TEMPORAL DISAGGREGATION OF PROBABILISTIC SEASONAL CLIMATE FORECASTS

Authors: Schneider, Jeanne; Garbrecht, Jurgen

Submitted to: American Meteorological Society Proceedings
Publication Type: Proceedings
Publication Acceptance Date: 10/1/2002
Publication Date: 10/1/2002
Citation: Schneider, J.M., Garbrecht, J.D. Temporal disaggregation of probabilistic seasonal climate forecasts. Available from: http://ams.confex.com/ams/annual2003/techprogram/paper_52785.htm Proceedings of the American Meteorological Society 14th Symposium on Global Change and Climate Variations. [2003].

Interpretive Summary: Seasonal climate forecasts are issued by NOAA for average temperature and total precipitation over 3-month overlapping periods covering the coming year. Many crop and hydrologic models employ weather generators based on monthly statistics to produce sequences of daily weather (e.g., air temperature, precipitation, solar radiation). To make the seasonal climate forecasts immediately useful for applications using weather generators, the forecasts for 3-month periods need to be reformatted in terms of 1-month forecasts. A previously proposed mathematical treatment of the problem produced physically unrealistic results. In this study, alternative approaches based on experience and observation were explored, and two were chosen for their ability to improve on the mathematical approach. Of those two methods, one is identified as the best current approach to transform the 3-month forecasts into a sequence of 1-month forecasts. That approach assumes that each of the 3-month forecasts is applicable to the middle month only and these are subsequently adjusted to best reflect the original 3-month forecasts. Two example applications demonstrate the good performance of this method.

Technical Abstract: Seasonal climate forecasts are issued by NOAA/CPC for average temperature and total precipitation over 3-month overlapping periods covering the coming year. Many crop and hydrologic models employ weather generators based on monthly statistics to produce stochastic realizations of daily weather (e.g., air temperature, precipitation, solar radiation). To make the forecasts immediately useful for applications employing weather generators, the forecasts for 3-month periods need to be disaggregated into 1-month, non-overlapping increments. Previous approaches based on algebraic inversion of the overlapping forecast sequence have produced unrealistic results, with strong forecast anomalies erroneously propagating into periods with zero forecast anomalies. These unphysical results are a consequence of the manner in which the forecasts are generated at this time, and are a characteristic feature of all such algebraic inversions. As an alternative approach to the problem, heuristic approaches to the disaggregation have been developed that produce physically plausible sequences of 1-month non-overlapping forecasts, which also re-aggregate to a good approximation of the original sequence of 3-month overlapping forecasts. Two heuristic approaches to disaggregation of 3-month overlapping seasonal climate forecasts are presented with illustrative examples.