Soil textures and nutrients estimation using remote sensing data in north india - Punjab region
Gaurav Dhiman, Jhilik Bhattacharya, Sangita Roy
Abstract
Estimating soil and plant nutrients are essential for crop growth and yield production. Quantifying nutrients by conventional methods is a time-consuming and hectic process. The hyperspectral imagery and multispectral sensors mounted on the UAV (Unmanned Aerial Vehicle) are costlier. This study used open-source remote sensing imagery and obtained ground truth data from the Punjab region to estimate nine nutrients and soil textures: Nitrogen (N), Phosphorous (P), Carbon (C), Ammonium (N-NH4), Nitrate Nitrogen (N-NO3), Calcium (Ca), Magnesium (Mg), Clay and Silt. To remove the noise from the dataset, this study applied DOS1 atmospheric correction and compared it with the raw dataset with four ensemble machine learning methods: Gradient Boosting (GB), Extreme Gradient Boosting (XGB), Random Forest Regressing (RFR), and Adaptive Boosting (ADA) for estimation of soil texture/nutrients content. The study has three main purposes; firstly, it focuses on the less studied agriculture-intensive Punjab region. Secondly, it analyses the impact of preprocessed vs raw multispectral images. Finally, it explores the correlation of multispectral images and different soil parameters via different machine learning techniques. Results of this study suggest that the raw dataset is in general significantly better than the DOS1 atmospheric correction. Also, some parameters have better correlation with the image data compared to others. This potential approach can estimate the soil texture/nutrient content in the same region.