Object-oriented Model to Predict Crop Yield Using Satellite-based Vegetation Index

Abstract

Different models had been developed to predict crop yields based on remotely sensed data. Most approaches were based on developing empirical relationships between the satellite-based normalized difference vegetation index (NDVI) data and the crop yield. This article is proposed to introduce a methodological framework for constructing an object-oriented yield prediction model using satellite data based on the two-level regression models. Here, the trends caused by the influence of technological improvements were considered. Regression models for the wheat and barley crop yield predictions have been developed. The two-level regression model, including the foreword stepwise regression (FSR) technique, firstly selects the set of features that reflect the spatial variations in crops, soil, and agriculture management within districts. After the steps of exploratory data analysis (EDA), object creation, and the zonal average of each object were carried out. The second level consists of yield prediction with multiple linear regression (MLR), least absolute shrinkage, and selection operator (Lasso), support vector machines (SVM) techniques. In the proposed model, the SVM technique outperforms the rest techniques by an average root mean square error (RMSE) of 5.59(4.51) for wheat(barley). The experiments showed that the proposed model provides stability and low prediction error in the vast majority of cases and the used techniques