The development of crop yield estimation model under abnormal climate conditions is of great significance to understand the impact of climate change on crop yield and improve the applicability of yield estimation model. This paper proposes a random forest regression yield estimation method based on meteorological disaster index, remote sensing vegetation index and trend yield. Based on this method, the wheat yield estimation models of five major wheat growing regions in China were constructed, and the accuracy of wheat yield estimation was verified at both county level and wheat growing region level in selected typical disaster years. The importance of input variables in different wheat growing regions to the establishment of yield estimation model was compared and analyzed. The results show that the R2 of fitting accuracy of yield estimation model in each wheat growing region was all above 0.95. The average relative error between actual and predicted yield in disaster years was below 0.060 at county level, and 0.049 at wheat growing region level. The importance of input variables differs in different wheat growing regions. The importance of standardized precipitation evapotranspiration index is high in all five wheat regions. The dry hot wind index is more important in the northwest spring wheat region and the north spring wheat region than in other three wheat growing regions. The importance of monthly average temperature anomaly and monthly average precipitation anomaly in each wheat growing region has little difference. The importance of Normalized Difference Vegetation Index (NDVI) at jointing stage and heading stage was higher than that at other stages. The yield estimation model constructed in this paper can meet the accuracy of yield estimation under abnormal climate conditions and provide reference for large-scale yield estimation methods under extreme climate conditions.