Examining MSU Instrumental Data - Lessons for the Future

National Oceanic & Atmospheric Administration (NOAA)

In an effort to advance research on global climate change, United States President George Bush announced in February 2002, the formation of the Climate Change Science Program (CCSP). The first CCSP report, Temperature Trends in the Lower Atmosphere: Steps for Understanding and Reconciling Differences, (Karl et al., 2006), concluded there is no longer a significant discrepancy between global temperatures measured at the surface with in-situ observing systems compared to those measured in the troposphere by satellites and weather balloons. Discrepancies in the rates of temperature change, however, remain to be resolved in the tropics (20°N to 20°S). Recommendations from that report can improve our ability to monitor climate variability and change.

Satellite data are available with nearly global coverage since 1979, but less than global coverage of upper air temperatures, winds, and moisture have been available since the end of the Second World War. The Microwave Sounding Units (MSU) operating on National Oceanic and Atmospheric Administration (NOAA) polar-orbiting platforms, along with the operational global radiosonde network have been the principal sources of multi-decadal temperature profiles. The satellite data represent average temperature over deep atmospheric layers rather than a particular level, which can create difficulty in interpreting MSU temperature trends as compared to other measurements, as different channels receive contributions from both the troposphere and stratosphere.

Complications

Specifically, MSU Channel 2 receives 10-15% of its emissions from the stratosphere (Spencer and Christy, 1992). This can make it difficult to interpret contributions of temperature change from the tropospheric temperature signal, as stratospheric cooling in recent decades has been relatively large in comparison to tropospheric warming. Research by Fu et al. (2004), suggests subtracting a suitable fraction of MSU 4 from MSU 2, in order to produce a tropospheric temperature value with less influence from the stratosphere. In contrast, the radiosonde data represent discrete levels in the atmosphere (Figure 1).

Figure 1 - Vertical profiles for the temperature products analysed in the CCSP 1.1 report. Radiosonde-based layer temperatures (T850-300, T100-50) are height-weighted averages of the temperature in those layers. Satellite-based temperatures (T2LT, T2, and T4) are mass-weighted averages with varying influence in the vertical as depicted by the curved profiles, i.e., the larger the value at a specific level, the more that level contributes to the overall satellite temperature average (Karl et al., 2006).

Additional complications that arise in developing a long-term record of upper tropospheric temperatures include time-varying biases that evolve while a satellite is in orbit. These include: orbital decay, diurnal drifting, inter-satellite biases and degradation of instrument calibration over time. A variety of experts have examined how these biases affect the data collected. For example, three expert groups - employing different methodologies - developed MSU climate data sets, which were used to distinguish dissimilarities in the MSU data and identify areas of improvement for future satellite-based observations (Figure 2).

Figure 2 - Three estimates of global mean temperature changes for MSU channel 2 (T2), expressed as anomalies relative to the 1979 to 1999 mean. Data are from: A, the University of Alabama in Huntsville (UAH); B, Remote Sensing Systems (RSS); and C, the University of Maryland (UMd). The estimates employ the same “raw” satellite data, but make different choices for the adjustments required to merge the various satellite records and to correct instrument biases. The statistical uncertainty is virtually the same for all three series. Differences between the series give some idea of the magnitude of structural uncertainties. The ± values define the 95% confidence intervals for the trends (Karl et al., 2006).

There are several ways to improve the decadal monitoring capability of present MSU instrumentation. There is a necessity to overlap old and new satellite instruments or configurations, as they evolve in time. This period of redundancy must provide experts sufficient time to calibrate for any small uncertainties that do not affect the analysis of climate trends. A full annual cycle of the climate should be made available within this period of overlap for optimised calibration efforts. Additionally, the launch of a replacement satellite should take place no later than a year prior to the projected time of failure for any key instrument. This recommendation is also emphasised by the National Research Council (NRC, 2000b), the Global Climate Observing System (GCOS) Climate Monitoring Principles (GCOS, 2004, Appendix 3), and the Global Earth System of Systems (GEOSS) 10 year Implementation Plan Reference Document (GEOSS, 2005).

Balloon-based radiosonde measurements have also been subject to scrutiny. Data discontinuities often result from periodic changes in station location, instrumentation and data processing methods. Major discontinuities may be related to solar heating of the temperature sensor and subsequent design change or data adjustments proposed to solve this problem. Other sources of measurement bias include sensor icing, software errors, poor calibration and operator errors (Karl et al. 2006).

Substantial Progress

Despite all these difficulties, there has been substantial progress in understanding both historical changes and their causes, as outlined in Karl et al. (2006). The report also provides a number of highest-utility recommendations for advancing our understanding of the vertical profile of temperature trends, in order to improve future monitoring efforts. Many of the difficulties in correcting for the non-climatic biases affecting climate data records are related to human decisions regarding either the errors in the assumptions used that underlie the production of climate data records or if important factors are ignored altogether. Sources of error may also result in estimating the parameters needed by algorithms used for producing climate data records because of finite sample sizes. As a result, when a single observing system or analysis team is used as the sole basis for estimating the total uncertainty, then the uncertainty is likely to be poorly estimated. When evaluating tropospheric temperatures, this problem is magnified due to the lack of high quality reference or “ground truth” data against which satellite observations can be compared, to assist in the rigorous removal of non-climatic influences (Karl et al., 2006). A collection of widely distributed reference sites could produce highquality data and strengthen the more globally-extensive satellite monitoring efforts.

In the forthcoming decades, there will be new, largely space-based observation platforms that will produce large increases in the volume and variety of data available. These data measurements will also be made with greater accuracy and resolution, particularly in the vertical direction, but to ensure we make the most efficient use of these data for climate change monitoring it will be also be critically important to address the issues identified by the CCSP Synthesis and Assessment Report (Folland et al., 2006).

Adam B. Smith, Thomas R. Karl, Gregory W. Withee National Oceanic and Atmospheric Administration NOAA’s National Climatic Data Center 151 Patton Avenue, Asheville, North Carolina 28801 NOAA’s National Environmental Satellite, Data & Information Service 1335 East West Hwy, Silver Spring, Maryland 20910 Web: www.noaa.gov