Thematically weighted regression models for identification of important drivers of environmental trends in lake survey data

The processes that drive environmental change are complex and often interwoven. Monitoring such processes and their effects on environmental variables is, at best, spatially and temporally incomplete. The Swedish Lake survey consists of several thousands of randomly selected lakes stretching over a wide range of potential influences, including regional and local scale pressures, such as recovery from acidification, changes in climate, and effects from various catchment characteristics. The main drawback for this monitoring data is that the temporal resolution is low and does not allow for the use of traditional trend evaluation methods at individual stations, much less the attribution of important drivers. In this study, we present a method that enables an evaluation of important drivers of change by defining trend coefficients as smoothed estimates over lakes that exhibit similar attributes, e.g., comparable levels or similar temporal changes in selected explanatory variables. The principles are the same as geographically weighted regression but replace the geographic coordinate system with a thematic one, based on principal (PCA) or partial least squares (PLS) components. We illustrate this method by evaluating trends in pH in Sweden from 2012 to 2023 and detected several regions where pH is decreasing, mainly in relation to changes in calcium.