Miscellaneous Plots

Click here for main page. Click here for SysII paper info. Click here for model forcing background.

System II Paper

Background stuff

1. Further investigation of results obtained from regression analysis: <>
   1. Correlation between gph500 ensemble mean and obs (DJF) here. Note: White space is not meant to be any hard and fast significance level...just a abs(0.4) cutoff for clarity.
   2. Time series of obs vs ensemble mean (gph500) for region of central Europe showing large negative correlation. Timeseries averaged over x=29E:34E/y=45N:52N spatial region.
   3. With interest in how spatially independent the data is: Teleconnection type plot wrt timeseries at 15E 40.1N for Europe and the world
   4. It should be noted that the minimum correlation point over Europe, is NOT co-located with the region of largest time variance difference (F-test). This is a trick of the eye.
      A plot of activity at the largest negative correlation point is show here, whilst that for the minimum f-value is show here. For the former, a significant obs/ensemble mean correlation (negative) is seen, and the individual members are quite spread. The f-test differences are only significant for two members, and in 5 cases the obs have a higher variance. In the latter, the correlation is not significant and the members show tighter clustering. Almost all the members are seen to have significant differences in time variance when compared with the obs. These plots can be seen in a loop here to aid comparison.

      A further plot looking at these two regions. Here, ensemble member correlations with their mean are calculated and their distribution fitted. Dot/dash lines represent +/- 2 dtandard deviations, and the thick dashed line represents the correlation of the observation with the ensemble mean.

      The same plot using the 9-ensemble DEMETER runs.

      Extending this analysis to a global sense. Shaded regions indicate correlations significantly (99%) outside the ensemble correlation cluster. Differences can be explained by:

      1. In the tropics, lack of ensemble variability.
      2. In the extra-tropics, it can be explained by model errors.

2. Further investigation of data table in paper:
   1. Comparisons between t_avg and probability data. (Prob=P_forecast-P_climate).Relationships between the two gross timeseries in each area are not seen to be significant. The t_avg values between the Atlantic and Pacific are highly correlated.
   2. Comparisons between t_avg and probability data. This is the raw comparison ie P_forecast values.

3. Combining anomaly (sign) agreement with t-value data. For each year, t-values are binned. ie find the area correspoding to a a t-value between 0-5 and calculate the probability of the ensemble mean anomaly agreeing (in sign) with the observations.
   Calculations are performed for Global and reduced area, and relate to t2m. Note:
   1. This is over ocean data only. Land has been masked out.
   2. Gaps indicate lack of data for that bin range.
   3. This is the first iteration... I need to check everything is correct (particularly the strange final season).

   The final season problems seem to be due to a generally bad simulation, coupled to a high predictability suggested in the South Pacific.

   Breaking the regions down and applying the same analysis leads to this. The longitude extents are 260W:110W (Pacific) and 110W:35E(Atlantic). Latitude bands (tropical and extra-tropical) are indicated in the plot titles. El Nino years are indicated by overplotted dashed line, La Nina by dot/dashed lines.

   Same data plotted again, this time with rval vs probability (each year represented by one line).

   The above results calculate a t-val using a time averaged intra-ensemble standard deviation in the demoninator. Producing the same plot again with a time varying demoninator produces only small differences. Thus, the temporal variability of the variability is much less than the intra-ensemble variance. This can further be illustrated by a plot of the t2m (stdev(stdev_i)/avg(stdev_i))*100 (giving a percentage). It can be seen that there are areas of significance in the tropical Pacific.

Model Forcing

Wind Stress Anomaly

Click here to compare ASO wind stress patterns with DJF.
Click here to compare ASO wind stress patterns with mean annual field.