Date of Award
Fall 11-5-2025
Document Type
Dissertation
Degree Name
Doctor of Philosophy (PhD)
First Advisor
Anthony Parolari
Second Advisor
Brook Mayer
Third Advisor
Kun Zhang
Abstract
Streams and rivers with excessive nutrients can compromise aquatic ecosystems, human health, and overall water resources. Precise, high-resolution measurements of nutrient concentrations are essential for pollution management programs; however, conventional monitoring methods frequently face limitations due to cost and sample restrictions. This dissertation aimed to advance the existing state-of-the-science nutrient monitoring and modeling approaches to reduce sampling requirements and model complexity. First, this study estimated the dimensionality and predictability of nutrient signals (NOx and TP signals), that describe the nutrient dynamics. The findings showed that nutrient data can be forecasted accurately for 1 to 5.5 months. Most of the signals displayed low to moderate dimensionality (5-10) and had some hidden order within their dynamics. This enabled the option of using reduced-order models to estimate concentration data. Second, this study implemented a data-driven sparse sensing (DSS) framework to estimate high-frequency nutrient data from minimal measurements. The simplest DSS models (training data consisting of only flow and nutrients) were able to estimate accurate daily concentrations and annual loads using as few as 40 samples per year (errors <±2% for NOx, <±9% for TP). Instead of relying on fixed-frequency sampling or the “storm chasing” approach, DSS models identified the optimal sampling times that contained the maximum information across the monitoring period. Finally, the DSS modeling framework was tested under real-world monitoring scenarios to address key challenges in designing nutrient monitoring programs. Results showed that accurate reconstructions (NSE > 0.5; load errors <±4%) are possible with only 2–3 years of data from 15 to 25 sites, especially when training data includes both flow and concentration. Overall, the results backed the use of reduced-order modeling approaches (i.e., DSS) that take advantage of the fact that nutrient signals are low-dimensional and sparse. Using DSS, daily nutrient concentration time-series can be generated with 80-90% reduction in the number of samples required, leading to a significant reduction in monitoring effort. These findings can support ongoing water quality management efforts by promoting monitoring strategies that are more resource-efficient and modeling frameworks that can operate with fewer parameters and limited data.
Comments
Doctor of Philosophy (PhD)