Canadian Climate Change Scenarios Network - National Node

EC - GC

Frequently Asked Questions

 

Click on a question below to view the answer. Links provided in the answers will guide you to further information on the website or from other sources. Should you have any further questions, please visit our Contact Us page.

 

Getting Started

• What is a climate change scenario?
• What is a ‘baseline’ climate?
• Where do I go to get data?
• How can I visualize GCM output?
• What is a Scenario Map?
• What is a Bioclimate Profile?
• What is a Scatterplot?
• What is Advanced Spatial Search?
• What is downscaling?
• How do I access Downscaling Tools (ASD, SDSM and LARS-WG)?
• What is a downscaling predictor?
• Where do I go to get statistical downscaling predictors?

 

Downscaling

• During the regression process, does SDSM do the regression with daily or monthly values?
• When using SDSM, how do I know which variables are dependent on regional-scale predictors?
• Is there a method within SDSM to merge or average the predictors for multiple adjacent grid cells?
• After calibration in SDSM, how do I match up the GCM reference period data with local observed data by date, given that the CGCM1 model is based on a 365-day year and the Hadley model is based on a 360-day year? 
• I am not sure when to make changes to the advanced settings in SDSM.  For example, how does one decide the amount of variance inflation or bias correction?
• It seems that the fourth root transformation is the default for downscaling precipitation in SDSM.  Why is this? 
• When I use either fourth root or others type of transformations for precipitation, the result is very unrealistic. Could you direct me to any documentation on the basic theory, including the math behind SDSM?
• I found out that the predictand data I am using "are for a day beginning at 0600 Greenwich (or Universal) Mean Time, which is within a few hours of midnight local standard time in Canada.” Does this mean I should lag all my NCEP values by a certain amount? If so, how do I decide which variables and how much I should lag them?
• Is there any literature available that would explain the following regridding technique: "A popular regridding technique is to compute the weighted average of neighbouring grid-points, where the weighting decreases with separation distance following a Gaussian curve up to a specific max separation?"
• Should my predictand data be normally distributed and standardized when I use it in SDSM?
• When running SDSM what is the best way to choose the predictors?
• I have been asked to downscale relative humidity using these programs. I am having a hard time finding any literature that refers to downscaling relative humidity and have yet to come across a study where SDSM or LARS-WG has performed the task.

 

What is a climate change scenario?

A climate change scenario is defined as:

• “… a coherent, internally consistent and plausible description of a possible future state of the world…” (Parry, 2002), and
• “… a plausible future climate that has been constructed for explicit use in investigating the potential consequences of anthropogenic climate change…” (IPCC TAR, 2001).

Climate change scenarios are the first step along the path to creating plausible representations of the future climate based on assumptions concerning future atmospheric composition, and on our understanding of the effect of increased atmospheric concentrations of, for example, greenhouse gases, particulates and other pollutants, on global climate. A number of different methods exist to construct climate change scenarios, including techniques utilizing arbitrary, analogue and global climate model (GCM) information. Each method has its own advantages and disadvantages. GCM-derived scenarios of climate change are by far the most common scenario type encountered, since they conform to most of the criteria proposed by the Intergovernmental Panel on Climate Change (IPCC) Task Group on Data and Scenario Support for Impact and Climate Assessment (IPCC-TGICA), an international body of experts convened to ensure consistency in climate impacts and adaptation research.

For more information, please see Scenarios Introduction and Constructing Scenarios.

 

What is a ‘baseline’ climate?

A meaningful assessment of the impacts of climate change will include a thorough assessment of the impact response to present-day or recent climate conditions, in addition to an assessment of its response to a number of climate futures. Specification of the present-day or baseline climate is, therefore, just as important as the specification of the scenarios of climate change.

Baseline climate information is important for:

• characterizing the prevailing conditions under which a particular exposure unit functions and to which it must adapt
• describing average conditions, spatial and temporal variability and anomalous events, some of which can cause significant impacts
• calibrating and testing impact models across the current range of variability
• identifying possible ongoing trends or cycles
• specifying the reference situation with which to compare future changes

The IPCC recommends that, where possible, the most recent 30-year climate 'normal' period should be adopted as the climatological baseline period in impact and adaptation assessments. The 1961-1990 normal period has been selected as the standard reference for many of these studies, as it is considered to:

• be representative of the present-day or recent average climate in the study region
• be of a sufficient duration to encompass a range of climatic variations, including a number of significant weather anomalies
• cover a period for which data on all major climatological variables are abundant, adequately distributed over space and readily available
• include data of sufficiently high quality for use in evaluating impacts
• be consistent or readily comparable with baseline climatologies used in other impact assessments

However, it is recognized that:

• in some cases there is better data access from an earlier time period (e.g. 1951-1980 or even 1931-1960)
• in some, but not all, regions more recent periods may already contain a significant warming trend which may be greenhouse gas related
• a 30-year period may not be of sufficient duration to reflect natural climatic variability on a multidecadal timescale, which could be important when considering long-term impacts

For more information, please see Baseline Climate Data.

