Adding heterogeneity to the soils through a NN

[taddyb]

What was done:

• Created a synthetic LGAR simulation using a 6-layer soil of:

Sandy -> sandy loam -> loam -> silty loam -> clay loam -> clay

• Created an MLP to turn POLARIS/SSURGO soils point data into lumped catchment soils data
• Added theta_e and theta_r as parameters
• Made sure the MLP weights are now the model parameters
• Updated the dpLGAR set initial states method

Issues addressed:

• The soil wetting front problem: NoneType Blow up after 8000 timesteps #14
  • By giving different values of theta to the soil layers, the wetting front drainage problem should fix itself.
• Model validation: Validate dpLGAR using Streamflow #8

Tested:

• Ran through the synthetic case and was able to generate data
• Ran the MLP to confirm parameter creation
• TODO: Train against synthetic data

Notes:

• This was not directly tested on the DDP agent, but should work based on preliminary checks