These hindcasts have been developed in stages, as part of a NOPP
project to improve the physics packages of wind-wave models. In
atmospheric modeling, a statistically homogeneous dataset can be
generated by performing a reanalysis with consistent model setup and
using all available data for the entire period.  There is not enough
data to develop a similar reanalysis for wind waves. However, wave
dynamics are different from atmospheric dynamics, in the sense that
they represent a forced/damped problem rather than an initial value
problem, and wind forcing is the dominant process driving wave
dynamics.  Therefore it's possible to produce an accurate hindcast
without assimilating wave data, but using a wind field from a
long-term reanalysis such as the Climate Forecast System Reanalysis
(CFSR, Saha et al 2010). This is a couple reanalysis of atmospheric,
oceanic, sea-ice, and land data, at much higher resolution than
previous reanalysis.  

Below are the main difference between the two hindcasts:

Phase 1 (released in 2012):
- Used the then-current operational WAVEWATCH III model (v3.14), with
the Tolman and Chalikov (1996) source term package which was last
tuned in 2000-2001.  
- Used wind forcing from CFSR (Saha et al, 2010)

- Phase 2 (released in 2017): 
Used the WAVEWATCH III model (v5.08), with the Ardhuin et al (2010)
source term package.
- Used bias corrected CFSR winds.
- Used an Arctic curvilinear grid.