Brief description and results of prediction using CAS method.
The Huang et al(1996) model has been integrated at 344 CDs over the US from 1932 to the present and has produced monthly data sets of E (evaporation), w (soil moisture), R (runoff), G(loss to groundwater), along with the observed inputs P (precipitation) and T (temperature). These data sets are up to date as of yesterday 12Z on each day. We can thus define an initial condition and launch forecast tools thereof. The current state of the national soil moisture anomaly is used to construct an analogue (for a description see article at
The constructed analogue allows us to make monthly and seasonal predictions of all of the above mentioned variables. We have done this in real time since early 1998. At CPC this tool is called CAS. Here we present results going back to 1981, where the forecast (hindcast really) for each year in the past prior to 1998 is done under cross validation, i.e. the data for the year we make forecasts for is not used in any way, except as verification. Correlation as a measure of skill in map form is given for both monthly and seasonal T&P forecasts for leads of -2 to +5. We only portray skill when the correlation is >=0.2, in the remaining areas there is, a-priori, no point believing the forecast. In order to increase sample size we pooled with neighbors, i.e. results for July are pooled with June and August. The real time forecast at the shortest possible lead would be lead +1. The -ve leads are specification problems, and are interesting for their own sake, and probably foreshadowing areas of sensitivity to soil moisture that may be seen in the forecast as well . For instance, at lead -2, seasonal, what do we know about JAS T&P given the soil moisture in August; at lead -1, monthly, what do we know about July T&P given soil moisture in July.