A useful expansion of the overland_routing.py module would be to include multiple routing frameworks of varying complexity which could be toggled from config.yaml. This would improve the DRL capability from both a hydrology and ML education perspective.
The routing options would be (in ascending complexity):
- Kinematic Wave
- Simple Diffusive Wave Approximation
- Full Diffusive Wave Approximation
- Shallow Water Equations
Where "simple" in the diffusive wave approximation refers to: no square root of the water surface slope term, hydraulic radius (hfx/hfy) is always taken as minimum water height between cells, water surface slope clipping.
The routing module would need to have another more restrictive stability criteria (Von Neumann?) included.
Worthwhile comparisons in performance between the different routing options could be made, including depth-per-cell, arrival time, and mass balance. Additionally, the performance of CNN's trained on options 1,2 & 3 compared against the CNN trained on the SWE's could be combined with DEM slope, roughness, and precipitation intensity to produce a metric defining the appropriate routing complexity required for a given hydrological region.
A useful expansion of the overland_routing.py module would be to include multiple routing frameworks of varying complexity which could be toggled from config.yaml. This would improve the DRL capability from both a hydrology and ML education perspective.
The routing options would be (in ascending complexity):
Where "simple" in the diffusive wave approximation refers to: no square root of the water surface slope term, hydraulic radius (hfx/hfy) is always taken as minimum water height between cells, water surface slope clipping.
The routing module would need to have another more restrictive stability criteria (Von Neumann?) included.
Worthwhile comparisons in performance between the different routing options could be made, including depth-per-cell, arrival time, and mass balance. Additionally, the performance of CNN's trained on options 1,2 & 3 compared against the CNN trained on the SWE's could be combined with DEM slope, roughness, and precipitation intensity to produce a metric defining the appropriate routing complexity required for a given hydrological region.