 

Where do I go to get data?

The CCCSN’s Download Data section provides all the basic data needed to construct climate scenarios, from climate models (both global and regional models), as well as data used as input variables for statistical downscaling tools. Additional data useful to validate climate models and/or to calibrate statistical downscaling tools are also provided here, as reanalysis products and observations. Other types of data are also supplied.

Please click on a link to go to the appropriate section:

• Climate models
  • Global Climate Model (GCM)
  • Canadian Regional Climate Model (CRCM)
• Reanalysis products (global and regional from NCEP)
• Statistical downscaling input
• Observations
• Other data

 

How can I visualize GCM output?

The CCCSN provides four tools for visualizing GCM data: Maps, Bioclimate Profiles, Scatterplots and Advanced Spatial Search. For information on each of these tools, please go to Visualization.

 

What is a Scenario Map?

The 20-year, 30-year or user-defined average (or average anomaly) of each variable from each GCM scenario can be mapped for a period between 2000 and 2100. These maps provide a visual image of how the climate will change in any particular scenario, compared to the baseline climate period. The resulting image can be downloaded onto your computer.

For more information, please see Scenario Maps Help

Go to Scenario Maps

 

What is a Bioclimate Profile?

Bioclimate Profiles provide a ‘climate at a glance’, graphical representation of climate and related indices on a site-by-site basis for historical and future time periods. A typical bioclimate profile consists of a number of elements that describe the temperature and moisture conditions at the site in question. The Bioclimate Profile information was developed to assist in multi-disciplinary studies of past and future climate regimes.

For more information, please see Bioclimate Profiles Help.

Go to Bioclimate Profiles

 

What is a Scatterplot?

In beginning any impact assessment, choosing which climate scenario to use is an important question that should be considered early on in the process. The Scatterplot is primarily used for the selection of useful scenarios. There are several considerations ranging from type of socio-economic futures that are of interest, as represented by the various IS92A and four major SRES scenarios, the availability of particular variables, model validity and model representativeness, among others. Scatterplots of various model scenarios will assist in identifying the range of values that are projected.

Using the Scatterplot tool, two variables can be plotted against each other for different locations in the country for a particular time slice from each scenario on the website. For example, if precipitation and temperature are plotted against each other for the 2020s, it is very easy to identify those scenarios that are both warmer and wetter and warmer and drier for the region. It is also possible to plot the trajectory of any single variable from the baseline period to the 2080 time period, identifying those scenarios that may diverge from what is typical later in the century.

For more information, please see Scatterplots Help.

Go to Scatterplots

 

What is Advanced Spatial Search?

Advanced Spatial Search finds and identifies locations on a map where the searched criteria (one or more) are found (for example, ‘what locations have a mean annual temperature greater than 10°C?’). All data in the database can be searched (historical and future climate scenario data). The search will return all places meeting a single or set of criteria on a map.

For more information, please see Advanced Spatial Search Help.

Go to Advanced Spatial Search

 

What is downscaling?

For many climate change studies, scenarios derived directly from global climate model (GCM) output may not be of sufficient spatial or temporal resolution to represent changes within a region, at a specific location or the climatic inputs to model a specific process. The spatial resolution of GCMs, in particular, means that the representation of, for example, orography and land surface characteristics, is greatly simplified compared to reality with consequent loss of some of the characteristics which may have important influences on regional climate (e.g. the Great Lakes system and Hudson Bay in North America). The need for detailed site or regional scenarios of climate change for impacts studies has existed for a number of years and has thus resulted in the development of a number of methodologies for deriving such information, generally from GCMs which, despite their shortcomings at finer resolution, are recognised as the best available method for determining internally-consistent scenarios of future climate. These methodologies are termed ‘downscaling.’ Downscaling techniques are generally divided into spatial and temporal classes.

For more information, please see Downscaling Help.

 

How do I access downscaling tools (ASD, SDSM and LARS-WG)?

The CCCSN offers three tools for downscaling: ASD, SDSM and LARS-WG. For access to these tools and for information on how to use them, please go to ASD, SDSM or LARS-WG.

 

What is a downscaling predictor?

To develop spatial downscaling, the use of daily predictors is needed. The predictor variables (e.g. mean temperature at 2m, mean sea level pressure, 500 hPa geopotential height, etc.) provide daily information concerning the large-scale state of the atmosphere, while the predictand describes conditions at the site scale (e.g. temperature or precipitation observed at a station).

The observed large-scale predictors have been derived from the NCEP Reanalysis data set (for use during calibration and validation of the downscaling model), while GCM data for the baseline and climate scenario periods have been derived from GCM experiments (presently CGCM1, CGCM2 and HadCM3). For more information about predictors, please see Reports and Publications and Note on Predictors.

