linearshallowwater.py

#!/usr/bin/env python
# -*- coding: utf-8 -*-

"""Linear Shallow Water Model

- Two dimensional shallow water in a rotating frame
- Staggered Arakawa-C grid
- fixed boundary conditions in the y-dimension (free slip)
- Linearised about a fluid depth H and u = 0

Dimensions are implied in comments on constants e.g. Lx is
the width of the domain in metres [m], however there is no
dependency on using specific units.  If all input values
are scaled appropriately other units may be used.

η = H + h

∂/∂t[u] - fv = - g ∂/∂x[h] + F                          (1)
∂/∂t[v] + fu = - g ∂/∂y[h] + F                          (2)
∂/∂t[h] + H(∂/∂x[u] + ∂/∂y[v]) = F                      (3)

f = f0 + βy
F is a forcing, default = (0, 0, 0)
"""

from __future__ import print_function
import time

import numpy as np
import matplotlib.pyplot as plt

experiment = '2d'           # set to '1d' or '2d'
plot_interval = 20          # plot every n steps

## CONFIGURATION
### Domain
nx = 128
ny = 129

H  = 100.0          # [m]  Average depth of the fluid
Lx = 2.0e7          # [m]  Zonal width of domain
Ly = 1.0e7          # [m]  Meridional height of domain

boundary_condition = 'periodic'  # either 'periodic' or 'walls'
if experiment is '1d':
    boundary_condition = 'walls'

### Coriolis and Gravity
f0 = 0.0              # [s^-1] f = f0 + beta y
beta =  0.0           # [m^-1.s^-1]
g = 1.0               # [m.s^-1]

### Diffusion and Friction
nu = 5.0e4          # [m^2.s^-1] Coefficient of diffusion
r = 1.0e-4          # Rayleigh damping at top and bottom of domain

dt = 1000.0         # Timestep [s]


## GRID
# Setup the Arakawa-C Grid:
#
# +-- v --+
# |       |    * (nx, ny)   h points at grid centres
# u   h   u    * (nx+1, ny) u points on vertical edges  (u[0] and u[nx] are boundary values)
# |       |    * (nx, ny+1) v points on horizontal edges
# +-- v --+
#
# Variables preceeded with underscore  (_u, _v, _h) include the boundary values,
# variables without (u, v, h) are a view onto only the values defined
# within the domain
_u = np.zeros((nx+3, ny+2))
_v = np.zeros((nx+2, ny+3))
_h = np.zeros((nx+2, ny+2))

u = _u[1:-1, 1:-1]               # (nx+1, ny)
v = _v[1:-1, 1:-1]               # (nx, ny+1)
h = _h[1:-1, 1:-1]               # (nx, ny)

state = np.array([u, v, h])


dx = Lx / nx            # [m]
dy = Ly / ny            # [m]

# positions of the value points in [m]
ux = (-Lx/2 + np.arange(nx+1)*dx)[:, np.newaxis]
vx = (-Lx/2 + dx/2.0 + np.arange(nx)*dx)[:, np.newaxis]

vy = (-Ly/2 + np.arange(ny+1)*dy)[np.newaxis, :]
uy = (-Ly/2 + dy/2.0 + np.arange(ny)*dy)[np.newaxis, :]

hx = vx
hy = uy

t = 0.0                 # [s] Time since start of simulation
tc = 0                  # [1] Number of integration steps taken


## GRID FUNCTIONS
# These functions perform calculations on the grid such as calculating
# derivatives of fields or setting boundary conditions

def update_boundaries():

    # 1. Periodic Boundaries
    #    - Flow cycles from left-right-left
    #    - u[0] == u[nx]
    if boundary_condition is 'periodic':
        _u[0, :] = _u[-3, :]
        _u[1, :] = _u[-2, :]
        _u[-1, :] = _u[2, :]
        _v[0, :] = _v[-2, :]
        _v[-1, :] = _v[1, :]
        _h[0, :] = _h[-2, :]
        _h[-1, :] = _h[1, :]


    # 2. Solid walls left and right
    #    - No zonal (u) flow through the left and right walls
    #    - Zero x-derivative in v and h
    if boundary_condition is 'walls':
        # No flow through the boundary at x=0
        _u[0, :] = 0
        _u[1, :] = 0
        _u[-1, :] = 0
        _u[-2, :] = 0

        # free-slip of other variables: zero-derivative
        _v[0, :] = _v[1, :]
        _v[-1, :] = _v[-2, :]
        _h[0, :] = _h[1, :]
        _h[-1, :] = _h[-2, :]

    # This applied for both boundary cases above
    for field in state:
        # Free-slip of all variables at the top and bottom
        field[:, 0] = field[:, 1]
        field[:, -1] = field[:, -2]

    	# fix corners to be average of neighbours
        field[0, 0] =  0.5*(field[1, 0] + field[0, 1])
        field[-1, 0] = 0.5*(field[-2, 0] + field[-1, 1])
        field[0, -1] = 0.5*(field[1, -1] + field[0, -2])
        field[-1, -1] = 0.5*(field[-1, -2] + field[-2, -1])


def diffx(psi):
    """Calculate ∂/∂x[psi] over a single grid square.

