that in historical disaster samples. The apple drought data that identified by the calculated of DI and trigger threshold were generally consistent with that disaster records in historical documents, with 85.58% of trigger threshold-based results completely consistent with historical records. The identification coincidence rate was 80.95% in the long-time series validation for typical sites. Generally, the trigger threshold of apple drought can provide a sound technical support for apple drought monitoring, early warning, and assessment in Northern China. The agrometeorological disaster trigger identification method based on small samples of historical disaster data can also offer a paradigm in the current research on the meteorological disasters of cash crops with insufficient historical disaster data.