South Florida 100 yr wind map results appear too low

[JWoodbury6]

Dear developers,

Over the last couple of weeks I’ve been running the model for the North Atlantic basin with a focus on the southeast US. I installed the software and ran the Port Hedland test case. I get similar results to what are shown in the installation guide, so I think everything is set correctly.

While the model seems to run fine for the North Atlantic basin and the genesis areas make sense (see below), the problem is that the windspeeds seem too low. This is particularly the case for south Florida (see 100 yr map of south Florida below, using 500,000 simulation years.). Areas around Miami show a 100 yr 0.2 sec gust of between 30 and 35 meters per second. I extracted all results from the wind field files for a location in downtown Miami and I find a 100 yr return period windspeed of about 34 m/s. While I cant find any maps that are directly comparable, based on the US HAZUS model and a study in the Caribbean (http://www.disaster-info.net/safehospitals_refdocs/documents/english/windhazardmaps/CaribbeanWindMaps_CAD.pdf), I would expect windspeeds to be a fair bit higher. I found a note in a TCRM paper (https://nhess.copernicus.org/preprints/nhess-2019-192/nhess-2019-192.pdf) indicating that the model underestimates hazard in parts of western Australia as compared to historical data. Might this be a similar issue? My configuration inputs are included below as well as the genesis map and the 100 yr windspeed results.

A few questions I am hoping you can answer:

1. Are my configuration settings reasonable?
2. Do you know of any studies using TCRM in the southeast US?
3. Is it possible that this is an issue from converting the 1 min mean to 0.2 second gusts? Is there an option to just output the 1 min mean?
4. It generally seems that there aren’t enough larger storms reaching southern Florida. Could it be that storms are too quickly killed off due to interaction with the Caribbean Islands?

Configuration file:
[Actions]
; TCRM modules to execute
DataProcess=True
ExecuteStat=True
ExecuteTrackGenerator=True
ExecuteWindfield=True
ExecuteHazard=True
CreateDatabase=False
PlotHazard=False

PlotData=False

ExecuteEvaluate=False
DownloadData=False

[DataProcess]
InputFile=ibtracs.NAsince1980.list.v04r00.csv
Source=IBTRACS
StartSeason=1980
FilterSeasons=True

[Region]
; Domain for windfield and hazard calculation
; gridLimit={'xMin':270,'xMax':290,'yMin':24,'yMax':37}
gridLimit={'xMin':275,'xMax':283,'yMin':23,'yMax':27}
gridSpace={'x':1.0,'y':1.0}
gridInc={'x':1.0,'y':0.5}

[StatInterface]
kdeType=biw
kde2DType=Gaussian
kdeStep=0.2
minSamplesCell = 100

[TrackGenerator]
NumSimulations=500000
YearsPerSimulation=1
SeasonSeed=403943
TrackSeed=89333

[WindfieldInterface]
TrackPath=./output/US_Southeast/tracks
Margin=2.0
Resolution=0.05
Source=TCRM
profileType=powell
windFieldType=kepert

[Hazard]
; Years to calculate return period wind speeds
InputPath=./output/US_Southeast/windfield
;Resolution=0.05
Years=2,5,10,20,25,50,100,200,250,500,1000,2000,2500,5000
MinimumRecords=100
ExtremeValueDistribution=power
CalculateCI=False

[Input]
LocationFile = input/stationlist.shp
landmask = input/landmask.nc
mslpfile = MSLP/slp.day.ltm.nc
datasets = IBTRACS,LTMSLP
MSLPGrid=1,2,3,4,12

[Output]
Path=./output/US_Southeast

[Logging]
LogFile=./output/US_Southeast/log/US_Southeast_track_500K_sample.log
LogLevel=INFO
Verbose=True

[Process]
ExcludePastProcessed=True
DatFile=./output/US_Southeast/process/dat/US_Southeast.dat

[RMW]
GetRMWDistFromInputData=False
mean=50.0
sigma=0.6

[TCRM]
; Output track files settings
Columns=index,age,lon,lat,speed,bearing,pressure,penv,rmax
FieldDelimiter=,
NumberOfHeadingLines=1
SpeedUnits=kph
PressureUnits=hPa

[IBTRACS]
; Input data file settings
url = https://www.ncei.noaa.gov/data/international-best-track-archive-for-climate-stewardship-ibtracs/v04r00/access/csv/ibtracs.since1980.list.v04r00.csv
path = input
filename = ibtracs.NAsince1980.list.v04r00.csv
columns = tcserialno,season,num,skip,skip,skip,date,skip,lat,lon,skip,pressure
fielddelimiter = ,
numberofheadinglines = 1
pressureunits = hPa
lengthunits = km
dateformat = %Y-%m-%d %H:%M:%S
speedunits = kph

[LTMSLP]
; MSLP climatology file settings
URL = ftp://ftp.cdc.noaa.gov/Datasets/ncep.reanalysis.derived/surface/slp.day.1981-2010.ltm.nc
path = MSLP
filename = slp.day.ltm.nc

[wcarthur]

Hi @JWoodbury6

Thanks for providing the configuration file & figures. I have done some basic testing with the track generation in the NATL basin, but have not recently done full assessments in the basin.

At present, TCRM only simulates up to 100,000 years of events. Your setting of 500,000 will see wind field files overwritten for each time past 100,000 years. This will skew the return period wind speeds since the fitting routine (whether that's the power or one of the other options in the Hazard section) will use the 500,000 as the number of simulated years, but there are only 100,000 years of simulated events. Can you check the output folders to see how many track files are generated?

Also, the Margin and Resolution settings in the WindfieldInterface section can have an effect on peak wind speeds. I typically use Margin=5 and Resolution=0.02 for hazard assessments (the settings from the example are such that the example runs quickly). This will increase the run time substantially (and memory requirements for the hazard calculations) - worth noting when you are running 500,000 years of events. Runing at a coarser resolution will likely underrepresent the peak wind speed, especially for more compact, high-intensity storms.

I'd suggest a larger gridLimit setting as well, since there is the possibility of events passing close to the edge of the domain not being captured correctly. I typically set a minimum of 10x10 degrees as a lower limit for this setting.

The landfall decay model is based on Australian region storms, which in my experience show more rapid decay - mainly due to interaction with steep topography or very dry environments, neither of which are present in southern Florida and the Caribbean.
The parameters controlling landfall decay are in TrackGenerator.TrackGenerator._singleTrack(). Vickery (2005) has some selected models for different parts of the US coastline that could readily be implemented for your use case.

[JWoodbury6]

Hi @wcarthur

Thank you for the suggestions.

To your first point regarding the simulations, it does show 500K tracks in the folder, so it doesn't appear to be overwriting them, but maybe I'm missing something. The installation document says that the model can be run up to 1,000,000 simulations (http://geoscienceaustralia.github.io/tcrm/docs/setup.html), so I picked 500K.

I changed the inputs to fit your suggestions above (fewer simulations, margin, resolution, gridlimit and landfall decay parameters). I ran the model for 50K simulations with the new inputs and I get the below results for Florida and parts of the Caribbean (100 yr return period). Some parts just north of Cuba look better, but south Florida has largely stayed the same. I did some work on investigating the tracks and it seems to me that not enough strong storms are crossing this area.

[richsdixon]

I did some work on investigating the tracks and it seems to me that not enough strong storms are crossing this area.

I posted in another thread about this - it seems as though the translational speed of the storms is only half of that seen in reality which may well mean that the storms meander more as they cross the Atlantic and never make landfall before dying out. Am intrigued to understand why this should be the case if the historical behaviour of the storms is informing their motion.