EVALUATING SOIL CARBON STOCKS IN AGROFORESTRY SYSTEMS USING DEEP LEARNING AND FIELD SAMPLING

Abstract

To answer this question, we require that the amount of carbon in the soil be well known to determine how well it could assist in dealing with climate change and how to go about using the land in a manner that is amicable to nature.  An integrated approach that involves deep learning, the field sampling of soils, and remote sensing are applied to predict soil organic carbon (SOC) in various types of agroforestry landscapes in the study.  The method of collection of the soil samples used was a stratified random sampling due to different depths and soils that are used as a method of management. Thereafter, we measured the SOC concentration by conventional laboratory examination.  A deep convolutional neural network (CNN) model integrated data of high resolutions Sentinel-2 and LiDAR with environmental conditions, such as conditions of climate and topography.  The CNN performed superiorly to both traditional models such as Random Forest, K-nearest neighbours. It also had lower RMSE / MAE and higher R 2, and that indicated that it was more effective as a predictive model.  Temperature, precipitation, elevation, and vegetation indices were the most important, with values of SHAP sensitivity analysis showing this.  The validated model developed spatial SOC maps indicating small-scale variations on the study area on where carbon is abundant and where carbon can be stored.  The natural deep learning strategy also cost less of time and money to the field than the conventional strategies even though it remains quite ecologically valid, since it was also tested on real world data.  These findings indicate that AI-based models may be used to assist with carbon tracking, improve carbon accounting models and contribute to specific agroecology interventions.  This research demonstrates the paradigm that may be employed repeatedly to enhancing the environmental assessment by opportunities of the new technologies. It provides policy-makers, conservationists, and land managers with powerful input in the era of digital and climate-smart agriculture.