Forecasting land use and land cover dynamics using combined remote sensing, machine learning algorithm and local perception in the Ago`enyiv´e Plateau, Togo

Abstract

Knowledge of land use and land cover (LULC) dynamics helps policymakers set sustainable land management plans. Population growth, migration, and rural exodus have led to spatial expansion in many cities worldwide, especially in coastal urban settings. This research aims at assessing and predicting LULC over the Ago`enyiv´e plateau (AP) in Togo, using remote sensing and Markov Chain along with local perceptions on LULC change, thus considering four scenarios (i.e., Business-as Usual, afforestation, wetland policy, and building policy). LULC was classified based on the maximum likelihood algorithm using Landsat images. The future scenarios maps were produced using the Multilayer Perceptron (MLP) neural networks-Markov chain modelling approach by considering five driver variables considered to affect future development. The historical change analysis was performed for the periods of 1986–2001, 2001–2011, and 2011–2020, while 1986–2020 was used for model validation and change prediction. Image accuracy assessment was performed using the error or confusion matrices, kappa coefficients (Kscores), Relative Operating Characteristics (ROC), and accuracy scores, with an overall accuracy (Kappa coefficient) of 87% (0.81), 88% (0.81), 90% (0.82), and 92% (0.84) for the years 1986, 2001, 2011, and 2020, respectively. The results showed an increase in built-up areas (38.71% land gain), while a decrease was observed in mixed vegetation/savannah (22.69% land loss), croplands/bare surfaces (22.89%), and wetlands (3.36% land loss) over the period of 1986–2020. An increase in built-up areas (7.79% and 19.79% land gain by 2030 and 2050, respectively) and a decrease in mixed vegetation (1.13% and 2.54% land loss by 2030 and 2050, respectively), croplands and bare surfaces (6.58% and 16.98% loss by 2030 and 2050, respectively), and wetlands (0.13% and 0.31% loss by 2030 and 2050, respectively), are expected under the BAU scenario. Contrary to the wetland policy, which induces change only in the wetlands (0.02% and 0.047% land loss by 2030 and 2050, respectively), the afforestation scenario results in a gain in forest/savannah (0.11% and croplands/bare surfaces (0.64% by 2050 compared to the BAU scenario), whereas the building policy scenario results in a long-term increase in croplands/bare surfaces (2.88% by 2050, compared to the BAU), and wetlands (0.027 and 2.88% by 2050) in addition to the mixed vegetation/savannah. LULC dynamics (including future urbanization trends) for the AP under various scenarios is a step ahead of time, highlighting the importance of the implementation of existing laws and policies, but also providing information to deal with the observed inharmonious spatial development of urban centres, and serving as a dataset source for LULC change impact assessments on related resources.