An official website of the United States government
Abstract The Ensemble Streamflow Prediction (ESP) framework combines a probabilistic forecast structure with process‐based models for water supply predictions. However, process‐based models require computationally intensive parameter estimation, increasing uncertainties and limiting usability. Motivated by the strong performance of deep learning models, we seek to assess whether the Long Short‐Term Memory (LSTM) model can provide skillful forecasts and replace process‐based models within the ESP framework. Given challenges inimplicitlycapturing snowpack dynamics within LSTMs for streamflow prediction, we also evaluated the added skill ofexplicitlyincorporating snowpack information to improve hydrologic memory representation. LSTM‐ESPs were evaluated under four different scenarios: one excluding snow and three including snow with varied snowpack representations. The LSTM models were trained using information from 664 GAGES‐II basins during WY1983–2000. During a testing period, WY2001–2010, 80% of basins exhibited Nash‐Sutcliffe Efficiency (NSE) above 0.5 with a median NSE of around 0.70, indicating satisfactory utility in simulating seasonal water supply. LSTM‐ESP forecasts were then tested during WY2011–2020 over 76 western US basins with operational Natural Resources Conservation Services (NRCS) forecasts. A key finding is that in high snow regions, LSTM‐ESP forecasts using simplified ablation assumptions performed worse than those excluding snow, highlighting that snow data do not consistently improve LSTM‐ESP performance. However, LSTM‐ESP forecasts that explicitly incorporated past years' snow accumulation and ablation performed comparably to NRCS forecasts and better than forecasts excluding snow entirely. Overall, integrating deep learning within an ESP framework shows promise and highlights important considerations for including snowpack information in forecasting.
more »
« less
Exploring the decision-making process in model development: focus on the Arctic snowpack
Abstract. The Arctic poses many challenges for Earth system and snow physics models, which are commonly unable to simulate crucial Arctic snowpack processes,such as vapour gradients and rain-on-snow-induced ice layers. These limitations raise concerns about the current understanding of Arctic warming and its impact on biodiversity, livelihoods, permafrost, and the global carbon budget. Recognizing that models are shaped by human choices, 18 Arctic researchers were interviewed to delve into the decision-making process behind model construction. Although data availability, issues of scale, internal model consistency, and historical and numerical model legacies were cited as obstacles to developing an Arctic snowpack model, no opinion was unanimous. Divergences were not merely scientific disagreements about the Arctic snowpack but reflected the broader research context. Inadequate and insufficient resources, partly driven by short-term priorities dominating research landscapes, impeded progress. Nevertheless, modellers were found to be both adaptable to shifting strategic research priorities – an adaptability demonstrated by the fact that interdisciplinary collaborations were the key motivation for model development – and anchored in the past. This anchoring and non-epistemic values led to diverging opinions about whether existing models were “good enough” and whether investing time and effort to build a new model was a useful strategy when addressing pressing research challenges. Moving forward, we recommend that both stakeholders and modellers be involved in future snow model intercomparison projects in order to drive developments that address snow model limitations currently impeding progress in various disciplines. We also argue for more transparency about the contextual factors that shape research decisions. Otherwise, the reality of our scientific process will remain hidden, limiting the changes necessary to our research practice.
more »
« less
- Award ID(s):
- 1931332
- PAR ID:
- 10563054
- Publisher / Repository:
- The Cryosphere
- Date Published:
- Journal Name:
- The Cryosphere
- Volume:
- 18
- Issue:
- 10
- ISSN:
- 1994-0424
- Page Range / eLocation ID:
- 4671 to 4686
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Abstract Predicting winter flooding is critical to protecting people and securing water resources in California’s Sierra Nevada. Rain-on-snow (ROS) events are a common cause of widespread flooding and are expected to increase in both frequency and magnitude with anthropogenic climate change in this region. ROS flood severity depends on terrestrial water input (TWI), the sum of rain and snowmelt that reaches the land surface. However, an incomplete understanding of the processes that control the flow and refreezing of liquid water in the snowpack limits flood prediction by operational and research models. We examine how antecedent snowpack conditions alter TWI during 71 ROS events between water years 1981 and 2019. Observations across a 500-m elevation gradient from the Independence Creek catchment were input into SNOWPACK, a one-dimensional, physically based snow model, initiated with the Richards equation and calibrated with collocated snow pillow observations. We compare observed “historical” and “scenario” ROS events, where we hold meteorologic conditions constant but vary snowpack conditions. Snowpack variables include cold content, snow density, liquid water content, and snow water equivalent. Results indicate that historical events with TWI > rain are associated with the largest observed streamflows. A multiple linear regression analysis of scenario events suggests that TWI is sensitive to interactions between snow density and cold content, with denser (>0.30 g cm−3) and colder (<−0.3 MJ of cold content) snowpacks retaining >50 mm of TWI. These results highlight the importance of hydraulic limitations in dense snowpacks and energy limitations in warm snowpacks for retaining liquid water that would otherwise be available as TWI for flooding. Significance StatementThe purpose of this study is to understand how the snowpack modulates quantities of water that reach the land surface during rain-on-snow (ROS) events. While the amount of near-term storm rainfall is reasonably predicted by meteorologists, major floods associated with ROS are more difficult to predict and are expected to increase in frequency. Our key findings are that liquid water inputs to the land surface vary with snowpack characteristics, and although many hydrologic models incorporate snowpack cold content and density to some degree, the complexity of ROS events justifies the need for additional observations to improve operational forecasting model results. Our findings suggest additional comparisons between existing forecasting models and those that physically represent the snowpack, as well as field-based observations of cold content and density and liquid water content, would be useful follow-up investigations.more » « less
