Geospatial land cover change analysis using the CA-Markov chain model in Chikwawa District, Malawi

Abstract

Context and Background: Changes in forest cover have profound implications for local climate regulation by altering the balance of energy and water exchanges. Over recent decades, anthropogenic and natural factors have significantly transformed land cover patterns, leading to potential ecological and hydrological consequences. In particular, understanding the extent and direction of these changes is essential for sustainable natural resource management and the protection of ecosystems vulnerable to climate-related risks. Goal and Objectives: This study aimed to assess historical and projected land cover changes, with a specific focus on forest loss, using a range of classification algorithms and prediction models. The objectives included evaluating the classification accuracy of multiple machine learning algorithms, projecting future land cover scenarios, and assessing the implications of land cover dynamics on local environmental stability. Methodology: The research utilized Landsat satellite imagery spanning from 1979 to 2023 to classify land cover using four classifiers: IsoData, Support Vector Machine (SVM), Random Forest (RF), and Maximum Likelihood (ML). Classification accuracy was assessed using the Kappa coefficient and overall accuracy metrics. Future land cover scenarios for the years 2035, 2045, and 2065 were modelled using the CA-Markov chain model. The Spearman rank correlation was applied to evaluate the statistical significance of temporal changes. Model performance and prediction reliability were further validated using Kappa-based indices including Kno, Klocation, and Kstandard. Results: Among the classifiers tested, Random Forest consistently yielded the highest accuracy, with Kappa coefficients of 85%, 86%, 86%, and 90%, and overall accuracies of 88%, 90%, 90%, and 93% for the years 1979, 1995, 2009, and 2023 respectively. Land cover analysis revealed a substantial decrease in forest (-21%), vegetation (-3%), and water (-0.3%) areas, alongside notable increases in built-up (+22%) and bare land (+16%) over the 44-year period. The CA-Markov projections indicated a continued decline in forest cover by-28.6%,-33.8%, and-51.0% for the periods 2035-2023, 2045-2035, and 2065-2045, respectively. Concurrently, built-up areas and bare land are expected to increase significantly. The CA-Markov model showed strong predictive agreement, with validation scores indicating good to excellent accuracy across all three tested years (e.g., 2023: CA-Markov = 81.6, Kno = 0.80, Klocation = 0.84, Kstandard = 0.83). These findings reveal a persistent trend of deforestation that, if unmitigated, may intensify runoff absorption deficits and increase flood risks in the district.