Communicating Uncertainty in Total Maximum Daily Load Modeling to Support the Application of Probabilistic Models in Practice

Degradation of groundwater and surface water quality by excess pollutant loading remains one of the most persistent and costly environmental management challenges in the US. To meet applicable water quality standards and prevent excess pollutants in waterbodies, the Clean Water Act requires that total maximum daily loads (TMDLs), which define the maximum allowable daily pollutant load to a waterbody, are implemented. Water quality models are used in TMDL development to determine appropriate pollutant load reduction amounts and inform pollution reduction strategies. Due to the approximate nature of models and limited observational data and errors, there is a level of uncertainty inherent in their load predictions. However, uncertainties represented as a pollutant load margin of safety (MOS) are often not formally quantified, communicated, or incorporated into TMDL decision-making primarily for two reasons. First, formal uncertainty analyses are perceived to be too computationally expensive. Second, communicating the nuances of mathematically complex models can be challenging. This insufficient—or altogether missing—understanding of modeling uncertainties can hinder the implementation of robust pollutant mitigation strategies, ultimately exacerbating risks to human health and creating avoidable financial and environmental harm. In this paper, we propose a step-by-step framework informed by approaches grounded in social science theory for aiding stakeholders in better understanding of uncertainties associated with developing TMDLs. Among various issues hindering the application of probabilistic models, we focus on the communication aspect of the water quality model uncertainty. This process can increase the inclusion of explicit MOS estimates via probabilistic models in practice and lead to better informed decisions and more reliable implementation strategies. To enhance effectiveness, the approaches described here can be developed through interdisciplinary collaboration among modelers and nonmodelers.