Replace daily lists in HydrologyModel with array implementation

[davidorme]

Is your feature request related to a problem? Please describe.

At the moment, the hydrology model tracks daily values for variables within a time step by appending daily arrays to a list and then using np.stack to combine them at the end to get statistics across the time step.

virtual_ecosystem/virtual_ecosystem/models/hydrology/hydrology_model.py

Lines 390 to 408 in 6812907

	# Create lists for output variables to store daily data
	daily_lists: dict = {name: [] for name in self.vars_updated}
	for day in np.arange(days):
	# Interception of water in canopy, [mm]
	interception = above_ground.calculate_interception(
	leaf_area_index=hydro_input["leaf_area_index_sum"],
	precipitation=hydro_input["current_precipitation"][:, day],
	intercept_parameters=self.model_constants.intercept_parameters,
	veg_density_param=self.model_constants.veg_density_param,
	)
	# TODO add canopy evaporation
	# Precipitation that reaches the surface per day, [mm]
	precipitation_surface = (
	hydro_input["current_precipitation"][:, day] - interception
	)
	daily_lists["precipitation_surface"].append(precipitation_surface)

Since the number of days is known (and will always be known?), this can be done by declaring a days x grid cell numpy array and inserting daily rows. If the code below is a fair comparison of the two approaches then this is faster and cleaner. I have to admit I was expecting it to be more faster, but it is faster and the advantage greater as the number of cells increases.

In [1]: import timeit
   ...: import numpy as np
   ...: 
   ...: 
   ...: def stacking_arrays(n_cells=10, n_days=30):
   ...: 
   ...:     val = []
   ...:     for day in np.arange(n_days):
   ...:         val.append(np.arange(n_cells))
   ...: 
   ...:     return np.sum(np.stack(val, axis=1), axis=1)
   ...: 

In [2]: def matrix_insertion(n_cells=10, n_days=30):
   ...: 
   ...:     val = np.empty((n_days, n_cells))
   ...:     for day in np.arange(n_days):
   ...:         val[day, :] = np.arange(n_cells)
   ...: 
   ...:     return np.sum(val, axis=0)
   ...: 

In [3]: 
   ...: assert np.allclose(stacking_arrays(),  matrix_insertion())
   ...: 
   ...: number = 1000

In [4]: np.min(timeit.repeat("stacking_arrays()", globals=globals(), number=number))
Out[4]: np.float64(0.02308262512087822)

In [5]: np.min(timeit.repeat("matrix_insertion()", globals=globals(), number=number))
Out[5]: np.float64(0.0157989589497447)

In [6]: np.min(timeit.repeat("stacking_arrays(n_cells=1000)", globals=globals(), number=number))
Out[6]: np.float64(0.05126216681674123)

In [7]: np.min(timeit.repeat("matrix_insertion(n_cells=1000)", globals=globals(), number=number))
Out[7]: np.float64(0.03652237495407462)

In [8]: np.min(timeit.repeat("stacking_arrays(n_cells=5000)", globals=globals(), number=number))
Out[8]: np.float64(0.2828623750247061)

In [9]: np.min(timeit.repeat("matrix_insertion(n_cells=5000)", globals=globals(), number=number))
Out[9]: np.float64(0.13297262508422136)

Describe the solution you'd like

Switch to matrix insertion, I think.