Hide the Decline – Data at Orland is Chopped

Various Updates (Aug 17-19)

It seems this is not so much “Hide the Decline” as “Drop the Estimates” (see comments)
– Graph of USHCN adjustments compared with GISS now added, then corrected.}

The data from the weather station at Orland, CA – ‘posterchild’ station for www.surfacestations.org has been modified again. The modification throws away early years of data that are warmer than more recent temperatures.  Is there any justification for this?

Figure 1. Photo courtesy of Anthony Watts http://www.surfacestations.org

Anthony Watts and a network of volunteers at http://www.surfacestations.org have so far examined 1003 of 1221 stations in the USA and rated them for their compliance with the NOAA/NCDC station quality ratings. Orland is featured as a stable rural station:

The Orland USHCN station is located behind the Orland Water Users Association off of 8th Street in Orland CA. It has the distinction of being well sited, and having been in the same location for over 100 years. It also has not been badly encroached upon by UHI as the community has not grown significantly during the period.

The version in USHCN and its modification by GIStemp were captured in the blink comparison below dated 29/12/2008 (full station data for Orland at Surfacestations.org). There is also a good history of investigating alteration of Orland data by Steve McIntyre, with his customary thoroughness, here and here.

Figure 2. Orland USHCN raw data/GISS homogenized data blink comparison. Animation courtesy of http://www.surfacestations.org Owner: Michael McMillan

However, the current GISS Station Data for Orland has been truncated, changing the station’s contribution to the surface record yet again.  Figure 3 below shows an overlay of current data on a graph of station data prepared using Alan Cheetham’s graphing software and 2009 GHCN download (Climate Data Visualiser) at Appinsys.

Figure 3. Current (July 2010) GIStemp data from Orland overlaid on plot of (GHCN V2) data from Appinsys (www.appinsys.com/GlobalWarming/)

With the current GISS adjustment the older data is warmed slightly, reducing the trend in line with compensation for UHI, but what the overlay shows is that the  unadjusted data had been made cooler in the past anyway and even when adjusted the data in 1905-1930 is cooler than the previous GHCN version of the data.  This creates a warming trend to present.

What about local stations?  Orland tracks quite closely with Chico Univ Farm and Willows 6W.  Willows 6W is truly rural, only ~17 miles away according to the GISS data, Chico Univ Farm just over 20 miles away is, just as its name suggests, on an agricultural station to the south of the city. Red Bluff/Mun (28 miles to the north) and Ukiah (68 miles south and west) are towns with populations just over 13,000 and 15,000 respectively.  Ukiah is outside of the Central Valley, closer to the coast, and cooler; it is included under the same 5-digit WMO Station number (72591). Oroville Usa, Redding Wso and Marysville (beside Yuba City in Figure 5 below) are all closer than Ukiah.

Figure 4. GHCN data for Orland and four neighbouring stations within the same WMO ID set (Graph generated by Appinsys Climate Data Visualizer)

Using the USHCN Map Interface to look at relative locations (Figure 5) of the stations, Red Bluff is not a USHCN station (nor is Oroville).

Figure 5. USHCN Stations in N. California around Orland

So how are the other stations treated? Well I’m sorry to say there is truncation here too, and not in a way that makes any sense, but more of that later. The start dates for each location’s station data by GISS are:

  • Orland: GISS – 1903 (USHCN – 1895)
  • Chico: GISS – 1902 (USHCN – 1895)
  • Willows 6W: GISS – 1903 (USHCN – 1903)
  • Red Bluff: GISS – 1890 (USHCN absent)
  • Ukiah: GISS – 1895 (USHCN – 1895)

So we have truncation of early data first by USHCN then further by GISS. The usual justification for truncation by GISS – that of lack of sufficient rural stations in early years –  cannot apply (even after changes due to Nightlights Radiance) as Red Bluff is not truncated, nor is Oroville (start date 1884).

The data and adjustments,  first by USHCN (for Orland)…..

  • RAW (TMEANRAW) is the unadjusted monthly mean temperature
  • TOBS (TMEANTOBS) is the monthly mean temperature adjusted only for the time of observation bias
  • Mean Temperature (TMEAN) is the fully adjusted monthly mean temperature (including homogeneity adjustment and urbanisation correction)

…then by GISS (After combining sources at the same location (unadjusted); After cleaning/homogeneity adjustment).  Although expecting GISS to use the USHCN output as input to GIStemp, a brief comparison suggests that none of the USHCN datasets fully match the GISS combined or adjusted data for Orland.  This could be a simple error on my part – I’ll check and update [update 17 Aug – see below].

The GISS adjusted anomaly plot for each station (Figure 6) shows how well the homogenisation has, er, homogenised the data.  I haven’t added trendlines but the adjusted trends are similar for all the N. Central Valley stations – 0.41°C/Century (Willows 6W); 0.55°C/Century (Chico); 0.58°C/Century (Orland); 0.65°C/Century (Red Bluff). Are microclimate responses to variations in temperature that uniform? I really don’t know but instinct says surely not?

