Use of partial least squares regression to identify factors controlling rice yield in Southern China

Abstract

Identification of the dominant factors controlling rice (Oryza sativa L.) yield is of paramount importance to improve fertilizer use efficiency, effectively promote rice productivity, and help ensure food security. The objective of this study was to quantify the influence of geographic attributes, soil properties, climatic, and fertilizer types on annual rice yield and to identify the dominant control factors in southern China. In total, 2010 soil samples were collected in the selected areas and were analyzed for 34 factors that potentially influenced rice yield. Because these factors exhibit multicollinearity, partial least squares regression (PLSR) was used to elucidate the linkages between rice yield and the 34 measured variables. The first‐order factors were identified by calculating the variable importance for the projection (VIP). The variables with high VIP values are the most relevant for explaining the dependent variables. Results indicated that the geographic attributes, climatic and fertilizer types exerted substantial influence on rice yield and explained 57 to 85% of the variation in rice yield. According to the VIP values, the following are the dominant first‐order factors controlling rice yield: longitude, latitude, organic potassium fertilizer, accumulated temperature ≥ 10°C, chemical potassium fertilizer, organic nitrogen fertilizer, straw incorporation, chemical nitrogen fertilizer, accumulated temperature ≥ 0°C, and organic phosphate fertilizer. These results indicate that the PLSR approach is beneficial as it partially eliminates the correlation of the variables and reduces bias regarding the contribution of the factors to rice yield. This approach could be applied to other climatic zones or cropping systems.