austriadownloader

• Github repository: https://github.com/Zerhigh/austriadownloader/
• Documentation https://Zerhigh.github.io/austriadownloader/

Introduction

This repository contains the code for developing and testing the austriadownloader package, capable of downloading temporally and spatially aligned Austrian Orthophoto (RGB & NIR) and Cadastral data for (among others) Deep Learning application. Available datasets include the Austrian Orthophoto series of 2024 (includes image tiles from 2021 to 2023) and the corresponding Cadastral datasets (eg. from 01.04.2021) published bi-anually. For a detailed analysis and description of datasources, processing steps, and methodology please refer to the corresponding publication (to be added upon release, contact authors for a pre-print until then).

Getting Started

All required meta-datasets are available in austriadownloader/austria_data/ and can be created by executing austriadownloader/austria_data/metadata_creation.py.

Provide sample image POIs as centroids in a dataframe with the following scheme in the WGS84 CRS (EPSG:4326). Image dimensions will be determined by other input parameters such as pixel_size and shape. An independent (but closely related) git-repository for automatically creating such a sample file is available under austriadownloader_sampler.

Sample file structure:

Column	Type	Description
id	str	Unique identifier for each location
lat	float	Latitude coordinate in decimal degrees
lon	float	Longitude coordinate in decimal degrees

An example for a sample file:

id	lat	lon
0	47.6615683485	15.9040047148
1	47.6730783029	15.9045680914
2	47.6845882247	15.9051317152
...	...	...

Code Example:

Refer to demo/demo.py for code, config, and sample files.

from pathlib import Path
from austriadownloader.downloadmanager import DownloadManager
from austriadownloader.configmanager import ConfigManager

config_path = Path("path_to_your_config.yml")
manager = DownloadManager(config=ConfigManager.from_config_file(config_path))
manager.start_download()

Input parameters are provided in the config file and include:

Column	Type	Description
data_path	Path or str	Input path for sampling POI table.
pixel_size	float	Pixel resolution in meters. Must be a predefined value from (0.2, 0.4, 0.8, ... 204.8)
shape	tuple[int, int, int]	Image dimensions as (channels, height, width). Channels must be 3 (RGB) or 4 (RGB & NIR).
outpath	Path or str	Directory path where output files will be saved.
mask_label	list, tuple[int] or int	Cadastral mask(s) to be extracted. A single cadastral label will result in a binary mask, if several cadastral classes are provided a multi-label mask is generated.
mask_remapping	Dict (default: None)	Allows the selection and merging of several cadastral classes.
create_gpkg	bool (default: False)	Indicates whether vectorized but unclipped tiles should be saved as .GPKG in addition to image tiles.
nodata_mode	str (default: 'flag')	Mode for handling no-data values ('flag' or 'remove').
nodata_value	int (default: 0)	Value assigned to no-data pixels in all image data products.
outfile_prefixes	Dict (default: input and target)	Custom name assignement for ouput files: raster -> input, vector -> target
verbose	bool (default: False)	Providing verbose comments during script execution.

Available Classes

To select your class labels, select one or more from the following list (Source: BEV, page 12 ff.):

Category	Code	Subcategory
Building areas	41	Buildings
	83	Adjacent building areas
Water body	59	Flowing water
	60	Standing water
	61	Wetlands
	64	Waterside areas
Agricultural	40	Permanent crops or gardens
	48	Fields, meadows or pastures
	53	Vineyards
	57	Overgrown areas
Forest	55	Krummholz
	56	Forests
	58	Forest roads
Other	42	Car parks
	62	Low vegetation areas
	63	Operating area
	65	Roadside areas
	72	Cemetery
	84	Mining areas, dumps and landfills
	87	Rock and scree surfaces
	88	Glaciers
	92	Rail transport areas
	95	Road traffic areas
	96	Recreational area
Gardens	52	Gardens
Alps	54	Alps

Results

Multi-label mask with all available cadastral classes selected (not all are present in the selected sample):

General overview of different cadastral classes merged into a binary mask:

Selection of unique cadastral classes:

Citation

This repository was created for a presentation at the AGIT 2025 conference.

Repository initiated with fpgmaas/cookiecutter-poetry.