Machine learning–based habitat mapping of the invasive Prosopis juliflora in Sharjah, UAE

Abstract

Prosopis juliflora, one of the most invasive trees, adversely affects the ecosystem and native plant communities in arid lands. This disrupts biodiversity and depletes water resources, posing significant ecological and economic challenges. Several attempts have been made to control this species in the United Arab Emirates (UAE) deserts but with little success. This study identifies and maps environmental variables influencing P. juliflora habitats using machine learning (ML); employs maximum entropy (MaxEnt) and statistical techniques to estimate its presence in Sharjah, UAE, home to one of its most intense populations; and conducts validation and sensitivity analysis. Eleven environmental variables representing geological, geomorphological, hydrological, eco-indicators, and climatological factors were selected to map the spread of the associated P. juliflora hazard. Variables were selected using collinearity and variance inflation factor (VIF) to eliminate bias, and ML techniques assigned weights based on overall accuracy (OA) and the Kappa coefficient before model implementation. Finally, a statistical comparison with MaxEnt was conducted to map P. juliflora habitats, classifying suitability as very high, high, low, and very low while estimating model accuracy. The results indicated that MaxEnt achieved a higher area under the curve (AUC 0.98) and more logical outcomes than statistical models (AUC 0.85) due to its superior handling of collinearity, complex environmental interactions, and capability of minimizing overfitting. The main findings show that the variable weights for MaxEnt and statistical models are primarily influenced by precipitation (27.0% and 18.18%), groundwater depth (14.9% and 26.8%), and total dissolved solids (TDS) (20.9% and 26.22%), respectively, indicating a shift in habitat distribution towards the eastern regions of the study area. Habitat mapping of P. juliflora is essential for local stakeholders and policymakers in decision-making regarding species conservation, sustainable land use, and climate adaptation. The findings conclude that ML offers a viable approach for habitat modeling of invasive species in similar arid regions worldwide.