William Keenan

William Keenan

Hydrologist and Geospatial Data Scientist

Publications

Characterizing the influence of remotely sensed wetland and lake water storage on discharge using LSTM models (in production)

Published in Hydrological Sciences Journal, 2026

Abstract: Globally, many wetlands and lakes are at risk for further loss, which can amplify downstream consequences of flood and drought events. We derived remotely sensed based time series of surface water storage (SWstorage) to determine when and where accounting for SWstorage dynamics improves predictions of river discharge. We trained four Long Short-Term Memory (LSTM) models, that differed in their inclusion of storage data and catchment characteristics, to simulate daily river discharge (2016-2023) for select watersheds across the conterminous United States. Adding SWstorage to a meteorology-only or meteorology-and-catchment characteristics model improved upon model Nash-Sutcliffe Efficiency (NSE) in 80.6% of the watersheds. Residuals during low-flow (Q70) events decreased by 47.6% when adding storage to meteorological data. Improvements were most consistent in ecoregions with a greater abundance of non-floodplain lakes and wetlands. This effort represents the first exploration to train a multi-watershed LSTM on landscape scale remotely sensed time series of SWstorage.

Recommended citation: Vanderhoof, M. K., Keenan, W., Dolan, W., Golden, H., Lane, C., Christensen, J., Solvik K., Rajib A. (2025). Characterizing the influence of remotely sensed wetland and lake water storage on discharge using LSTM models. Hydrological Sciences Journal. https://doi.org/10.1080/02626667.2025.2593333 https://doi.org/10.1080/02626667.2025.2593333

Relating surface water dynamics in wetlands and lakes to spatial variability in hydrologic signatures

Published in Wetlands Ecology and Management, 2025

Abstract: The retention of surface water in wetlands and lakes can modify the timing, duration, and magnitude of river discharge. However, efforts to characterize the influence of surface water on discharge regimes have been generally limited to small, wetland-dense watersheds. We developed random forest models to explain spatial variability in six hydrologic signatures, reflecting flashiness, high, and low flow conditions, at 72 gaged watersheds with variable water storage capacity across the conterminous United States. In addition to variables representing meteorology and landscape characteristics, we also tested the inclusion of surface water dynamics, derived from Sentinel-1 and Sentinel-2. Models for all six signatures improved with the addition of catchment characteristics, including surface water dynamics, relative to models with only climate variables. Percent improvement in model adjusted R2, mean square error, and Akaike information criterion ranged from 4.00 to 14.33%, 5.00 to 20.30%, and 2.75–8.14, respectively. Automated variable selection can be indicative of the relative importance of certain variables over others. Using a forward selection process, five of the six signature models selected remotely sensed inundation or wetland variables (p < 0.05). For example, the variable semi-permanent and permanent (SP + P) floodplain inundation (i.e., lakes along rivers) was associated with lower annual flashiness. Further, SP + P non-floodplain waters and geographically isolated wetlands significantly contributed to explaining variability in the low flow signatures. Our findings underscore the capacity of wetlands to stabilize and maintain flows during dry periods. Improved understanding of how surface water dynamics influence hydrologic signatures can inform wetland restoration efforts and facilitate improved resilience to extreme flow conditions.

Recommended citation: Vanderhoof, M. K., Nieuwlandt, P., Golden, H. E., Lane, C. R., Christensen, J. R., Keenan, W., & Dolan, W. (2025). Relating surface water dynamics in wetlands and lakes to spatial variability in hydrologic signatures. Wetlands Ecology and Management, 33(4), 53. https://doi.org/10.1007/s11273-025-10066-z https://doi.org/10.1007/s11273-025-10066-z

A Geospatial and Machine Learning Framework for Forecasting Ground Level Ozone Pollution

Published in Electronic Theses and Dissertations, 2024

Abstract: The major detrimental health effects of ground-level ozone (GLO) pollution make it imperative that both policy makers and ordinary citizens have access to high accuracy, high-resolution forecasts of their local area. Recently, advancements in computing power have made it possible to apply artificial intelligence (AI) techniques to a variety of big data modelling problems, including GLO forecasting and estimation. Of these AI methods, deep neural networks (DNN) have demonstrated the highest accuracy due to their ability extract non-linear relationships from high dimensional, noisy data inputs. This research effort uses novel data sources, namely NOAA’s High Resolution Rapid Refresh (HRRR) meteorology model, and a long-short-term-memory (LSTM) neural network to forecast and interpolate ozone values at high spatiotemporal resolution of 1 hour and 3 km. The accuracies of the LSTM models are analyzed using lagged ozone at various forecast horizons and across the varying geographies of eleven ground sensors. I use Denver, Colorado as my study area due to its long-standing GLO pollution problem and relatively high density of EPA ozone monitoring stations.

Recommended citation: Keenan, William J., "A Geospatial and Machine Learning Framework for Forecasting Ground Level Ozone Pollution" (2024). Electronic Theses and Dissertations. 2398. https://digitalcommons.du.edu/etd/2398 https://digitalcommons.du.edu/etd/2398