Ice jam formation, breakup and prediction methods based on hydroclimatic data using artificial intelligence: A review

https://doi.org/10.1016/j.coldregions.2020.103032Get rights and content

Abstract

In cold regions, the high occurrence of ice jams results in severe flooding and significant damage caused by a rapid rise in water levels upstream of ice jams. These floods can be critical hydrological and hydraulic events and be a major concern for citizens, authorities, insurance companies and government agencies. In the past twenty years, several studies have been conducted in ice jam modelling and forecasting, and it has been found that predicting ice jam formation and breakup is challenging, due to the complexity of the interactions between the hydroclimatic variables leading to these processes. At this time, several mathematical models have been developed to predict breakup processes. The current methods of breakup prediction are highly empirical and site-specific. The information on the progress of the methods and the variables used to predict the occurrence, severity, and timing of the breakup ice jams still remains limited. This study summarizes the different processes contributing to ice jam formation and breakup, the various existing ice jam prediction models, and their potential and limitations regarding the improvement in ice jam predictions. An overview of the application of artificial neural networks and fuzzy logic systems in ice-related problems is presented. Genetic programming is also explained as a possible mean for ice-related problems. Although genetic programming shows promising results in hydrological modelling, it has not yet been used in ice-related problems. The review of literature highlights that data-driven and machine learning techniques provide promising means in predicting ice jams with better confidence, but more scientific research is needed.

Introduction

In the past twenty years, flooding has been the most frequent and costly natural hazard worldwide, accounting for 47% of all recorded events (Wahlstrom and Guha-Sapir, 2015). Floods may result from the swelling of rivers due to intense rainfall or snowmelt, the failure of a natural or artificial dam, or to backflow due to jamming caused by wood debris or ice. Ice jams cause some of the most devastating and recurrent flood events in the northern rivers (Lagadec et al., 2015; Lindenschmidt et al., 2018; Pawłowski, 2016).

Almost 60% of rivers in the northern hemisphere experience significant effects of river ice (Prowse, 2006). River ice plays a vital role in the hydrology of cold regions, notably Canada, northern USA, Russia, northern Europe (especially Nordic countries), Japan, Korea, and China. In Canada, moving or stationary ice exists in almost every river over periods ranging from days to many months (Beltaos and Prowse, 2001). For many cold-region rivers, ice freeze-up, breakup, and jamming occur each year, causing a multitude of socio-economic impacts such as flooding, endangering human safety, damaging property and infrastructure, impeding navigation, inhibiting hydropower generation, as well as morphological and biological effects such as scouring and erosion of the river bed, sediment transport, biochemical exchanges, and aquatic productivity and diversity (Beltaos, 2002; Beltaos and Prowse, 2001; Carr and Vuyovich, 2014; Chokmani et al., 2007; Mahabir et al., 2006; Massie et al., 2002; Newton et al., 2017).

Ice jam formation is a sudden natural phenomenon in the rivers of cold regions that usually occurs during mid-winter (in the temperate areas) and spring ice cover breakups and can cause a rapid rise in water levels and severe flooding upstream of ice jams (Das et al., 2018; Guo et al., 2018). It is caused by the constriction of the channel by natural or anthropogenic obstacles such as existing ice blocks, river narrowing, slope break, meanders and bridges. To minimize the damages caused by these floods, it is becoming necessary to increase response time by developing efficient ice jam forecasting tools. Although floods are generally forecastable, ice-induced floods associated with both the formation and release of ice jams are highly variable and challenging to predict (Lindenschmidt et al., 2018; Rokaya et al., 2018).

Ice jam prediction is a problem of major practical significance, which is only partially resolved in specific cases, and it remains arduous to predict whether and when an ice jam will form at a particular location. However, there have been a large number of studies focused on other ice-related issues such as modelling the water levels caused by an ice jam (Beltaos et al., 2012). Although relatively rare and limited, the development of data-based models to predict the occurrence of ice jams is considered essential given the need to protect residents in at-risk areas.

