A global spatial meta-regression analysis of mangrove valuation studies

• We address spatial dependency in meta-data of mangroves from the Ecosystem Services Valuation Database (ESVD). • We disentangle and quantify spatial autocorrelation from other sources of autocorrelation. • The Spatial Autoregressive Combined, based on a Weight matrix which combines geographical and ecological distances, provides the best performance. • Using M-estimators, the SAC model produces satisfactory benefit transfer errors. Meta-regression analyses of environmental valuation studies often reveal spatial dependence, which must be addressed to properly explain and predict variation in valuation estimates. This paper develops meta-regression models that explicitly incorporate spatial dependence using a dataset of mangrove ecosystem service (ES) values, comprising 506 estimates from 106 primary studies extracted from the Ecosystem Services Valuation Database (ESVD). We first estimate conventional aspatial models with explanatory variables accounting for multiple sources of heterogeneity and autocorrelation. Subsequently, we specify and estimate a series of spatial regression models that integrate spatial processes directly. Different spatial weighting matrices are tested, based on geographic distances and attribute-based similarity reflecting socioeconomic and biophysical characteristics of study sites. In addition, hybrid matrices are developed to combine these spatial dimensions and to relax the assumption that identical spatial processes affect all components of the model. After selecting the best aspatial and spatial models, we evaluate their predictive performance using in-sample and out-of-sample validation. Results show that the random-effects model, which clusters observations by site, offers a theoretically sound framework that effectively captures latent spatial dependence. However, Moran’s I tests applied to OLS residuals reveal remaining spatial autocorrelation, indicating that explicit spatial modelling is still needed to uncover underlying spatial processes. Among the alternative spatial specifications, the Spatial Autoregressive Combined (SAC) model performs best, as it allows distinct spatial processes to influence both the dependent variable and the error term, also through correlations captured by hybrid matrices. While spatial dependence does not substitute the explanatory contribution of site-level heterogeneity, it marginally improves out-of-sample predictive accuracy. This suggests that spatially explicit meta-regression models can yield more reliable and spatially consistent predictions for benefit-transfer applications.