sdtaylor / phenology_forecasts_data Goto Github PK

in phenology_model_metadata.csv there are version numbers, these are the notes for that.
All models use NPN data from 2009-2017.

version -1

These are the naive models using doy ~ latitude, see here. These are also bootstrapped 50 times to get parameter uncertainty.

redone on 2018-07-13 to use the updated data mentioned in version 4 below. The -1 version number was kept the same.

version 1:
These were models selected via AIC of 20% held out data from either ThermalTime,Alternating, and Uniforc. Built 2018-01-02

version 2:
Same as version 1, but the winning model was built with a Bootstrap version to have parameter uncertainty. built 2018-03-16

version 3:
new ensemble models using the stacking method to derive weights. see here
For phenophases 371 and 501 the following models were included.
m1 = pyPhenology.models.Alternating()
m2 = pyPhenology.models.ThermalTime()
m3 = pyPhenology.models.Uniforc()
m4 = pyPhenology.models.Linear(parameters={'time_start':(-30,60), 'time_length':(1,120)})

For newly added phenophase 498 (fall coloring) the following were used
m1 = pyPhenology.models.Linear(parameters={'time_start':(180,300), 'time_length':(10,90)})
m2 = pyPhenology.models.FallCooling()

Note for 498 the doy was defined as when the intensity was >= 50%
built 2018-05-21

version 4:
Same as 3, but using the NPN formated data, instead of doing it myself.
see sdtaylor/phenology_forecasts#34.
Also because of this the phenophase 498 models are no longer
using "intensity was >= 50%" as used in version 4, but the
first 'yes' doy as used in other phenophases

For phenophases 371 and 501 the following models were included.
m1 = pyPhenology.models.Alternating()
m2 = pyPhenology.models.ThermalTime()
m3 = pyPhenology.models.Uniforc()
m4 = pyPhenology.models.Linear(parameters={'time_start':(-30,60), 'time_length':(1,120)})

For newly added phenophase 498 (fall coloring) the following were used
m1 = pyPhenology.models.Linear(parameters={'time_start':(180,300), 'time_length':(10,90)})
m2 = pyPhenology.models.FallCooling()

built 2018-07-09

version 5:
An un-weighted ensemble using otherwise the same specs as V4
For phenophases 371 and 501 the following models were included.
m1 = pyPhenology.models.Alternating()
m2 = pyPhenology.models.ThermalTime()
m3 = pyPhenology.models.Uniforc()
m4 = pyPhenology.models.Linear(parameters={'time_start':(-30,60), 'time_length':(1,120)})

For phenophase 498 (fall coloring) the following were used
m1 = pyPhenology.models.Linear(parameters={'time_start':(180,300), 'time_length':(10,90)})
m2 = pyPhenology.models.FallCooling()

built 2019-05-24