Design and development of novel data analysis techniques for crop yield mapping using remote sensing satellite imagery

Abstract

Timely information about crop yield at the regional level is essential to ensure food security newlinefor the country. A decrease in arable land and crop yield variability due to climate change newlineinfluences agricultural production and is a concern for the government, highlighting the newlineimportance of early crop yield prediction to develop strategic decisions on imports and newlineexports. Crop yield forecasted using approaches like sample surveys are not feasible on a newlinelarge scale, and results are available after harvest only. Prediction techniques using crop newlinegrowth modelling based on the crop s phenological stages and environmental conditions newlineare challenging as they have intensive data requirements. Statistical models were also used newlineto map crop yield with climatic indices, fertiliser and soil information. However, the newlinestatistical model requires sufficient and reliable data for accurate predictions, while newlinecollecting soil and fertiliser data is expensive and unavailable for all locations. Remote newlinesensing is an effective technique for building empirical models for crop yield prediction. newlineRemote sensing satellite images are available free of cost to the users and available newlineglobally. Researchers mainly focused on the visible and near-infrared regions of the newlineelectromagnetic spectrum for crop yield mapping. Until recently, researchers used newlinevegetation indices extracted from raw satellite images and climatic indices with machine newlinelearning and deep learning techniques for crop yield mapping. They combine soil data, newlinefertiliser information and other environmental variables with vegetation indices. However, newlinethey discard other spectral bands which contain a wealth of information in many studies. newlineRecent advances in deep learning for computer vision-based applications and the newlineavailability of hyperspectral and multispectral images with better spatial and temporal newlineresolutions have led to a significant increase in the number of studies related to crop yield newlineprediction utilising all spectral bands.