Modelling Future Vegetation-Cover Change Scenarios in Baringo County, Kenya, up to 2055 Using MSAVI and CA-Markov- MLP

Abstract

Vegetation-cover change is an important component of land degradation in arid and semi-arid lands because it affects forage availability, soil protection, ecosystem function, and pastoral livelihood options. This study quantified historical changes in vegetation condition across the arid and semi-arid lands of Baringo County, Kenya, and developed a scenario-based projection to 2055. Historical and contemporary vegetation-condition rasters for 1994 and 2024 were classified using the Modified Soil Adjusted Vegetation Index (MSAVI) into sparse vegetation, grass, and dense vegetation. Observed 1994-2024 transitions were summarised with a Markov transition matrix and spatially allocated using a Cellular Automata-Markov framework supported by a Multi-Layer Perceptron transition-potential model. Climatic and biophysical covariates were used to spatially constrain the SSP5-8.5 scenario projection. The study area comprised 21,086,319 cells on a common 20-m analysis grid, equivalent to 8,434.53 km². Between 1994 and 2024, dense vegetation declined from 4,179.64 to 3,508.28 km² (-16.1%), while grass increased by 408.77 km² (+12.2%) and sparse vegetation by 262.59 km² (+29.2%). Under the modelled 2055 scenario, dense vegetation decreases to 3,185.63 km², representing a cumulative reduction of approximately 23.8% from the 1994 baseline; sparse vegetation increases by 52.1%, and grass by 15.7%. Summary validation statistics (R2 = 0.693, RMSE = 0.047 MSAVI units, overall accuracy = 71.48%, mean bias = +0.003) indicate moderate landscape-scale predictive performance. The results should be interpreted as scenario projections rather than deterministic forecasts, particularly because the historical baseline predates Sentinel-2 and because Markov-based projections assume persistence in the broad structure of observed transition processes. The projected decline in dense vegetation and continued expansion of sparse cover identify priority areas for rangeland restoration, vegetation monitoring, and spatially targeted land-management planning.