Network research focuses on the big questions: Where and when is permafrost thaw occurring in Canada and what are the hazards arising from such change?

The research is organized in five interwoven themes requiring a critical mass and diversity of expertise that no single research group or government agency has.

Background and objectives

Permafrost maps, even those representing current conditions, are based on prediction because permafrost cannot easily be observed remotely. For example, the 1995 Permafrost Map of Canada is based on delineating combinations of physiographic regions and climate that are known to predict the extent of permafrost and ground-ice content. This network aims to inform the adaptation to climate change by prototyping new knowledge products and simulation is a key element in this.

The objective of Theme 3 is to improve the accuracy and delivery of transient permafrost simulation so that its results can support stakeholder needs at local and national scales.

The aim of predicting changes in a future beyond human experience and scientific observation combined with the lack of permafrost data make process-based models more appropriate for future predictions than statistical models. Additionally, simulating variables related to impacts, for example subsidence rates or soil mechanical characteristics, and simulating them at a scale that is suitable for comparison with observations and for interpretations at specific locations is required to understand and support adaptations to climate change. A wide variety of simulation scales and strategies exist including one-dimensional columns, one-dimensional columns with parameterised lateral interaction, and partial and full three-dimensionality. Similarly, models differ with respect to their sophistication in representing the soil-vegetation-atmosphere continuum. These aspects are important for scientific progress. However, we intend to make accurate predictions for all of Canada and for this we are focusing on using a suite of one-dimensional models with complementary strengths in representing vegetation, snow and water – the most important drivers of permafrost characteristics and change. With these models, we build a framework of suitable inputs (meteorological variables, surface cover, subsurface materials) and to enable evaluation of simulation results with ground and remote observations. This is important because accurate predictions and quantification of uncertainties in remote areas of a large landmass depends on the ability to supply suitable inputs as much, or more, than the sophistication of the model used. Assigning considerable resources to the evaluation of model results will thereby increase our understanding of limitations in our modelling systems and allow us to prioritise further model refinements.

Claude Duguay

Theme 3 is jointly led by Claude Duguay (University of Waterloo) and Joe Melton (University of Victoria). Joe coordinates the modelling activities, including the quantification of confidence in model outputs. Claude supports the application and evaluation of model output using remote sensing derived products. By the end of the network, seven researchers will be trained under Theme 3, consisting of 2 MSc, 3 PhD.

Joe Melton

Joe Melton

Stephan Gruber has expertise in the measurement and simulation of permafrost and contributes to the evaluation of permafrost models and their use for stakeholders.  

Trevor Lantz studies the ecological impacts of thermokarst through remote-sensing and Traditional Knowledge and contributes Traditional Knowledge perspectives on past land-cover change.   

Bernhard Rabus has expertise in remote-sensing of surface subsidence and helps to evaluate simulated ground-ice loss.  

Pascale Roy-Léveillée is an expert in the measurement and simulation of thermokarst processes and contributes to the improved simulation of phenomena and feedbacks related to standing water. 

Oliver Sonnentag  investigates land surface-atmosphere interactions in high latitudes and supports the evaluation of simulated carbon fluxes with field-based and ABoVE data products. data products.

Sub-theme research

Most climate-driven permafrost (sub-)models have high complexity limiting progress in model refinement to incremental advances through individual projects. The theme 3 approach is to focus on improving only one climate model; the Canadian Land Surface Scheme including Biogeochemical Cycles (CLASSIC) which includes processes related to permafrost dynamics. It is reasonably well evaluated in cold-regions and forms the terrestrial component of the Canadian Earth System Model (CanESM) which is operated by Environment and Climate Change Canada (ECCC). As a result, the improvements in process representation generated by NSERC PermafrostNet will improve simulations of permafrost-climate interactions in the CanESM and, therefore, future climate scenarios for Canada and the world. This work with CLASSIC also creates an important linkage between the climate modelling community, field geocryologists, engineers and stakeholders.

Two sub-themes structure the research in Theme 3.  

This subtheme improves the representation of key processes related to permafrost dynamics in CLASSIC in order to better link long-term climate change not only with ground temperature change, but also with variables such as terrain subsidence or terrain wetting.

