Response to “letter to the editor: ‘Design of an integrated climatic assessment indicator (ICAI) for wheat production: A case study in Jiangsu Province, China’ by Xiangying Xu, ping Gao, Xinkai Zhu, Wenshan Guo, Jinfeng Ding, Chunyan Li, Min Zhu, Xuanwei Wu”

Abstract

Abstract We thank Dr. Mohammadi and other colleagues for their interest in our research and valuable comments about the integrated climatic assessment indicator (ICAI) of wheat production (doi: https://doi.org//10.1016/j.ecolind.2019.01.059). Dr. Mohammadi brought up a discussion concerning the modeling details of artificial intelligence (AI) algorithms used in our study in a recent letter to the editor (doi: https://doi.org//10.1016/j.ecolind.2019.04.055). In our research, we use AI algorithms to construct the nonlinear relationships between wheat yields and meteorological factors. In addition, by using the algorithms and transformed indicators, the yield levels of rain-fed wheat were analyzed and predicted. Support Vector Machine (SVM) and Random Forest (RF) are powerful modeling methods used in agro-ecosystems and were applied in the study. The modeling details of SVM and RF in the design of ICAI for wheat production are provided in this discussion including the determination of input factors of the models and the selection process of hyper parameters of the models. The climatic factors such as temperature, precipitation, solar radiation, etc. were selected as the input of models during key growth stages of winter wheat by means of correlation analysis between de-trended wheat yields and meteorological factors. Grid search and experience-based search combined with 5 times of 5-fold cross validation were used during parameter selection. The spatial and temporal prediction accuracy of classification models achieved over 80% and 56% respectively in our research. The ICAI transformed from model output can be used to predict yield levels of winter wheat in Jiangsu and it will help decision makers to take effective measures in the process of disaster prevention. In the future, more efficient and precise parameter selection methods need to be investigated to simplify the modeling process and improve the prediction accuracy.