Effective management of water quality requires rigorous scientific understanding of the causal mechanisms driving changes over time, but reliably inferring and modeling these mechanisms is challenging…

Effective management of water quality requires rigorous scientific understanding of the causal mechanisms driving changes over time, but reliably inferring and modeling these mechanisms is challenging due to the diffuse nature of non-point source pollution and the dynamic complexity of pollution and its downstream effects. Recently developed empirical methods based on nonlinear dynamics theory offer a reliable means of inferring causality by tracing downstream conditions (e.g., eutrophication) to upstream pollution sources, to guide targeted investments in restoration projects and water infrastructure upgrades. Studies published by the PI demonstrate the methods scientific validity and practical utility in systems whose flows are controlled by engineered structures, but the method is untested in naturally flowing rivers (unregulated flow). During phase i, ecco scientific will evaluate the validity of the methodology in this setting, through a case study benchmarked against existing results of a conventional method (watershed modeling or stable isotope tracing). Further, existing software implementations lack a cohesive, computationally efficient, quality-controlled workflow conducive to performing causal analyses on a large scale. During phase i, ecco scientific will evaluate the technical benefits of optimizing the code base for parallel execution on a graphics processing unit (GPU), by estimating the associated reduction in execution time. Ecco scientific anticipates that the phase i effort will (1) demonstrate the scientific validity and practical utility of the causal inference methodology in the unregulated-flow setting and (2) estimate execution-time acceleration on the order of 100X with GPU parallelization (compared to execution on a CPU). The GPU-accelerated software will massively expand the number and scope of feasible causal investigations, in terms of the number of river/hydrological systems that can be analyzed, the number of variables per river system, and the size of the data per variable. Application of the proposed software will provide scientific insights to (1) inform the development and targeting of interventions for effective management and restoration of water quality, (2) identify patterns of unintended consequences of past and ongoing management actions, and (3) inform refinement of existing processbased models by identifying real-world causal mechanisms that are inadequately represented or entirely absent in model formulations. Potential clients include public-sector agencies on local, state, and federal levels; universities and non-profit research institutions; and engineering/environmental consulting firms in the private sector. Commercialization will be pursued through deployment of the proposed software as part of ecco scientifics consulting/research practice and potentially through software licensing.