granular-channel-hydro

Subglacial hydrology model for sedimentary channels
git clone git://src.adamsgaard.dk/granular-channel-hydro
Log | Files | Refs | README | LICENSE

commit 63f99a6d5a5fdb264a19d29281a8b8a373358632
parent d0ec40ca4dde58229f1c40d4fc0d3789a3d679a2
Author: Anders Damsgaard <andersd@riseup.net>
Date:   Wed,  8 Mar 2017 19:42:12 -0800

solve for water pressure

Diffstat:
M1d-channel.py | 66++++++++++++++++++++++++++++++------------------------------------
1 file changed, 30 insertions(+), 36 deletions(-)

diff --git a/1d-channel.py b/1d-channel.py @@ -24,10 +24,11 @@ import sys Ns = 25 # Number of nodes [-] # Ls = 100e3 # Model length [m] Ls = 1e3 # Model length [m] +# Ls = 1e3 # Model length [m] total_days = 60. # Total simulation time [d] t_end = 24.*60.*60.*total_days # Total simulation time [s] tol_Q = 1e-3 # Tolerance criteria for the normalized max. residual for Q -tol_N_c = 1e-3 # Tolerance criteria for the norm. max. residual for N_c +tol_P_c = 1e-3 # Tolerance criteria for the norm. max. residual for P_c max_iter = 1e2*Ns # Maximum number of solver iterations before failure print_output_convergence = False # Display convergence statistics during run safety = 0.01 # Safety factor ]0;1] for adaptive timestepping @@ -44,15 +45,13 @@ sand_fraction = 0.5 # Initial volumetric fraction of sand relative to gravel D_g = 1. # Mean grain size in gravel fraction (> 2 mm) D_s = 0.01 # Mean grain size in sand fraction (<= 2 mm) -Q_terminus = 0.01/2. # Desired water flux at terminus [m^3/s] -m_dot = Q_terminus/Ls # Water source term [m/s] - -mu_w = 1.787e-3 # Water viscosity [Pa*s] -friction_factor = 0.1 # Darcy-Weisbach friction factor [-] +# Boundary conditions +P_terminus = 0. # Water pressure at terminus [Pa] +m_dot = 1.0e-5 # Water source term [m/s] # Channel hydraulic properties manning = 0.1 # Manning roughness coefficient [m^{-1/3} s] -#F = rho_w*g*manning*(2.*(numpy.pi + 2)**2./numpy.pi)**(2./3.) +friction_factor = 0.1 # Darcy-Weisbach friction factor [-] # Channel growth-limit parameters c_1 = -0.118 # [m/kPa] @@ -70,15 +69,15 @@ s = numpy.linspace(0., Ls, Ns) ds = s[1:] - s[:-1] # Ice thickness and bed topography -H = 6.*(numpy.sqrt(Ls - s + 5e3) - numpy.sqrt(5e3)) + 1.0 # glacier +H = 6.*(numpy.sqrt(Ls - s + 5e3) - numpy.sqrt(5e3)) + 10.0 # glacier # slope = 0.1 # Surface slope [%] # H = 1000. + -slope/100.*s b = numpy.zeros_like(H) N = H*0.1*rho_i*g # Initial effective stress [Pa] -p_w = rho_i*g*H - N # Initial guess of water pressure [Pa] -hydro_pot = rho_w*g*b + p_w # Initial guess of hydraulic potential [Pa] +#p_w = rho_i*g*H - N # Initial guess of water pressure [Pa] +#hydro_pot = rho_w*g*b + p_w # Initial guess of hydraulic potential [Pa] # Initialize arrays for channel segments between nodes S = numpy.ones(len(s) - 1)*S_min # Cross-sect. area of channel segments[m^2] @@ -88,9 +87,7 @@ W = S/numpy.tan(numpy.deg2rad(theta)) # Assuming no channel floor wedge Q = numpy.zeros_like(S) # Water flux in channel segments [m^3/s] Q_s = numpy.zeros_like(S) # Sediment flux in channel segments [m^3/s] N_c = numpy.zeros_like(S) # Effective pressure in channel segments [Pa] -e_dot = numpy.zeros_like(S) # Sediment erosion rate in channel segments [m/s] -d_dot = numpy.zeros_like(S) # Sediment deposition rate in chan. segments [m/s] -c_bar = numpy.zeros_like(S) # Vertically integrated sediment concentration [-] +P_c = numpy.zeros_like(S) # Water pressure in channel segments [Pa] tau = numpy.zeros_like(S) # Avg. shear