Synthetic Contact and Purification Plant Network Data for Austria

Objectives

This data contains synthetic contact networks among Austrian households, which was created using an agent-based COVID-19 model. The data is foundation to the work published in Model-Based Optimisation of Outbreak Detection via Wastewater Probing where it was used as input to a multipopulation-SIR compartment model.

Beyond its original purpose, the data is intended to be used for being used in dynamic models in which pysical inter-household contacts are required. This includes, but is not restricted to epidemiological models. The data should not be used for static network analysis, considering that it is not sufficiently validated in this respect.

Methods

The synthetic household coordinates, their inhabitant counts as well as their contacts with other households via school, workplace and leisuretime are randomly sampled by the agent-based model, but are statistically representative for Austria for that date w.r. to data about population census, household distribution, regional number and size of work-places and school(classes), and average number (Polymod survey) and location (mobile phone data) for leisuretime contacts. For more details on the underlying data sources we refer to the SI of Evaluation of Contact-Tracing Policies against the Spread of SARS-CoV-2 in Austria: An Agent-Based Simulation where the agent-based epidemics model and its underlying data sources are described. For the generation of the household-household contact data, the agent-based epidemic model is run for 10 days counting the number of sampled contacts between members of different households.

In addition, (almost) every synthetic household is linked to a wastewater treatment plant. The assignment of these plants is based on actual Austrian wastewater treatment plants and their catchment areas - see data.gv.at for details.

Content

The different datasets differ in terms of the network's scale:

• Network_100 is an unscaled version with roughly 4.5 Million households. This reflects quite accurately the actual number of households in Austria on 2020-01-01.
• Network_0XX are scaled-down versions (factor 0.XX), wherein the number of agents (as well as work-places, schools, etc.) in the agent-based are scaled-down by the correponding factor.

Each network dataset is available in two versions: a human-readable CSV version and a version specifically designed to be loaded with Python3 and numpy.

csv

• nodes_households.csv contains data about the synthetic household nodes.
  • id: id of the household
  • latitude/longitude: sampled place of residence (WSG84 coordinate)
  • regionId: municipality ID corresponding to the place (area status 2020).
  • inhabitants: number of household members
• nodes_wastewater.csv contains data about the purification plant nodes.
  • id: id of the plants
  • latitude/longitude: place of the plant (WSG84 coordinate)
  • regionId: municipality ID corresponding to the place (area status 2020).
• edges_household_XXXX.csv for school, work and leisuretime contains the edges of the household-household-network corresponding to this contact location.
  • id1, id2: ids of the contact household (match with the *id* column in nodes_households.csv)
  • avgContacts: average number of daily contacts between the two households within the 10-day simulation period of the agent-based model.
•  edges_wastewater.csv contains the edges of the household-purification plant-assignment.
  •  id1: id of the household (match with the *id* column in nodes_households.csv)
  •  id2: id of the purification plant associated with the household (match with the *id* column in nodes_wastewater.csv)

numpy

The following data was exported with Python (pickle) 3.12.2 and numpy 2.4.2

• adjacency_households_households.npz contains an adjacency matrix of the household-household contacts (size NxN) as sparse-numpy array (.npz format). The non-zero entry for index (i,j) means that household i has contacts with household j, whereas the value of the matrix is the sum of all corresponding avgContacts fields in edges_household_XXXX.csv for school, work and leisuretime. Note that the indices in the matrix are not identical with the ids of the household as defined in the CSV - see metadata.pickle.
• adjacency_households_wastewater.npz contains an adjacency matrix of the household-plant assignment (size NxM) as sparse-numpy array (.npz format). A non-zero (always 1) entry for index (i,j) means that household i is assigned to plant j. Note that the indices in the matrix are not identical with the ids of the household/plant as defined in the CSVs - see metadata.pickle.
• metadata.pickle contains a tuple of two pickled dictionaries:
  • the first one maps the (numpy-)index of the household to the corresponding metadata - itself a dict with keys:
    • id: id of the household
    • lat/long: sampled place of residence (WSG84 coordinate)
    • regionId: municipality ID corresponding to the place (area status 2020).
    • inhabitants: number of household members
  • the second one maps the (numpy-)index of the purification plant to the corresponding metadata - itself a dict with keys:
    • id: id of the plants
    • lat/long: place of the plant (WSG84 coordinate)
    • regionId: municipality ID corresponding to the place (area status 2020).