CAP Meeting

2 December 1998

Scripps Institution of Oceanography

Predicting ENSO Impacts on Intraseasonal Precipitation Statistics in California: The 1997-98 Event

T.P. Barnett, A. Gershunov and D.C. Cayan

Climate Research Division

Scripps Institution of Oceanography


ENSO has a significant effect on seasonal statistics of high frequency precipitation and temperature (e.g. frequencies of extremes) in both observations and GCMs. GCMs, however, do not simulate realistically the spatial features over the United States of the ENSO signal in high frequency weather. On the other hand, they do a reasonable job simulating ENSO signals in monthly or seasonal averages of such well-behaved variables as geopotential heights. The question we pose is can the large scale abilities of the GCMs be used with observations, in a statistical downscaling scheme, to produce useful GCM-derived intraseasonal climate statistics.

Observed 700mb geopotential heights have been used to to create a Hybrid model that estimates precipitation total, intensity and frequency. This model shows considerable skill during extreme ENSO events ( Figure 1 ).

Forecasts of heavy rainfall frequency are made by two models of differing complexity: (a) straight statistical forecast based on a simple ENSO index and daily observations from 1950-1995 (Statistical) and (b) statistically downscaled GCM forecast going from geopotential height anomalies to the same set of intraseasonal metrics as in (a) (Hybrid). Looking at the El Nino JFMs on record, the Hybrid model contributes significantly to the forecast skill when looking at the precipitation total, intensity and and frequency ( Figure 2 ).

As a case study, we will present forecasts of heavy rainfall frequency for JFM 1998. We compare the Statistical and Hybrid models with a dynamically downscaled GCM forecast made by a nested regional model. To the left is shown the precipitation intensity for the Hybrid model compared to observations. The green region shows where the model is in good agreement with observations (within 50 percent). The yellow and red areas show where the model underestimates precipitation and the blue and purple areas show where the model overestimates precipitation. This image illustrates the good agreement between the Hybrid model and observations for JFM 1998. In this case the use of the nested regional model does not add to the forecast skill ( Figure 3 ). A forecast for JFM 1999 from the statistical model indicates a drier-than-normal season for southern California ( Figure 4 ).


Figure 1

Figure 1 shows the skill of the observed 700mb geopotential height model in explaining precipitation total, intensity and frequency in percent variance for all years, warm years and cold years.

Figure 2

Figure 2 shows the skill in proportion of variance explained by the purely Statistical model (tropical Pacific SST - California rainfall) and the Hybrid model (GCM 700mb heights and statistical downscaling to the three precipitation variables) for the El Nino JFMs on record. This is a cross-validated measure, so negative values are possible where no skill is present. Only positive values are plotted in color and with contours at .1 intervals.

Figure 3

Figure 3 shows the ratio of the JFM 1998 forecasts to observations from the three approaches (Statistical, Hybrid and Dynamical - columns) for the three rainfall variables (rows). The heavy contour is ratio = 1. Green color represents forecast agreement to within 50 percent of observations. Yellow and red (blue and purple) are underestimations (overestimations) of between 50-100 percent and over 100 percent, respectively.

Figure 4

Figure 4 shows the JFM 1999 forecast from the Statistical model. Anomalies are shown in the left column in units of probability of being larger than a random observation from the local 1950-1995 climatology. The 0.5 (the median) contour is thickened. Larger (wetter) values are solid, smaller (drier) values are dashed at 0.1 intervals. The middle column represents total forecast fields (as opposed to anomalies). The right column shows the cross-validated skill associated with this La Nina forecast.