MACA
What is MACA?
Multivariate Adaptive Constructed Analogs (MACA) is a statistical method for downscaling Global Climate Models (GCMs) from their native coarse resolution to a higher spatial resolution that captures reflects observed patterns of daily near-surface meteorology and simulated changes in GCMs experiments. This method has been shown to be slightly preferable to direct daily interpolated bias correction in regions of complex terrain due to its use of a historical library of observations and multivariate approach.
Datasets
We have produced two different downscaled datasets covering CONUS-plus. Both use a common set of 20 CMIP5 GCMs from models that provided daily output of requisite variables for historical (1950-2005) and future experiments under RCP4.5 and RCP8.5. A requirement for any statistically downscaled product is historical data, or training data, that GCM output is bias corrected to. The two products we produced use different training data. Briefly, one dataset uses the 6-km (1/16th degree) daily product of Livneh et al. (2013) from 1950-2011 that also incorporates the Canadian portion of the Columbia River Basin. The other uses the gridMet daily dataset at a ~4-km grid (1/24th degree) from 1979-2012.
Maximum temperature, minimum temperature, maximum and minimum relative humidity, precipitation accumulation, downward surface shortwave radiation, wind-velocity, and specific humidity.
Maximum temperature, minimum temperature, maximum and minimum relative humidity, precipitation accumulation, downward surface shortwave radiation, wind-velocity, and specific humidity.
Methods
MACA is a multi-step procedure that uses bias correction procedures and a constructed analogs approach for developing the fine-scale spatial pattern using a library of observed patterns. The analog concept uses the idea that large-scale patterns such as those adequately simulated by GCMs can be matched using a combination of observed patterns. The fine-scale "footprint" accompanying these observed patterns can then serve as the basis for the fine-scale features corresponding to a model "day". A more thorough explanation and videos can be found here.
accuracy
MACA was validated using reanalysis (a surrogate for a "best case" GCM) across the western US. MACA was compared to a statistical downscaling process that used linear interpolation as opposed to analogs. We demonstrated slightly improvements in overall measures of skill using MACA than interpolation based approaches.
Data limitations
- Bias correction methods can not account for all aspects of bias. Choices made in the seasonal window of bias correction will preserve GCM signals for the time period of bias correction (e.g., a 30-day window), but may not at longer timescales.
- Statistically downscaled data generally do not effectively capture how changes in land-surface feedbacks (e.g., snow-albedo) influence local climate;
- Imperfections in the training dataset associated with non-climatic factors will be passed on to the downscaled dataset.
There are several options for acquiring MACA data outlined below.
Interfaces for specifically extracting the data
Directly access netCDF files
Data Services
Interfaces for specifically extracting the data
- Download data for specific locations as a csv file
- Download gridded data for specific region(s) as netCDF files
Directly access netCDF files
Data Services
- All MACAv2 datasets are available on the Northwest Knowledge Network (University of Idaho). A tutorial of how to use OPENDAP with code snippets for various programming languages is provided here.
- The gridded 1/16-deg(~6km) MACAv2-LIVNEH dataset
- The gridded 1/24-deg(4km) MACAv2-METDATA dataset
- The gridded 1/16-deg(~6km) MACAv2-LIVNEH dataset is available on North Carolina Climate Office's THREDDS server.
- The gridded 1/24-deg(4km) MACAv2-METDATA dataset is available on the Geo Data Portal(GDP). A tutorial we developed for extracting data from GDP can be viewed here
- The gridded 1/24-deg(4km) MACAv2-METDATA dataset is available on the Google Cloud through Google Earth Engine.
Question: Are the MACA data for year 1987 saying something about the weather in the actual year 1987?
Answer: No. The historical period of the MACA data is from 1950-2005, but these years do not correspond to the actual years 1950-2005. What is important here is that these years have the same statistics as the actualy years 1979-2009 (from the training data). The year 1987 of MACA data is not meant to be a hindcast of the weather from that year.
Question: Can I use the MACA data to look at projected future changes for 2040-2044 compared to the years 1990-2000?
Answer: GCM experiments are meant to say something about the future projections of climate. To assess climate, you should be taking averages of the weather over at least 30 years of data. So 5 future years is too short to assess the climate and even 10 years of historical is too short.