There are three critical components of ice jams; occurrence, severity (i.e. water level) and timing of breakup. There is no analytical solution available to predict these aspects of ice jams due to the complexity of the interactions between hydrologic, hydraulic, and meteorological processes (Mahabir et al., 2006; Massie et al., 2002; McDonald et al., 2002; Zhao, 2012). Hence, to date, a small number of ice jam prediction models have been developed employing highly site-specific, empirical as well as statistically based methods such as threshold methods, multi-regression models, logistic regression models, and discriminant function analysis (Barnes-Svarney and Montz, 1985; Mahabir et al., 2006; Massie et al., 2002; White, 2003; White and Daly, 2002; Zhao, 2012). However, these simple models are not entirely successful due to the complexity of the interactions between the hydroclimatic variables involved. White (2003) reviews prediction methods for the occurrence of breakup ice jams including empirical models, threshold models, and discriminant function analysis, and also includes a short description of the application of artificial intelligence models to the occurrence of breakup ice jams. She concludes that existing models are limited regarding their transferability to other locations, and there is no better performing model in ice jam forecasting. White and Daly (2002) reviewed current methods and proposed a three-step process using a discriminant function analysis to predict breakup ice jams.

Numerical models are capable of solving a wide variety of river ice problems. For instance, 1-D and less commonly, 2-D models have been used to predict the thickness of ice jams and the corresponding water levels. Some numerical models have been developed to simulate ice jam induced flooding, including ICEJAM, RIVJAM, HEC-RAS, ICEPRO, ICESIM, RIVICE, River2D, and DYNARICE models (Beltaos, 2008; Brayall and Hicks, 2012). However, these models show limitations in predicting ice jam occurrence. Also, the mathematical formulations used in these models are complex and require many parameters related to upstream and downstream gauging stations that are often unavailable to support and calibrate the model, as they are challenging to measure. Too many assumptions and simplifications corresponding to these parameters may degrade model accuracy and limit their transferability to other rivers (Chen and Ji, 2005).

Given the advances in the development of artificial intelligence systems and their successful application to recent hydrologic problems, certain recommendations can be formulated to better employ these tools when predicting the occurrence and severity of ice jams. However, information on the progress of the applied prediction methods and the variables influencing the occurrence, severity, and timing of the breakup remains limited. Hence, the present review paper summarizes the progress of artificial intelligence systems and their application to river-ice problems. A theoretical background on the hydroclimatic factors influencing the occurrence of ice jams is presented. This is followed by an overview of artificial neural networks, genetic programming, and fuzzy logic as prediction tools of ice jams. Then, a discussion on the application of data-driven models (DDMs) in ice-related problems as well as a comparison between these models is given. The review concludes with a brief summary and future research directions.

Section snippets

Theoretical background on types of ice jams and flooding

An ice jam is a stationary accumulation of frazil ice or randomly oriented ice blocks that resists the incoming river flow, raising water levels and overflowing the river banks (Beltaos, 1995, Beltaos, 2008). Beltaos (1995) provides comprehensive information on different types of ice jams. According to the International Association for Hydro-Environment Engineering and Research (IAHR), ice jams can be classified based on four criteria; dominant formation processes, vertical extent, state of

Hydroclimatic variables influencing the formation of ice jams

Forecasting an ice jam flood includes predicting the occurrence of the jam, the peak discharge during the presence of the jam, the jam location, and the resulting water levels. Forecasting the occurrence of an ice jam each year is challenging, and it is not yet possible to model the complex hydroclimatic interactions that lead to jam occurrence. Although variations of hydroclimatic factors may not be similar from one ice jam event to another in the same river section, there are some

Numerical ice-jam prediction models

Shen (2010) reviews the existing models of various river ice processes taking place during the winter, from freeze-up to breakup. Mathematical models are available for various river ice processes, including heat exchanges, water temperature, skim ice, frazil and anchor ice, border ice, freeze-up ice run and cover formation, thermal growth and decay of ice cover, surface ice jams, undercover transport and frazil jam, as well as comprehensive river ice models. Models of freeze-up ice runs and

