Predicting streamflow drought in the conterminous United States using machine learning and a donor-gage approach, 1982–2020

Aaron Joseph Heldmyer; Roy Sando; Caelan Simeone; Michael Wieczorek; Scott Douglas Hamshaw; Phillip J. Goodling; Ryan R. McShane; Jeremy Alejandro Diaz; David Watkins; Bryce Alan Pulver; Apoorva Shastry; Konrad C. Hafen; John C. Hammond

doi:10.5194/hess-30-5925-2026

USGS · 70283423

Predicting streamflow drought in the conterminous United States using machine learning and a donor-gage approach, 1982–2020

Aaron Joseph Heldmyer·Roy Sando·Caelan Simeone·Michael Wieczorek·Scott Douglas Hamshaw·Phillip J. Goodling·Ryan R. McShane·Jeremy Alejandro Diaz·David Watkins·Bryce Alan Pulver·Apoorva Shastry·Konrad C. Hafen·John C. Hammond

View original source ↗Export metadata ↓

Abstract

Drought is a highly consequential natural disaster that may likely increase in both severity and extent across the conterminous United States (CONUS). The mechanisms affecting the propagation of drought from the atmosphere to streamflow are complex and interactive, making the prediction of streamflow drought difficult with current modeling approaches. Machine learning is an emerging tool in the field of hydrology that may be well-suited to prediction of streamflow drought across large and topographically diverse areas. Here, we train and analyze 3198 random forest models at U.S. Geological Survey streamgages to understand common meteorological drivers of streamflow drought and to define physiographic characteristics of basins sensitive to these drivers. We also develop a novel dynamic regionalization approach using donor gages to predict daily streamflow drought at pseudo-ungaged locations. Our results show that teleconnections, temperature, evaporative demand, and snow–water equivalent are important drivers of streamflow drought in the West, Southwest, and Northern Rocky Mountains (Northern Rockies) regions of the United States, and precipitation and soil moisture are primary drivers of streamflow drought in the Northeast, Southeast, and the Northwest regions. Prediction using dynamic regionalization shows comparable performance to at-site models. This method may be applicable to other hydrologic prediction problems requiring transferability to ungaged locations.

Explore related subjects

Hydrology and Earth System Sciences Article

Source-reported bounding extent: 25.08° to 49.38905° latitude; -124.68721° to -66.96466° longitude. This indicates report coverage, not an exact sampling location. View area on OpenStreetMap.

Keep this discovery

Explore connections, maps & timelines

BibTeX RIS

Aaron Joseph Heldmyer, Roy Sando, Caelan Simeone, Michael Wieczorek, Scott Douglas Hamshaw, Phillip J. Goodling, Ryan R. McShane, Jeremy Alejandro Diaz, David Watkins, Bryce Alan Pulver, Apoorva Shastry, Konrad C. Hafen, John C. Hammond. 2026-09-23. Predicting streamflow drought in the conterminous United States using machine learning and a donor-gage approach, 1982–2020. https://doi.org/10.5194/hess-30-5925-2026

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related USGS reports

USGS2026

Proposing sources for discrete groundwater discharges to patterned pools in three regional raised northern peat bogs

Raised northern peat bogs are generally assumed to be entirely precipitation-fed (ombrogenous), suggesting they are void of groundwater (minerogenous) inflow from underlying sediments. Patterned pools in raised bogs have often been attributed to surficial flow filling depressions along the peat surface that produces subtle differences in peat pore-water chemistry. However, we present evidence that certain patterned pools may be partially fed by localized upwelling of minerogenous groundwater from underlying glacial sediments in three northern peat bogs of Maine, USA. Underlying permeable glacial deposits, embedded in hydraulically confining glacio-marine clay deposits, were delineated using ground-penetrating radar and transient electromagnetic surveys. Paired point measurements of temperature and specific conductance (SpC) were surveyed around pools, and statistical relationships indicative of groundwater upwelling were established. Uncrewed aerial systems (UAS) thermal infrared (TIR) mapping was conducted, augmented by handheld TIR imaging, to examine characteristic groundwater temperatures in cold and warm seasons across pool surfaces. Surface water samples were acquired to assess the relationship between SpC / temperature signals and elevated iron and manganese concentrations that could be indicative of glacial aquifer sources. The combined datasets present evidence for localized upwelling in pools underlain by glacial structures, and the possibility of minerogenous groundwater contributions. Upwelling through the peat matrix is possibly partially facilitated by macropore, “peat pipe” features that serve as preferential flowpaths of varying lengths to the surface. Such upwelling could drive a positive feedback loop where elevated concentrations of ionic constituents repeatedly dilate the peat matrix and/or terminal electron acceptors enhance anaerobic respiration and microbial activity, accelerating the humification of peat around patterned pools and potentially magnifying carbon loss.

