CIMSS Satellite Consensus (SATCON) Tropical Cyclone Intensity Estimation Algorithm
Derrick Herndon and Chris Veldon email@example.com
Cooperative Institute for Meteorological Satellite Studies (CIMSS)
Satellite-based Tropical Cyclone (TC) intensity estimates can differ by 30 knots or more. Such a large spread among estimates is problematic for the TC forecaster, especially when a TC is approaching landfall. On October 3, 2002 Hurricane Lili had just completed a period of rapid intensification with the pressure falling to 938 mb by 2200 UTC on October 2. However by 0800 UTC on October 3 the satellite presentation had significantly deteriorated with an eye no longer visible and an asymmetric appearance to the coldest cloud tops. At this time subjective Dvorak and objective ADT (Advanced Dvorak Technique) estimates indicated a minimum sea-level pressure of 935 mb, primarily due to weakening constraints. The CIMSS AMSU (Advanced Microwave Sounding Unit) method indicated a pressure of 954 mb and the CIRA AMSU method indicated a pressure of 970 mb. At the high end of these estimates the estimated maximum sustained wind was 125 knots while at the low end the estimate was 83 knots, a spread of more than 40 knots. Lili was only 5 hours from landfall at this point.
The current method for merging the information provided by objective intensity estimates is ad- hoc and differs between forecasters and between TC forecasting agencies. SATCON (SATellite CONsensus) is an algorithm that produces a weighted consensus of objective satellite-based estimates to determine TC current intensity. The contemporary version of SATCON employs the CIMSS ADT, the CIMSS AMSU method, and the CIRA AMSU method as consensus members. Future versions plan on adding an SSMI-based estimate as a potential fourth member. Each of the SATCON members have well- documented error characteristics which are situational dependant. For example the ADT skill is dependant on the scene type while the AMSU methods performance is dependant on whether the storms core is well resolved. The consensus uses these error characteristics to assign weights to each member. The result is an estimate that is superior in skill to the individual members
In 2006, SATCON was run in real-time using weights developed from a training sample of 2001- 2004 data. Statistical performance for the 2005-2006 Hurricane Seasons (Table 1 below) indicates skill versus the individual members as well as a simple average of the three SATCON members. The developers plan to continue testing the algorithm in 2007 in real-time, and then validate the performance using recon. The current SATCON web page is hosted at CIMSS and can be found at: http://cimss.ssec.wisc.edu/tropic/satcon/satcon.html
N = 128
AVG ERROR RMSE
Table 1. SATCON independent MSLP performance for 2005-2006 versus reconnaissance MSLP within 3 hours of the SATCON estimate.