We are aware that users of the MACA dataset will try many different approaches to come to conclusions about changes in future climate. Unfortunately, some of the possible ways of using the dataset are not obviously erroneous or non-ideal. Therefore, we have come up with a short list of advice on using the MACA dataset. First, here are some advised general steps for the calculation of a change projected by a set of GCMs.
The following order for an analysis of a change in metric is advised:
Climate is usually determined by taking an average of at least 30 years of data, where more is better.
The MACA dataset is gridded on a 4-km scale. The region you are interested in may overlap more than one grid point in the dataset. Likely, you will want some average value of a metric over your area. Since even in 4-km, there can be quite a bit of spatial variability (i.e. over more complex landscapes or near the coastline) so that one needs to be careful about averaging meteorological variables over the spatial grid. It all cases envisioned, it is suggested that your target metric be derived at each grid point in your target area. Then after you look at the results over the spatial grid, decide which grid points are really of concern to your goals and average your metric over those grid points.
To establish a historical baseline for the statistics of a certain area, the full 56 year historical period (1950-2005) from each GCM should be utilized. The MACA process maps GCM values to values from the training observation dataset, so that only the full 56 years will give the proper statistics held in the training dataset. Utilizing a subset of the 56 year historical period, will not guarantee that the extremes typical of the data will be seen or that the resulting statistics will be typical of the historical period.
To ascertain the changes predicted by each GCM, the same statistical quantity should be derived for both a future period(i.e. 2040-2070) and for the full historical period (1950-2005) for each GCM. The difference between these quantities for these periods then reflects the future anomaly predicted by the GCM.
We provide future predictions from several different GCMs and for different future scenarios. It is never good practice to obtain statistical information from a sample size of only 1 or 2. It is ideal to use as many of the models as possible in getting a good signal on the predicted change for the future (as well as some information on error from the uncertainty between the models).
The GCMs are not expected to get a correct value on any day/year of either the historical or future periods. Therefore, the meteorological values on any day/year should not be averaged over the GCMs before determining a metric. Instead, the metric should be calculated for each model independently for the season and time period and only metrics should be averaged over the GCMs.
The MACA statistical downscaling process applied to daily CMIP5 data guarantees that the statistical distribution of the historical period of the downscaled data agrees with the training dataset at the daily time scales. However, aggregating the daily downscaled data to longer time scales (i.e. monthly, seasonal or even yearly) does not guarantee that the statistical distribution of these quantities for the historical period will be preserved. This may have unintended consequences.