Radiosonde descent reports look promising

Bruce Ingleby (ECMWF), David Edwards (UK Met Office)


The radiosonde descent after balloon burst offers the possibility of an extra atmospheric profile at little or no extra cost. Quality checks suggest that it should be possible to start assimilating some of the descent data at ECMWF over the next couple of years.

The Vaisala RS41 radiosonde software can generate descent reports, currently in BUFR dropsonde format. Descent data from Germany, Finland and the UK for January and June 2018 have been processed and compared with the ECMWF short-range forecast (background). The descent rate is faster than the ascent rate, especially at upper levels. Over 95% of the radiosondes generating ascent reports also generate descent reports. Most of the descents reach 700 hPa, but the numbers available decrease below that level. This is because the signal is lost when the radiosonde goes below the horizon as seen from the launch station.

Encouraging results

The figure shows observation-minus-background (O–B) statistics for June 2018 for German radiosonde ascents/descents, which have the largest sample size and the smoothest O–B statistics. Mostly, the descent statistics look similar to the ascent statistics with two main differences: larger temperature biases at upper levels and smaller vector wind root-mean-square (rms) differences, especially at upper levels. It is known that the ECMWF background has a cold bias at lower stratospheric levels, but it seems that the ascent data here is closer to the truth than the descent data. The Finnish descent data looks even warmer (not shown), possibly due to higher descent speeds (Finnish radiosondes do not use a parachute), but further investigation is required. The descent wind rms differences are smaller than those for ascent winds because the measured descent winds are somewhat smoother. Vaisala apply a digital filter to the winds before producing the meteorological reports, to remove pendulum motion and for smoothing in general. The same time filter is applied to both ascent and descent data. It seems that the filtering should depend on the ascent/descent rate. The ascent data may contain residual noise from imperfect handling of the pendulum motion.

Steps towards operational use

The World Meteorological Organization (WMO) has recently approved a new BUFR sequence specifically for descent data. We hope to see this in use in 2019. For numerical weather prediction (NWP), any biased temperatures are problematic and we need to consider correcting or rejecting the data. As the stratospheric descent data is fairly close in space and time to the ascent, rejecting those temperatures would not be a major loss. Other aspects of processing and quality control would also need attention. Vaisala is changing the RS41 casing, which makes the radiosonde lighter, so assessment should be repeated on the new version. It should be possible to make fairly minor changes relatively soon and assimilate much of the descent data with hopefully positive impact, and at minimal cost. In the longer term, the interface to NWP could be rethought, with NWP using a raw version of the measurements.

O–B statistics for Germany, June 2018. Temperature and relative humidity mean and standard deviation and vector wind root-mean-square (rms) O–B statistics.