Skip to main content

B. Gailleton

💧Getting Started on Graphflood

/softwares/graphflood/getting_start/
Table of Contents

## Quick Start

While I am still constructing the documentation I provide on this page quick ways to use graphflood.

You need to have:

  • 2D topography (DEM file, premade function, anything that provide a 2D regular grid of elevation)
  • Precipitation rates (scalar, 2D array)
  • Manning coefficient (scalar, 2D array) -> 0.033 is a commonly accepted value in rivers
  • [if needed] 2D Boundary conditions
  • [If not done yet] scabbard installed

You need to now

  • Every backend has different pros and cons and may produce slightly different results.
  • gpu backend is a lot faster, but you need a GPU with enough memory
  • graphflood is iterative, you need to decide on a convergence criterion (easy solution: run until the flow depth stabilises)
  • Higher dt = faster convergence, but more instabilities. I am working on adding automatic dt and convergence detection.

**

# Simple Setup

DEM has no NoData, flow can out from every edges of the model

Default Boundary Conditions

import matplotlib.pyplot as plt
import numpy as np
import taichi as ti
import scabbard as scb


# Loading the DEM
grid = scb.io.load_raster('dem.tif')


# Change this variable to change the backend:

## GPU backend with taichi 
### (if you have a MAC this remains untested but theoretically works)
### (If you are on Linux/Windows WITHOUT Nvidia GPU, you need to install the vulkan SDK/drivers -> https://vulkan.lunarg.com/)
# backend = 'gpu'
# ti.init('gpu')
## CPU backend with DAGGER (OG graphflood - both options work)
# backend = 'cpu'
# backend = 'dagger'

## CPU libTopoToolbox backend
backend = 'ttb'

# running Graphflood dor 1000 iterations
results = scb.graphflood.std_run(
    grid, # Sting or grid so far
    P = 1e-4, # precipitations, 2D numpy array or scalar
    BCs = None, # Boundary codes
    N_dt = 500,
    backend = backend,
    dt = 1e-2,
    init_hw = None) # init_hw can be a 2D array of DEM dimension with pre-existing flow depth. For example if you want to continue a run

fig,ax = scb.visu.hillshaded_basemap(grid)


hw = results['h'].Z.copy() # if ttb is used, results arrays are topotoolbox GridObj
hw[hw<0.01] = np.nan
im = ax.imshow(hw, cmap = 'Blues', vmax = 3., extent = grid.geo.extent, alpha = 0.75)
plt.colorbar(im, label = 'Flow Depth (m)')
plt.show()

# Isolating sub-section of a DEM

DEM has NoData, or for any other reason will not consider some nodes (e.g. bellow sea level, only the main watershed, ...)

Use of automatic functions to set up the Boundary Conditions

import matplotlib.pyplot as plt
import numpy as np
import taichi as ti
import scabbard as scb


# Loading the DEM
grid = scb.io.load_raster('dem.tif')

# Isolating the main drainage basin
## this function returns a mask of 0,1 showing the main watershed
mask = scb.flow.mask_main_basin(grid, MFD = False)
## You could also, alternatively, mask seas in case you wanna remove all the data below a certain values
# mask = mask_seas(grid, sea_level = 0.)

## Encode the mask into boundary conditions
BCs = scb.flow.mask_to_BCs(grid,mask)

# Change this variable to change the backend:

## GPU backend with taichi 
### (if you have a MAC this remains untested but theoretically works)
### (If you are on Linux/Windows WITHOUT Nvidia GPU, you need to install the vulkan SDK/drivers -> https://vulkan.lunarg.com/)
backend = 'gpu'
## CPU backend with DAGGER (OG graphflood - both options work)
# backend = 'cpu'
# backend = 'dagger'

## CPU libTopoToolbox backend
# backend = 'ttb'

# running Graphflood dor 1000 iterations
results = scb.graphflood.std_run(
    grid, # Sting or grid so far
    P = 1e-4, # precipitations, numpy array or scalar
    BCs = BCs, # Boundary codes
    N_dt = 5000,
    backend = backend,
    dt = 1e-2,
    init_hw = None)

fig,ax = scb.visu.hillshaded_basemap(grid)


hw = results['h'].copy()
hw[hw<0.01] = np.nan
im = ax.imshow(hw, cmap = 'Blues', vmax = 3., extent = grid.geo.extent, alpha = 0.75)
plt.colorbar(im, label = 'Flow Depth (m)')
plt.show()

# Reach Mode: modelling a local trunk with entry fluxes

Coming soon: how to finely manage boundary conditions in a case where I want to input flow on the left edge of my DEM, and force it to only output at the right edge

Graphflood output