Datasets:

Languages:
English
ArXiv:
License:
planetswe / generation /gen_SWE_from_ic_file.py
LTMeyer's picture
Add files using upload-large-folder tool
0ea62c2 verified
"""
Adapted from Dedalus SWE example:
https://dedalus-project.readthedocs.io/en/latest/pages/examples/ivp_sphere_shallow_water.html
"""
import argparse
import logging
import multiprocessing as mp
import os
from glob import glob
import dedalus.public as d3
import numpy as np
os.environ["OMP_NUM_THREADS"] = "1"
os.environ["NUMEXPR_MAX_THREADS"] = "1"
logger = logging.getLogger(__name__)
# Simulation units
meter = 1 / 6.37122e6
hour = 1
second = hour / 3600
day = hour * 24
year = (
hour * 1008
) # 42 day years - Chosen based on the fact that the sim gets boring ~ 4000 hours
def run_ic_file(ic_file, output_dir):
output_dir = output_dir + ic_file.split("/")[-1].split(".")[0]
# Parse IC file for PM xfer
print(output_dir)
# Parameters
Nphi = 512
Ntheta = 256
dealias = 3 / 2
R = 6.37122e6 * meter
Omega = 7.292e-5 / second
nu = 1e5 * meter**2 / second / (160) ** 2 # Hyperdiffusion matched at ell=96
g = 9.80616 * meter / second**2
timestep = 60 * second
burn_in = 0.25 * year
stop_sim_time = burn_in + 3 * year # 1*year
dtype = np.float64
# Bases
coords = d3.S2Coordinates("phi", "theta")
dist = d3.Distributor(coords, dtype=dtype)
basis = d3.SphereBasis(
coords, (Nphi, Ntheta), radius=R, dealias=dealias, dtype=dtype
)
# Fields
u = dist.VectorField(coords, name="u", bases=basis)
h = dist.Field(name="h", bases=basis)
# Substitutions
zcross = lambda A: d3.MulCosine(d3.skew(A)) # noqa: E731
# Copy ICs from hpa 500 fields
ICs = np.load(ic_file)
ICs = np.swapaxes(ICs, 1, 2)
ICs = np.flip(ICs, 2)
u0 = ICs[:2] * meter / second # * .3
h0 = ICs[2] * meter / g # Conversion from geopotential to gp height
H = h0.mean() # Should be about 5500 meters
h0 = h0 - H
hs0 = ICs[3] * meter - H
hs = dist.Field(name="hs", bases=basis)
hs.load_from_global_grid_data(hs0)
hs.low_pass_filter((128, 256))
u.load_from_global_grid_data(u0)
h.load_from_global_grid_data(h0)
# # Initial conditions: balanced height
c = dist.Field(name="c")
problem = d3.LBVP([h, c], namespace=locals())
problem.add_equation("g*lap(h) + c = - div(u@grad(u) + 2*Omega*zcross(u))")
problem.add_equation("ave(h) = 0")
solver_init = problem.build_solver()
# solver_init.solve()
# Momentum forcing - seasonal
def find_center(t):
time_of_day = t / day
time_of_year = t / year
max_declination = 0.4 # Truncated from estimate of earth's solar decline
lon_center = time_of_day * 2 * np.pi # Rescale sin to 0-1 then scale to np.pi
lat_center = np.sin(time_of_year * 2 * np.pi) * max_declination
lon_anti = np.pi + lon_center
return lon_center, lat_center, lon_anti, lat_center
def season_day_forcing(phi, theta, t, h_f0):
phi_c, theta_c, phi_a, theta_a = find_center(t)
sigma = 2 * np.pi / 3
# Coefficients aren't super well-designed - idea is one side of the planet increases
# the other side decreases and the effect is centered around a seasonally-shifting Gaussian.
# The original thought was to have this act on momentum, but this was harder to implement in a stable way
# since increasing/decreasing by same factor is net energy loss.
coefficients = np.cos(phi - phi_c) * np.exp(
-((theta - theta_c) ** 2) / sigma**2
)
# coefficients = np.exp(-(phi - phi_c)**2 / sigma) * np.exp(-(theta-theta_c)**2 / sigma**2)
forcing = h_f0 * coefficients
return forcing
phi, theta = dist.local_grids(basis)
t = dist.Field(name="t")
lat = np.pi / 2 - theta + 0 * phi
phi_var = dist.Field(name="phi_var", bases=basis)
phi_var["g"] += phi
theta_var = dist.Field(name="theta_var", bases=basis)
theta_var["g"] += lat
h_f0 = (
2 * meter
) # Increasing this starts leading to fast waves (or maybe it just looks that way at 60 FPS/ 2.x day per sec)
h_f = season_day_forcing(phi_var, theta_var, t, h_f0)
# Problem
problem = d3.IVP([u, h], namespace=locals(), time=t)
problem.add_equation(
"dt(u) + nu*lap(lap(u)) + g*grad(h) + 2*Omega*zcross(u) = - u@grad(u)"
)
problem.add_equation("dt(h) + nu*lap(lap(h)) + (H)*div(u) = - div(u*(h-hs)) + h_f")
# Init to remove fast waves in sim - should probably just filter in time here, but this works.
solver = problem.build_solver(d3.RK222)
solver.stop_sim_time = burn_in
CFL = d3.CFL(
solver,
initial_dt=10 * second,
cadence=1,
safety=0.1,
threshold=0.05,
max_dt=1 * hour,
)
CFL.add_velocity(u)
logger.info("Trying init loop to get rid of fast waves")
for i in range(10):
logger.info("Starting init cycle %s" % i)
solver_init.solve()
for j in range(10 + i * 30):
timestep = CFL.compute_timestep()
solver.step(timestep)
solver_init.solve()
# Now do burn-in
try:
logger.info("Starting burn-in loop")
while solver.proceed:
timestep = CFL.compute_timestep()
solver.step(timestep)
# print(uf.evaluate()['g'])
if (solver.iteration - 1) % 10 == 0:
logger.info(
"Burn-in Iteration=%i, Time=%e, dt=%e"
% (solver.iteration, solver.sim_time, timestep)
)
except:
logger.error("Exception raised, triggering end of burn loop.")
raise
# Now define real problem
solver = problem.build_solver(d3.RK222)
solver.stop_sim_time = stop_sim_time
# Analysis
snapshots = solver.evaluator.add_file_handler(
output_dir, sim_dt=1 * hour, max_writes=1 * year
)
snapshots.add_tasks(solver.state, layout="g")
# CFL
CFL = d3.CFL(
solver,
initial_dt=10 * second,
cadence=1,
safety=0.1,
threshold=0.05,
max_dt=1 * hour,
)
CFL.add_velocity(u)
# Main loo
logger.info("Starting main loop")
while solver.proceed:
timestep = CFL.compute_timestep()
solver.step(timestep)
if (solver.iteration - 1) % 10 == 0:
logger.info(
"Iteration=%i, Time=%e, dt=%e"
% (solver.iteration, solver.sim_time, timestep)
)
solver.log_stats()
if __name__ == "__main__":
parser = argparse.ArgumentParser()
n_cores = mp.cpu_count()
parser.add_argument("--index", type=int, default=0)
parser.add_argument("--ic_dir", default="../data_stubs/")
parser.add_argument(
"--output_dir",
default="/mnt/home/polymathic/ceph/the_well/testing_before_adding/earthswe/",
)
args = parser.parse_args()
ind = int(args.index)
all_files = sorted(glob(f"{args.ic_dir}IC_*.npy"))
output_dir = args.output_dir
print("Processing IC", ind)
run_ic_file(all_files[ind], output_dir)