prepared by Henrik Fisser within the upscaling part of the EuroDataCube COVID-19 Custom Script Contest.
This indicator is based on the detection of moving trucks on primary roads and trunks in the EU. The detection uses data from the Copernicus Sentinel-2 satellite.
Transportation on roads is by far the most important mode of cargo logistics in Europe. The amount of roaming trucks may indicate how the economy of a country is performing. Germany for instance uses truck counts from toll statistics as a short-term economic indicator. Sentinel-2 data may be used for capturing some of these economic dynamics through the EU-wide truck detection. Detecting trucks with satellite imagery has traditionally not been attempted at relatively coarse resolution, but a new detection approach now facilitates extracting moving trucks from Sentinel-2 imagery. The truck indicator shows if there has been an unnormal low or high number of trucks on primary roads and trunks within the respective region of interest. Truck counts are aggregated in a period of approximately three months. Currently available information was derived in April, May and June for the area of the EU.
This indicator is based on a method that enables to detect trucks on a large scale using Sentinel-2 data. The method exploits an effect related to the Sentinel-2 Multispectral Instrument (MSI) geometry. Sentinel-2 does not see a moving truck once but three times in the red-blue-green wavelengths. As the truck keeps travling during this short time offset, it appears spectrally disassembled. This pattern may be used for detecting roaming trucks on roads. Although visual inspection cannot confirm that the objects are trucks, this is implied by the ratio between size of different vehicles and a Sentinel-2 pixel (see validation for details). However, a confusion with moving vehicles of similar size such as buses may occur. In order to generally reduce false detections the computation is constrained to road data from Open Street Maps (OSM). The blue dots in Fig. 1 are examples of identified trucks in these subsets.
Fig. 1: Truck detection examples in Madrid
Detected trucks in 2020 are compared with the median value of the baseline years 2017, 2018 and 2019. In each period the number of detected trucks in an area is divided by the number of observations in order to make the figures comparable across the continent. The counts of each year are compared with the baseline value, which is the median of the three baseline years. The median is used in order to reduce arbitrary oscillations of the baseline. Truck counts higher or lower 5 % than the baseline are denoted as 'High' or 'Low' respectively.
In average, 7 % less trucks than normal were observed on primary roads and trunks in the EU during the considered period. However, the figures vary across the EU as depicted in Fig. 2. The map might already suggest some patterns, nevertheless it is worth looking into the details. Lockdown measures have been diverse across the EU both in duration and intensity. Depending on area-specific data availability and the lockdown periods effects directly induced by the measures as well as potential catch-up effects might be captured.
Fig. 2: Trucks on primary roads / trunks - indicator results in EU regions
The Sentinel-2 truck detection method has been validated with German traffic count stations on the two motorways A3 and A61. This validation implies that the Sentinel-2 method detects in average 78 % of the trucks counted at the four used station (Fig. 3), though with some variety. It also suggests that detected objects are mostly trucks and not cars as the number of counted cars at the count stations is approximately three to four times higher than the number of trucks. Since the station data is hourly, its mean per 15 minutes was calculated for the date and time of the respective Sentinel-2 acquisition. It has to be kept in mind that the validation is yet limited to Germany and that an area-specific uncertainty is introduced by potential haze during large-scale processing.
Fig. 3: Trucks detected by Sentinel-2 vs. station counts [%]