Hydrology and Earth System Sciences, Volume 25, Issue 4

Anthology ID:
Copernicus GmbH
Bib Export formats:

pdf bib
Summary and synthesis of Changing Cold Regions Network (CCRN) research in the interior of western Canada – Part 2: Future change in cryosphere, vegetation, and hydrology
C. M. DeBeer | H. S. Wheater | John W. Pomeroy | Alan Barr | Jennifer L. Baltzer | Jill F. Johnstone | M. R. Turetsky | Ronald E. Stewart | Masaki Hayashi | Garth van der Kamp | Shawn J. Marshall | Elizabeth M. Campbell | Philip Marsh | Sean K. Carey | William L. Quinton | Yanping Li | Saman Razavi | Aaron Berg | Jeffrey J. McDonnell | Christopher Spence | Warren Helgason | A. M. Ireson | T. Andrew Black | Mohamed Elshamy | Fuad Yassin | Bruce Davison | Allan Howard | Julie M. Thériault | Kevin Shook | M. N. Demuth | Alain Pietroniro

Abstract. The interior of western Canada, like many similar cold mid- to high-latitude regions worldwide, is undergoing extensive and rapid climate and environmental change, which may accelerate in the coming decades. Understanding and predicting changes in coupled climate–land–hydrological systems are crucial to society yet limited by lack of understanding of changes in cold-region process responses and interactions, along with their representation in most current-generation land-surface and hydrological models. It is essential to consider the underlying processes and base predictive models on the proper physics, especially under conditions of non-stationarity where the past is no longer a reliable guide to the future and system trajectories can be unexpected. These challenges were forefront in the recently completed Changing Cold Regions Network (CCRN), which assembled and focused a wide range of multi-disciplinary expertise to improve the understanding, diagnosis, and prediction of change over the cold interior of western Canada. CCRN advanced knowledge of fundamental cold-region ecological and hydrological processes through observation and experimentation across a network of highly instrumented research basins and other sites. Significant efforts were made to improve the functionality and process representation, based on this improved understanding, within the fine-scale Cold Regions Hydrological Modelling (CRHM) platform and the large-scale Modélisation Environmentale Communautaire (MEC) – Surface and Hydrology (MESH) model. These models were, and continue to be, applied under past and projected future climates and under current and expected future land and vegetation cover configurations to diagnose historical change and predict possible future hydrological responses. This second of two articles synthesizes the nature and understanding of cold-region processes and Earth system responses to future climate, as advanced by CCRN. These include changing precipitation and moisture feedbacks to the atmosphere; altered snow regimes, changing balance of snowfall and rainfall, and glacier loss; vegetation responses to climate and the loss of ecosystem resilience to wildfire and disturbance; thawing permafrost and its influence on landscapes and hydrology; groundwater storage and cycling and its connections to surface water; and stream and river discharge as influenced by the various drivers of hydrological change. Collective insights, expert elicitation, and model application are used to provide a synthesis of this change over the CCRN region for the late 21st century.

pdf bib
Rainfall–runoff prediction at multiple timescales with a single Long Short-Term Memory network
Martin Gauch | Frederik Kratzert | Daniel Klotz | Grey Nearing | Jimmy Lin | Sepp Hochreiter

Abstract. Long Short-Term Memory (LSTM) networks have been applied to daily discharge prediction with remarkable success. Many practical applications, however, require predictions at more granular timescales. For instance, accurate prediction of short but extreme flood peaks can make a lifesaving difference, yet such peaks may escape the coarse temporal resolution of daily predictions. Naively training an LSTM on hourly data, however, entails very long input sequences that make learning difficult and computationally expensive. In this study, we propose two multi-timescale LSTM (MTS-LSTM) architectures that jointly predict multiple timescales within one model, as they process long-past inputs at a different temporal resolution than more recent inputs. In a benchmark on 516 basins across the continental United States, these models achieved significantly higher Nash–Sutcliffe efficiency (NSE) values than the US National Water Model. Compared to naive prediction with distinct LSTMs per timescale, the multi-timescale architectures are computationally more efficient with no loss in accuracy. Beyond prediction quality, the multi-timescale LSTM can process different input variables at different timescales, which is especially relevant to operational applications where the lead time of meteorological forcings depends on their temporal resolution.