The long awaited surfacestations paper

 Our paper

Fall, S., A. Watts, J. Nielsen-Gammon, E. Jones, D. Niyogi, J. Christy, and R.A. Pielke Sr., 2011: Analysis of the impacts of station exposure on the U.S. Historical Climatology Network temperatures and temperature trends. J. Geophys. Res., in press. Copyright (2011) American Geophysical Union.

has been accepted and is now in press. Below I have presented a summary of the study and its major messages from my perspective.  While the other authors of our paper have read and provided input on the information given below, the views presented below are mine. I will be posting on the history of my involvement on this subject in a follow-up post in a few days.

Volunteer Study Finds Station Siting Problems Affect USA Multi-Decadal Surface Temperature Measurements

We found that the poor siting of  a significant number of climate reference sites (USHCN) used by NOAA’s National Climatic Data Center (NCDC) to monitor surface air temperatures has led to inaccuracies and larger uncertainties in the analysis of multi-decadal surface temperature anomalies and trends than assumed by NCDC.

NCDC does recognize that this is an issue. In the past decade, NCDC has established a new network, the Climate Reference Network (CRN), to measure surface air temperatures within the United States going forward. According to our co-author Anthony Watts:

“The fact that NOAA itself has created a new replacement network, the Climate Reference Network, suggests that even they have realized the importance of addressing the uncertainty problem.”

The consequences of this poor siting on their analyses of multi-decadal trends and anomalies up to the present, however, has not been adequately examined by NCDC.

We are seeking to remedy this shortcoming in our study.

The placement of the USHCN sites can certainly affect the temperatures being recorded—both an area of asphalt (which is warmer than the surroundings on a sunny day or irrigated lawns (which is cooler than surrounding bare soil on a sunny day) situated near a station, for example, will influence the recorded surface air temperatures.

NOAA has adopted siting criteria for their climate reference stations: CRN 1 stations are the least likely to being influenced by nearby sources of heat or cooling, while CRN 5 stations are the most likely to be contaminated by local effects.  These local effects include nearby buildings, parking lots, water treatment plants irrigated lawns, and other such local land features.

To determine how the USHCN stations satisfied the CRN siting criteria and also whether the station siting affected temperature trend characteristics, Anthony Watts of IntelliWeather set up the Surface Stations project in 2007. More than 650  volunteers nationwide visually inspected (and rated) 1007 of the 1221 USHCN stations.  The volunteers wrote reports on the surroundings of each station and supplemented these reports with photographs.  Further analysis by Watts and his team used satellite and aerial map measurements to confirm distances between the weather station sensors and nearby land features.

The Surface Stations project is truly an outstanding citizen scientist project under the leadership of Anthony Watts!  The project did not involve federal funding. Indeed, these citizen scientists paid for the page charges for our article.  This is truly an outstanding group of committed volunteers who donated their time and effort on this project!

Analyzing the collected data, as reported in our paper, we found that only 80 of the 1007 sites surveyed in the 1221 station network met the criteria of CRN 1 or CRN 2 sites – those deemed appropriate for measuring climate trends by NCDC.  Of the remaining, 67 sites attained a CRN 5 rating – the worst rating.  While the 30-year and 115-year trends, and all groups of stations, showed warming trends over those periods,  we found that the minimum temperature trends appeared to be overestimated and the maximum warming trends underestimated at the poorer sites.

This discrepancy matters quite a bit. Wintertime minimum temperatures help determine plant hardiness, for example, and summertime minimum temperatures are very important for heat wave mortality. The use of temperature trends from poorly sited climate stations, therefore, introduces an uncertainly in our ability to quantify these key climate  metrics.

While all groups of stations showed warming trends over those periods, there is evidence to suggest a higher level of uncertainty in the trends since it was found, as one example, that according to the best-sited stations, the 24 hour temperature range in the lower 48 states has no century-scale trend, while the poorly sited locations have a significantly smaller diurnal temperature range. This raises a red flag to avoid poorly sited locations since clearly station measurement siting affects the quality of the surface temperature measurements.

The inaccuracies in the maximum and minimum temperature trends do matter also in the quantification of global warming. The inaccuracies of measurements from poorly sited stations are merged with the well sited stations in order to provide area average estimates of surface temperature trends including a global average.  In the United States, where this study was conducted, the biases in maximum and minimum temperature trends are fortuitously of opposite sign, but about the same magnitude, so they cancel each other and the mean trends are not much different from siting class to siting class. This finding needs to be assessed globally to see if this also true more generally.

However, even the best-sited stations may not be accurately measuring trends in temperature or, more generally, in trends in heat content of the air which includes the effect of water vapor trends (which is the more correct metric to assess surface air warming and cooling; see). Also, most of the best sited stations are at airports, which are subject to encroaching urbanization, and/or use a different set of automated equipment designed for aviation meteorology, but not climate monitoring. Additionally, the NCDC corrections for station moves or other inhomogeneities use data from poorly-sited stations for determining adjustments to better-sited stations, thus muddling the cleaner climate data. We are looking at these issues for our follow-on paper.

However, we know from our study that the use of these poorly sited locations in constructing multi-decadal surface temperature trends and anomalies has  introduced an uncertainty in our quantification of the magnitude of how much warming has occurred in the United States during the 20th and early 21st century.

One critical question that needs to be answered now is; does this uncertainty extend to the worldwide surface temperature record? In our paper

Montandon, L.M., S. Fall, R.A. Pielke Sr., and D. Niyogi, 2011: Distribution of landscape types in the Global Historical Climatology Network. Earth Interactions, 15:6, doi: 10.1175/2010EI371

we found that the global average surface temperature may be higher than what has been reported by NCDC and others as a result in the bias in the landscape area where the observing sites are situated. However, we were not able to look at the local siting issue that we have been able to study for the USA in our new paper.

