An end-to-end pipeline based on a Two-Dimensional Convolutional Neural Network for monitoring desertification in Africa using remote sensing images

Desertification is a major environmental challenge in Africa, influenced by climate change, deforestation, and unsustainable land use. Effective monitoring is crucial for sustainable land management. This paper presents an end-to-end pipeline based on a Two-Dimensional Convolutional Neural Network (2D-CNN), achieving a classification accuracy of over 91% across a dataset derived from MODIS imagery collected over Africa between 2015 and 2023. The pipeline encompasses data acquisition, preprocessing, model development, evaluation, and prediction, facilitating large-scale desertification sensitivity analysis. The model’s performance was rigorously assessed using multiple metrics, including precision (90%), recall (89%), F1-score (89.5%), balanced accuracy (74.32%), and Matthews Correlation Coefficient (MCC) (0.86). Our proposed 2D-CNN consistently outperforms traditional machine learning models, including Random Forest (RF), XGBoost, Recurrent Neural Network (RNN), and Variational Autoencoder (VAE), demonstrating superior classification performance. The analysis reveals significant desertification expansion in the Sahel and Southern Africa regions, emphasizing the urgency for proactive intervention strategies.