Temperature Index Drought

Following up on the post from yesterday, I test the assumption underpinning the regional climate change work in Australia.

The most common approach has been to assess how well each of the available models simulates the present climate of the region (e.g. Dessai et al. 2005), on the assumption that the more accurately a model is able to reproduce key aspects of the regional climate, the more likely it is to provide reliable guidance for future changes in the region.

As far as I can see this is an untested assumption, and may be a case of ‘accident chasing’.

The danger of accident chasing is that there is no advantage to selecting or weighting global climate models on the basis of concordance with regional climate, because better concordance is a statistical accident. Predictions outside the concordant region will have lower accuracy that expected. When models’ concordances are random, there is no reason to expect them to perform better on that region in the future; in fact they do worse. If so the whole basis for improvement of regional predictions by regional concordance would be flawed.

The analysis that follows is a first cut, and would be put into a more standard framework, such as estimating the change in out-of-sample confidence interval as a function of GCM weighting function. Steve McIntyre’s work in applying Brown and Sundberg of climate proxies seems like a reasonable direction for estimating the real confidence intervals of predictions with ensembles of GCMs. So please bear with me.

I needed a test to show concordant models continue to be concordant, while non-concordant continue to be non-concordant, thus justifying selecting the most concordant models. For example, the good estimates of frequency of droughts in 1900-1949 would also be good in the period 1950-1998. Models that are biased high or low in 1900-1949 would also be biased in 1950-1998.

I used the 95th percentile of temperature, and the 5th percentile of rainfall data provided for 13 models and 7 regions in the Drought Exceptional Circumstances report. These data show the area affected by exceptional high temperatures or low rainfalls, and typically have a lot of zeros. These 13 models had already been slected for concordance at the continental scale, from the 22 models in the IPCC. Ideally, I would run this analysis on these data too. However, this is the data I have at hand.

I calculated the frequency of these exceptional events for the periods 1900-1949 and 1950-1998, both on real data (observed) and the climate models (expected) for each model and region. I then correlated the difference between the observed and expected frequencies over the two periods, for each model, in each region. This gave me a correlation (or lack thereof) for each region in both temperature and rainfall (see Table below).

Above is a figure illustrating the origin of one value in the Table for the Murray-Darling Region. Each point on the figure is a difference between frequency between observed and expected frequencies. The X value of the point is the difference in frequency over the 1900-1949 period, and the Y axis is the difference in frequency over the 1950-1998 period. The value in the table is the significance of the correlation; scattered points like this one have a high value indicating no correlation between the two periods.

The idea is that a high correlation would indicate that estimates of the frequency of droughts by models are consistent by region. Models with good estimates in 1900-1949 are also good in the period 1950-1998. Models that are biased high or low in 1900-1949 are also biased in 1950-1998.

The Table below shows the significance of correlation of frequency between the two time periods. The results in bold are significant at the 95% confidence level.

95%Temp 5%Rain
MDB 0.649 0.253
NSW 0.923 0.025
NWAust 0.00072 0.002
Qld 0.097 0.057
SW-WA 0.550 0.612
SWAust 0.110 0.00007
VicTas 0.397 0.115

For reference, below is the scatter plot for the best correlated region, 95% temperature in the NWAust.

I appears from the general lack of correlations that concordance is not constant in the regionalized GCMs. This would suggest the strategy of selecting or weighting more concordant models would not increase concordance outside the observed interval. While these results don’t necessarily apply to concordance at the continental scale (to select the 13 models from the 22 models) I don’t think the result would be much different at that scale.

Of interest too, there are fewer significant concordances on extreme temperature than rainfall.

To recap, it states in Climate Change in Australia 2007, Technical Report – Chapter 5: Regional climate change projections (temperature and precipitation) that:

The most common approach has been to assess how well each of the available models simulates the present climate of the region (e.g. Dessai et al. 2005), on the assumption that the more accurately a model is able to reproduce key aspects of the regional climate, the more likely it is to provide reliable guidance for future changes in the region.

Has this core assumption ever been checked? These results suggest the assumption is not justified. To quote Lubos at The Reference Frame again:

And perhaps, most people will prefer to say “I don’t know” about questions that they can’t answer, instead of emitting “courageous” but random and rationally unsubstantiated guesses.

Here is the script. Get the table by running rank_table(). Don’t forget to uncomment the data inputs read*() first time.

Advertisements

0 thoughts on “Temperature Index Drought

  1. Pingback: wynajem samochodow

  2. Pingback: wypozyczalnia lawet

  3. Pingback: depilacja obejrzyj

  4. Pingback: link

  5. Pingback: oszczedzanieonline.blog4u.pl

  6. Pingback: filmowanie wesel Lublin

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s