pypsa-eur/scripts/build_solar_thermal_profiles.py
Fabian Neumann 013b705ee4
Clustering: build renewable profiles and add all assets after clustering (#1201)
* Cluster first: build renewable profiles and add all assets after clustering

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* correction: pass landfall_lengths through functions

* assign landfall_lenghts correctly

* remove parameter add_land_use_constraint

* fix network_dict

* calculate distance to shoreline, remove underwater_fraction

* adjust simplification parameter to exclude Crete from offshore wind connections

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* remove unused geth2015 hydro capacities

* removing remaining traces of {simpl} wildcard

* add release notes and update workflow graphics

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

---------

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: lisazeyen <lisa.zeyen@web.de>
2024-09-13 15:37:01 +02:00

87 lines
2.4 KiB
Python

# -*- coding: utf-8 -*-
# SPDX-FileCopyrightText: : 2020-2024 The PyPSA-Eur Authors
#
# SPDX-License-Identifier: MIT
"""
Build solar thermal collector profile time series.
Uses ``atlite.Cutout.solar_thermal` to compute heat generation for clustered onshore regions from population layout and weather data cutout.
The rule is executed in ``build_sector.smk``.
.. seealso::
`Atlite.Cutout.solar_thermal <https://atlite.readthedocs.io/en/master/ref_api.html#module-atlite.convert>`_
Relevant Settings
-----------------
.. code:: yaml
snapshots:
drop_leap_day:
solar_thermal:
atlite:
default_cutout:
Inputs
------
- ``resources/<run_name/pop_layout_<scope>.nc``:
- ``resources/<run_name/regions_onshore_base_s<simpl>_<clusters>.geojson``:
- ``cutout``: Weather data cutout, as specified in config
Outputs
-------
- ``resources/solar_thermal_<scope>_base_s<simpl>_<clusters>.nc``:
"""
import atlite
import geopandas as gpd
import numpy as np
import xarray as xr
from _helpers import get_snapshots, set_scenario_config
from dask.distributed import Client, LocalCluster
if __name__ == "__main__":
if "snakemake" not in globals():
from _helpers import mock_snakemake
snakemake = mock_snakemake("build_solar_thermal_profiles", clusters=48)
set_scenario_config(snakemake)
nprocesses = int(snakemake.threads)
cluster = LocalCluster(n_workers=nprocesses, threads_per_worker=1)
client = Client(cluster, asynchronous=True)
config = snakemake.params.solar_thermal
config.pop("cutout", None)
time = get_snapshots(snakemake.params.snapshots, snakemake.params.drop_leap_day)
cutout = atlite.Cutout(snakemake.input.cutout).sel(time=time)
clustered_regions = (
gpd.read_file(snakemake.input.regions_onshore).set_index("name").buffer(0)
)
I = cutout.indicatormatrix(clustered_regions)
pop_layout = xr.open_dataarray(snakemake.input.pop_layout)
stacked_pop = pop_layout.stack(spatial=("y", "x"))
M = I.T.dot(np.diag(I.dot(stacked_pop)))
nonzero_sum = M.sum(axis=0, keepdims=True)
nonzero_sum[nonzero_sum == 0.0] = 1.0
M_tilde = M / nonzero_sum
solar_thermal = cutout.solar_thermal(
**config,
matrix=M_tilde.T,
index=clustered_regions.index,
dask_kwargs=dict(scheduler=client),
show_progress=False
)
solar_thermal.to_netcdf(snakemake.output.solar_thermal)