Laws inputs

This tutorial showcases the different inputs that can be used inside the laws.

using ODINN

If we represent the A simple rheological law that we already presented in the Laws tutorial, we can see that it depends on an input T, which is the long term air temperature:

params = Parameters() # Dummy parameters
nn_model = NeuralNetwork(params)
A_law = LawA(nn_model, params)

(:T,) -> Array{Float64, 0}   (↧@start  custom VJP  ✅ precomputed)

The inputs of the laws are represented as objects that derive from AbstractInput. We have defined a subset of inputs that can be used to parameterize laws.

Implementation

It is also possible to define new inputs by creating a new struct type and defining the method for this specific type. On top of type of the input, we also need to override three methods using multiple dispatch:

• default_name: It returns a symbolic representation of the input, which is convenient to have the complete name of it.
• get_input: It computes the value of the input at a given time t for a specific glacier inside the simulation.
• Base.zero: It returns the zero value of the input for a specific glacier, used to generate an empty cache.

For example the scalar long term air temperature is defined with the following code:

struct iAvgScalarTemp <: AbstractInput end
default_name(::iAvgScalarTemp) = :averaged_scalar_long_term_temperature
function get_input(temp::iAvgScalarTemp, simulation, glacier_idx, t)
    glacier = simulation.glaciers[glacier_idx]
    return mean(glacier.climate.longterm_temps_scalar)
end
function Base.zero(temp::iAvgScalarTemp, simulation, glacier_idx)
    glacier = simulation.glaciers[glacier_idx]
    return zero(glacier.climate.longterm_temps_scalar)
end

An input can compute a physical quantity which is not already defined in the iceflow model like the long term air temperature above, but it can also re-use existing variables. This is the case of the surface slope ∇S. It is retrieved simply by returning the cached variable that lives in the iceflow model:

struct i∇S <: AbstractInput end
default_name(::i∇S) = :∇S
function get_input(::i∇S, simulation, glacier_idx, t)
    return simulation.cache.iceflow.∇S
end
function Base.zero(::i∇S, simulation, glacier_idx)
    (; nx, ny) = simulation.glaciers[glacier_idx]
    return zeros(nx - 1, ny - 1)
end

List of available inputs

For the moment we support the following list of inputs:

• Scalar averaged long term air temperature

iAvgScalarTemp <: AbstractInput

iAvgScalarTemp()

averaged_long_term_temperature: iAvgScalarTemp()

• Gridded averaged long term air temperature

iAvgGriddedTemp <: AbstractInput

iAvgGriddedTemp()

gridded_long_term_temperature: iAvgGriddedTemp()

• Cumulative positive degree days (PDD)

iCPDD <: AbstractInput

iCPDD()

CPDD: iCPDD{Dates.Week}(Dates.Week(1))

• Ice thickness on the dual grid in the SIA

iH̄ <: AbstractInput

iH̄()

H_dual_grid: iH̄()

• Surface slope

i∇S <: AbstractInput

S = B + H
∇S = (avg_y(diff_x(S) / Δx) .^ 2 .+ avg_x(diff_y(S) / Δy) .^ 2) .^ (1/2)

i∇S()

∇S: i∇S()

• Topographic roughness

iTopoRough{F<:AbstractFloat} <: AbstractInput

iTopoRough()

topographic_roughness: iTopoRough{Float64}(200.0, :scalar, :flow, :bed)

This page was generated using Literate.jl.