A conceptual overview of data-driven methods

Ice jam studies can be categorized into three main types: (1) descriptive (2) physical, and (3) predictive and frequency modelling (Barnes-Svarney and Montz, 1985). Descriptive studies characterize the hydrological events involving ice jams, detailing their formation and breakup, while physical studies focus on identifying and mapping river sections with the potential risk of ice jams. The most complicated type of ice jam studies are predictive and frequency analysis. Descriptive and physical

Application of ANNs

The review of literature shows that ANNs have been successfully applied to predict river ice jam flooding (e.g., Mahabir et al., 2007), river ice breakup (e.g., Mahabir et al., 2006; McDonald et al., 2002; Zhao et al., 2012), timing of ice formation and breakup (e.g., Hu et al., 2008; Chen and Ji, 2005; Sun and Trevor, 2018a; Wang and Li, 2009; Wang et al., 2008), river ice thickness (Chokmani et al., 2007), water level (Sun and Trevor, 2018b), and ice jam thickness (Wang et al., 2010). They

Discussion and conclusions

Predicting ice jam occurrence is important, as it gives an early warning of possible consequent flooding events in cold regions. There is no analytical method available to predict ice jam occurrence. Hence, to date, a small number of highly site specific empirical and statistical jam prediction methods have been developed, but they all resulted in a high rate of false-positive errors.

The literature review shows that many forces drive ice jams and many parameters (e.g., porosity and thickness of

Future directions

Despite the progress that has been made in forecasting river ice processes, this field still has much research potential, especially in the prediction of ice jam formation using hydroclimatic and geomorphological information. This needs comprehensive field observations, hydrometeorological, geomorphological, ice jam data collection and testing with robust models. Many useful numerical models have been developed to predict river ice jams and consequent floods, but this field still needs much

Declarataion of Competing Interest

None.

Acknowledgments

This work is part of the project “DAVE: A geospatial tool to better anticipate ice jams,” funded by the Canadian Safety and Security Program (CSSP), led by the Defence Research and Development Canada's Centre for Security Science (DRDC CSS).

References (86)

  • D.D. Massie et al.

    Application of neural networks to predict ice jam occurrence

    Cold Reg. Sci. Technol.

    (2002)
  • H.T. Shen

    Mathematical modelling of river ice processes

    Cold Reg. Sci. Technol.

    (2010)
  • H.T. Shen et al.

    SPH simulation of river ice dynamics

    J. Comput. Phys.

    (2000)
  • W. Sun

    River ice breakup timing prediction through stacking multi-type model trees

    Sci. Total Environ.

    (2018)
  • W. Sun et al.

    Combining k-nearest-neighbour models for annual peak breakup flow forecasting

    Cold Reg. Sci. Technol.

    (2017)
  • W. Sun et al.

    A stacking ensemble learning framework for annual river ice breakup dates

    J. Hydrol.

    (2018)
  • L. Zhao et al.

    Applicability of multilayer feed-forward neural networks to model the onset of river breakup

    Cold Reg. Sci. Technol.

    (2012)
  • S. Abdollahi et al.

    Daily mean streamflow prediction in perennial and non-perennial rivers using four data driven techniques

    Water Resour. Manag.

    (2017)
  • S. Appelbaum

    Nonstructural Flood Protection Study, Technical Reports

    (1982)
  • P.L. Barnes-Svarney et al.

    An ice jam prediction model as a tool in floodplain management

    Water Resour. Res.

    (1985)
  • S. Beltaos

    Numerical computation of river ice jams

    Can. J. Civ. Eng.

    (1993)
  • S. Beltaos

    River Ice Jams

    (1995)
  • S. Beltaos

    Effects of climate on midwinter ice jams

    Hydrol. Process.

    (2002)
  • S. Beltaos

    River ice breakup processes: recent advances and future directions

    Can. J. Civ. Eng.

    (2007)
  • S. Beltaos

    River Ice Formation

    (2013)
  • S. Beltaos et al.

    Climatic change and river ice breakup

    Can. J. Civ. Eng.

    (2003)
  • S. Beltaos et al.

    Climate impacts on extreme ice-jam events in Canadian rivers

    Hydrol. Sci. J.