    i.e. d/dx(psi)[i,j] = (psi[i+1/2, j] - psi[i-1/2, j]) / dx

    The derivative is returned at x points at the midpoint between
    x points of the input array."""
    global dx
    return (psi[1:,:] - psi[:-1,:]) / dx

def diff2x(psi):
    """Calculate ∂2/∂x2[psi] over a single grid square.

    i.e. d2/dx2(psi)[i,j] = (psi[i+1, j] - psi[i, j] + psi[i-1, j]) / dx^2

    The derivative is returned at the same x points as the
    x points of the input array, with dimension (nx-2, ny)."""
    global dx
    return (psi[:-2, :] - 2*psi[1:-1, :] + psi[2:, :]) / dx**2

def diff2y(psi):
    """Calculate ∂2/∂y2[psi] over a single grid square.

    i.e. d2/dy2(psi)[i,j] = (psi[i, j+1] - psi[i, j] + psi[i, j-1]) / dy^2

    The derivative is returned at the same y points as the
    y points of the input array, with dimension (nx, ny-2)."""
    global dy
    return (psi[:, :-2] - 2*psi[:, 1:-1] + psi[:, 2:]) / dy**2

def diffy(psi):
    """Calculate ∂/∂y[psi] over a single grid square.

    i.e. d/dy(psi)[i,j] = (psi[i, j+1/2] - psi[i, j-1/2]) / dy

    The derivative is returned at y points at the midpoint between
    y points of the input array."""
    global dy
    return (psi[:, 1:] - psi[:,:-1]) / dy

def centre_average(phi):
    """Returns the four-point average at the centres between grid points."""
    return 0.25*(phi[:-1,:-1] + phi[:-1,1:] + phi[1:, :-1] + phi[1:,1:])

def y_average(phi):
    """Average adjacent values in the y dimension.
    If phi has shape (nx, ny), returns an array of shape (nx, ny - 1)."""
    return 0.5*(phi[:,:-1] + phi[:,1:])

def x_average(phi):
    """Average adjacent values in the x dimension.
    If phi has shape (nx, ny), returns an array of shape (nx - 1, ny)."""
    return 0.5*(phi[:-1,:] + phi[1:,:])

def divergence():
    """Returns the horizontal divergence at h points."""
    return diffx(u) + diffy(v)

def del2(phi):
    """Returns the Laplacian of phi."""
    return diff2x(phi)[:, 1:-1] + diff2y(phi)[1:-1, :]

def uvatuv():
    """Calculate the value of u at v and v at u."""
    global _u, _v
    ubar = centre_average(_u)[1:-1, :]
    vbar = centre_average(_v)[:, 1:-1]
    return ubar, vbar

def uvath():
    global u, v
    ubar = x_average(u)
    vbar = y_average(v)
    return ubar, vbar

def absmax(psi):
    return np.max(np.abs(psi))


## DYNAMICS
# These functions calculate the dynamics of the system we are interested in
def forcing():
    """Add some external forcing terms to the u, v and h equations.
    This function should return a state array (du, dv, dh) that will
    be added to the RHS of equations (1), (2) and (3) when
    they are numerically integrated."""
    global u, v, h
    du = np.zeros_like(u)
    dv = np.zeros_like(v)
    dh = np.zeros_like(h)
    # Calculate some forcing terms here...
    return np.array([du, dv, dh])

sponge_ny = ny//7
sponge = np.exp(-np.linspace(0, 5, sponge_ny))
def damping(var):
    # sponges are active at the top and bottom of the domain by applying Rayleigh friction
    # with exponential decay towards the centre of the domain
    global sponge, sponge_ny
    var_sponge = np.zeros_like(var)
    var_sponge[:, :sponge_ny] = sponge[np.newaxis, :]
    var_sponge[:, -sponge_ny:] = sponge[np.newaxis, ::-1]
    return var_sponge*var

def rhs():
    """Calculate the right hand side of the u, v and h equations."""
    u_at_v, v_at_u = uvatuv()   # (nx, ny+1), (nx+1, ny)

    # the height equation
    h_rhs = -H*divergence() + nu*del2(_h) - r*damping(h)

    # the u equation
    dhdx = diffx(_h)[:, 1:-1]  # (nx+1, ny)
    u_rhs = (f0 + beta*uy)*v_at_u - g*dhdx + nu*del2(_u) - r*damping(u)

    # the v equation
    dhdy  = diffy(_h)[1:-1, :]   # (nx, ny+1)
    v_rhs = -(f0 + beta*vy)*u_at_v - g*dhdy + nu*del2(_v) - r*damping(v)

    return np.array([u_rhs, v_rhs, h_rhs]) + forcing()

_ppdstate, _pdstate = 0,0
def step():
    global dt, t, tc, _ppdstate, _pdstate

    update_boundaries()

    dstate = rhs()

