Developing an Automated Uncertainty Quantification Tool to Improve Watershed-Scale Predictions of Water and Nutrient Cycling

Abstract

The main objective of this SBIR Phase I project is to demonstrate the technical feasibility of performing a formal uncertainty-based data-worth analysis using a state-of-the-art watershed model. This objective is accomplished by (1) developing an uncertainty-based data-worth analysis approach and workflow, (2) implementing it into an existing simulation-optimization framework, and (3) demon¬strating its performance by applying the developed computer code to a watershed use case. The 12-months Phase I project started April 9, 2018. As demonstrated in this SBIR Phase I Final Project Report, the technical objectives have been accomplished. Specifically, the uncertainty-based data-worth analysis approach as implemented in the model-independent simulation-optimization framework of iTOUGH2 was combined with the PEST protocol and linked to the ECOSYS terrestrial ecosystem model. The software package, associated toolsets, and external watershed simulation program was successfully installed on various platforms (multi-processor Mac, PC, and Linux cluster). A use case was defined for an ECOSYS model, a comprehensive plant ecosystem model of natural and managed terrestrial ecosystems that can account for surface energy exchange, microbial metabolism, vegetation phenology/physiology, as well as vertical and lateral hydrologic and biogeochemical fluxes. Existing data collected along a hillslope transect were combined with potentially useful monitoring data to perform a sensitivity analysis, data-worth analysis, uncertainty propagation analysis, and data inversion. In particular, the data-worth analysis calculates how much each data point collected in the field can reduce the uncertainty in target predictions requested by the decision-makers. An uncertainty analysis determines whether the prediction uncertainties are sufficiently low, i.e., acceptable for the decision-maker. If so, the data-worth analysis indicates which existing data could be removed to arrive at a cheaper watershed monitoring design without substantially increasing prediction uncertainty. If uncertainties are unacceptably high, the data-worth analysis suggests which additional data should be collected to effectively reduce the prediction uncertainty. All of these applications were run using the code’s Expert Mode, i.e., taking advantage of the domain experts knowledge and experience in using the underlying iTOUGH2 framework and its command structure. (The development of a user-friendly, marketable Automatic Model of the toolset is one of the main objectives of the Phase II SBIR/STTR work.) The theoretical basis, overall workflow, and key command structure needed to perform a data-worth analysis in Expert Mode are being documented along with the ECOSYS use case in a report entitled “Uncertainty-Based Data-Worth Analysis for Watershed Management” and associated user’s manual, with relevant sections reproduced in this final report. In summary, we believe we have successfully demonstrated the technical feasibility of the project. In the remainder of the Phase I project period, we will expand this demonstration to a computationally more demanding watershed model, and finalize the documentation of the toolset’s Expert Mode application.