    (2001)
  • S. Beltaos et al.

    River-ice hydrology in a shrinking cryosphere

    Hydrol. Process.

    (2009)
  • S. Beltaos et al.

    Ice jam modelling and field data collection for flood forecasting in the Saint John River, Canada

    Hydrol. Process.

    (2012)
  • M. Brayall et al.

    Applicability of 2-D modelling for forecasting ice jam flood levels in the Hay River Delta, Canada

    Can. J. Civ. Eng.

    (2012)
  • T. Buffin-Bélanger et al.

    Ice formation in small rivers

  • R.W. Carson et al.

    Tests of river ice jam models

  • R.W. Carson et al.

    Tests of river ice jam models–phase 2

  • R. Carson et al.

    Comparative testing of numerical models of river ice jams

    Canadian Journal of Civil Engineering

    (2011)
  • S.Y. Chen et al.

    Fuzzy optimization neural network approach for ice forecast in the Inner Mongolia reach of the yellow river

    Hydrol. Sci. J.

    (2005)
  • K. Chokmani et al.

    Estimation of river ice thickness using artificial neural networks

  • A. Das et al.

    Sustainable Ice-Jam Flood Management for Socio-Economic and Socio-Ecological Systems

    Water

    (2018)
  • DHI

    MIKE11 reference manual

    (2005)
  • R.W. Dudley et al.

    Trends in streamflow, river ice, and snowpack for coastal river basins in Maine during the 20th century (no. 2002–4245)

  • G.M. Flato et al.

    Calculation of ice jam profiles

    Proceedings, 4th Workshop on River Ice, Montreal, Paper C-3. CGU-HS Committee on River Ice Processes and the Environment, Edmonton, Canada.

    (1986)
  • I. Goodfellow et al.

    Deep Learning

    (2016)
  • N. Gulley et al.

    Fuzzy Logic Toolbox User’s Guide

    (1995)
  • Cited by (16)

    • PCA-based SVM classification for simulated ice floes in front of sluice gates

      2022, Polar Science
      Citation Excerpt :

      Combining PCA with the ML model can improve the accuracy and efficiency of the model (Chen and Jian, 2015; Noori et al., 2011). Therefore, the combination of SVM and PCA is an important technology to address problems under the influence of multiple factors, such as river ice phenomena (Madaeni et al., 2020; Strofer et al., 2019). The purpose of this study is to combine PCA and SVM models to form an integrated model.

    • Modelling of ice jam floods under past and future climates: A review

      2022, Journal of Hydrology X
      Citation Excerpt :

      Likewise, several other machine learning techniques have also been employed (e.g. Wang et al., 2010; Aleshin and Malygin, 2018). A comprehensive review on IJF modelling using artificial intelligence is available from Madaeni et al. (2020). Fuzzy logic models use if-then-else rules to define the relationship between inputs and outputs.

    • An automated procedure to map breaking river ice with C-band HH SAR data

      2021, Remote Sensing of Environment
      Citation Excerpt :

      From 2012 to 2016, Canada's federal government expended about C$111 million to evacuate First Nation communities at risk of flooding—it is reasonable to assume that ice jams were involved in ⅓ of the cases (Gerard and Davar, 1995; INAC, 2017). A range of literature demonstrates that river ice processes and characteristics represent an active field of study for scientists and engineers (e.g. Ashton, 1986; Beltaos, 1995; Beltaos, 2000; Beltaos, 2007; Buffin-Bélanger and Bergeron, 2011; Beltaos, 2013; Beltaos and Burrell, 2015; Morin et al., 2015; Ansari et al., 2017; Kovachis et al., 2017; Morse and Turcotte, 2018; Nafziger et al., 2019; Nyantekyi-Kwakye et al., 2019; Chen and She, 2020; Das and Lindenschmidt, 2020; Krylenko et al., 2020; Madaeni et al., 2020; Rokaya et al., 2020). According to Beltaos (2000), the breakup process represents the least understood aspect of river ice science.

    • Practical AI Cases for Solving ESG Challenges

      2023, Sustainability (Switzerland)
    View all citing articles on Scopus
    View full text