README.md

---
license: other
language:
- en
- sv
tags:
- climate
pretty_name: Low-to-High Resolution Weather Forecasting using Topography
---
# Dataset: Low-to-High-Resolution Weather Forecasting using Topography

The dataset is intended and structured for the problem of transforming/interpolating low-resolution weather forecasts into higher resolution using topography data.


![Dataset diagram: 4 ECMWF grid points at the corners, 1 SMHI observation station, topographical data in the background](./assets/high_res_weather_forecasting_dataset.png)

The dataset consists of 3 different types of data (as illustrated above):
- **Historical weather observation data ([SMHI](https://opendata.smhi.se/apidocs/metobs/index.html))**
  - Historical weather observation data from selected SMHI observation stations (evaluation points)
- **Historical low-resolution weather forecasts ([ECMWF](https://www.ecmwf.int/en/forecasts/datasets/set-i))**
  - For a given SMHI station: Historical, (relatively) low-resolution ECMWF weather forecasts from the 4 nearest ECMWF grid points
- **Topography/elevation data ([Copernicus DEM GLO-30](https://spacedata.copernicus.eu/collections/copernicus-digital-elevation-model))**:
  - Topography/elevation data around a given SMHI station, grid enclosed by the 4 ECMWF points
  - Mean elevation around each of the nearest 4 ECMWF grid points for a given SMHI station (mean over +/-0.05 arc degree in each direction from the grid point center)

The dataset is meant to facilitate the following modeling pipeline:
- Weather forecasts for a set of 4 neighboring ECMWF points are combined with topography/elevation data and turned into higher resolution forecasts grid (corresponding to the resolution of the topography data)
- SMHI weather observation data is used as a sparse evaluation point of the produced higher-resolution forecasts

## License and terms

The terms of usage of the source data and their corresponding licenses:
- SMHI weather observation data: (Creative Common Attribution 4.0)
  - [SMHI terms and conditions](https://www.smhi.se/data/oppna-data/information-om-oppna-data/villkor-for-anvandning-1.30622)
  - [SMHI: Creative Commons Attribution 4.0](https://www.smhi.se/polopoly_fs/1.93273!/Menu/general/extGroup/attachmentColHold/mainCol1/file/CreativeCommons-Erk%C3%A4nnande-4.0.pdf)
  - [SMHI: terms](https://www.smhi.se/polopoly_fs/1.164900!/Villkor%20f%C3%B6r%20konsekvensbaserade%20v%C3%A4dervarningar%20och%20meddelanden%20%28G%C3%A4ller%20fr%C3%A5n%20oktober%202021%29.pdf)
- ECMWF historical weather forecasts:
  - [ECMWF License](https://www.ecmwf.int/sites/default/files/ECMWF_Standard_Licence.pdf)
- Copernicus DEM GLO-30:
  - [Copernicus: Data access license](https://spacedata.copernicus.eu/documents/20123/121286/CSCDA_ESA_User_Licence_latest.pdf/)
  - [Licence for the use of the Copernicus WorldDEM-30](./topography/eula_F.pdf)

## Acknowledgment

The dataset has been developed as part of the [OWGRE project](https://www.smhi.se/en/research/research-departments/meteorology/owgre-1.197929), funded within the ERA-Net SES Joint Call 2020 for transnational research, development and demonstration projects. 

## Data details & samples

### SMHI weather observation data

SMHI weather observation data is structured in csv files, separately for each weather parameter and weather observation station.
See the sample below:

```csv
Datum;Tid (UTC);Lufttemperatur;Kvalitet
2020-01-01;06:00:00;-2.2;G
2020-01-01;12:00:00;-2.7;G
2020-01-01;18:00:00;0.2;G
2020-01-02;06:00:00;0.3;G
2020-01-02;12:00:00;4.3;G
2020-01-02;18:00:00;4.9;G
2020-01-03;06:00:00;6.0;G
2020-01-03;12:00:00;2.7;G
2020-01-03;18:00:00;1.7;G
2020-01-04;06:00:00;-4.6;G
2020-01-04;12:00:00;0.6;G
2020-01-04;18:00:00;-5.9;G
2020-01-05;06:00:00;-7.9;G
2020-01-05;12:00:00;-3.1;G

