Oceanography | Vol. 38, No. 3
whereas the average spacing between cutting-edge OGCM grid
cells is 1 km. In this sense, cutting-edge OGCMs are becoming
unconstrained by data because the data are sparse compared to
the OGCM degrees of freedom (and notice that this is not true for
the ocean components of cutting-edge IPCC models). The unequal
growth of OGCM resolution and data density reflects the so-called
maturation of computational oceanography (Haine et al., 2021).
Cutting-edge OGCMs are thus becoming more and more valuable
as a resource in oceanography.
OGCM SOLUTIONS AND DATA ACCESS
LLC4320
For example, the 2016 black dot in Figure 1 is a model solution
called LLC4320 (the name refers to the latitude-longitude-cap
horizontal grid with 4320 × 4320 grid cells in each of 13 faces
that tile the global ocean; Rocha et al., 2016; Arbic et al., 2018).
The LLC4320 simulation provides hourly output for one year
in 2011–2012 using the Massachusetts Institute of Technology
OGCM code. A few similar solutions exist using other circula-
tion models and different configurations. Collectively, such solu-
tions are called “nature runs” or “digital twins” of the ocean cur-
rents (Boyes and Watson, 2022; Chen et al., 2023; NASEM, 2024;
Vance et al., 2024). They are useful for many purposes that include
understanding ocean dynamics, designing observing systems,
and machine learning.
Indeed, the oceanographic community is eagerly adopting
these cutting-edge OGCM solutions. To illustrate, the red dots
in Figure 2 show the number of papers each year that utilize the
LLC4320 solution. As in Figure 1, the y-axis of Figure 2 is loga-
rithmic, and straight lines indicate exponential growth. Thus,
Figure 2 shows that the number of LLC4320 papers per year has
grown roughly as an exponential with a doubling time of around
3 yr; dozens of papers now employ the LLC4320 simulation per year.
Despite this growing popularity, the data from LLC4320-type
cutting edge simulations are very challenging to use. The main
problem is the massive size of the datasets, which means that
access to these data is difficult and time-consuming. For LLC4320,
the total uncompressed data volume is four petabytes (one peta-
byte is 1015 bytes), and it takes many months to obtain accounts
on the NASA supercomputers where the LLC4320 simulation
was run. Moreover, the datasets are far too massive for individual
researchers to download and analyze personal copies.
POSEIDON PROJECT
Making the LLC4320 (and similar) simulation data easy to use is
therefore an important priority. Evidence from a neighboring field
in fluid mechanics shows the benefits of opening massive simula-
tion datasets to easy community access. Specifically, the blue dots
in Figure 2 show the number of papers each year that utilize the
Johns Hopkins Turbulence Database (JHTDB; Li et al., 2008). The
JHTDB is an open numerical turbulence laboratory that provides
free access to benchmark numerical solutions for various canonical
turbulence problems. Figure 2 shows that the number of JHTDB
papers per year has also grown exponentially, with a doubling time
of 3.0 yr. In total, more than 6 × 1014 individual model grid cells
have been queried using the JHTDB. A recent paper states that
FIGURE 1. Growth over time of the number of
horizontal grid cells in global ocean general cir-
culation models (OGCMs, see the black dots), the
number of horizontal grid cells in the global cou-
pled climate model from the Intergovernmental
Panel on Climate Change (IPCC, see the colored
dots), and the number per year of deep (greater
than 1,000 m depth) CTD stations. Note that the
y-axis is logarithmic and the straight red lines
indicate exponential growth (the doubling times,
τ2× are shown). The black dot in 2016 is for the
LLC4320 OGCM (see text and Figures 2 and 3).
The three-letter abbreviations in color refer to
the IPCC assessment reports. Modified from
Figure 2 in Haine et al. (2021)
2.5 yr
τ2× =
3.7 yr
τ2×
IPCC model
horizontal
grid cell #
Deep CTD
stations
per year
OGCM horizontal
grid cell #
SAR
FAR
AR4
AR5
AR6
TAR
109
108
107
106
105
104
103
103
104
105
Horizontal Grid Scale (m)
1980
1990
2000
2010
Number
2020