Forecasting crop yield with deep learning based ensemble model. (2022)

• Record Type: Journal Article
• Authors: Divakar, M. Sarith; Elayidom, M. Sudheep; Rajesh, R.
• Is Part Of: Materials today. Volume 58:Part 1(2022)
• Page Start: 256
• Page End: 259
• Publication Date: 2022
• Subjects: Remote sensing -- Crop yield prediction -- Deep learning -- Ensemble model -- Convolutional long short term memory; Materials science -- Congresses -- Periodicals; 620.1
• DOI: 10.1016/j.matpr.2022.02.109
• Languages: English
• ISSNs: 2214-7853

Abstract

Abstract: Early prediction of crop yield before harvest is essential in agriculture for taking various policy decisions related to crop production to ensure food availability. Traditional approaches are based on expensive survey data that are not scalable, and results are usually available after harvest only. Techniques based on climatic indices and soil information are expensive to collect and not available for all locations. Remote sensing data archives are available free of cost and can be used with historical crop yield data for building forecasting systems. Until recently, techniques based on crop simulation models and machine learning approaches used derived indices from remote sensing satellites, discarding many spectral bands that carry prominent information. Recent studies have used deep learning models with feature engineered remote sensing data. This study used multiple feature engineering techniques to reduce the input imagery dimension and proposed an ensemble model based on LSTM and Convolutional LSTM to predict soybean and rice yield. Results show that the proposed model shows comparable performance to existing approaches with fewer parameters.