```

### ECMWF historical weather forecasts

Historical ECMWF weather forecasts contain a number of forecasted weather variables at 4 nearest grid points around each SMHI observation station:

```
<xarray.Dataset>
Dimensions:                 (reference_time: 2983, valid_time: 54,
                             corner_index: 4, station_index: 275)
Coordinates:
  * reference_time          (reference_time) datetime64[ns] 2020-01-01 ... 20...
    latitude                (corner_index, station_index) float64 55.3 ... 68.7
    longitude               (corner_index, station_index) float64 ...
    point                   (corner_index, station_index) int64 ...
  * valid_time              (valid_time) int32 0 1 2 3 4 5 ... 48 49 50 51 52 53
  * station_index           (station_index) int64 0 1 2 3 4 ... 271 272 273 274
  * corner_index            (corner_index) <U3 'llc' 'lrc' 'ulc' 'urc'
    station_names           (station_index) <U29 ...
    station_ids             (station_index) int64 ...
Data variables:
    PressureReducedMSL      (reference_time, valid_time, corner_index, station_index) float32 ...
    RelativeHumidity        (reference_time, valid_time, corner_index, station_index) float32 ...
    SolarDownwardRadiation  (reference_time, valid_time, corner_index, station_index) float64 ...
    Temperature             (reference_time, valid_time, corner_index, station_index) float32 ...
    WindDirection:10        (reference_time, valid_time, corner_index, station_index) float32 ...
    WindSpeed:10            (reference_time, valid_time, corner_index, station_index) float32 ...

```

### Topography data

The topography data is provided in the chunks cut around each of the SMHI stations.
The corners of each chunk correspond to ECMWF forecast grid points.

Each chunk consists approximately 361 x 361 points, spanning across 0.1° x 0.1°. (Some of the values across longitudes are NaN since apparently the Earth is not square [citation needed]).

![Sample topography map](./assets/elevation_data_sample_0_1_degree.png)

## Loading the data

The dependencies can be installed through `conda` or `mamba` in the following way:

```bash
mamba create -n ourenv python pandas xarray dask netCDF4
```

---

Below, for a given SMHI weather observation station, we read the following data:
- weather observations
- historical ECMWF weather forecasts
- topography/elevation

```python
import pickle
import pandas as pd
import xarray as xr

smhi_weather_observation_station_index = 153
smhi_weather_observation_station_id = pd.read_csv(
    './smhi_weather_observation_stations.csv',
    index_col='station_index'
).loc[smhi_weather_observation_station_index]['id'] # 102540

weather_parameter = 1 # temperature

# NOTE: Need to unzip the file first!
smhi_observation_data = pd.read_csv(
    './weather_observations/smhi_observations_from_2020/'
    f'parameter_{weather_parameter}'
    f'/smhi_weather_param_{weather_parameter}_station_{smhi_weather_observation_station_id}.csv',
    sep=';',
)
print('SMHI observation data:')
print(smhi_observation_data)

ecmwf_data = xr.open_dataset(
    './ecmwf_historical_weather_forecasts/ECMWF_HRES-reindexed.nc'
).sel(station_index=smhi_weather_observation_station_index)
print('ECMWF data:')
print(ecmwf_data)

topography_data = xr.open_dataset(
    'topography/sweden_chunks_copernicus-dem-30m'
    f'/topography_chunk_station_index-{smhi_weather_observation_station_index}.nc'
)
print('Topography chunk:')
print(topography_data)

mean_elevations = pickle.load(open('./topography/ecmwf_grid_mean_elevations.pkl', 'rb'))
print('Mean elevation:')
print(mean_elevations[0])
```