 

Where do I go to get statistical downscaling predictors?

CGCM1, CGCM2 and HadCM3 predictors can be downloaded from Statistical Downscaling Input.

 

During the regression process, does SDSM do the regression with daily or monthly values?

To produce daily simulated values (y), the regression is made with daily predictands (occurrence and amount of precipitation independently, i.e. conditional process, and unconditional process for temperature) and daily predictors (x1, x2 and x3), whereas the model parameters (or regression coefficients, a, b, c) are calculated for each month. The same series of predictors is used over all the year but the regression coefficient is calculated for each month, i.e. the weight of each predictor in the regression can vary from a month to another. A correction term is also included in the equation (i.e. a bias correction and an inflation factor).

 

When using SDSM, how do I know which variables are dependent on regional-scale predictors?

The key is to know if, in the correlation process, the predictand is directly related to the predictor or is there some intermediate value or parameter that is in play when developing that correlation. The SDSM model is set up to handle temperature and precipitation and has been designed to handle conditional models such as precipitation as it relates to predictors such as mid-level flow.

 

Is there a method within SDSM to merge or average the predictors for multiple adjacent grid cells?

There is no method within SDSM to average or aggregate the predictors from multiple adjacent grid cells. This must be done externally.

 

After calibration in SDSM, how do I match up the GCM reference period data with local observed data by date, given that the CGCM1 model is based on a 365-day year and the Hadley model is based on a 360-day year?

After calibration with NCEP predictors, which contains the right calendar year (corresponding to 365 or 366 days, i.e. equivalent of the real calendar), some of the options in the Settings menu must be changed prior to scenario generation. Click on the Settings button at the top of the screen and check the appropriate Year Length box. Also, amend the Standard Start/End Date in line with the GCM data time–slices. For example, HadCM3/CGCM1 has year lengths of 360/365 days and the period 1961-1990 is generally used to represent current climate forcing. Once necessary changes have been made to the Settings, click on Back to return to the Generate Scenario screen. This automatically produces the “dropped” conditions from predictand datasets to be compatible with the 360/365 days of the GCM model (i.e. to match up with the corresponding Year length of the GCM and observed predictand data).

 

I am not sure when to make changes to the advanced settings in SDSM. For example, how does one decide the amount of variance inflation or bias correction?

When the model is first calibrated, the variance inflation and bias correction are set to recommended values. Once the validation occurs, the user can see how well the model matches the real data. At that point adjustment of these two values can occur to maximize that validation.

 

It seems that the fourth root transformation is the default for downscaling precipitation in SDSM. Why is this?

The fourth root transformation is effectively the default to transform precipitation as this variable is not normally distributed. Other transformation methods have been tested without any added value to the downscaling result.

 

When I use either fourth root or others type of transformations for precipitation, the result (after several steps of downscaling and climate change scenario projection) is very unrealistic. Could you direct me to any documentation on the basic theory, including the math behind SDSM?

First, the quality and data transformation procedure in SDSM gives the opportunity to transform the predictand (i.e. local temperature or precipitation from the observed station) prior to model calibration is using logarithm, power, fourth root, etc. The choice for precipitation by default is fourth root.

During the model calibration (this operation takes a user-specified predictand along with a set of predictor variables, i.e. NCEP ones, and computes the parameters of multiple linear regression equations), the process can be chosen as unconditional (i.e. for temperature) or conditional (i.e. for precipitation). In a conditional process, there is an intermediate process between regional forcing and local weather (e.g. local precipitation amounts depend on the occurrence of wet-days, which in turn depend on regional-scale predictors such as humidity or atmospheric pressure, for example).

We generally use the fourth root transformation and select the unconditional process for precipitation, and the results appear to be realistic according to the relevant choice of predictors related to the atmospheric forcing responsible to the occurrence of precipitation process.

Perhaps your choice of predictors is incorrect and/or the unconditional process has not been selected, but the problem is not likely related to fourth root transformation. If you have more specific questions related to mathematics or basic theory about SDSM, you could contact Rob Wilby and/or Christian Dawson as these two scientists have developed this software and are more familiar with the methodology behind SDSM.

 

Regarding lagging NCEP data, I read the following on the CCCSN website:

“The daily NCEP values are the average of 4 values taken at 0Z, 6Z, 12Z and 18Z (Universal Time/Greenwich Mean Time). You should ensure that the time of observation of the predictand data, which you have to supply yourself, corresponds with the time of the NCEP daily values. It may be necessary to lag (both forward and backward options exist in SDSM) the NCEP predictor data so that it corresponds more closely with the timing of the observed predictand data.”