Maine↗

USGS2025

Technical note: A low-cost approach to monitoring relative streamflow dynamics in small headwater streams using time lapse imagery and a deep learning model

Despite their ubiquity and importance as freshwater habitat, small headwater streams are under-monitored by existing stream gage networks. To address this gap, we describe a low-cost, non-contact, and low-effort method that enables organizations to monitor relative streamflow dynamics in small headwater streams. The method uses a camera to capture repeat images of the stream from a fixed position. A person then annotates pairs of images, in each case indicating which image has more apparent streamflow or indicating equal flow if no difference is discernible. A deep learning modeling framework called streamflow rank estimation (SRE) is then trained on the annotated image pairs and applied to rank all images from highest to lowest apparent streamflow. From this result a relative hydrograph can be derived. We found that our modeled relative hydrograph dynamics matched the observed hydrograph dynamics well for 11 cameras at 8 streamflow sites in western Massachusetts. Higher performance was observed during the annotation period (median Kendall's Tau rank correlation of 0.75, with a range of 0.6–0.83) than after it (median Kendall's Tau of 0.59, with range 0.34–0.74). We found that annotation performance was generally consistent across the 11 camera sites and 2 individual annotators and was positively correlated with streamflow variability at a site. A scaling simulation determined that model performance improvements were limited after 1000 annotation pairs. Our model's estimates of relative flow, while not equivalent to absolute flow, may still be useful for many applications, such as ecological modeling and calculating event-based hydrological statistics (e.g., the number of out-of-bank floods). We anticipate that this method will be a valuable tool to extend existing stream monitoring networks and provide new insights on dynamic headwater systems.

Massachusetts↗

USGS2025

Interrogating process deficiencies in large-scale hydrologic models with interpretable machine learning

Large-scale hydrologic models are increasingly being developed for operational use in the forecasting and planning of water resources. However, the predictive strength of such models depends on how well they resolve various functions of catchment hydrology, which are influenced by gradients in climate, topography, soils, and land use. Most assessments of hydrologic model uncertainty have been limited to traditional statistical methods. Here, we present a proof-of-concept approach that uses interpretable machine learning techniques to provide post hoc assessment of model sensitivity and process deficiency in hydrologic models. We train a random forest model to predict the Kling–Gupta efficiency (KGE) of National Water Model (NWM) and National Hydrologic Model (NHM) streamflow predictions for 4383 stream gauges in the conterminous United States. Thereafter, we explain the local and global controls that 48 catchment attributes exert on KGE prediction using interpretable Shapley values. Overall, we find that soil water content is the most impactful feature controlling successful model performance, suggesting that soil water storage is difficult for hydrologic models to resolve, particularly for arid locations. We identify nonlinear thresholds beyond which predictive performance decreases for NWM and NHM. For example, soil water content less than 210 mm, precipitation less than 900 mm yr −1 , road density greater than 5 km km −2 , and lake area percent greater than 10 % contributed to lower KGE values. These results suggest that improvements in how these influential processes are represented could result in the largest increases in NWM and NHM predictive performance. This study demonstrates the utility of interrogating process-based models using data-driven techniques, which has broad applicability and potential for improving the next generation of large-scale hydrologic models.

conterminous United States↗