Figure 6. Anomaly plots of GISS adjusted data (Orland data is now in black, which is easier to see)

Going back to Figure 4, USHCN current data is altered from what had been previously captured there (and in Figure 2) – the initial RAW Mean Temperature for Orland is 17.4°C (1904); the Adjusted Mean in 1904 is 16.8°C, but the Adjusted data goes back to 1895 which is now reported as 16.3°C. I can only presume this is using infill from other stations. Perhaps I need to do a bit more reading on this.

In the meantime, here’s why I have a problem with the way the data for Orland and nearby stations is now reported. If a single station shows a response for several years that is not reflected in nearby stations this can be symptomatic of some alteration of the station location, equipment or surroundings; that, I agree should, be adjusted. But when there is a very similar response in nearby stations, that tends to suggest short-term or relatively rapid alteration in the local climate (or weather).  Orland’s location history (see first few comments) seems unaltered; Chico and Willows 6W show reasonable agreement with these warmer temperatures prior to 1903.  Oroville also shows a similar pattern and this is kept by GISS, even after adjustment (Figure 7), although there is a ~0.5°C reduction in the oldest temperatures due to the adjustment.

Figure 7. GISS adjusted data for Oroville

If station and instrument are stable, what is the justification for hiding the decline? I can only think that land use change in early settlement and commencement of agriculture in the Central Valley might explain the temperature drop, but would this account for ~2°C drop in ~25 years at Orland, Chico, Oroville and Willows?  I have found some papers related to this and will update when I have the time to read them properly.

Updates 17 Aug.

First I have plotted the Raw, TOBS and Adjusted USHCN data with GHCNV2 data for Orland:

Figure 8. Comparison of USHCN adjustments with GHCNv2 (unadjusted) data for Orland.

Secondly I can’t find any flaw in my file comparing GIStemp and USHCN so here is the graph:

Figure 9. Overlay of GISS Unadjusted and Adjusted data on USHCN data for Orland.

If you look closely, the GISS data (from here) is not an exact match to any of the stages of USHCN adjustment (from here) . This is was a surprise, but see Figure 10.

Also from David Jones in comments, if the early data for Orland is estimated – would that not have been valuable to know before now? What other early data is estimated?

Update 19 Aug.

Davidd Jones also pointed out that the likely reason I did not get a match between the two data sets was because the USHCN annual mean was is calculated with January-December annual data, whereas I had used the same method as GISS of first calculating seasons DJF, MAM, JJA, SON, then deriving a Dec-Nov annual mean. He was correct (Figure 10). Minor differences that remain are likely to be because of (my) lack of attention to detail in handling years with missing months (1907, 1995).

Figure 10. Replot of Figure 9 with GISS Annual Mean recalculated (Jan-Dec)

This is a confirmation, if anyone needed it, that GISS unadjusted data is the same as USHCN Adjusted data.

About these ads
This entry was posted in GIStemp, Station Data and tagged , , . Bookmark the permalink.

19 Responses to Hide the Decline – Data at Orland is Chopped

  1. KevinUK says:

    verity,

    The Orland, Chico, Oroville etc charts show clear evidence of the cyclic cooling (1980 to 1910) followed by warming (1910 to 1940), followed by cooling (1940 to 1970) followed by warming (1970 to 2010) trends that correlate very well with the ocean cycle particularly ENSO and PDO and not very well with atmospheric CO2 concentration over the same multi-decadal period (1880 to 2010).

    Its therefor eno surprise to me that GISS are cherry picking the data from these stations in order to ‘hide’ the clear evidence of teh fact that these cooling/warming trends are primarily if not wholely due to nature climatic variability.

    • David Jones says:

      Kevin, what cherry picking do you think GISS are doing? They are just using publicly available temperature datasets (GHCN, USHCN, SCAR READER). They are not responsible for producing or maintaining these datasets.

  2. David Jones says:

    I can explain why the USHCN record for station 046506 (Orland) starts in 1895 but the GISTEMP data starts in 1903. That’s because the USHCN values from 1895 through to 1902 (and some of 1903) are flagged with an ‘E’, which according to the documentation for USHCN v2 indicates “value is an estimate from surrounding values; no original value is available;”. GISTEMP discards such values, thus the GISTEMP record starts in 1903.

    You wouldn’t want GISTEMP to use estimated data, right?

    • Verity Jones says:

      Thanks – oh LOL – it’s OK for GIStemp to make estimates, but they don’t like to use those produced by NCDC ;-)

      But actually that just throws up more questions (which I don’t expect you can answer – they are probably just rhetorical – after all I think I spend most of my time talking to myself anyway!)

      First of all the estimates must be based on something, I assume some original data, as there is a record of the station since 1883 (http://climateaudit.org/2009/06/29/orland-ca-and-the-new-adjustments/#comment-186449). What has happened to it – has it basically been discarded completely? Clearly someone in a professional capacity at NCDC must have used skilled judgement or algorithms to base estimate on this for inclusion.

