13th Conference on Applied Climatology and the 10th Conference on Aviation, Range, and Aerospace Meteorology

Tuesday, 14 May 2002: 8:45 AM
Multi-variate specification and prediction of rainfall over Devon and Cornwall, South West England
Ian D. Phillips, Univ. of Birmingham, Birmingham, United Kingdom; and G. R. McGregor
The aim of this paper is to understand mechanisms underlying precipitation variability over South West England at monthly and seasonal timescales by exploring statistical links between a regional rainfall anomaly (RRA) index for Devon and Cornwall and North Atlantic atmosphere and ocean variables. These variables include indices of large-scale surface wind flow and vorticity over the British Isles; zonal (Z), meridional (M) and vector (V) water vapour flux (VF) anomalies calculated at five western European radiosonde stations; gridded sea-surface temperature anomalies (SSTA) for the North Atlantic domain 10oN - 70oN, 80oW - 20oE; and the North Atlantic Oscillation index (NAOI). Multiple linear regression models were used to explore RRA – ocean/atmosphere associations, and develop one and two month and season lead RRA prediction models. Retrospective testing of the stability of the predictive equation sets over the model development period 1950-97 was used to isolate a sub-set of models worthy of further investigation. Independent predictions were then made using these models over the period 1998-2001.

Stepwise regression equations were able to explain between 52.8 - 88.4% (72.3 – 87.5%) of the variance in concurrent monthly (seasonal) RRA values; explanation was generally highest for the winter months. Large-scale pressure (MSLP) and vorticity anomalies emerged as the most important controls on concurrent RRA values. The inclusion of SST terms does, however, confirm that SSTs in certain regions preferentially favour above or below average rainfall over South West England and exert a greater influence on rainfall anomalies than the concurrent NAO state. In terms of their explanatory power, SSTAs dominate the predictive equation sets. Retrospective testing of the stability of the one (two) monthly lead models over the 1950-97 calibration period revealed that the one month (t+1) models for February, March, October, November and December (F, M, O, N, D) and the two month (t+2) models for March, April, May, August and November (M, A, M, A, N) had ‘at least, satisfactory performance’ and so were deemed worthy of further investigation; monthly-averaged correlation co-efficients during this period were of the order 0.4-0.5. Predictive models for spring and summer (Sp, Su) were found to consistently outperform their autumn and winter counterparts. Independent two-month lead predictions for M, A, M, A, N made over the period 1998-2001 revealed a moderate drop in model skill (r=0.342, n=19), and hence demonstrate some utility for operational long-lead prediction of the RRA. The one-season lead (t+1) summer model was found to be the most accurate when the seasonal Sp, Su models were validated for the years 1998-2001. Whilst SST variations hold the key to long-range RRA forecasting, modest increases in skill were evident when SSTAs were combined with VF, demonstrating that a multi-variate approach is essential for the specification and prediction of rainfall anomalies at monthly and seasonal timescales. Furthermore, it would appear that atmospheric water vapour flux forms an atmospheric bridge that links North Atlantic ocean variability with rainfall variations over South West England.

Supplementary URL: