Global in situ upper air data for climate change research

Long, homogeneous observed time series of climate state quantities are needed for climate monitoring not only at the Earth`s surface but also in the free atmosphere, since climate anomalies and climate change have a three- dimensional spatial structure. In situ upper air data, mostly radiosonde and tracked balloon data, are available over the northern hemisphere from the late 1930s onward and global coverage is given from the International Geophysical Year 1958 onward. To exploit their full potential for climate research, it is essential (i) to remove spurious biases and shifts from the station records and (ii) to assimilate the data with proper dynamic data assimilation systems in the framework of so-called reanalyses. The task of removing temporal changes in observation biases is referred to as homogenization. The lack of homogenized upper air records back to the 1970s or even the 1930s has been identified by the Intergovernmental Panel on Climate Change (IPCC) 4th assessment report as a major source of uncertainty that causes serious limitations in our ability to diagnose climate change. Within FWF project P18120-N10, ending in early 2009, the PI`s group has developed leading homogenization methods for radiosonde temperature and wind data. The calculated homogeneity adjustments are already used for bias correction in ongoing reanalysis efforts of the satellite era (1979-) at the European Centre for Medium-Range Weather Forecasts (ECMWF) and National Aeronautics and Space Administration (NASA). A new project is proposed with three major aims: (i) Development of a unified automatic homogenization system that analyzes and adjusts upper air temperature, wind and humidity dataset together. It should yield homogenized datasets of these parameters back to 1958. (ii) Development of bias estimation methods for radiosonde data within the data assimilation process. (iii) Investigation and, as far as possible, homogenization of the 1938-1958 radiosonde dataset. Aim (i) addresses the fact that radiosonde coverage is global from 1958 onward, but no homogenized global radiosonde humidity and wind data set comprehensive enough for use as input in climate data assimilation yet exists. Homogenized temperature data from University of Vienna and other sources between 1958-1978 exist but still contain substantial inconsistencies that must be removed, as recent research has shown. Treating all these parameters together is a novel approach that should yield improved detection efficiency, since breaks in different parameters often occur synchronously, but are not detectable in all time series. The homogenization of data from later periods may be further improved through optimization of the underlying statistical methods and through comparison with new datasets (e.g. recalibrated satellite radiances). Aim (ii) is the development of methods which estimate observation biases together with the atmospheric state vector without the need for externally specified bias correction tables as provided by the PI. Such methods work well at ECMWF for satellite radiances but have not yet been applied to conventional upper air data. If these could be used for radiosonde data, at least from 1979 onward, the reanalysis process and possibly also the operational assimilation of radiosonde data would be substantially facilitated. Aim (iii) should be feasible for the first time with the increasing number of digitized upper air data together with background departure time series from available surface observation only reanalyses and test assimilations of the early period. If successful, the project should lead to noticeably reduced uncertainty of the estimated upper air climate state of the past 8 decades. The resulting homogenized datasets are expected to be used as in situ upper air input for future reanalyses and will be available via public climate data portals. The project would also support a EC 7th framework project toward a new global reanalysis led by ECMWF, which has been recently submitted and where the PI is project partner. The close cooperation with leading data assimilation and climate research centres, should ensure maximum scientific impact of the proposed work.

The global radiosonde network has been an important observing system for meteorologists since the late 1930s up to now. Some (about 1000) of these records are potentially useful for climate studies but frequent changes of the radiosonde equipment caused many inhomogeneities (spurious stepwise changes) in the records. They caused serious misinterpretation of the records, especially regarding the amplification of surface temperature variations in the tropics in the upper troposphere (around 8km). While climate models predict strong amplification, mostly with factors 2-3, the raw radiosonde observations show no amplification during some decades (e.g. 1979-1999). During this project (lasting from 2009 to 2012) the homogenization system developed in earlier projects (called RAOBCORE/RICH), that already allowed for more comprehensive bias correction than all other homogenized radiosonde datasets so far, has been further improved and extended. The uncertainty estimates have also been improved by varying the most important parameters of the adjustment system in a sensible range. The amplification of trends in the upper troposphere, though weaker than in climate models, could be revealed and was found to be robust.This was a main reason why several versions of the adjusted radiosonde dataset have been used in major reanalysis projects, such as the European Interim Reanalysis (ERA-Interim), the Modern Era Retrospective-Analysis for Research and Applications (MERRA) and the Japanese Reanalysis (JRA-55). Thus, the RAOBCORE dataset has quite an impact on global climate science, since thousands of scientists use reanalysis data. In the second part of the project more attention was paid to adjusting also older radiosonde t data as well as tracked balloon data before 1958, using surface data only reanalyses (20CR) as reference. First results indicate the feasibility of the approach despite much larger background errors compared to reanalyses where upper data are assimilated. All the above work is done offline, i.e. outside a data assimilation system. There exist, however, elegant variational bias correction schemes that can be employed for this purpose. We tried to develop those for radiosonde wind and temperature in the ECMWF model. While significant progress has been made this task has yet to be completed. Not only the homogeneity of reanalysis input is of interest but also the homogeneity of the reanalyses themselves and of global moisture and energy budgets derived from them. In the project the budget calculation as well as analysis methods have been considerably refined in collaboration with a leading climate analysis group at the National Center for Atmospheric Research (USA). On the one hand discontinuities in the reanalyses could be clearly identified, but also quite interesting patterns in the budgets of the tropical Pacific in association with El Niño Southern Oscillation (ENSO) could be revealed, leading to two respectable papers in the Journal of Climate.