Talk on WUWT about snow cover tripped a look at the datasets indicated there.
The weekly data turns out to be corrupt, something is radically wrong with the sampling, more than an erratic timebase. Tried twice to fix it, no go. Read more…
Seems a good time to make some decisions.
- Concluded that for the time being leave GISStemp alone, it is too different and liable to change.
- A lot has been achieved and so consolidating into something useful actually with users is the right move now.
- Adding fairly standard gridded datasets is easy, just added ghcn merged which took about 5 minutes, with one change needed, had to turn input scaling off. (only because I copied a template which had it on) Seems it might be one of the more interesting datasets.
What really needs doing? Read more…
Progress continues. These are the datasets already done and locally available here.
In directory listing order
- anom-grid2-1880-current GHCN
- channel_tls_tb_anom_v03_2 RSS tls
- channel_tlt_tb_anom_v03_2 RSS tlt
- channel_tmt_tb_anom_v03_2 RSS tmt
- channel_tts_tb_anom_v03_2 RSS tts
- grid_1880-2009_RVose GHCN R Vose
- tlsmonamg_5.1 UAH tls
- tltmonamg_5.2 UAH tlt
- tmtmonamg_5.1 UAH tmt
Converting most was not too difficult but RSS proved a spot of “fun”. Binary is not a problem but the documents obfuscate. It gave in but on doing a quick render to eyeball sanity, huh? Data runs from the zero meridian, not as everyone else -180 to 180. A quick and dirty hack fixes that, accumulate into two halves and write out switched over.
RSS supply data in unrounded floats, also unique. That is not how I am storing data so it is chopped to hundredths, good enough for the intended purposes. It wouldn’t be a problem storing full precision except I care about compact and fast enough. All the above database on disk uncompressed is 127M
Point here being the local database version is invariant, always same data, same place.
Computing the global mean temperature data seems to work for RSS too. I still don’t know how best to handle this for the datasets with missing data: perhaps there is no one right answer. Have a play with various schemes something to try and get a feeling for what works well.
Someone asked in private if a dataset subtraction could be done. Yes that is a todo sometime, it would be interesting to see the differences on a global basis.
Automatic plot range setting proved very easy to do since the total range in the entire dataset is always known.
(ceiling(max(abs(tmin),abs(tmax)))/5)*5 and then + and – the result as the range
This seems to give +-25C for land, surprisingly, +-10 for mid troposphere and +-30C for lower stratosphere.
Hanging question: why is the mid trop range small? Also why such bizarre looking figures?
The max/min figures, not grabbed them all yet
- UAH lower strat +26C 2009-02, -20.81C 1997-03
- UAH mid trop +6.99C 2006-01, -6.66C 1989-02
- UAH lower trop +10.39C 1981-02, -9.75C 1982-01
- GHCN R Vose +19.27 1950-01, -20.21 1917-12 (Fort Yukon, Alaska see footnote)
- Hadcrut3 +19.5 1950-01, -20.9 1917-12
- Crutem3 +19.5 1950-01, -20.9 1917-12 (should be the same, land only)
Now in the above we start to see the commonality of datasets.
The Vose all time minimum coincides with the lat/long grid cell for Fort Yukon
Since I can pull out a timeseries here it is
1917.792 -0.12 1917.875 -2.87 1917.958 -20.21 1918.042 4.94 1918.125 -3.07
Aspects of Winter Temperatures in Interior Alaska, N.A. Streten, undated, last reference 1967
Things came together. A rewrite of the software including automatic handling of different data grid sizes and new data.
As things in the lap of the Gods do, the gremlins suddenly vanished, total database rebuild of all 8 datasets went perfectly and all functions work. Earlier I had emailed UAH asking if they could unlock outside access to the new data and a couple of hours later, voila. Pulled the three files directly into the directory here, rebuild UAH, accepted it perfectly and deep breath, render. It works. Time then for a single malt and bed, in that order. A bar in the bed is very uncomfortable. (and no I am not a drinker but once in a while is nice)
Today I had a closer look at the plot, yup, makes sense. Colder than normal over central Asia, north west Europe, south east USA. Warmer over north east USA/Canada and that makes some sense given what I have read about lack of snow etc. there.
The rather large residual annual cycle in UAH and RSS might play a part. This can be removed via lookup table post 2003, too warm Dec/Jan, too cold late spring. This is unpublished. Part of the work on gridded is about trying to work out what is wrong.
Click for full size version.
I’ve change quite a few things. The aspect ratio is now about right. Minimal graticule. White seems a sensible zero colour.
What I have not done yet is sort out temperature scaling. When the dataset is translated and inserted into the database the max and min in all time is recorded. That will help.