Fig 1.
Visual representation of the methodological approach.
For every one of the 116 districts in Austria the epidemiological curve of the district is compared to the epidemiological curve of the corresponding federal state. This figure shows the example of the district Innsbruck (orange) in the federal state Tyrol. The highlighted blue areas are the districts of Tyrol. More information on the districts is available on Statistics Austria. The epidemiological curves are calculated employing an age-structured compartmental model. The red curve is the epidemiological curve for Innsbruck assuming transmission rates that equal those observed in Tyrol (blue). By including a dependence of the transmission rate on weather, interventions and mobility (green), the district-independent effect sizes of these district-specific input variables can be calculated. The map layer of this figure was taken from open government data.
Fig 2.
Results for the cross-validated hyperparameter search.
For different recovery times, 1/β, we show the percent change in RSS, ΔRSS, between null and augmented model for training (blue) and test (yellow) data. For recovery times of more than 20d, the augmented model explains more than 60% of the regional variations in both test and training data. The probability distribution for 1/β [39] is shown on the axis to the right-hand side.
Fig 3.
Summary of effect sizes of the input variables.
Impacts on the transmission rate are shown in percent for weather variables (blue), NPIs (green) and mobility (magenta). Results for weather and mobility timeseries refer to changes in α for a unit change of one SD in the input. NPIs targeting large gatherings, temperature and humidity show the strongest transmission rate reductions whereas cloudiness leads to the strongest increase. Error bars denote the CI.
Table 1.
Summary of effects of meteorological and mobility time series on the transmission rate.
For each variable we give its unit, the standard deviation (SD) of the input time series and the percent change with its weighted SD of the transmission rate associated with a unit SD change in the input.
Table 2.
Summary of effects of NPIs on the transmission rate.
For each category of NPIs we give the number of implementations observed in our data, list typical examples of what the NPI consists of and the percent change of the transmission rate associated with a unit SD change in the input.
Table 3.
Result of the re-fit and re-run with variable group dropout for investigation of impact of the different variable groups on explained variation.