Important Data Notes and Limitations

Data Limitations

The following caveats should be reviewed by anyone intending to use data from the Climate Atlas for decision-making. These caveats are especially pertinent for engineers, planners and other professionals looking for climate data to inform their work.

Downscaling Method

The climate model data we used (from 12 models) was statistically downscaled by the Pacific Climate Impacts Consortium (PCIC) from a monthly time resolution to a daily time resolution using the Bias Corrected Spatial Disaggregation (BCSD) method. The BCSD method is a relatively simple downscaling method compared to, for example, dynamical downscaling. Because PCIC started with monthly values, it used an historical climate dataset produced by Natural Resources Canada (NRCan) to simulate daily variability within the monthly data. Daily values were sampled randomly from historical months in the NRCan dataset and then rescaled so that the monthly means match the original climate model projections. Importantly, this means that the downscaled daily data in the Atlas reflect changes in the monthly means, but not changes in daily variability within the original model output. In some situations, the downscaled data may produce results that are inconsistent with the original model data.

Temporal Resolution: Timing and Persistence of Weather Events

A further consequence of the downscaling approach used to construct the data in the Atlas is that the simulated daily data are not well positioned to be used to count the number of days in a row with some threshold temperature or precipitation amount. They are better suited to assessments of the frequency of temperature or precipitation conditions over months, seasons and years.

For example, the data from an individual model should not be used to compute the frequency of heat waves as measured by the number of consecutive days on which some threshold temperature was observed; but we can more reasonably compute the total number of hot days over a long period (i.e. a season, year). Similarly, the data should not be used to determine the persistence of rainfall events (or days without rain), which means that the data is not well suited for many water management and engineering applications that require daily inputs.


Unless otherwise indicated, the data presented in the Atlas depicts the average, or ensemble, of the statistically downscaled data from 12 climate models. It is standard practice in climatological studies to use data not from one model but a range of models, to represent the uncertainty associated with the modeling process. For the sake of simplicity, we chose to use the mean of the ensemble, rather than the median values, as is done in some studies. Importantly, for some of our depictions, the mean value is accompanied by the highest and lowest values from the 12 models, to represent the range of projections across the models.


We define the four seasons in the standard climatological manner: Winter = December / January / February; Spring = March / April / May; Summer = June / July / August; and Fall = September / October / November. For almost any Canadian, one or more of these sets of months will fail to capture the true start and end dates of their seasons. Furthermore, choosing specific months to define seasons does not take into account the inevitable shifts in climate that will come to define our seasons in the future.

Spatial Resolution and Interpolation

The spatial downscaling process used by PCIC relies upon the Natural Resources Canada (NRCan) gridded dataset of historical daily meteorological observations to represent how temperature and precipitation varies in space. Although the gridded data is generally of high quality, the quality is spatially inhomogenous. In particular, the quality of the NRCan data is reduced where there are i) temporal gaps in the weather station data; ii) large geographic distances between weather stations; iii) mountains present; or iv) large contrasts in microclimate in a region. In short, the climate model data we use is most prone to interpolation error over mountainous regions, and areas with few long term weather stations, especially northern Canada.

Nearest Neighbour Analysis

Climate change values for municipalities (cities and towns) were computed by finding the nearest neighbouring grid point. That is, downscaled data from single grid points representing a 10 km by 10 km area were used to represent the climate of cities and towns, even though those places may be larger than this area. This choice was made for the sake of computational simplicity.

Areal Averaging

The Atlas allows users to explore climate change values across all of Canada, including remote and rural areas. However, due to computational and internet server limitations, we could not provide all the original model data at its native 10 km resolution. For the initial version of the Atlas, we have opted to areally average the data at two scales:

i) National Topographic Service (NTS) of Canada 1:250,000 topographic map areas;

ii) Provinces and Territories.

Only data points completely contained within the spatial domains of these regions were included in their areal averages.

Terrestrial Data Only

It is important to note that the model data is available for the terrestrial regions of Canada only (including lakes, but not including all of the Great Lakes, which extend into the United States). No data was available across the Oceans. Many of the NTS grid squares along Canada’s coasts contain large areas of open ocean, sometimes with only a tiny fraction of the area being land. In these cases, the areal averaged value for these coastal grid squares only represent conditions over the terrestrial portion of the grid.

Confidence and Impacts

It is impossible to state with certainty that a specific projected climate change will occur. In fact, any expectation of absolute confidence from climate models is an ill-fated quest; there will always be uncertainty associated with what the models ‘say’ about how the climate will change. Furthermore, it is impossible to say with any certainty that the climate change that does occur will have a particular impact. However, the level of confidence one might have in possible impacts will vary from case to case, depending on the kind of climate change associated with the impact, and the nature of the relationship between the change and the impact of that change.

For example, it is one thing to compute the number of +30 °C days projected for some time in the future, but quite another to state how an increase in +30 °C days will likely impact human health, forest fire frequency or thunderstorm intensity, and another still to calculate the cumulative impact these impacts have on the economy. We might have much higher confidence in stating that an increase in the number of hot days will likely increase the number of heat strokes in urban environments (unless adaptation occurs), but lower confidence that the increase in these hot days will increase the severity of thunderstorms (which require specific kinds of meteorological conditions). That is, there is a quite direct relationship between heat-related illnesses and high outdoor temperatures, whereas thunderstorm development depends on many environmental factors, with temperature being just one. Similarly, we are likely to have much lower confidence in stating that these impacts will negatively affect the nation’s economy. Our level of confidence decreases as the system becomes more complex and more variables come into play.

