Wheat yield modeling using climatic indices and hierarchical clustering

The study of climatic indices with relation to the meteorological data is one of the effective factors in decision making of climate, agriculture, water resource engineering planning and determination of management strategies.The aim of this research is improvement of climatic indices- crop yield modeling with emphases on the model inputs based on the clustering analysis. Data derived from clusters of climatic indices was conducted with mean calculation of each cluster. The investigated indices were 11 climatic indices (Lang, De Martonne, Koppen 1, Koppen 2, Koppen 3, Angstrom, Ivanov, Selyaninov, PEI, VCI and aridity). Simple regression and artificial neural networks were used as modeling of climatic indices- wheat yield in Gilan, Esfahan, Kermanshah and West Azerbaijan provinces. The modified indices with correct structure led to increase of accuracy in crop yield estimation, for example the agreement index of Kermanshah province was increased 12.82% from De Martonne to Koppen 2 index. RMSE related to crop yield eastimatiom using climatic indices compared to the direct use of meteorological data in all provinces decreased 36.66%. The clustering analysis regard to the models input determination increased the models accuracy (Average RMSE of all provinces with clustering=0.7 and without clustering= 1.15, Average RRMSE with clustering=0.29 without clustering= 0.5). Therefore, the synthesize of climatic indices as the model input with proper analysis led to improvement of crop yield modeling.