Valentina Radic
Associate Professor
The long-term objectives of my research program are to use state-of-the-art field, modeling and data analysis methods to quantify the response of mountain glaciers to climate change on regional and global scales, and to narrow the uncertainties in projections of glacial contributions to regional streamflow as well as to global sea level rise.
Significant Past Contributions
World-wide melting of mountain glaciers and ice caps (i.e., excluding the Antarctic and Greenland ice sheets) is a well-established significant contributor to current and future sea-level rise. Early in my career (2004-2012), I developed and applied novel numerical and statistical methods to project the contribution of glacier melt to sea-level rise in response to future climate forcing scenarios from global climate models (GCMs). In Radic and Hock (2011), we provided the first detailed and regionally resolved projections of glacier mass change on a global scale.
Since 2012, the global inventory of glaciers has grown significantly and better climate projections from reliable GCMs have become available. To harness these, in Radic et al. (2014) I led a team of collaborators in a major initiative to update regionally resolved projections of glacier contribution to sea level rise. Our results featured in Chapter 13 (Sea Level Rise) of the Intergovernmental Panel on Climate Change Fifth Assessment Report. Critically, our updated results identify the regions with the largest potential contribution to future sea level rise (Arctic Canada, Alaska and Antarctic glaciers) and the regions most vulnerable to glacier wastage (European Alps, New Zealand, Caucasus, Western Canada). In particular, the latter are projected to lose more than 75% of their current ice volume by 2100.
Beyond future global sea level rise, deglaciation has major implications for regional hydrology. Changes in seasonal runoff impact drinking water supply, agriculture, hydropower generation and freshwater ecosystems. As part of a long-term collaborative project, I contributed to the development of the Regional Glaciation Model (RGM) for western Canada. This was the first model for regional glacier evolution to couple surface mass-balance with an ice dynamics model of high complexity (Clarke et al., 2015). In this study, I developed a statistical downscaling method to convert coarsely-resolved climate fields from GCMs to finely-resolved ones that drive melt processes at the scale of individual glaciers.
Insufficiently resolved or incomplete data on glacier geometries as well as computational limitations prevent us from implementing an RGM-like model globally at present. However, we have shown that reliable projections of the global glacier response to climate change are possible with simplified models for the geometric evolution of glaciers (e.g. volume-area scaling models). To do so, we used a simplified glacier-dynamics model to assess sensitivity and response time of glaciers to climate perturbations on regional and global scales (Bach et al., 2018). We show that volume sensitivity is mainly driven by the glacier size distribution in a given region, as well by the distribution of surface slopes, while response time scales are dominated by the ablation gradient: mountain ranges with small glaciers and low slopes are the most vulnerable to volume loss, while those with high mass turnover (and in particular, high ablation rates) will adjust most rapidly to changes in climate.
Ongoing research
My research on the projections of glacier mass changes on regional and global scales has identified three major sources of uncertainties in glacier melt models: (1) sensitivity to poorly constrained parameters in the semi-empirical models of surface mass balance; (2) sensitivity to the downscaling methods used to convert the climate variables at coarse spatial resolution (e.g. 100 x 100 km grid) to sub-kilometer spatial resolution; and (3) poor parameterization of turbulent heat fluxes, an important contributor to energy available for melting, at glacier surfaces. One strand of my research has aimed to narrow each of these uncertainties. To achieve this goal, we not only need better models and more observations, but also to test novel approaches (e.g. deep learning) in addressing 'old' problems. I further elaborate on the sources of uncertainty and give an overview of the ongoing research projects.
Models of Surface Energy Balance and Climate Downscaling
As part of regional and global mass balance models, the simulation of glacier melt is often based on semi-empirical temperature-index models that require calibration with available observations. Since the observations needed for calibration are available for fewer than 1% of mountain glaciers worldwide, these models cannot be adequately constrained for more than 99% of glaciers, making the projections of glacier melt and runoff highly uncertain. By contrast, physics-based models based directly on surface energy balance (SEB), are universally applicable at any glacier surface. To force these models, however, meteorological variables and energy fluxes are needed at or in the vicinity of the glaciers in question. In the absence of observations detailing these variables, the required forcing must be derived from downscaling the coarse-resolution output from GCMs and/or reanalysis datasets to the local glacier scale. Despite the advantage of physics-based modeling over the empirical models, it remains to be shown whether downscaled meteorological fields from GCMs can successfully resolve the local processes driving surface melting for the majority of glaciers. To address this question, we focus on evaluating the performance of an SEB model forced with dynamically downscaled meteorological fields. The first step in this analysis, presented in Mekdes Tessema's MSc thesis, evaluated the performance of Weather Research and Forecasting (WRF) model in downscaling the energy fluxes relevant for modeling surface melt at three glaciers in Canadian Rockies. Our ultimate goal is to develop a dynamical downscaling coupled with statistical downscaling (machine learning) approach to simulate spatially-resolved regional glacier melt based on SEB models (ongoing PhD projects of Christina Draeger and Hannah Phelps).
Assessment of Turbulent Heat Fluxes under Katabatics
As part of the physics-based models of glacier melt, turbulent heat fluxes generated by wind drag and temperature gradients in the near-surface atmospheric layer can contribute significantly to the energy available for melt. These fluxes, however, are poorly understood as their direct measurements, which require specialized instrumentation and continuous maintenance, are hard to obtain at glacier surfaces. Current models of turbulent heat fluxes at glacier surfaces rely on simplified bulk or closure methods that use mean meteorological variables such as near-surface air temperature, wind, and relative humidity. Although these methods are widely used, they are based on assumptions such as a boundary layer flow driven primarily by wind in the free atmosphere, which is rarely the case near sloped glacier surfaces where katabatic flow is prevalent. Katabatic winds are negatively buoyant downslope winds that are characterized by a near surface wind speed maximum (< 50 m height above the surface). Since 2014, my team has conducted a series of field experiments measuring near-surface meteorological variables and turbulent fluxes using eddy-covariance technology at a total of seven stations at three mountain glaciers in western Canada. These measurements represent one of the longest uninterrupted time-series of eddy-covariance measurements at sloped glacier surfaces. Analysis of the data revealed that the bulk methods for turbulent fluxes fail in the presence of katabatic flow (Radic et al., 2017, Noel Fitzpatrick's PhD thesis). To improve the method’s performance, it is necessary to develop a katabatic flow model in which the two key parameters (eddy viscosity and diffusivity) are represented as functions of height, wind shear, and vertical temperature gradients (ongoing PhD project of Cole Lord-May and Hannah Phelps).
Applications of Machine Learning Methods
A new way of thinking about big data and empirical models (machine learning) can make progress in ways that were previously not possible. Since 2013, I have been teaching a graduate course in data analysis and machine learning in the geosciences (EOSC510) and was able to contribute to a large number of studies led by grad students across the science and engineering communities at UBC. Unglert et al., (2016), for example, was the first study in volcano monitoring to develop an automated and unsupervised method for identifying patterns in seismicity potentially indicative of eruptive behavior. Sam Anderson's PhD project included detecting and projecting the changes in glacier-fed streamflows across western Canada to provide a first ever risk assessment of freshwater availability for the region (Anderson and Radic, 2020). Here we developed a novel framework for data analysis and machine learning in order to asses the local water resource vulnerability to rapid deglaciation in Canadian Rockies. As part of his PhD thesis, Sam also developed a deep learning model to simulate regional streamflow across Western Canada.