Appendix- Summary of Trend Analysis Results

Temperature trend estimates do indeed vary according to site classification. Assuming trends from the better-sited stations (CRN 1 and CRN 2) are most accurate:

• Minimum temperature warming trends are overestimated at poorer sites
• Maximum temperature warming trends are underestimated at poorer sites
• Mean temperature trends are similar at poorer sites due to the contrasting biases of maximum and minimum trends
• The trend of the “diurnal temperature range” (the difference between maximum and minimum temperatures) is most strongly dependent on siting quality. For 1979-2008 for example, the magnitude of the linear trend in diurnal temperature range is over twice as large for CRN 1&2 (0.13ºC/decade) as for any of the other CRN classes. For the period 1895-2009, the adjusted CRN 1&2 diurnal temperature range trend is almost exactly zero, while the adjusted CRN 5 diurnal temperature range trend is about -0.5°C/century.
• Vose and Menne[2004, their Fig. 9] found that a 25-station national network of COOP stations, even if unadjusted and unstratified by siting quality, is sufficient to estimate 30-yr temperature trends to an accuracy of +/- 0.012°C/yr compared to the full COOP network. The statistically significant trend differences found here in the central and eastern United States for CRN 5 stations compared to CRN 1&2 stations, however, are as large (-0.013°C/yr for maximum temperatures, +0.011°C/yr for minimum temperatures) or larger (-0.023°C/yr for diurnal temperature range) than the uncertainty presented by Menne at al (2010).

More detailed results are found in the paper, including analyses for different periods, comparisons of raw and adjusted trends, and comparisons with an independent temperature data set.

Questions and Answers

Q: So is the United States getting warmer?

A: Yes in terms of the surface air temperature record. We looked at 30-year and 115-year trends, and all groups of stations showed warming trends over those periods.

Q: Has the warming rate been overestimated?

A: The minimum temperature rise appears to have been overestimated, but the maximum temperature rise appears to have been underestimated.

Q: Do the differing trend errors in maximum and minimum temperature matter?

A: They matter quite a bit. Wintertime minimum temperatures help determine plant hardiness, for example, and summertime minimum temperatures are very important for heat wave mortality. Moreover, maximum temperature trends are the better indicator of temperature changes in the rest of the atmosphere, since minimum temperature trends are much more a function of height near the ground and are of less value in diagnosing heat changes higher in the atmosphere; e.g see .

Q: What about mean temperature trends?

A: In the United States the biases in maximum and minimum temperature trends are about the same size, so they cancel each other and the mean trends are not much different from siting class to siting class. This finding needs to be assessed globally to see if this also true more generally.

However, even the best-sited stations may not be accurately measuring trends in temperature or, more generally, in trends in heat content of the air which includes the effect of water vapor trends.  Also, most are at airports, are subject to encroaching urbanization, and use a different set of automated equipment. The corrections for station moves or other inhomogeneities use data from poorly-sited stations for determining adjustments to better-sited stations.

Q: What’s next?

A:  We also plan to look specifically at the effects of instrument changes and land use issues, among other things.  The Surface Stations volunteers have provided us with a superb dataset, and we want to learn as much about station quality from it as we can.

================================================================

UPDATE: Since some people seemed unable to divine the link upstream, the pre-print version of the paper, posted on Dr. Pielke’s website is available here:

http://pielkeclimatesci.files.wordpress.com/2011/05/r-3671.pdf

– Anthony

UPDATE2: Dr. John Nielsen-Gammon, Texas State Climatologist and co-author, weighs in with his post:

Something for Everyone: Fall et al. 2011

As you may have heard, the long-awaited peer-reviewed analysis of the results of the SurfaceStations.org project has finally been released. I can’t wait to see the dueling headlines. Some will argue that the take-home message should be: Poor Station Siting Strongly Effects Temperature Trend Measurements, and will laugh at the idea that we can say with sufficient accuracy what has happened to our climate. Others will argue that the take-home message should be: Poor Station Siting Has No Effect on Temperature Trend Measurements, and will laugh at all the effort expended on a null result. Both sides will find solid evidence for their points of view in the paper. How can that be? How can one paper support opposing conclusions?

…

Here, in brief, are the answers: The poorest sites tend to be warmer. The minimum temperatures are warming faster at poorer sites than at better sites. The maximum temperatures are warming slower at poorer sites than at better sites. The adjustments reduce the differences by about half. The two effects are roughly equal and opposite so the mean temperature is rising at about the same rate across sites of different quality while the diurnal temperature range shows the biggest difference across sites.

On the one hand, this seems to be confirmation of the quality of the temperature record. All types of sites show the same mean temperature trend, so there’s no change necessary to our estimates of observed historical temperature trends in the United States.

On the other hand, there are several warning flags raised by this study. First, station siting is indeed important for the maximum and minimum temperature measurements. Second, the adjustments are only partly correcting the temperature record. Third, since the adjustments use data from all surrounding stations, there’s the danger that the mean trends are dominated by data from the poorer stations. (Less than ten percent of the USHCN stations are sited well enough to be considered appropriate for climate trend measurements.) Finally, and perhaps most important, are we really so lucky that the rest of the world would also have its poorly-sited stations have erroneous maximum and minimum temperature trends that just happen to be equal and opposite to each other?

Read his full essay here