Prediction of cotton yield reduction after hail damage using a UAV‐based digital camera

Abstract

AbstractTo elucidate different performances of cotton (Gossypium hirsutum L.) yield reduction prediction model at multiple damage stages under different hailstorm damage levels, six times (I–VI, no repeated damage) of hail damage simulation treatments with four damage levels (0, 30, 60, and 90%) for modeling and natural hailstorm tracking experiment for model verification were conducted in 2017–2019. Eleven image parameters derived from an unmanned aerial vehicle‐based digital camera were analyzed. In addition, four regression methods (traditional regression analysis, partial least squares regression (PLSR), support vector regression, and back‐propagation neural network) were employed to develop a prediction model of cotton yield reduction. The results indicate that cotton yield prediction model at each damage stage performed better than a general model developed for all damage stages. Considering the bloom stage (75 d after sowing) as a boundary, three PLSR models based on multiple image parameters were developed for pre‐bloom hail damage. For post‐bloom hail damage, a general quadratic polynomial model based on relative canopy cover performed the best among all regression models. In conclusion, for pre‐bloom hail damage, the yield prediction model for each damage stage should be developed, whereas for post‐bloom hail damage, a general yield prediction model should be developed. The results of this work might serve as a guide for farmers and agricultural insurance companies to estimate hailstorm damages quickly and accurately.