Modeling the drivers of eutrophication in Finland with a machine learning approach

Abstract

Anthropogenic eutrophication is one of the most common threats to inland water quality, often causing toxic algal blooms and loss of aquatic biodiversity. Mitigating the harmful impacts of eutrophication requires managing nutrient inputs from the catchment focusing on the major local drivers of eutrophication. These drivers can be identified using models that predict lake trophic state based on characteristics of the lake and its catchment. In this study, we aimed to extend the spatial scope of these models by identifying drivers of eutrophication in a large sample of lakes (1547) distributed across Finland. Moreover, we used satellite-observed chlorophyll a (chl a) concentration as trophic state indicator, instead of site-sampled data, which is commonly used in existing research. We identified major drivers of eutrophication on river basin district to country scale based on 11 catchment and lake characteristics, applying the random forest algorithm. On country scale, the catchment and lake characteristics explained 41% of the variation in lake chl a concentrations, and on river basin district scale, 20%–44%. Catchment and lake hydromorphology were the most important explanatory characteristics. Especially, high natural eutrophication level, shallow mean depth of lake, and small share of lake area in the catchment were related to increased lake chl a concentration. Moreover, depending on the dominant land use type in the model area, share of agricultural area and share of peatland area in the catchment were ranked among the most important drivers of increased lake chl a concentration. The results suggest that trophic state predictive models utilizing satellite-observed chl a concentration could provide an additional, cost-effective tool for addressing lake eutrophication, especially in areas without and extensive on-site monitoring network.