<p>Mangroves are among the most productive and carbon-rich coastal ecosystems, yet their carbon dynamics in arid regions remain poorly understood. This study estimates above-ground biomass (AGB) and above-ground carbon (AGC) in the arid mangroves of Nayband National Park, Persian Gulf, Iran, using a combination of field-based allometric measurements and Landsat 8-OLI remote sensing data integrated with machine learning (ML) algorithms. Thirty sample plots were established, and tree height, diameter at breast height (DBH), and wood density were measured to calculate AGB and AGC. Four ML algorithms, Neural Network Regression (NNR), eXtreme Gradient Boosting Regression (XGBR), Support Vector Regression (SVR), and CatBoost Regression (CBR) were evaluated. we applied rigorous leave-one-out cross-validation (LOOCV) during model training and optimized hyperparameters using grid search to prevent overfitting. SVR demonstrated superior predictive accuracy (R² = 0.998), outperforming other models, followed closely by NNR, XGBR, and CBR models (R² = 0.996, 0.966, and 0.948, respectively). In contrast, the Linear Regression model displayed poor performance (R² = 0.084). Among the indices, RVI was identified as the most significant predictor, followed by NDVI, while LAI contributed less to carbon prediction.The estimated mean AGB and AGC were 54 t ha⁻¹ and 24 t C ha⁻¹, respectively, reflecting the limitations imposed by arid and hypersaline conditions. Spatiotemporal analysis of mangrove cover from 1990 to 2019 revealed significant expansion in both Bidkhon and Basatin regions despite anthropogenic pressures. Comparisons with global mangrove forests indicate that Nayband mangroves store lower carbon relative to humid tropical systems, emphasizing the critical need for targeted conservation and restoration strategies in arid coastal environments. The integration of remote sensing and ML provides a robust and cost-effective approach for large-scale mangrove carbon assessment in challenging climatic regions.</p>