Recollection of this day in history - How I got hooked on computer modeling

Recollection of this day in history - How I got hooked on computer modeling

August 24, 2022

RAMS simulation of Hurricane Andrew

Responding to the 30-year anniversary of Hurricane Andrew. One of my favorite accomplishments (Eastman, J.L., 1995: Numerical simulation of Hurricane Andrew - Rapid intensification. 21st Conference on Hurricanes and Tropical Meteorology, 24-28 April 1995, Miami, Florida, AMS, Boston, 111-113.) Made me think about how old I am getting and how I am in a sense the father of modern-day hurricane forecasting. Amazingly the "good ole boys’ network" at the HRC sucked up all funds and they pursued it. We gave up getting funding despite having Bill Gray on our team (the father of statistical seasonal hurricane forecasting). I remember he walked in one day when our team was viewing the animation I produced. He piped in loudly, "when the hell did that hurricane form, I missed it!". We all roared with laughter.

I have been addicted to numerical modeling ever since. Thanks Dr. Roger Pielke Sr. for letting me go off in tangents during my grad school days.

Original Tweet: Know it well. A couple years later I became the first person to ever simulate an actual hurricane with a mesoscale model (hi-res, sophisticated), took 10 days to run. Today this is the norm for forecasting hurricanes thanks to CPU power. https://twitter.com/MattDevittWINK/status/1562409999883812864

Comments

MAPS Model Description

Multiple Algorithm Pandemic Software MAPS is a valuable tool for forecasts or research that simulates pandemics or epidemics. It is written so that for a given virus the use can choose what type of differential equation solver that is most appropriate. The level of sophistication allows the simulation of the common flu to more complicated viral pathways. It is written in modular FORTRAN and can be run in serial (one processor) or parallel (unlimited processors). This provides a platform that can model a single point all the way up to global scales. If run in parallel the domain is decomposed to accommodate the number of processors specified. In this mode the domain is specified by latitude and longitude bounds and a desired grid spacing to produce a mesh, where the differential equations are solved for each square contained in the mesh. The individual points and MPI decomposed sub-domains are not independent. They communicate through ground and air transportation. The model has several solver algorithms, from a simple SI compartment model (flu) to the most sophisticated SEIRDCV (Susceptible Exposed Infected Recovering Death reCovered Vaccinated) compartment model. Variants are all subject to their own characteristic settings, from varying transmission to death rates. They also have settings for hospitalization rates. Hospitalization has its own algorithm that can mimic capacity effects like stress, triage mode, transport to adjacent hospitals, and ICU vs normal beds. Each variant also has its own mask/Shelter In Place settings reflecting country/state/local. mitigation policies including vaccination/booster rates. Output can range from global, country, state, local average files with the S E I R C D V fields or optional spatial output (GrADS, EWB, netCDF, ascii). The model also can be run in a "normal" or "forecast" modes. Post-processing can be run as a simple statistical analysis or a more involved training mode for the forecast system. The assimilation system can ingest observations at multiple scales with a multitude of methodologies. Seperate files are created depending on the number of processors deployed.