Print Version

Chapter III: Short description of the algorithm

Table of Contents

Short description of the algorithm

The algorithm first retrieves the temperature and humidity profiles and then calculates precipitable water and instability parameters from them. The retrieval is only possible in cloud-free areas.

The so-called 'optimal estimation' method is used to retrieve temperature and moisture profiles from satellite observations – in this case the SEVIRI IR brightness temperatures. This is an inversion technique, which tries to find an atmospheric profile that best reproduces the observations (Koenig and de Coning, 2009). The main concept of optimal estimation are explained in greater detail in the 'Global Instability Index' EUMeTrain webcast http://www.eumetrain.org/resources/global_instability_index_2011.html).

In general, the optimal estimation is a multi-solution problem (Eyre, 1991), and a "first guess profile" is used as a constraint to be fed to an iteration scheme as an initial proposal for the solution. The original first guess is then modified step by step until the simulated radiance field at the top of the atmosphere matches the satellite observations, or in practice, until its difference to the observations is minimal.

The details of the SPhR algorithm are described thoroughly in the NWCSAF material (Li, 2007 and 2008; Martinez, Romero and Li, 2012a, 2012b), and discussed in the ‘Physical Retrieval’ EUMeTrain webcast ( http://www.eumetrain.org/resources/physical_retrieval_2011.html). The main inputs of the SPhR product are SEVIRI IR brightness temperature data (WV6.2, WV7.3, IR8.7, IR10.8, IR12.0 and IR13.4 channels) and NWP forecast fields. The first guess profiles are calculated by a nonlinear regression equation based on NWP profiles and SEVIRI infrared (IR) measurements.

Using geostationary measured SEVIRI data the temperature and humidity profile retrieval is based on data of limited number of (spectrally broad) IR channels, so SPhR can only slightly improve the humidity profiles compared to the background NWP data, mainly at middle and high levels.