Estimating the value of ecosystem services: A machine learning approach for Missouri wetlands

Ecosystem services provide critical environmental, economic, and social benefits, yet their monetary valuation remains constrained by spatial heterogeneity and the cost of primary studies. This paper advances wetland ecosystem service valuation with a supervised machine learning approach, integrating gradient boosting with a large dataset from the Ecosystem Services Valuation Database (ESVD). Benchmarked against ordinary least squares under study-grouped holdout evaluation, the GBM achieves roughly double the out-of-sample explanatory power while avoiding the implausible extrapolations characteristic of linear models. We apply the framework to five Missouri wetland conservation areas spanning diverse ecological gradients, with block-bootstrap prediction intervals quantifying the substantial uncertainty inherent in cross-site transfer. Regulating and maintenance services dominate estimated values at all sites, highlighting the large share of wetland economic value that is invisible to market-based decision-making. SHAP-based decomposition reveals that the same covariates can have opposing effects across service categories: a form of heterogeneity that linear benefit transfer cannot capture. Beyond the Missouri application, the framework offers a globally transferable methodology for ecosystem service valuation using publicly available geospatial data and internationally standardized classification.