I found out that the predictand data I am using "are for a day beginning at 0600 Greenwich (or Universal) Mean Time, which is within a few hours of midnight local standard time in Canada.” Does this mean I should lag all my NCEP values by a certain amount? If so, how do I decide which variables and how much I should lag them?

The largest lag in your case would be 6 hours. A six hour lag may not be significant enough to worry about.

We have lagged data in the case where we want the predictors to emulate persistence. For example, the previous day's value of maximum temperature may be a valid predictor of the next day's maximum temperature. When I tried this in a specific case I found no significant difference in the explained variance by lagging the data. But it may work in some cases.

Another approach that we've tried is to choose predictors upstream of the area of concern. In this case, there has to be a meteorological (physical) link between the predictor in the upstream location and the predictand in the area of concern. One example is mid-level atmospheric flow (wind) and precipitation occurrence. In one unique experiment, we chose upstream predictors for a pollution predictand, reasoning that most pollution results from long range transport.

 

Is there any literature available that would explain the following regridding technique:

“A popular regridding technique is to compute the weighted average of neighbouring grid-points, where the weighting decreases with separation distance following a Gaussian curve up to a specific max separation.”

The following reference discusses interpolation and regridding techniques:

Wilby, R.L. and T.M. Wigley (1997): Downscaling general circulation model output: a review of methods and limitations. Progress in Physical Geography 21: 530-548.

 

Should my predictand data be normally distributed and standardized when I use it in SDSM?

Yes, to the point where the precipitation data has to be transformed utilizing techniques offered in SDSM. The fourth root transformation has been used for precipitation data in Atlantic Canada.

 

When running SDSM what is the best way to choose the predictors?

As suggested in various studies, the choice of predictor variables is one of the most influential steps in the development of statistical downscaling procedure. The ideal predictor must be strongly correlated with the target variable (i.e. predictand), physically sensible and plausible, well represented in the GCM control run, and capture multi-year variability. In other words, predictors relevant to the local predictand should be adequately reproduced by the host climate model at the temporal and spatial scales used to condition the downscaled response. Prior knowledge of climate model limitations is necessary when screening potential predictors to prevent the introduction of biases in the downscaling procedure. Other complementary work must be done to systematically evaluate the accuracy of other GCM predictors, but this work is time-consuming as the size and positioning of the predictor field vary seasonally and spatially.

There is a lot of guidance on choosing predictors in the SDSM manual. Here's one approach.

Choose all the predictors and run the explained variance on groups of six or eight at a time. Out of those groups there are typically one or two predictors with the highest explained variance and those are chosen, creating a list of six or eight total predictors. You then have to see how correlated they are with each other (techniques are in SDSM). There could be a predictor with a high explained variance (say, greater than 80%) but it may be very highly correlated with another predictor. This means that you can't tell if this predictor will add information to the process and you'll have to drop it from the list. Some people have suggested aiming for no more than six predictors, since beyond that number, the additional predictors do not add to the model process in a significant way (this has been tested in several cases).

Explained variances you should expect from the predictors are: temperatures values from 60% to 85% and precipitation values from 10% to 15% (and in some cases below 10%). Don't be discouraged by the low explained variances for precipitation. This is very typical for precipitation and is most likely due to the fact that we have to transform the data and the resulting distribution, while normal, is not as statistically "elegant" as the temperature distribution.

There are other SDSM users who have tried statistical techniques to aid predictor selection with varying success. Some people endorse the above approach due to its simplicity, especially for anyone trying SDSM for the first time.

 

I have been asked to downscale relative humidity using these programs. I am having a hard time finding any literature that refers to downscaling relative humidity and have yet to come across a study where SDSM or LARS-WG has performed the task.

In our downscaling and climate impacts historical analyses, we avoid using relative humidity. For historical analysis, we prefer to use absolute humidity (e.g. dew point temperature or specific humidity) rather than relative humidity since relative humidity is less conservative on a diurnal level and less conservative among various micro-environments than dew point temperature. For example, no matter when – summer or winter – a reading of relative humidity in the early morning remains a relatively high constant (most of days with ~100%). However, dew point temperature between summer and winter early morning is much different. As a result, the use of specific humidity to develop downscaling transfer functions should obtain stronger models than the use of relative humidity.

In statistical downscaling, you should check the reliability of GCM outputs to decide which variables should or should not used in the development of downscaling transfer functions. We decided not to use specific humidity or dew point temperature as a predictor in development of the downscaling transfer function due to poor data quality of GCM-modelled specific humidity. Approximately 40–60% of the daily mean dew point temperatures calculated directly from the GCM output of specific humidity are greater than the GCM outputs of temperatures. We know that dew point temperature should be less than or equal to temperature. As a result, we have used other weather variables to downscale dew point without dew point itself.

Environment Canada - Government of Canada Important Notices and Disclaimers Created : 2007-02-20 Modified : 2007-03-21 Reviewed : 2007-02-20 Url of this page :