Estimating Total Rice Biomass and Crop Yield at Field Scale Using PlanetScope Imagery Through Hybrid Machine Learning Models

• DOI: 10.1007/s41748-024-00481-2
• Publication Date: 2024-10-18T16:02:24Z
• Authors: Kishore Chandra S...; Chiranjit Singha; Biswajeet Pradhan
• Keywords: Field crop

Abstract Near real-time crop monitoring has been a challenging due to the lack of high-resolution remote sensing images suitable for agricultural applications. The PlanetScope constellation, comprising approximately 130 Dove satellites, collects entire Earth daily, with resolution 3.7 m. from satellite, along vegetation indices, geo-environmental data, and soil parameters, were utilized analysed using machine learning models enhance accuracy predicting total biomass rice yield at field scale. study area, covering nearly 214 sample plots, was located in Tarekswar block Hooghly, West Bengal, India. Alongside ten indices three Principal Component Analysis (PCA) nutrient levels, thirty-six factors analyzed predict (ML) models, namely Random forest (RF), Extreme Gradient Boosting (XGB), Support Vector Machine (SVM), Bagging Tree (Treebag), Generalized Additive Models (gamSpline), Elastic Net (enet), Ordinary regression LASSO penalty (rqlasso), Genetic Algorithm (evtree), Bayesian Regularized Neutral Networks ( brnn), cubist there hybrid ensembles. Boruta multi-collinearity analysis also conducted selected explore their influence levels. area exhibited robust yields ranging 5 10 t/ha, accompanied by healthy growth. Four ML ─cubist, random forest, enet, model—showed promising predictions R 2 > 0.88. Most classified less than 20 ha as falling into “very-low class”, showing region’s suitability cultivation its highly fertile alluvial soil. sensitive revealed that 24 individual significantly influenced final including, organic carbon (OC), phosphorus (P), electrical conductivity (EC), mechanization level, majority indices. A critical carried out through Map query tool showed five estimated via displayed strong correlations (exceeding 89%) identifying areas high very yields. can serve guideline near-real-time near future, images.