Xiaolan L. Wang

The Arctic has experienced the most rapid warming on Earth in recent decades. This affects Canada's landmass, which extends well into the Arctic. Nevertheless, limited spatial and temporal observational coverage, combined with large climate model uncertainties, pose challenges to understanding both past and future climate changes in these regions relative to preindustrial conditions. This is particularly challenging in a place like Canada that has insufficient historical data to determine preindustrial reference conditions. Emergent constraints can overcome this limitation by using historical observations for the modern post‐industrial era to constrain estimates of both preindustrial reference levels and future warming. Here we apply a carefully tested Bayesian observational constraint method to simultaneously assess the externally forced historical and future warming in Canada. Testing indicates that the approach reduces bias and uncertainty in historical and future warming estimates, increasing confidence that it may also serve as a basis for developing a broader understanding of climate change in other high‐latitude regions. We estimate that external forcing from human activity, has warmed Canada by 2.2 [1.3, 3.1]°C between the 1850–1900 pre‐industrial period and the recent 2015–2024 decade. Applying these same observational constraints to future climate conditions indicates that Canada will warm to 5.1 [3.2, 7.0]°C above pre‐industrial levels by the end‐of‐century under an intermediate emissions scenario SSP 2‐4.5, and to 6.7 [4.6, 8.9]°C under a high‐emissions scenario SSP 3‐7.0, with the largest warming projected for Northern Canada, followed by Quebec.

In situ station data records usually have different periods of data coverage, with individual station records starting and stopping at different times. These records may also have different missing data rates and missing data rates may change over time. These changes in data availability (i.e., the inhomogeneous sampling of data), can bias trend estimates using a gridded dataset if left untreated. Reanalysis data, such as the 20CRv3, ERA20C, OCADA, and ERA5 datasets can be used to estimate the effects of changes in data availability. This can be done by comparing results from a spatially and temporally complete version of reanalysis dataset with results that are obtained from an incomplete reanalysis dataset that has the same availability in space and time as the station data. Results obtained from the incomplete data can then be compared with the correct results from the corresponding complete dataset.This study uses reanalysis datasets to estimate the effects of changes in data availability on the gridded precipitation and surface air temperature datasets and their representativeness of the climate and changes therein. The estimated effects are then used to correct sampling biases in the gridded datasets, producing sampling bias-corrected gridded datasets for use to assess climate change in Canada. The results show that sampling biases are larger in precipitation data than in temperature data due to the higher temporal and spatial variability of precipitation and lower precipitation station density. The biases are also larger in the period and regions of sparse observations (such as the early period and northern Canada).

Using the changepoints identified and adjusted to produce the homogenized monthly precipitation dataset, this study developed a homogenized daily precipitation dataset for Canada, in which all data gaps are infilled for the period back to 1900 for southern stations (south of 60˚˚ N), and back to the first day of 1948 or the first day of observation before 1948 for northern stations using advanced spatial interpolation of both monthly and daily values from other stations in the region. The homogenized daily precipitation dataset was then used to assess trends in a set of precipitation extreme indices, including annual maximum one-day (RX1day) and five-day (RX5day) precipitation, as well as annual number of heavy precipitation days, R10mm (annual count of days when precipitation >=10mm). The results show that both annual maximum one-day and five-day precipitation have increased significantly at most stations across Canada over their data record periods, with most stations in the Rocky Mountains and southern Prairies showing insignificance decreases over the period of 1948-2022. Increases in the annual maximum one-day and five-day precipitation are largest and significant in central to northern Canada and in the Maritimes provinces. Annual number of heavy precipitation days has increased significantly at most stations in northern Canada.

Using hourly surface wind speed data from 155 stations in Canada, this study first developed a homogenized monthly mean wind speed dataset for the period of 1953-2023, which was then used to characterize observed changes in surface wind speed in Canada. The hourly data were first quality controlled and adjusted for non-standard anemometer heights before being used to calculate monthly mean wind speed series. To identify artificial discontinuities, the monthly mean wind speed series were subject to a semi-automated comprehensive data homogenization procedure, which uses a combination of station metadata and multiple statistical tests with and without using reference series. Reference series used include up to four best significantly-correlated neighbour stations’ data series, the ensemble mean series of monthly wind speed taken from the Twentieth Century Reanalysis version 3 (20CRv3), and monthly mean geostraphic wind speeds derived from homogenized surface pressure data. The results from the automated procedure were then reviewed manually using metadata and visual inspection of the multiphase regression fits with expert judgement. As a result, all the 155 data series were identified to have one or more artificial discontinuities, which were diminished by quantile matching adjustments. Anemometer height change, station joining, relocation, instrument changes/problems were found to be the main causes of data inhomogeneities. The homogenized dataset for 1953-2023 shows wind stilling in region from northern British Columbia (BC) to southern Yukon-Northwest Territories and from southern Prairies to Quebec-Labrador, which was matched with wind strengthening in the region from southern-central BC to the Rocky Mountains, and in Newfoundland and the high Arctics. The trend pattern of in-situ wind speed data bear substantial similarity to that of both the ERA5 and 20CRv3 reanalysis wind speed data.