"Probabilistic" Forecast Maps

The maps of forecast probabilities consider categorical forecasts, in which climate anomalies are classified as "Above Normal", "Near Normal", or "Below Normal". The probability maps are based on the forecast from a particular climate model as well as on the historical performance of that model (i.e. how well the climate model simulated the observed climate variability when forced with the observed sea surface temperatures (SSTs)). We wish to know the likelihood of a given forecast, but we don't have observations for the future. So, we look to the information we have on how well the model agreed with the observed climate variability during the historical simulations.

FORECASTS BASED ON THE ENSEMBLE MEANS We start by examining how well the model-simulated year-to-year variability, covering the years 1950 forward for some given season, agrees with the observed variability in those years. As of early 2001, the climatological base period is designated as 1969-1998 (close to the 1971-2000 climatology period used by IRI's Monitoring Group and many National Meteorological Services, but still incorporating the higher quality, higher resolution observed data available from the Climate Research Unit of the University of East Anglia). The seasonal temperature and precipitation anomalies, which are the deviation from average 1969-1998 seasonal conditions, are categorized as "Above Normal", "Near Normal", or "Below Normal" for each [model] point on the globe. At each point, the wettest 1/3 of the years from 1969-1998 define the "Above Normal" category for precipitation at that point for a particular season, the driest 1/3 of the years from 1969-1998 define the "Below Normal" category, and the values during the other 10 years in the 1969-1998 period define the "Near Normal" category. The categorical values for temperature are similarly determined. These categorical determinations are made for both the observed climate variability and the historical model-simulated climate variability, using the ensemble mean of the model simulations.

Next we consider how often the model correctly simulated the conditions for each category, during the historical simulation. For example, when the model ensemble mean says it will be "Above Normal" precipitation, how often was the observed precipitation "Above Normal"? At a particular location, it may be that for 60% of the years when the model indicated "Above Normal" precipitation, the location was observed to receive "Above Normal" precipitation. This means that 40% of the years when the model indicated "Above Normal" precipitation, it was not observed to be "Above Normal" precipitation. It may be that 30% of those years the observed precipitation was "Near Normal", and in 10% of the years the precipitation was actually "Below Normal". From this information, one can tell that there is very low probability that there would be "Below Normal" rainfall at this location if the model were to show "Above Normal" conditions.

The forecast maps show the probabilities of each of the three categories ("Above Normal", "Near Normal", "Below Normal"), based on the category indicated by the ensemble mean forecast, and the model's past performance with respect to the observations when predicting that category.

The information given in the "Above","Normal", and "Below" maps is then combined into a "Re-built Forecast". This forecast gives the most likely qualitative forecast based on the actual quantitative climate forecast from the model and the historical performance of the model. There are 5 categories for the "Re-built Forecast".

 
For precipitation, the categories are: 
Dry (D), Not Wet (NW), Normal (N), Not Dry (ND), and Wet (W).  
For temperature the categories are: 
Cold (C), Not Hot (NH), Normal (N), Not Cold (NC), and Hot (H).

The cut-off percentiles chosen to rebuild the forecast are given below. They are very general, and may be inappropriate for the purposes of some users. Thus some users may prefer to focus only on the probabilities given in the "Above","Normal", and "Below" maps.

For each [model] point, 
if       A>=50% --> RB fcst = A 
else if  N>=50% --> RB fcst = N
else if  B>=50% --> RB fcst = B
else if  A<30% and B>A  --> RB fcst = NA
else if  B<30% and A>B  --> RB fcst = NB
else --> RB fcst = none

where,
RB fcst = Re-built forecast
A=Above Normal
N=Near Normal
B=Below Normal
NA=Not Above Normal (NW for precipitation; NH for temperature)
NB=Not Below Normal (ND for precipitation; NC for temperature)

There is no forecast in the regions where the model (because of skill or the current climate signal) can't distinguish the categorical forecast beyond guessing (i.e. about 33% probability of "above", "normal", and "below"). The rebuilt map is shaded grey in these regions.