    # take adams-bashforth step in time
    if tc==0:
        # forward euler
        dt1 = dt
        dt2 = 0.0
        dt3 = 0.0
    elif tc==1:
        # AB2 at step 2
        dt1 = 1.5*dt
        dt2 = -0.5*dt
        dt3 = 0.0
    else:
        # AB3 from step 3 on
        dt1 = 23./12.*dt
        dt2 = -16./12.*dt
        dt3 = 5./12.*dt

    newstate = state + dt1*dstate + dt2*_pdstate + dt3*_ppdstate
    u[:], v[:], h[:] = newstate
    _ppdstate = _pdstate
    _pdstate = dstate

    t  += dt
    tc += 1


## INITIAL CONDITIONS
# Set the initial state of the model here by assigning to u[:], v[:] and h[:].
if experiment is '2d':
    # create a single disturbance in the domain:
    # a gaussian at position gx, gy, with radius gr
    gx =  2.0e6
    gy =  0.0
    gr =  2.0e5
    h0 = np.exp(-((hx - gx)**2 + (hy - gy)**2)/(2*gr**2))*H*0.01
    u0 = u * 0.0
    v0 = v * 0.0

if experiment is '1d':
    h0 = -np.tanh(100*hx/Lx)
    v0 = v * 0.0
    u0 = u * 0.0
    # no damping in y direction
    r = 0.0

# set the variable fields to the initial conditions
u[:] = u0
v[:] = v0
h[:] = h0

## PLOTTING
# Create several functions for displaying current state of the simulation
# Only one is used at a time - this is assigned to `plot`
plt.ion()                         # allow realtime updates to plots
fig = plt.figure(figsize=(8*Lx/Ly, 8))  # create a figure with correct aspect ratio

# create a set of color levels with a slightly larger neutral zone about 0
nc = 12
colorlevels = np.concatenate([np.linspace(-1, -.05, nc), np.linspace(.05, 1, nc)])

def plot_all(u,v,h):
    hmax = np.max(np.abs(h))
    plt.clf()
    plt.subplot(222)
    X, Y = np.meshgrid(ux, uy)
    plt.contourf(X/Lx, Y/Ly, u.T, cmap=plt.cm.RdBu, levels=colorlevels*absmax(u))
    #plt.colorbar()
    plt.title('u')

    plt.subplot(224)
    X, Y = np.meshgrid(vx, vy)
    plt.contourf(X/Lx, Y/Ly, v.T, cmap=plt.cm.RdBu, levels=colorlevels*absmax(v))
    #plt.colorbar()
    plt.title('v')

    plt.subplot(221)
    X, Y = np.meshgrid(hx, hy)
    plt.contourf(X/Lx, Y/Ly, h.T, cmap=plt.cm.RdBu, levels=colorlevels*absmax(h))
    #plt.colorbar()
    plt.title('h')

    plt.subplot(223)
    plt.plot(hx/Lx, h[:, ny//2])
    plt.xlim(-0.5, 0.5)
    plt.ylim(-absmax(h), absmax(h))
    plt.title('h along x=0')

    plt.pause(0.001)
    plt.draw()

im = None
def plot_fast(u,v,h):
    # only plots an imshow of h, much faster than contour maps
    global im
    if im is None:
        im = plt.imshow(h.T, aspect=Ly/Lx, cmap=plt.cm.RdBu, interpolation='bicubic')
        im.set_clim(-absmax(h), absmax(h))
    else:
        im.set_array(h.T)
        im.set_clim(-absmax(h), absmax(h))
    plt.pause(0.001)
    plt.draw()

def plot_geo_adj(u, v, h):
        plt.clf()

        h0max = absmax(h0)
        plt.subplot(311)
        plt.plot(hx, h[:, ny//2], 'b', linewidth=2)
        plt.plot(hx, h0[:], 'r--', linewidth=1,)
        plt.ylabel('height')
        plt.ylim(-h0max*1.2, h0max*1.2)

        plt.subplot(312)
        plt.plot(vx, v[:, ny//2].T, linewidth=2)
        plt.plot(vx, v0[:, ny//2], 'r--', linewidth=1,)
        plt.ylabel('v velocity')
        plt.ylim(-h0max*.12, h0max*.12)

        plt.subplot(313)
        plt.plot(ux, u[:, ny//2], linewidth=2)
        plt.plot(ux, u0[:, ny/2], 'r--', linewidth=1,)
        plt.xlabel('x/L$_\mathsf{d}$',size=16)
        plt.ylabel('u velocity')
        plt.ylim(-h0max*.12, h0max*.12)

        plt.pause(0.001)
        plt.draw()

plot = plot_all
if experiment is '1d':
    plot = plot_geo_adj


## RUN
# Run the simulation and plot the state
c = time.clock()
for i in range(100000):
    step()
    if i % plot_interval == 0:
        plot(*state)
        print('[t={:7.2f} u: [{:.3f}, {:.3f}], v: [{:.3f}, {:.3f}], h: [{:.3f}, {:.2f}]'.format(
            t/86400,
            u.min(), u.max(),
            v.min(), v.max(),
            h.min(), h.max()))
        #print('fps: %r' % (tc / (time.clock()-c)))