<p>Monitoring nutrient levels, particularly nitrate (NO₃⁻) and orthophosphate (PO₄³⁻), in wetlands is essential for preserving water quality and functional ecosystem services. Traditional ground-based sampling is accurate but suffers from limited temporal and spatial coverage. Satellite remote sensing provides a complementary pathway, enabling long-term, large-scale assessments. In this study, we evaluate and compare six machine learning (ML) models for estimating nitrate and orthophosphate concentrations in the Anzali Wetland, northern Iran, using input data from Sentinel-2 MSI, ERA5-Land meteorology, and IMERG V06 precipitation, along with concurrent field measurements. The Anzali Wetland is a vital, internationally recognized Ramsar site; however, regular in situ water quality monitoring is severely restricted due to logistical challenges and access limitations. To address this critical gap, this study presents a novel approach to water quality mapping by integrating multi-source remote sensing data, combining high-resolution optical imagery (Sentinel-2) with hydro-meteorological datasets (ERA5, IMERG), driven by advanced machine learning models. For nitrate, although random forest (RF) and gradient boosting regressor (GBR) achieved near-perfect fits in training, the Gaussian process regressor (GPR) outperformed all others in testing (MSE = 0.0056 (mg/L)², NSE = 0.7937). Orthophosphate estimation was more challenging: RF achieved the best testing performance (MSE = 0.1089 (mg/L)², NSE = 0.7582), while GPR performed poorly (MSE = 0.4917 (mg/L)², NSE = −0.0941). Error distribution and Taylor diagrams confirmed nitrate’s more predictable behavior compared to orthophosphate. Feature importance analysis revealed that Sentinel-2 bands B5, B9, B4, and B6 were most relevant for nitrate, whereas B2, precipitation, B3, and B8 dominated orthophosphate prediction. This integration of multi-source data and ML demonstrates the feasibility of frequent nutrient monitoring for Ramsar wetlands.</p>