A hierarchical interannual wheat yield and grain protein prediction model using spectral vegetative indices and meteorological data

Abstract

The use of remote sensing data for predicting wheat yield and quality is becoming a more feasible alternative to destructive and post-harvest laboratory-based test methods. However, most prediction models which make use of remote sensing data are statistical rather than mechanistic, therefore difficult to extend at interannual and regional scales. In this work, an interannual expandable wheat yield and quality predicting model using hierarchical linear modeling (HLM) was developed, integrating hyperspectral and meteorological data. The results showed that the ordinary least squares (OLS) regression for predicting wheat yield and grain protein content (GPC), one key indicator of grain quality, had low stability at the interannual extension. The predictive power for yield by HLM method was higher than OLS, with R2, RMSEv and nRMSE values of 0.75, 1.10 t/ha, and 20.70 %, respectively. GPC prediction by the HLM method was enhanced when the gluten type was considered, with R2, RMSEv and nRMSE values of 0.85, 1.02 %, and 6.87 %, respectively. The results of this study confirmed that HLM can be a robust method for improving yield and GPC predicting stability under various growing seasons in winter wheat.