In 2006, NOAA's National Weather Service (NWS) introduced the Local 3 Month Temperature Outlook (L3MTO), which extended the NWS's probabilistic outlooks for the average 3 month temperature at the local level. The next step in creating local climate products is to develop a similar outlook for 3 month precipitation totals at a specific location. Creating a local precipitation outlook is more complex than for temperature because of the high spatial and temporal variability of precipitation. These characteristics of the observation data (1) make it difficult to fit a single probability distribution to data over a large spatial area and (2) may lead to questionable predictability even in the case of an adequate distribution fit. This complexity partially explains why NWS large scale precipitation outlooks have varied predictability.

To extend the NWS 3 Month Precipitation Outlook to local sites, different statistical downscaling techniques are being tested. The simplest way is to apply the methods used in the development of the L3MTO, which used a linear regression model to identify the statistical relationship between station data and data for the corresponding CPC forecast division (FD). Many linear regression tests assume that dependent and independent variables are normally distributed. We tested the station and FD data for normality at 234 sites in the Western U.S. The study included goodness-of-fit tests for Normal, Lognormal and Gamma distributions for 2 time periods: 1950-2003 and 1971-2000, the latter is the present climatological period. The test results were stratified by potential predictability expressed as FD forecast Heidke skill score (HSS) over 1994 – present. The tests were done for 12 overlapping 3 month periods, e.g. January through March, February through April, etc. The results indicate that approximately 5% to 60% (higher in the cold seasons, lower in the late spring early summer) of stations were normally distributed when the corresponding FD HSS was above 8%. There were about 10% to 80% Lognormally distributed stations and roughly 10% to 95% of stations matching a Gamma distribution, for those with potential predictability. The tests for 2 time periods were different: more stations were viable for downscaling if trained for 1971-2000 (usual NWS practice) than for a longer period of observations.

We also report on tests of an additional downscaling method, which makes use of a regression model with normal-quantile transformation of the data. The transformation includes use of the climatological underlying distribution (Normal, Lognormal or Gamma) expressed as normal quantiles. We hope to complete development of these techniques for operational implementation of an L3MPO product in the future.