The lack of a representation of excess ground ice and its melt is a significant gap in many permafrost models. While excess ground ice has been incorporated into both site-level, regional and global models, the representation remains crude. Additionally, there is presently no dataset of excess ground ice suited for use as a model input and simulation results have been difficult to test because data on subsidence are sparse.

Sub-theme 3.1 interfaces with Theme 1 to use site-level data from PINGO to run simulations based on GRIP (Sub-theme 3.2) and works with Theme 4 to determine the most valuable model outputs from future scenarios for use in hazard assessment. One of our present projects introduces into CLASSIC plant functional types that are specific to boreal and Arctic permafrost regions. CLASSIC explicitly simulates the fluxes of carbon dioxide and methane between the land surface and atmosphere. It is well suited to determine the future carbon fluxes of the Canadian permafrost region whether dominated by enhanced plant productivity (uptake by the land surface) or dominated by aerobic (carbon dioxide) and anaerobic (methane) respiration by the thawing land surface. This is relevant for informing policy because climate feedbacks from natural wetland and permafrost feedbacks have been estimated to reduce permissible emission budgets of carbon dioxide from fossil fuels by 9–15%.

Rose Lefebvre (T3-MSc1)  

Title: Simulating land cover change and its influence on permafrost with CLASSIC.

Supervisors: Joe Melton and Oliver Sonnentag with Elyn Humphreys, and Gesa Meyer. 

The principal influences on ground heat changes in cold regions are vegetation, hydrology, snow cover and topography. The vegetation in permafrost affected regions varies from boreal to Arctic biomes. To simulate the influence of changing land cover, this project will incorporate new cold-region specific plant functional types (PFTs) into CLASSIC using literature values and remotely-sensed datasets of land cover change. This is an important prerequisite for producing scenario simulations of future permafrost change within the network. Specific cold-region PFTs such as shrubs have been parameterized using site-level observations and literature reviews. Rose will be incorporating the influence of bryophytes (mosses) into CLASSIC to improve model thermal performance and to account for their carbon cycle impacts. Other processes will be included as required to ensure the model is able to adequately capture observed behaviour with the simplest representation possible.

Maria Shaposhnikova (T3-MSc2)

Title: A temporal deep learning approach to bedfast and floating thermokarst lake ice mapping using SAR imagery: Old Crow Flats, Yukon, Canada.

Supervisors: Claude Duguay and Pascale Roy-Léveillée.

Many shallow arctic lakes and ponds of thermokarst origin freeze to bed in the winter months maintaining the underlying permafrost in its frozen state. Synthetic aperture radar (SAR) offers a unique opportunity to monitor lake ice regimes remotely. Taking advantage of the growing temporal resolution of microwave remote sensing, Maria applied a temporal deep learning approach to lake ice regime mapping. Her project combined imagery for the Old Crow Flats (OCF), Yukon, Canada to create an extensive annotated dataset of SAR time-series labeled as either bedfast ice, floating ice, or land, to train a temporal convolutional neural network (TempCNN). The trained TempCNN, in turn, allowed automatic mapping of lake ice regimes over a 29-year period (1993-2021). The classified maps aligned well with the available field measurements and Canadian Lake Ice Model (CLIMo) simulated ice thickness.

This sub-theme focuses on improved application of existing knowledge and models to produce output useful for stakeholders. Rather than creating new modeling methods, we use existing models, knowledge and data in combination and evaluate it against observational data in PINGO.

Subsurface characteristics, such as the amount and stratigraphic distribution of excess ice and organic matter, are the most important determinants of climate impacts on permafrost systems and can be partially predicted based on observable surface phenomena. We use data and knowledge generated in Theme 1, as well as heuristic and statistical models, to extend current spatial maps and conceptual models of ground-ice content into quantitative data sets suitable for use as input to numerical simulations. In addition to GRIP, this sub-theme is working to produce the Permafrost Surface Characteristics map (PESC). Together, GRIP and PESC will allow key terrain types to be distinguished.