Where we have lower confidence in our ability to make meaning from the data, we surveyed the most up-to-date peer-reviewed literature and connected with experts from across Canada to better inform our work. However, we will never be able to predict future impacts with absolute certainty. In spite of this, we strongly believe that society needs to move forward to address the risks of climate change in the face of uncertainty. Indeed, this is one of the goals of our work: to initiate discussions about how the climate is likely to change, what the impacts of these changes may be, and how we should respond.

Specific Climate Variables and Indices


Precipitation is much more difficult to model than temperature. Moreover, the downscaling technique used to produce the precipitation values presented in the Atlas is not conducive to the representation of daily precipitation events. Consequently, we have low confidence in the precipitation projections for daily timescales. More specifically, it is likely that the projections underestimate the frequency and intensity of heavy precipitation events. We have higher confidence in the long-term averages of monthly, seasonal and annual precipitation totals presented in the Atlas.

Note that the range of projected precipitation values across the models used to create our ensemble values is relatively high compared to the range of values in the temperature projections; this is typical of all climate model studies.

Heavy Precipitation Days

The Climate Atlas shows relatively little change in the number of heavy precipitation days in the future; however, there is much uncertainty regarding how well climate models capture these intense and often localized events (such as a thunderstorm).

Nevertheless, climate scientists are quite confident that the number of heavy precipitation events — and especially rainfall events – will increase in the future, as the atmosphere becomes more energetic and moist. Indeed, there is an abundance of evidence that rainstorms in many parts of the world, including North America, are becoming more frequent and intense.

Furthermore, the downscaling technique used to produce the precipitation values represented in the Climate Atlas is not conducive to the representation of daily precipitation events and the frequency of Heavy Precipitation Days (HPD). Consequently, we have low confidence in the HPD values. More specifically, it is likely that the projections underestimate the frequency and intensity of heavy precipitation events. We have higher confidence in the long-term averages of monthly, seasonal and annual precipitation totals represented in the Atlas.

Mean Temperature

In meteorology, the daily mean temperature reflects the average of all temperature measurements made in a day (typically the average of 24 hourly measurements). It is common in climatological research to calculate daily mean temperature as the average of the daily maximum and minimum temperatures. On daily time-scales, the two methods for computing the daily mean can yield slightly different results.

Because we only had access to modelled daily maximum and daily minimum temperatures, we defined the daily mean temperature as the average of the daily max and min temperature. All of the mean temperature values in the Atlas (including historical observed values) have been calculated using the same method, making them internally consistent and comparable.

Summer Days, Very Hot Days, Tropical Nights

The daily temperature values summarized in the Atlas were generated using a statistical downscaling technique applied to monthly climate model data. They should not, therefore, be used to calculate the persistence or timing of hot weather events (e.g., the frequency and length of heat waves).

Very Cold Days

The daily temperature values summarized in the Atlas were generated using a statistical downscaling technique applied to monthly climate model data. They should not, therefore, be used to calculate persistence or timing of Very Cold Days (e.g., the frequency and length of cold snaps).

Date of First Fall Frost, Date of Last Spring Frost, Frost-Free Season

The daily temperature values summarized in the Atlas were generated using a statistical downscaling technique applied to monthly climate model data. Consequently, they are not ideally suited for estimations of the average date of last spring frost, date of first fall frost, and length of frost-free season.

However, for the period 1950-2005 we compared observed frost-free season values (extracted from NRCan’s gridded dataset derived from meteorological data) with model-generated values, and found a relatively high level of agreement. Therefore, we are quite confident that it is reasonable to use the statistical climate model data to estimate the frost-free season from the projected data.

In addition to this caveat, the following two points should be kept in mind by anyone using our frost-free season values:

1) Over the period 1950-2005 we compared observed frost-free season values (from NRCan’s data) with model-generated values and found that the two values matched very well; thus, we have a relatively high level of confidence that the statistical, downscaled climate model data is appropriate for estimating the frost-free season length.

2) The temperatures we use to compute the frost-free season are based on standard weather station observations, which are captured 1.2 m above the ground. Minimum temperatures at or near ground level, where plants emerge and grow, can vary quite a bit from the surface air temperature. Thus, this method of identifying the presence or absence of frost is an approximation. As the minimum temperatures measured at ground level are often colder than those 1.2 m above the ground, the estimated length of the frost-free season presented here is likely longer than the actual length of the season.

Corn Heat Units

Corn Heat Units (CHU) are a measure of heat accumulation when temperatures are within the optimal range for corn growth. CHU only begin to accumulate after a mean temperature of 12.8 °C is observed on at least three consecutive days. Likewise, CHU stop accumulating after two or more days in a row with minimum temperatures of -2 °C or lower. The statistical nature of the climate model data means it is ill-suited for analyzing temperature persistence across a string of days; however, given that the beginning and end of the CHU season contribute relatively little to the overall total, we elected to compute CHU. We stress, however, that real-world CHU values may differ somewhat from the modelled CHU values displayed in the Atlas.