Specification and prediction of monthly and seasonal precipitation amounts in California and Arizona river drainage basins

A forward selection procedure of stepwise regression is applied to 30 years of data to develop a set of multiple regression equations for percentiles of monthly and seasonal precipitation amounts (P) in high altitude river basins of California and Arizona during periods of 1–4 months in duration. Two main methods are used. The first is specification of P from simultaneous values of the field of 700 hPa mean height anomalies (H). The second is pure prediction, which uses only previous (observed) data as potential predictors. On average, 61 per cent of the monthly P variance is explained by specification equations containing about three concurrent values of H; seasonal specification equations explain about 2 per cent more of the variance by means of slightly fewer heights. In contrast, only 16 per cent of the prediction equations are significant at the 95 per cent level determined by a Monte Carlo test. All these equations select the local value of P observed during the month prior to the target period as the most important predictor.