As more observational data becomes available in PINGO, the quality of GRIP will improve, and more processes and phenomena will be represented in CLASSIC. We use these advances to quantify confidence in simulation products and to make simulation products that are best suited to inform decisions by stakeholders. Models are driven by meteorology derived from atmospheric re-analyses and de-biased CMIP6 climate scenarios. Given the size of Canada and the variability expected at the scale of tens of metres, we use sub-grid simulation, (or ‘tiling’) to represent terrain types efficiently. Sub-grid simulation allows meaningful comparison of simulation results with observations (PINGO, Theme 1) and thereby support more informed regional or national products as well as the extraction of simulation results that represent specific locations, for example a peatland just north of Yellowknife. The concept of terrain types is extended to include infrastructure and to derive boundary conditions to impose climate change effects, including changes in snow and vegetation, efficiently on engineering simulations.

Charles Gauthier (T3-MSc3)

Title: Fate of carbon in Canadian permafrost-affected soils.

Supervisors: Oliver Sonnentag and Joe Melton

Permafrost regions are estimated to contain over 1000 Pg carbon (1015g) in the upper three meters, approximately double the carbon contained in either terrestrial vegetation or the atmosphere. At present, it is unclear whether these regions will be a net source of carbon to the atmosphere, due to respiration of the soil carbon, or a net sink of carbon, due to enhanced photosynthetic activity. Understanding carbon emissions better is relevant for risk reduction in the long term. It is important to connect how permafrost environments are modeled (T3-MSc1 and T3-PhD1) and the input data available (PINGO, T3-PhD3) with national and international efforts to understand the carbon dynamics of permafrost regions. This project uses available datasets of soil carbon stocks and respiratory fluxes in a Bayesian optimization framework to produce refined parameter values for CLASSIC’s soil carbon scheme.  Charles’ project will result in more accurate simulations of global carbon stocks and fluxes and their response to future climate changes. CLASSIC enhancements from this work will inform major international bodies such as the IPCC and the Global Carbon Project.

Gabriel Karam (T3-MSc4)

Title: The effect of environmental controls on the thermal contraction-cracking process of ice wedges.

Supervisor: Stephan Gruber

Ice wedges are a widespread landform in the continuous permafrost region. They grow over decades or centuries due to winter thermal contraction cracking. In spring, meltwater seeps into the cracks and freezes to form ice veins. With repeated cycles, these veins widen and can reach widths of several metres. This project is developing a numerical model to simulate the contraction-cracking process in permafrost using frozen soil mechanics. Environmental controls such as soil characteristics and climate were investigated to determine their effects on crack width, depth, and spacing.

Gabriel’s project focuses on the mechanical modeling of ice wedge formation using the extended finite element method (XFEM) in ABAQUS. The model focuses on the initial cracking events in existing permafrost that form new epigenetic wedges. It uses a thermo-mechanical simulation, with temperature data driving a thermal model that is then applied to a mechanical model to cause cracking. The model incorporates temperature-dependent properties for four soil types: coarse sand, fine sand, silt, and clay.

Former student Bingqian Zhang (T3-PhD3)

Title: Mapping and parameterising permafrost terrain types.

Supervisors: Bernhard Rabus with Trevor Lantz, Duane Froese, Joe Melton and Stephan Gruber.

Transient simulation driven by climate data is well suited for understanding the evolution of ground temperature and ground-ice loss over time. At the same time, this requires input data on surface (e.g. vegetation) and subsurface (e.g. ground ice) characteristics in order to produce meaningful results. Subsurface conditions need to be inferred and carefully spatialized based on sparse observations and knowledge about Quaternary history. To produce output data and maps with enough spatial detail to be relevant for local applications and to be tested with ground observations, the data on surface and subsurface characteristics must have significantly higher resolution (~10–100 m) than the typically coarser grids for the climate variables (~10–200 km). These base maps can be delivered in spatially explicit (gridded) form or follow a sub-grid approach where for coarser grids, a list of typical terrain types along with their characteristics and abundance is provided.

The two base maps produced by the network are the GRound Ice Potential and geotechnical permafrost base map of Canada (GRIP) and the PErmafrost Surface Characteristics base map (PESC). GRIP will contain information on variables such as excess ice content, soil type, salinity, or organic content as a function of depth. PESC will be based, and contain information, on surface cover such as vegetation, as well as derivatives of topography relating to the tendencies to accumulate snow and water or to receive solar radiation. The base maps are derivable from primary spatial data capturing generally different moments in time around a chosen “base map datum”. These primary data include optical and radar remote sensing layers in their original form or as derived products (e.g., elevation models, land cover), polygon data such as maps of surficial geology and reconstructed Quaternary history and finally point evidence (e.g. boreholes, outcrops). Bingqian’s project designs methods of spatial analysis for deriving GRIP and PESC to make these maps available in versions of increasing sophistication. This project provides a key linkage in the flow of information and knowledge between themes and Bingqian is working closely with Theme 1, whose insight is turned into spatial products here, and with the other Themes as potential users of GRIP and PESC. 

Successive improvements of GRIP are being investigated, and these versions are valuable products for network members and stakeholders and inform the strategy and timeline for production of an improved version for large areas or nationally. Using deep learning methods and higher resolution (primary) remote sensing data, statistical approaches, including neural networks, incremental products are being generated for network participants to use and evaluate in the chosen test areas. 

Former student

Hannah Macdonell (T3-PhD4, started as MSc project)

Title: Quantifying confidence in simulations of permafrost change.

Supervisors: Stephan Gruber with Joe Melton, Trevor Lantz, and Steve Kokelj.

Hannah’s project is investigating how well ground temperature models are validated by developing methods and tools for evaluating existing models with existing data. Statistical approaches to error and uncertainty will be used to inform model development (Where and when does the model perform least well? Which model performs better?) and the use of model results (How well do permafrost simulations perform at a particular location?). In a follow-up MSc project, this quantitative evaluation will be complemented by investigating face validity, a concept developed to capture the trust that diverse experts, such as model developers, permafrost field scientists and northerners, place in simulation results based on subjective assessment. Investigating face validity will utilise the diverse expertise in the network to identify and assess unknowns in simulations and to develop ways of communicating these to both model developers and end users. The combination of statistical evaluation with face validity will improve the dialogue between model developers and users of simulation results and thereby reduce barriers to the acceptance and uptake of simulation products. Hannah is closely working with and underpinning the research of Theme 3 and Theme 4 projects.

Galina Jonat (T3-PhD5)

Title: Simulation-based climate services for permafrost environments.

Supervisors: Stephan Gruber with Alex Cannon, Fabrice Calmels and Shawn Kenny.

Galina’s project will produce and evaluate simulation based permafrost data products such as ground temperature or subsidence intended to be useful for a broad range of stakeholders. Simulations will be driven by de-biased/downscaled climate data and use GRIP produced in T3- PhD3 and supported by work in Theme 1. Galina’s project differentiates three types of user interest: (1) Detailed information that can be related to individual sites will be provided by simulating permafrost under typical terrain types in an area such as peatlands or low-shrub tundra on till. Users will be able to select terrain types and output variables, such as ground temperature or subsidence, based on their needs and investigate their temporal evolution and uncertainty. One, or several projects, from Themes 4 and 5 will be selected as test cases for site-level simulation and allow interaction with application projects and partner needs. (2) Regional or national maps will show best estimates and confidence intervals for key variables and time slices (e.g., the years 2030–2050) with a resolution of several kilometers, only. This will be based on sub-grid simulation of differing terrain types. (3) Boundary conditions that can be used to impose the combined effects of climate and surface changes onto geotechnical and hydrologic models. The boundary conditions must provide a simple and sufficiently truthful representation of ground-atmosphere interaction. This project produces novel output and insight through constrained model confidence, simulating at a scale fine enough to predict local phenomena and through incorporating an improved data set of ground ice potential and geotechnical characteristics. The products developed here can serve as prototypes for informing future climate services provided in Canada.

Collaborators and partners

  • C. Spence 
  • E. Humphreys 
  • P. Lipovsky 
  • C. Stevens 
  • M. Packalen 
  • J. McLaughlin 
  • S. Wolfe 
  • S.L. Smith 
  • P. Morse 
  • A. Bevington 
  • Y. Zhang 
  • L. Arenson 
  • A. Cannon
  • G. Meyer

Peer-reviewed research findings are listed on our Publications page.