More Like this

1. Projecting Flood Frequency Curves Under Near‐Term Climate Change

   https://doi.org/10.1029/2021WR031246  

   Awasthi, C.; Archfield, S. A.; Ryberg, K. R.; Kiang, J. E.; Sankarasubramanian, A. (August 2022, Water Resources Research)

   Abstract Flood‐frequency curves, critical for water infrastructure design, are typically developed based on a stationary climate assumption. However, climate changes are expected to violate this assumption. Here, we propose a new, climate‐informed methodology for estimating flood‐frequency curves under non‐stationary future climate conditions. The methodology develops an asynchronous, semiparametric local‐likelihood regression (ASLLR) model that relates moments of annual maximum flood to climate variables using the generalized linear model. We estimate the first two marginal moments (MM) – the mean and variance – of the underlying log‐Pearson Type‐3 distribution from the ASLLR with the monthly rainfall and temperature as predictors. The proposed methodology, ASLLR‐MM, is applied to 40 U.S. Geological Survey streamgages covering 18 water resources regions across the conterminous United States. A correction based on the aridity index was applied on the estimated variance, after which the ASLLR‐MM approach was evaluated with both historical (1951–2005) and projected (2006–2035, under RCP4.5 and RCP8.5) monthly precipitation and temperature from eight Global Circulation Models (GCMs) consisting of 39 ensemble members. The estimated flood‐frequency quantiles resulting from the ASLLR‐MM and GCM members compare well with the flood‐frequency quantiles estimated using the historical period of observed climate and flood information for humid basins, whereas the uncertainty in model estimates is higher in arid basins. Considering additional atmospheric and land‐surface conditions and a multi‐level model structure that includes other basins in a region could further improve the model performance in arid basins. 

   more » « less
2. Recent (1980 to 2015) Trends and Variability in Daily‐to‐Interannual Soluble Iron Deposition from Dust, Fire, and Anthropogenic Sources

   https://doi.org/10.1029/2020GL089688  

   Hamilton, Douglas S.; Scanza, Rachel A.; Rathod, Sagar D.; Bond, Tami C.; Kok, Jasper F.; Li, Longlei; Matsui, Hitoshi; Mahowald, Natalie M. (September 2020, Geophysical Research Letters)

   Abstract The iron cycle is a key component of the Earth system. Yet how variable the atmospheric flux of soluble (bioaccessible) iron into oceans is, and how this variability is modulated by human activity and a changing climate, is not well known. For the first time, we characterize Satellite Era (1980 to 2015) daily‐to‐interannual modeled soluble iron emission and deposition variability from both pyrogenic (fires and anthropogenic combustion) and dust sources. Statistically significant emission trends exist: dust iron decreases, fire iron slightly increases, and anthropogenic iron increases. A strong temporal variability in deposition to ocean basins is found, and, for most regions, dust iron dominates the absolute deposition magnitude, fire iron is an important contributor to temporal variability, and anthropogenic iron imposes a significant increasing trend. Quantifying soluble iron daily‐to‐interannual deposition variability from all major iron sources, not only dust, will advance quantification of changes in marine biogeochemistry in response to the continuing human perturbation to the Earth System. 

   more » « less
3. Estimating Post‐Fire Flood Infrastructure Clogging and Overtopping Hazards

   https://doi.org/10.1029/2023WR036522  

   Jong‐Levinger, Ariane; Houston, Douglas; Sanders, Brett_F (August 2024, Water Resources Research)

   Abstract Cycles of wildfire and rainfall produce sediment‐laden floods that pose a hazard to development and may clog or overtop protective infrastructure, including debris basins and flood channels. The compound, post‐fire flood hazards associated with infrastructure overtopping and clogging are challenging to estimate due to the need to account for interactions between sequences of wildfire and storm events and their impact on flood control infrastructure over time. Here we present data sources and calibration methods to estimate infrastructure clogging and channel overtopping hazards on a catchment‐by‐catchment basis using the Post‐Fire Flood Hazard Model (PF2HazMo), a stochastic modeling approach that utilizes continuous simulation to resolve the effects of antecedent conditions and system memory. Publicly available data sources provide parameter ranges needed for stochastic modeling, and several performance measures are considered for model calibration. With application to three catchments in southern California, we show that PF2HazMo predicts the median of the simulated distribution of peak bulked flows within the 95% confidence interval of observed flows, with an order of magnitude range in bulked flow estimates depending on the performance measure used for calibration. Using infrastructure overtopping data from a post‐fire wet season, we show that PF2HazMo accurately predicts the number of flood channel exceedances. Model applications to individual watersheds reveal where infrastructure is undersized to contain present‐day and future overtopping hazards based on current design standards. Model limitations and sources of uncertainty are also discussed. 

   more » « less
4. Tracking the risk of a deployed model and detecting harmful distribution shifts

   Podkopaev, Aleksandr; Ramdas, Aaditya (April 2022, Tenth International Conference on Learning Representations)

   When deployed in the real world, machine learning models inevitably encounter changes in the data distribution, and certain—but not all—distribution shifts could result in significant performance degradation. In practice, it may make sense to ignore benign shifts, under which the performance of a deployed model does not degrade substantially, making interventions by a human expert (or model retraining) unnecessary. While several works have developed tests for distribution shifts, these typically either use non-sequential methods, or detect arbitrary shifts (benign or harmful), or both. We argue that a sensible method for firing off a warning has to both (a) detect harmful shifts while ignoring benign ones, and (b) allow continuous monitoring of model performance without increasing the false alarm rate. In this work, we design simple sequential tools for testing if the difference between source (training) and target (test) distributions leads to a significant increase in a risk function of interest, like accuracy or calibration. Recent advances in constructing time-uniform confidence sequences allow efficient aggregation of statistical evidence accumulated during the tracking process. The designed framework is applicable in settings where (some) true labels are revealed after the prediction is performed, or when batches of labels become available in a delayed fashion. We demonstrate the efficacy of the proposed framework through an extensive empirical study on a collection of simulated and real datasets. 

   more » « less
5. Trend Analyses of Baseflow and BFI for Undisturbed Watersheds in Michigan—Constraints from Multi-Objective Optimization

   https://doi.org/10.3390/w13040564  

   Hagedorn, Benjamin; Meadows, Christina (February 2021, Water)

   null (Ed.)

   Documenting how ground- and surface water systems respond to climate change is crucial to understanding water resources, particularly in the U.S. Great Lakes region, where drastic temperature and precipitation changes are observed. This study presents baseflow and baseflow index (BFI) trend analyses for 10 undisturbed watersheds in Michigan using (1) multi-objective optimization (MOO) and (2) modified Mann–Kendall (MK) tests corrected for short-term autocorrelation (STA). Results indicate a variability in mean baseflow (0.09–8.70 m3/s) and BFI (67.9–89.7%) that complicates regional-scale extrapolations of groundwater recharge. Long-term (>60 years) MK trend tests indicate a significant control of total precipitation (P) and snow- to rainfall transitions on baseflow and BFI. In the Lower Peninsula Rifle River watershed, increasing P and a transition from snow- to rainfall has increased baseflow at a lower rate than streamflow; an overall pattern that may contribute to documented flood frequency increases. In the Upper Peninsula Ford River watershed, decreasing P and a transition from rain- to snowfall had no significant effects on baseflow and BFI. Our results highlight the value of an objectively constrained BFI parameter for shorter-term (<50 years) hydrologic trend analysis because of a lower STA susceptibility. 

   more » « less