Hurricane Dorian forecasts

hammad93 / hurricane-net

A deep learning framework for forecasting Atlantic hurricane trajectory and intensity.

Creative Commons Zero v1.0 Universal

45 stars 11 forks source link

Hurricane Dorian forecasts #5

Closed hammad93 closed 2 years ago

hammad93 commented 5 years ago

https://www.nhc.noaa.gov/

Using these advisories, create a track with proper vectors and run it through the model https://www.nhc.noaa.gov/archive/2019/al05/al052019.public.009.shtml? https://www.nhc.noaa.gov/archive/2019/al05/al052019.public.013.shtml? https://www.nhc.noaa.gov/archive/2019/al05/al052019.public.017.shtml? https://www.nhc.noaa.gov/archive/2019/al05/al052019.public.021.shtml? https://www.nhc.noaa.gov/archive/2019/al05/al052019.public.025.shtml?

hammad93 commented 5 years ago

made some progress here https://github.com/hammad93/hurricane-net/commit/1b4be3ddf60dd5516684b4b495da6e25e6332f68 where i was able to recompile the model and debug it against google colab

hammad93 commented 5 years ago

input = {
  0 : {
      'entry_time' : parse('Fri Aug 30 2019 1100 PM'),
      'lat' : 25.5,
      'long' : 71.4,
      'max_wind' : 140 / 1.51 , # mph to knots
      'min_pressure' : 948.0
    },
  -24 : {
      'entry_time' : parse('Thu Aug 29 2019 1100 PM'),
      'lat' : 23.3,
      'long' : 68.4,
      'max_wind' : 105 / 1.51 , # mph to knots
      'min_pressure' : 977.0
    },
  -48 : {
      'entry_time' : parse('Wed Aug 28 2019 1100 PM'),
      'lat' : 19.7,
      'long' : 66.0,
      'max_wind' : 85 / 1.51 , # mph to knots
      'min_pressure' : 986.0
    },
  -72 : {
      'entry_time' : parse('Tue Aug 27 2019 1100 PM'),
      'lat' : 16.0,
      'long' : 63.0,
      'max_wind' : 50 / 1.51 , # mph to knots
      'min_pressure' : 1006.0
    },
  -96 : {
      'entry_time' : parse('Mon Aug 26 2019 1100 PM'),
      'lat' : 13.2,
      'long' : 59.7,
      'max_wind' : 50 / 1.51 , # mph to knots
      'min_pressure' : 1003.0
    },
  -120 : {
      'entry_time' : parse('Sun Aug 25 2019 1100 PM'),
      'lat' : 11.7,
      'long' : 55.3,
      'max_wind' : 50 / 1.51 , # mph to knots
      'min_pressure' : 1003.0
    }
}

hammad93 commented 5 years ago

reminder of the units,

'lat' : #.##N
'long' : #.##W

'max-wind' : knots per hours

output = {
24: {
'lat': 21.56690102815628,
'long': 67.65919329077005,
'max_wind': 69.06800121068954},
48: {
'lat': 24.110200852155685,
'long': 70.70270804166793,
'max_wind': 76.14463597536087},
72: {
'lat': 27.2433926820755,
'long': 70.24173901826143,
'max_wind': 82.61742889881134},
96: {
'lat': 30.90182771682739,
'long': 66.2531468346715,
'max_wind': 79.1596582531929},
120: {
'lat': 34.78565542697906,
'long': 59.178523366525766,
'max_wind': 74.8192635178566
}
}

analysis of initial operational use has forecasts being not advisable. for example, the output 24 hours delta was 69 knots per hour which is 79 mph or just a category 1. the truth was hurricane delta had maximum sustained winds of 150 mph or category 5.

hammad93 commented 5 years ago

we can do great improvement because there exists some bugs and the model is already deprecated in keras

hammad93 commented 4 years ago

After re analyzing the forecasts, the knots to mph conversion was done using 1.51 instead of 1.15 which caused the output to be off.

The forecasts are much more accurate than previously analyzed. With the addition of sea surface temperatures it may be very good.

hammad93 commented 4 years ago

# mph to knots
mph * 0.868976