Also, note that the choice of the baseline period (here 1990-2000) may have other consequences on your comparison. The baseline period is a reference period for calculating the respective future climate change. Choosing Different reference periods for the baseline will result in different projections for change. Further, baseline periods that are chosen as subsets of the period 1950-2005 (i.e. 1990-2000 or 1971-2000) will result in variations in the baseline between the models since the MACA process maps the statistics of the period 1950-2005 to the statistics of the training data. If you wish to avoid inter-model variations in the baseline, choose 1950-2005 as a baseline. As seen in the 'Analysis - Bias Maps' tab on this webpage, this will result in minimal biases between the models and the training data.
Also, note that the choice of the baseline period (here 1990-2000) may have other consequences on your comparison. The baseline period is a reference period for calculating the respective future climate change. Choosing Different reference periods for the baseline will result in different projections for change. Further, baseline periods that are chosen as subsets of the period 1950-2005 (i.e. 1990-2000 or 1971-2000) will result in variations in the baseline between the models since the MACA process maps the statistics of the period 1950-2005 to the statistics of the training data. If you wish to avoid inter-model variations in the baseline, choose 1950-2005 as a baseline. As seen in the 'Analysis - Bias Maps' tab on this webpage, this will result in minimal biases between the models and the training data.
Question: I like the projections I've seen with the model CNRM-CM5. Can I use only this model in my study?
Answer: The intended use of the CMIP5 project is to get statistical information on future climates from the many different models available. You should use as many models as possible in looking at your study in order to get a good signal on the predicted change for the future (as well as some information on errors or uncertainties between the models). We suggest using at least 10 models.
Question: I aggregated the daily MACA data to annual values but I'm confused that their statistics do not match up with annual values from the training data.
Answer: The downscaling process is performed on the daily GCM data using the daily training data. This ensures that the distribution of data in 15-45 day windows is mapped to the training data and does not guarantee that the the distribution of annual data matches the training data. Also, each GCM has its own sequencing of daily data, so that aggregations of the daily data to monthly/seasonal/annual values are not likely to match similar aggregations of the training data.
QUESTION: I am having problems with the netcdf files.
Problem: Users may encounter problems using MACA netCDF data using CDO (Climate Data Operators) tools from the Max-Planck-Institut fur Meteorologie. These problems stem in that the MACAv2-METDATA/LIVNEH netCDF datasets are in the netcdf4 format and that these files do not have time as the UNLIMITED dimension. In general, I believe that CDO does not support netcdf4, but needs netcdf3 (classic) format.
Solution: The easiest thing to do to fix this problem is to convert the netcdf4 files to netcdf3-classic format. This can be done using either of the following options which require NCO:
Solution: The easiest thing to do to fix this problem is to convert the netcdf4 files to netcdf3-classic format. This can be done using either of the following options which require NCO:
- nccopy is available as part of the netCDF software distribution. This can be used to convert a netcdf file from netcdf4 to netcdf3 with:
nccopy -k classic filenetcdf4.nc filenetcdf3.nc using this guidance - ncks is available as part of the NCO (netCDF Operators) library and can also be used to convert a netcdf file from netcdf4 to netcdf3 format with:
ncks -3 filenetcdf4.nc filenetcdf3.nc using this guidance
- ncdump -k filenetcdf3.nc to check the netCDF version, i.e. classic vs netCDF-4
- ncdump –h filenetcdf3.nc to show that the time has unlimited dimensions giving time = UNLIMITED ;
Reference
Abatzoglou J.T. and Brown T.J. A comparison of statistical downscaling methods suited for wildfire applications, International Journal of Climatology (2012), 32, 772-780
Licensing
The MACA datasets were created with funding from the US government and are in the public domain in the United States. For further clarity, unless otherwise noted, the MACA datasets are made available with a Creative Commons CC0 1.0 Universal dedication. In short, John Abatzoglou waives all rights to the work worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law. You can copy, modify, distribute, and perform the work, even for commercial purposes, all without asking permission. John Abatzoglou makes no warranties about the work, and disclaims liability for all uses of the work, to the fullest extent permitted by applicable law.
To the extent possible under law, John Abatzoglou has waived all copyright and related or neighboring rights to MACA Datasets. This work is published from: United States.
To the extent possible under law, John Abatzoglou has waived all copyright and related or neighboring rights to MACA Datasets. This work is published from: United States.