      Even if the data is too fragmented to produce even a useable year, I hope it can be used for something. If the period around 1900 was as warm as suggested by Orland and surrounding regions that information is valuable – not least becuse the changes in the intervening years may be induced by man’s influence (land use changes), or by the climate changing, or by some combination of the two. Local effects and microclimates are important (of course it is now the aspiration to start collecting more data on this – ironic).

      Then there is the issue of other stations in the area also reporting warmer temperatures in the 1900s – or are these based on estimates too? Oroville is still used by GIStemp, but it is not part of USHCN, so it is not trucated. Does that mean that the estimates were there in GHCN but not flagged as estimates? If not flagged in GHCN if this part of the new code for USHCN?

      • David Jones says:

        I went to NCDC’skinda cool kinda clunky tool for downloading (scans of) original coop reporting forms. The first form is for March 1903 (and because I looked at the form I can say that the monthly mean for March is computed using 29 days of data). If there are original data, NCDC don’t seem to have it. Perhaps it was lost by South Pacific Railroad. So, I have no reason to suppose that the data flagged as ‘E’ are based on original data (at that station). It seems like that “no original value is available”, just like it says in the documentation (referred to in my comment above).

        Reply – Well I am shocked. That suggests the other stations that are similarly treated are also estimated. How widespread is this? No don’t answer I can now look for myself. Thanks for providing the links. I think I may have stumbled on it previously but, as always there are so many information sources it is very easy to overlook one that you’ll subsequently find useful.

  3. David Jones says:

    In figure 4 your plotting of the “average temperature” is obviously wrong for station 42572591000 (“RED BLUFF/MUN”). The huge spike you see in 1988 is because that year is missing data for the months of January, February, and March. Your plotting program is computing the average for a year by averaging all of the present monthly temperatures. If you’re going to do something that naive, you should only do it when there are 12 present values for a year. It is much better to compute a yearly mean via monthly anomalies; GISS describe the method that GISTEMP uses.

    Reply – If you are going to call me something try ‘lazy’ not ‘naive’. Learn to read! I have very limited control over the data in someone else’s graphical interface. Compare Figure 5, which is my original, just colour matched.

    • David Jones says:

      Okay, put in another way: the plotting software you have chosen to use is too naive and I would not trust it to produce yearly averages.

      What figure 5 are you talking about? Your figure 5 is a map. Your figure 6 shows the same stations but shows yearly anomalies, not yearly temperatures.

      • Verity Jones says:

        Actually I think the Appinsys database is quite useful. It is not perfect granted, and there are a few glitches in it, but it is useful for an instant comparison of data as each is treated in the same way. In this case it was useful for the shape of the data.

        For yearly averages perhaps this is better? – this database excludes any years with one or more missing months and does calculate yearly averages from seasonal ones.

        My you are being nitpicky – I meant Figure 6 and the intention was to show that the style was different from Figure 4.

      • David Jones says:

        Oh. Nesting overload.

        That climateapplications.com thing looks pretty. The exact data sources in question (raw and adjusted) could do with being clarified.

      • David Jones says:

        Thanks.

        Some of the numbers for Red Bluff on climateapplications.com are suspect. I know you’re not responsible for the website, but you know a man who is?

        The graph showing the trend from 1880 to 1909 (the top left of the 4 little graphs) has the wrong legend for the raw data. The plot looks okay, but the text says “Raw data 1970 to 2010 (0.08 C/Century)”; that’s the write trend value for the period 1970 to 2010, but it shouldn’t appear on that chart.

        Below the charts the trends appear as tables. The slope for the 1880 to 2010 period for the MAM mean for the raw data is listed as -87.79. That’s obviously not right. And given that that one is obviously it makes the MAM mean for the raw data for 1880 to 1909 look suspect at -5.11. Though I haven’t actually checked.

      • Verity Jones says:

        I’m sure Kevin will pick up on your comment soon. There are a few glitches in the software that haven’t been ironed out yet and they only get looked at when time permits.

  4. David Jones says:

    As to why USHCN data does not match GISTEMP “raw” data: That’s because GISTEMP adjusts the USHCN so that the recent seasonal cycle matches the seasonal cycle of the GHCN data for the same station. That’s important to do prior to the next step of processing where records for the same station are combined; failing to match seasonal cycles would result in a step change in temperatures. When considering a single record this has no effect on anomaly or trend, when considering multiple records for a single station it removes spurious jumps.

    This is documented in GISS’s description of their analysis.

    The net effect on yearly temperatures, as in your Figure 9, should be that “GISS unadjusted” is a constant offset from “USHCN adjusted” (except for ‘E’ flags and years with missing months, as previously noted). Eyeballing the graph, that looks extremely plausible.

  5. AFPhys says:

    Excellent study.

    Please, if you are able, replace the green for Orland in “Figure 4. GHCN data for Orland and four neighbouring stations…” and “Figure 6. Anomaly plots…” with black.

    Thank you.

  6. Pingback: TWAWKI » The IPCC science revisionists

Comments are closed.