The effect of radiosonde instrument changes on climate trends of global atmospheric precipitable water

While global surface warming since the 1970s is quite well established, efforts by different researchers to determine upper air temperature and moisture trends produce diverging results, using either radiosonde data, satellite data, or reanalyses. In the radiosonde record, most earlier instruments have a warm and wet bias compared to modern models, so trends are biased because all long-term stations have had several instrument changes. This provides an update of progress toward developing unbiased global upper air climate trends using radiosonde data, focusing mainly on total precipitable water.

The goals of this project are to identify all radiosonde models which have been used, to develop complete location and instrument metadata for all upper air stations, and to determine instrument differences so unbiased global and regional upper air temperature and moisture trends can be computed. The uncorrected global precipitable water trend since 1973 shows substantial moistening in the late 1970s, relatively steplike drying around 1990, and no noticeable trend since then. The net result is slight moistening since 1973 but slight apparent drying after the 1980s. While instrument changes will not affect interannual changes substantially (such as a very large ENSO-related moistening and drying oscillation from 1997 to 2000), correcting for the wet bias of most earlier instruments is likely to intensify the 1970s moistening and cause some moistening since the early 1990s.

By examining statistics based on data series at each upper air station, down to the level of individual observations, data characteristics which most reliably identify instrument types are moisture variables (such as the average dew point depression in various layers or the lowest relative humidity reported) and indicators of data reporting (such as the lowest pressure or temperature at which the dew point is reported). Attempting to identify instrument changes by searching for temperature discontinuities is only partially successful because natural variations (especially in the stratosphere after volcanic eruptions) greatly exceed most instrument differences. If an instrument change is confirmed by discontinuities in any variables, especially at a large number of stations, then the size of the temperature discontinuity can be estimated even if it is smaller than natural fluctuations in that period.

So far, over 1600 instrument model variations have been documented, and inferring complete instrument metadata at over 2400 stations is in progress. The last major countries using older types of instruments, China, India, and the Russian Federation, are gradually transitioning to modern radiosonde types. Some preliminary findings are as follows: (1) Not all instruments in the 1970s and earlier have a wet bias, because several countries built radiosondes based on 1960s VIZ models with carbon hygristors, which were radiatively heated and produced too-dry readings in the daytime. (2) Most stations have complex histories and change back and forth between models before transitioning completely to a new model. (3) Sometimes, it is possible to infer the existence of an undocumented instrument by consistent data characteristics seen at several stations, such as a transition in France to an unknown instrument in late 2004. (4) The most widespread changes from older humidity sensors (with a large wet bias) occurred in 3 periods, including the 1960s (for most instruments based on USA VIZ designs), 1975 to 1985 (as Vaisala switched from a hair to a capacitive sensor), and since 2000 (in China, India, and Russia), so in between these periods there is probably little change in the global net moisture bias from instrument changes.