a) Project Background
Many governmental and academic institutions have developed global climate models (GCMs) to provide fundamental climate information to assess the climate change impacts on terrestrial and ecological systems at global and regional scales. Since 1995, a working group of Climate and Ocean-Variability, Predictability, and Change (CLIVAR), part of the World Climate Research Program has established the Coupled Model Intercomparison Project (CMIP) to provide multi-model outputs. In the last decades, the climate change assessment work completed in the province was based on CMIP5 or older climate projections. However, as the climate science is rapidly evolving, more comprehensive and sophisticated global and regional climate model are now available to provide reliable climate forecast. As a result, the Intergovernmental Panel on Climate Change (IPCC) has started releasing the new set of climate projections under CMIP Phase 6 (CMIP6) project. These new climate projections are based on newly developed emission scenarios. However, applications for domain-specific impact models, e.g., hydrologic and water-quality models at a watershed-scale, have been impeded by coarse resolutions of climate models that have limited capability to capture climate variability at a local scale. Therefore, the requirement of a tool to downscale GCMs’ output has been growing to evaluate the impacts of climate change at local or regional scales. Further, a full range of climate projections is required to take into account uncertainty in impact studies. However, sufficient resources are seldom available for such comprehensive assessments and it is not practical due to a high computational cost. Therefore, a methodology is desirable to systematically select a subset of climate projections that sufficiently capture the full range of future climate variability.
Downscaling techniques nested by GCMs over a specific region have been developed to produce a higher resolution climate dataset by dynamical and statistical downscaling models. The dynamical downscaling methods primarily rely on physical-process descriptions of atmospheric phenomena by Regional Climate Models (RCMs) which require highly skilled experts and a high-end computation facility to run RCM and to store enormous amount of output produced by RCMs. However, many studies have found that systematic errors remain in climate models’ outputs. In contrast, statistical downscaling methods are a cost and time effective way by employing the relationships between observations and outputs of climate models to downscale and remove systematic errors in GCMs’ outputs. Therefore, statistical downscaling methods have actively been applied to bridge the gap of spatial resolution between GCMs’ output and requirement of impact studies at a local scale. Over last decades, univariate statistical downscaling methods have been applied to fit the GCMs’ output to marginal distributions of observations for individual variables, which may result in spurious consequences from impact models as they do not capture inter-dependence between climate variables. On the contrary, multivariate statistical downscaling methods preserve an inter-variables dependence structure, which is crucial especially for impact studies in snow-dominated watersheds. Therefore, it is necessary to apply the multivariate statistical downscaling methods to enhance the reliability climate change impacts by applying numerical models which are forced by downscaled climate projections.
Environmental Monitoring and Science Division (EMSD) has already developed following main quantitative approaches for the selection of climate change scenarios: (i) clustering and (ii) envelope methods. The clustering approach narrows the number of GCM scenarios by identifying very different sub-clusters of future projections using a automated cluster analysis algorithm and then retaining a single representative GCM scenario from each sub-cluster. The envelope approach relies on selecting extreme GCM scenarios to capture a wide range of simulated changes for a number of climatological variables. In order to fully address the challenges associated with GCM selections, the climate model performance needs to be evaluated by comparing historical model simulations with observations, known as hindcasting, and then integrating, the hindcasting, clustering, and envelope methods on a single framework. As the current developed algorithm focuses only on extreme climate conditions, there is a need for further enhancement of the initial work to select climate change scenarios for different applications in Albert and its major watersheds to provide a spatially representative understanding of potential variability in climate change effects. As a result, a combination of extreme and representative climate change projections are required to provide essential and reliable climate change impact assessments (for example, the lower and upper extreme bounds of future climate condition as well as the average changes expected in future climate condition).
b) Project Objectives
This project involves statistical downscaling to generate high resolution (0.1° ˜ 10 km) climate projections (for select representative GCMs) for the province of Alberta to support several Alberta Environment and Parks (AEP) priority projects requiring climate change impact assessments. A set of representative climate change projections will be identified for each major watersheds of Alberta to capture the full range of climate variability in future. As a reliable historical climate dataset is one of prerequisites to implement the statistical downscaling for the province, a performance-based historical hybrid climate dataset for the province has already been developed and will be used as a reference climate data for statistical downscaling. In summary, the broad objectives of this project are:
(i) Compile/download GCM projections
(ii) Evaluate GCMs’ skill in simulating historical climatology,
(iii) Identify a set of representative GCM scenarios for Alberta and its major watershed, and
(iv) Generate high-resolution climate projections using a state-of-the-art statistical downscaling method.
c) Project structure
As a first step, the performance of GCMs (global climate models) is measured by a hindcasting approach to screen GCMs exhibiting poor-performance during a historical period. Then cluster and envelop approaches will be applied for selecting a subset of climate projections that fully capture a range of future climate variability for watershed and province scales. Subsequently, statistical downscaling methods will be applied to the selected climate projections.
As a reliable historical climate dataset is one of prerequisites to implement the statistical downscaling, a REFerence Reliability Evaluation System (REFRES) has already been developed and applied to produce a performance-based historical hybrid climate dataset for Alberta by combining multiple historical gridded climate datasets. Employing the generated hybrid climate dataset as a reference data, a package of programs for state-of-the-art multivariate statistical downscaling methods will be applied to generate high-resolution climate projections in Alberta.
