Advancing Soil Assessment Quality for Crop Yield Prediction Using an Optimized Category Integrated Dual Task Graph Neural Network Approach

Abstract

Soil quality assessment plays a role in improving agricultural productivity and sustainability, as it is essential for making informed decisions in precision farming. This study proposes a new model for soil assessment quality (SAQ) and crop yield prediction (CYP) based on a category integrated dual task graph neural network (CIDTGNN). The designed model estimates various soil health parameters (SHP) such as soil salinity, soil moisture (SM) and soil organic carbon (SOC) for the Rupnagar district of Punjab, India, and predicts crop yield by fusing remotely sensed Sentinel‐1 and Sentinel‐2 satellite data with field observations. Design SAQ‐CIDTGNN model learns to predict SHP while estimating the crop yield by adopting graph‐based data representations. The weight parameters of the SAQ‐CIDTGNN are optimized using the banyan tree growth optimization (BTGO) approach, leading to improved accuracy. The technique outperforms existing machine learning (ML) replicas, including deep learning methods, attaining an R 2 value of 0.78 in CYP with a significantly lower error rate. This proposed method outperformed the ordinary least squares (OLS) regressor in R 2 by 42% and reduced Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) by 38% and 39%, respectively.