Statistical Interpretation in Germany - the last 50 Years (Invited Presentation)
The main developments in statistical post-processing techniques and verification in the former East Germany are connected with the work and philosophy of K. Balzer, a major intellectual guide of K. Knüpffer. Balzer designed and improved the Perfect Prog system AFREG (analogue cases and regression) in the 1960s and 70s which paralleled the work of Lönnqvist in Sweden, and, additionally, was further motivated by the work of Bill Klein at TDL. During the 1970s the breakthrough of MOS in the United States, due to the pioneering work of Bob Glahn and others at TDL, combined with the worldwide efforts of various developers led to the 'First WMO Symposium on statistical interpretation of broad scale NWP products into local weather elements' in Warsaw - with L. Bengtsson being 'spiritus rector'.
During the 1980s the AFREG principle had been extended to a Multi- Model AFREG-Mix. It became evident that for more and more elements - starting with wind speed and direction - statistical forecasts could on the average not be improved any longer by the forecaster.
K. Knüpffer was encouraged by all these successes but lacked access to sufficient computer capacity to test his ideas. Instead, he drew up plans specifying how an ideal weather forecasting system would look like in an ideal world (that is one with unlimited computing power).
After the German unification K. Balzer implemented both AFREG-Mix and his comprehensive national verification system, which was inspired by the work of A. Murphy, at Deutscher Wetterdienst (German National Weather Service, DWD).
K. Knüpffer used his first opportunity to visit Bob Glahn at TDL in summer of 1990. I appreciated Bob's kindness and hospitality from the very first meeting which was followed by more visits later in the 1990s. Encouraged by both the success of his first MOS system - developed for the private Dutch company Meteo Consult Wageningen (now: Meteo Group) - and the idea of TDL, Knüpffer made a proposal to affiliate a 'European Centre for Statistical Post-Processing' to ECMWF. Unfortunately, this effort was in vain. Instead, in 1994, Knüpffer founded together with his Dutch partner Diederik Haalman the private company Meteo Service which specialized in all aspects of MOS. It provides MOS services worldwide, based on the GFS model (and subsequent others), for any interested customer. One of them was DWD. K. Balzer verified the DWD-MOS forecasts for one year before they were offered to forecasters. This verification was unique in the sense that it provided information about the performance of the human forecasters as compared to a MOS guidance that they did not know when making these forecasts (in normal operations the forecaster knows the guidance in advance). It revealed the superiority of MOS over the operational Kalman-Bucy Filtering and also demonstrated a significant advantage of MOS over man-made forecasts. In our century, weather forecasting competitions are an efficient tool not just to play the weather forecasting game every weekend but also to compare and analyze the performance of different numerical models and different MOS algorithms. Results will be presented.
Present applications of MOS range from standard weather element forecasts to aviation weather forecasts, to water level forecasts, as well as to wind and solar energy forecasts. Whenever there is a direct numerical model output (DMO) for any observed element, we achieve a reduction of error variance of MOS of about 50% as compared to such a DMO. This has turned out to be an almost constant value during the last 20 years. The benefit of using MOS at observing stations has been extended to the area by orography-dependent coefficient interpolation. This way NWP information in low spatial resolution (e.g. 1 deg Lat/Lon) can be transformed into site-specific weather forecasts of any desired spatial density (e.g. a 1-km grid) with minimum loss of MOS forecast quality.
Bob once asked how a three-man company can do all this MOS stuff. A comparison (similarities and differences) of the approaches of TDL and Meteo Service shall provide an answer.
Last but not least it should be mentioned that MOS is a perfect way to combine all the valuable predictive information we gather. The better the numerical models the better the MOS, and the better the MOS the better the final forecast. That means: Co-operation between these three system components is the key. Comparative verification and competitions are powerful tools to explore strengths and weaknesses of the components with the goal tu further optimize the forecasting system. We are grateful to Bob with gratitude for the important milestones he has accomplished to push our community further into that direction.