Logo image
Geospatial Artificial Intelligence Algorithms for the Assessment of Commiphora Myrrha Habitats in Semi-Arid Landscapes
Journal article   Open access   Peer reviewed

Geospatial Artificial Intelligence Algorithms for the Assessment of Commiphora Myrrha Habitats in Semi-Arid Landscapes

Khalifa M. Al-Kindi, Saif Al-Hatmi, Laila Al Harthy, Abdurahman Al-Hniai, Ahmed Hamood Albusaidi, Salim Al Rahbi, Nasser Al Rashdi, Omar Al Amri, Anthony B. Cunningham and Ahmed Al-Harrasi
Geomatica (Ottawa), Vol.78(1), 100110
2026
pdf
Commiphora myrrha14.37 MBDownloadView
Open Access CC BY V4.0

Abstract

Commiphora myrrha Multiscale Geographically Weighted Regression Oman Random Forest semi-arid ecology species distribution modeling XGBoost
Understanding species distribution in arid ecosystems requires modelling frameworks capable of capturing spatial heterogeneity, nonlinear ecological interactions, and data limitations. This study presents a novel comparative multi-model assessment that applies Multiscale Geographically Weighted Regression (MGWR), Random Forest (RF), and Extreme Gradient Boosting (XGBoost) in parallel to model the habitat suitability of Commiphora myrrha (C. myrrha), a dioecious tree species at the easternmost edge of its natural range in southern Oman. MGWR was applied as an explanatory spatial regression, while RF and XGBoost were used for predictive habitat suitability mapping. Field survey data collected between 2023 and 2025 across Wadi Amat, Harweeb, and Aydam were combined with high-resolution environmental layers to calibrate and validate the models. MGWR captured strong spatially structured relationships between C. myrrha occurrence and environmental predictors, revealing pronounced multiscale spatial variability in environmental controls (in-sample R² = 0.978; reported as an explanatory diagnostic, with predictive performance assessed separately via cross-validated RF and XGBoost). RF and XGBoost demonstrated robust predictive performance, with AUC values of 0.96 and 0.95, respectively. Suitability mapping revealed a distinct gradient from low to high suitability, with mid-elevation zones exhibiting optimal conditions for the species. Temperature, slope, and geology were identified as the dominant predictors, with spatial variation in their influence captured through MGWR’s multiscale bandwidth calibration. This study contributes to the literature by demonstrating the value of comparing spatial regression with ensemble machine learning for ecological modelling in data-scarce, topographically complex regions. Practically, it offers a transferable framework for identifying ecological hotspots and informing conservation strategies under climate and land use pressures. These findings have direct implications for biodiversity management in Oman and similar semi-arid ecosystems, supporting evidence-based policy development and adaptive conservation planning. •This study presents a novel hybrid approach that integrates Multiscale Geographically Weighted Regression (MGWR), Random Forest (RF), and Extreme Gradient Boosting (XGBoost) to model the habitat suitability of Commiphora myrrha (Burseraceae), a dioecious tree species at the easternmost edge of its natural range in southern Oman.•Field survey data collected between 2023 and 2025 across Wadi Amat, Harweeb, and Aydam were combined with high-resolution environmental layers to calibrate and validate the models.•MGWR provided spatially explicit insights into localized ecological drivers, explaining 97.8% of the variance in habitat suitability.•RF and XGBoost demonstrated robust predictive performance, with AUC values of 0.96 and 0.95, respectively. Suitability mapping revealed a distinct gradient from low to high suitability, with mid-elevation zones exhibiting optimal conditions for the species.•Temperature, slope, and soil type were identified as the dominant predictors, with spatial variation in their influence captured through MGWR’s multiscale bandwidth calibration.•This study contributes to the literature by demonstrating the value of integrating spatial regression with ensemble machine learning for ecological modelling in data-scarce, topographically complex regions.•Practically, it offers a transferable framework for identifying ecological hotspots and informing conservation strategies under climate and land use pressures. These findings have direct implications for biodiversity management in Oman and similar semi-arid ecosystems, supporting evidence-based policy development and adaptive conservation planning.

Details

Metrics

1 Record Views
Logo image