Ensemble Member Early Trends vs. Later Trends

There’s been a lot of work done with respect to evaluating GCM projected temperature trends vs. observed trends, particularly in determining whether individual models or the ensemble mean can be rejected based on these trends. 

What I’ve been a bit curious about is whether the model runs with lower trends in this first decade of comparison (2001-2010) generally correspond to lower overall projected temperature increases for the remainder of the century (2011-2099).  No, it’s not an incredibly deep question, and has probably been answered elsewhere, but this gives me an excuse to work with some time series stuff in R and dive into model data.

I got the 54 ensemble members for SRES A1B from Climate Explorer. 

I’ve also included the data, along with my R code, in the link here.

Anyhow, here is a histogram of the global temperature trends for the ensemble members in the first 10 years, starting in 2001:

Next, here is what the model run trends look like for the remainder of the century:

As is clear, the variability in the first 10 years in much greater than in the remaining 90, which would be consistent with the idea of weather noise dominating the short-term trends while having the GHG forcings take over in the long haul.  14 of the 54 runs seem to show less than an 0.1 C rise in the first decade, whereas none showed a trend this low continuing over the the remainder of the century.

So, back to the original question — do early trends in these model runs give us an indicator of what the remainder of the century for that run is going to look like?

From this chart, it would appear that no, it does not. 

Of course, it may be that it is better to look at individual models instead of the ensemble members, since two runs from the same model may have different early trends but are likely to eventually converge (and hence we would not gain anything by looking at the separate runs in the early stages), whereas two different models are more likely to have different underlying processes whereby differing early trends may theoretically suggest a difference in climate sensitivity.

I’m also curious about how well these individual runs compare to the 20th century observations vs. the multi-model mean.

Population Growth and UHI – other 30 year periods

After taking a close look at the 1970-2000 period, I wanted to post some results from other 30 year periods (first available NOAA gridded pop data for the US is 1930).  There is some overlap, since we don’t have 90 years of population data, so I’m choosing to use 1930-1960 and 1950-1980 in addition to the 1970-2000 tests already done.  One benefit is that since most of the CRS to MMTS transitions occurred after 1980, I did not need to control for these changes in my pairwise comparisons (although there may be other inhomogeneities causing issues).  

Anyhow, here are the results for using log (popWithin10km+25)  and log(PopWithin1km+25) for each of the two periods:

I suppose the interesting notes here are that there is a much clearer signal in the 1950-1980 period than in that 1930-1960 period, and that neither of these show a UHI signal to the degree that’s present in 1970-2000.  Furthermore, in both of these scenarios, the 1km radius for population density results in a slightly better correlation.