stress from current [Pa] porosity = numpy.ones_like(S)*0.3 # Sediment porosity [-] res = numpy.zeros_like(S) # Solution residual during solver iterations @@ -240,40 +237,40 @@ def flux_solver(m_dot, ds): def pressure_solver(psi, f, Q, S): # Iteratively find new water pressures - # dN_c/ds = f*rho_w*g*Q^2/S^{8/3} - psi (Kingslake and Ng 2013) + # dP_c/ds = psi - f*rho_w*g*Q^2/S^{8/3} (Kingslake and Ng 2013) it = 0 max_res = 1e9 # arbitrary large value - while max_res > tol_N_c or it < Ns: + while max_res > tol_P_c or it < Ns: - N_c_old = N_c.copy() + P_c_old = P_c.copy() # P_downstream = P_upstream + dP - # N_c[1:] = N_c[:-1] \ + # P_c[1:] = P_c[:-1] \ # + psi[:-1]*ds[:-1] \ # - f[:-1]*rho_w*g*Q[:-1]**2./(S[:-1]**(8./3.))*ds[:-1] \ # Dirichlet BC (fixed pressure) at terminus - N_c[-1] = 0. + P_c[-1] = P_terminus # P_upstream = P_downstream - dP - N_c[:-1] = N_c[1:] \ - + psi[:-1]*ds[:-1] \ - - f[:-1]*rho_w*g*Q[:-1]**2./(S[:-1]**(8./3.))*ds[:-1] + P_c[:-1] = P_c[1:] \ + - psi[:-1]*ds[:-1] \ + + f[:-1]*rho_w*g*Q[:-1]**2./(S[:-1]**(8./3.))*ds[:-1] # + psi[:-1]*ds[:-1] \ # - f[:-1]*rho_w*g*Q[:-1]**2./(S[:-1]**(8./3.))*ds[:-1] - max_res = numpy.max(numpy.abs((N_c - N_c_old)/(N_c + 1e-16))) + max_res = numpy.max(numpy.abs((P_c - P_c_old)/(P_c + 1e-16))) if print_output_convergence: print('it = {}: max_res = {}'.format(it, max_res)) if it >= max_iter: raise Exception('t = {}, step = {}:'.format(time, step) + - 'Iterative solution not found for N_c') + 'Iterative solution not found for P_c') it += 1 - return N_c + return P_c def plot_state(step, time, S_, S_max_, title=True): @@ -286,7 +283,7 @@ def plot_state(step, time, S_, S_max_, title=True): # ax_Pa.plot(s/1000., N/1000., '--r', label='$N$') ax_Pa.plot(s_c/1000., N_c/1e6, '-k', label='$N$') ax_Pa.plot(s_c/1000., H_c*rho_i*g/1e6, '--r', label='$P_i$') - #ax_Pa.plot(s_c/1000., P_c/1e6, ':y', label='$P_c$') + ax_Pa.plot(s_c/1000., P_c/1e6, ':y', label='$P_c$') ax_m3s = ax_Pa.twinx() # axis with m3/s as y-axis unit ax_m3s.plot(s_c/1000., Q, '.-b', label='$Q$') @@ -300,9 +297,9 @@ def plot_state(step, time, S_, S_max_, title=True): ax_m3s.set_ylabel('[m$^3$/s]') ax_m3s_sed = plt.subplot(3, 1, 2, sharex=ax_Pa) - ax_m3s_sed.plot(s_c/1000., Q_g, ':', label='$Q_g$') - ax_m3s_sed.plot(s_c/1000., Q_s, '-', label='$Q_s$') - ax_m3s_sed.plot(s_c/1000., Q_t, '--', label='$Q_t$') + ax_m3s_sed.plot(s_c/1000., Q_g, ':', label='$Q_{gravel}$') + ax_m3s_sed.plot(s_c/1000., Q_s, '-', label='$Q_{sand}$') + ax_m3s_sed.plot(s_c/1000., Q_t, '--', label='$Q_{total}$') ax_m3s_sed.set_ylabel('[m$^3$/s]') ax_m3s_sed.legend(loc=2) @@ -352,12 +349,6 @@ def print_status_to_stdout(time, dt): sys.stdout.flush() s_c = avg_midpoint(s) # Channel section midpoint coordinates [m] - -# Find gradient in hydraulic potential between the nodes -hydro_pot_grad = gradient(hydro_pot, s) - -# Find field values at the middle of channel segments -N_c = avg_midpoint(N) H_c = avg_midpoint(H) # Find water flux @@ -421,8 +412,11 @@ while time <= t_end: # Find hydraulic roughness f = channel_hydraulic_roughness(manning, S, W, theta) - # Find new effective pressures consistent with the flow law - N_c = pressure_solver(psi, f, Q, S) + # Find new water pressures consistent with the flow law + P_c = pressure_solver(psi, f, Q, S) + + # Find new effective pressure in channel segments + N_c = rho_i*g*H_c - P_c # Find new maximum normalized residual value max_res = numpy.max(numpy.abs((S - S_prev_it)/(S + 1e-16)))