gfdl's home page > about us > FY2001 milestones > an improved 40-year atmospheric temperature database
An Improved 40-Year Atmospheric Temperature Data Base
An Improved 40-Year Atmospheric Temperature Data Base
John R. Lanzante and Stephen A. Klein, GFDL
Dian J. Seidel, ARL
Purpose
The primary sources for long-term global upper-air temperature measurements are satellites and radiosondes. Satellite records are limited to the last two decades and yield information regarding the average temperature in a few very broad vertical layers. Radiosondes have a much longer period of coverage, at least the last 4 decades, and routinely measure temperature in more detail in the vertical, at more than a dozen levels above the surface.
However, it has been found that the long-term continuity of radiosonde temperature measurements may have been significantly compromised by changes in instruments and measurement practices, which vary from country to country, and are often not well documented. These changes sometimes introduce artificial components to the temperature measurements such that long-term trends computed from these data are subject to considerable uncertainty. This project is an extension of earlier NOAA Research at ARL and GFDL and is aimed at improving the upper-air temperature data base by applying adjustments to enhance the long-term continuity and by determining how the data quality varies from region to region. With these improvements, more reliable estimates of long-term trends can be made, varying by geographic region and height above the surface.
Efforts
To try to sample as much of the global domain as possible, a network of 87 stations ( Figure 1 ) has been selected from among those with the longest periods of record. Due to the complexity and laborious nature of the adjustment process, it was necessary to limit the total number of stations. Based on experience from past work, a two step procedure was developed to make improvements to these data. The first step involves determining at which times abrupt artificial changes occurred, separately for each station and vertical level. This was accomplished by way of a critical review of the data along with ancillary information including statistical transformations of the data, independent indicators of climate variability, as well as any available documentation on possible changes in instruments.
The second step of the procedure is to make adjustments to remove the major part of the artificial change, while retaining the natural variability. At times it is possible to utilize information from a number of vertical levels simultaneously to help retain the natural component of the variations. As part of our data enhancement strategy, in some cases we have found it necessary to delete limited portions of the data which were too poor to allow useful adjustments to be made.
An example of one of the more extreme problems is shown in Figure 2, which displays the time series of temperature for one upper-tropospheric level (300 hPa or approximately 9 km above the surface) at the Russian station Pechora (65N, 57E). The red curve is a smoothed time series of the original data. Taken at face value it would appear that a sudden warming of monumental proportions took place in the mid 1970's, with the warm period lasting for about 8 years, until an equally dramatic cooling occurred. After careful examination of all relevant information we concluded that these discontinuous temperature changes were artificial, due to some undocumented instrumental or procedural changes. We chose to adjust the data at the times of discontinuity (as indicated by the blue vertical lines) resulting in the adjusted time series displayed as the green curve. Similar adjustments were made for most of the other tropospheric levels at this station.
In addition to the new adjusted data set, detailed records of all of our findings have been created. These metadata records document the specific times and vertical levels at which we identified problems, on a station by station basis. These records may prove useful to future users of these data. Based on our experiences we are also able to make recommendations regarding the relative reliability of data from different geographical regions or countries of control.
Customers
The new data set created will soon be available for use by the climate research community. Scientists will have access to an improved data set as well as better understanding of its strengths and weaknesses. To the extent that more confident conclusions may be drawn regarding climate variability and change, policy makers and the public at large will benefit as well.
Significance
Long-term observations of atmospheric temperature are crucial for the study of both natural and anthropogenically induced climate variations. Much has been learned from the study of temperatures measured near the surface, both from instrumental records from the last century or two as well as proxy measurements during the last millennium or more. In so much as scientific consensus has solidified behind the notion that global warming is in progress, attention has turned more towards attribution of the cause(s). In this regard the profile of temperature above the earth's surface up to the lower stratosphere can play a significant role in differentiating among the candidate causes.
Success
Adjustment of the temperatures from the 87-station network has been completed and the problems encountered have been recorded in our metadata documentation. In an attempt to evaluate the veracity of our data modifications a comparison has been made with temperatures from an independent data source, the Microwave Sounding Unit (MSU), which has flown on NOAA polar-orbiting satellites since 1979. Shown in Figure 3 is a plot of the time series representing temperature vertically averaged over the lower troposphere, from both satellites and radiosondes for the same location as in Figure 2. While both the MSU (black) and original radiosonde (red) series capture the more rapid swings of about 5 years, the long-term trends, as given by lines using the same color conventions, indicate a cooling based on the radiosonde data but warming based on satellite measurements. The explanation for this disparity is the artificial warm period during most of the 1980's. After adjustment (green) both the radiosonde time series and trend line agree fairly well with the MSU data.
A broader comparison with MSU, based on the full 87-station network, suggests that overall our adjustments have reduced the disparity between radiosonde and MSU temperature trends. However, since we have not utilized MSU or any other satellite measures in identifying problems or making adjustments, our data set is a completely independent one. The convergence of MSU and radiosonde trends by way of adjustment yields more credibility to these estimates during the last two decades, as well as to our methodology, thus rendering more confidence in our pre-satellite era trend estimates as well.
Next Steps
Having adjusted the radiosonde data for 87 stations world wide, the next steps are geared towards creating climate data records for retrospective analysis of global temperature variability and trends, and for near real time climate monitoring. With support from the Office of Global Programs, and in collaboration with the National Climatic Data Center (NCDC), ARL and GFDL will create global, hemispheric, and zonal upper-air temperature time series based on the adjusted data. We will investigate whether statistical methods developed at NCDC to homogenize surface temperatures are applicable to radiosonde data. These methods are aimed at obtaining better spatial sampling, using a larger network of radiosonde stations. If successful, these techniques should allow for relatively straightforward, operational updating of the data and monitoring of upper-air temperature change. The new climate data records will be compared with other spatially-averaged data products based on satellite and radiosonde data. They will be archived and made public by NCDC, which will also assume responsibility for updates. Several manuscripts documenting our work will be prepared for submission to peer-reviewed journals.