-
Abstract. Sea salt aerosols play an important role in the radiationbudget and atmospheric composition over the Arctic, where the climate israpidly changing. Previous observational studies have shown that Arctic sea ice leads are an important source of sea salt aerosols, and modeling efforts have also proposed blowing snow sublimation as a source. In this study,size-resolved atmospheric particle number concentrations and chemicalcomposition were measured at the Arctic coastal tundra site ofUtqiaġvik, Alaska, during spring (3 April–7 May 2016). Blowing snow conditions were observed during 25 % of the 5-week study period andwere overpredicted by a commonly used blowing snow parameterization based solely on wind speed and temperature. Throughout the study, open leads werepresent locally. During periods when blowing snow was observed, significantincreases in the number concentrations of 0.01–0.06 µm particles(factor of 6, on average) and 0.06–0.3 µm particles (67 %, on average) and a significant decrease (82 %, on average) in 1–4 µmparticles were observed compared to low wind speed periods. These size distribution changes were likely caused by the generation of ultrafineparticles from leads and/or blowing snow, with scavenging of supermicronparticles by blowing snow. At elevated wind speeds, both submicron andsupermicron sodium and chloride mass concentrations were enhanced,consistent with wind-dependent local sea salt aerosol production. Atmoderate wind speeds below the threshold for blowing snow as well as during observed blowing snow, individual sea spray aerosol particles were measured.These individual salt particles were enriched in calcium relative to sodiumin seawater due to the binding of this divalent cation with organic matter in the sea surface microlayer and subsequent enrichment during seawaterbubble bursting. The chemical composition of the surface snowpack alsoshowed contributions from sea spray aerosol deposition. Overall, theseresults show the contribution of sea spray aerosol production from leads onboth aerosols and the surface snowpack. Therefore, if blowing snowsublimation contributed to the observed sea salt aerosol, the snow beingsublimated would have been impacted by sea spray aerosol deposition rather than upward brine migration through the snowpack. Sea spray aerosol production from leads is expected to increase, with thinning and fracturingof sea ice in the rapidly warming Arctic.more » « less
-
A Lagrangian Snow‐Evolution System for Sea‐Ice Applications (SnowModel‐LG): Part I—Model DescriptionAbstract A Lagrangian snow‐evolution model (SnowModel‐LG) was used to produce daily, pan‐Arctic, snow‐on‐sea‐ice, snow property distributions on a 25 × 25‐km grid, from 1 August 1980 through 31 July 2018 (38 years). The model was forced with NASA's Modern Era Retrospective‐Analysis for Research and Applications‐Version 2 (MERRA‐2) and European Centre for Medium‐Range Weather Forecasts (ECMWF) ReAnalysis‐5th Generation (ERA5) atmospheric reanalyses, and National Snow and Ice Data Center (NSIDC) sea ice parcel concentration and trajectory data sets (approximately 61,000, 14 × 14‐km parcels). The simulations performed full surface and internal energy and mass balances within a multilayer snowpack evolution system. Processes and features accounted for included rainfall, snowfall, sublimation from static‐surfaces and blowing‐snow, snow melt, snow density evolution, snow temperature profiles, energy and mass transfers within the snowpack, superimposed ice, and ice dynamics. The simulations produced horizontal snow spatial structures that likely exist in the natural system but have not been revealed in previous studies spanning these spatial and temporal domains. Blowing‐snow sublimation made a significant contribution to the snowpack mass budget. The superimposed ice layer was minimal and decreased over the last four decades. Snow carryover to the next accumulation season was minimal and sensitive to the melt‐season atmospheric forcing (e.g., the average summer melt period was 3 weeks or 50% longer with ERA5 forcing than MERRA‐2 forcing). Observed ice dynamics controlled the ice parcel age (in days), and ice age exerted a first‐order control on snow property evolution.more » « less
-
null (Ed.)Snow cover plays a key role in the water and energy budgets over cold regions. Understanding and parameterizing water and heat exchange over snow surfaces in hydrologic models remains a major challenge. An innovative approach based on the theory of maximum entropy production (MEP) was developed for modeling energy budgets for snow-covered surfaces. This study generalizes the MEP model to simulate surface water vapor (latent heat) fluxes over an entire snowpack lifecycle, including snow accumulation and melting during the early growing season. The expanded MEP model combines soil evaporation, canopy transpiration, and snow sublimation to evaluate snow water loss during the lifecycle of the snowpack. Two hypotheses are tested: (1) sublimation becomes negligible during snowmelt when snowpack is isothermal (0°C) and (2) transpiration is progressively activated as a function of the air temperature during vegetation awakening. The proposed approach is shown to be effective for modeling the total surface water vapor fluxes over the snowpack's lifecycle. Both the hypotheses are supported by field observations.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Journal Article:
- https://doi.org/10.5194/tc-18-4671-2024
