This work introduces a novel solution to measure economic activity through remote sensing for a wide range of spatial areas. We hypothesized that disturbances in human behavior caused by major life-changing events leave signatures in satellite imagery that allows devising relevant image-based indicators to estimate their impacts and support decision-makers. We present a case study for the COVID-19 coronavirus outbreak, which imposed severe mobility restrictions and caused worldwide disruptions, using flying airplane detection around the 30 busiest airports in Europe to quantify and analyze the lockdown's effects and post-lockdown recovery. Our solution won the Rapid Action Coronavirus Earth observation (RACE) upscaling challenge, sponsored by the European Space Agency and the European Commission, and now integrates the RACE dashboard. This platform combines satellite data and artificial intelligence to promote a progressive and safe reopening of essential activities. [arXiv] [DOI]
- Best contribution (super-prize) -- RACE upscaling challenge
- Mauricio Pamplona Segundo (USF)
- Allan Pinto (UNICAMP)
- Rodrigo Minetto (UTFPR)
- Ricardo Da Silva Torres (NTNU)
- Sudeep Sarkar (USF)
There are three ways of obtaining the data employed in this research. The first one is through a Sentinel Hub account by using the script download.py . Make sure you update the following variables with your Sentinel Hub credentials:
config.instance_id = '***REMOVED***'
config.sh_client_id = '***REMOVED***'
config.sh_client_secret = '***REMOVED***'These credentials can be obtained in the Sentinel Hub dashboard.
The second option is to download the data from IEEE Dataport.
The third option is to download the data from our institutional repository.
The last two options consist of one tar/zip file for all cloud probability masks from all airports (cloud_masks.zip (11.8 GB)) and one tar/zip file per airport with color images and valid pixel masks as follows:
| file (size) | file (size) | file (size) | file (size) | file (size) |
|---|---|---|---|---|
| AGP.zip (22.3 GB) | CDG.zip (14.5 GB) | HEL.zip (12.2 GB) | MAD.zip (33.4 GB) | PMI.zip (19.4 GB) |
| AMS.zip (10.2 GB) | CPH.zip (13.0 GB) | IST.zip (18.2 GB) | MAN.zip (7.4 GB) | STN.zip (14.7 GB) |
| ARN.zip (16.2 GB) | DUB.zip (9.0 GB) | LGW.zip (14.2 GB) | MUC.zip (18.3 GB) | TXL.zip (15.7 GB) |
| ATH.zip (17.5 GB) | DUS.zip (13.0 GB) | LHR.zip (11.1 GB) | MXP.zip (19.4 GB) | VIE.zip (20.0 GB) |
| BCN.zip (18.9 GB) | FCO.zip (20.6 GB) | LIS.zip (27.0 GB) | ORY.zip (14.7 GB) | WAW.zip (16.2 GB) |
| BRU.zip (14.2 GB) | FRA.zip (9.1 GB) | LTN.zip (9.8 GB) | OSL.zip (12.4 GB) | ZRH.zip (18.6 GB) |
Access to the institutional repository will be made available upon request to the e-mail mauriciop@usf.edu. Requesters will be granted one month of access to download the aforementioned files.
All the files in our dataset contain modified Sentinel-2 data processed by Euro Data Cube. If you use any of these images for any kind of media (publications, reports, videos, etc), please add the message "Contain modified Sentinel-2 data processed by Euro Data Cube" close to the image.
To download images from Sentinel Hub, run the following command:
$ mkdir images
$ python3 download.py
This command will download all images from the chosen airports within the specified time interval and save them in the folder images. If you decide to download the images from our institutional repository, just unzip all airport files inside the folder images (cloud masks are not necessary for training/inference).
To detect airplanes, run:
$ python3 inference.py
This code will use our pre-trained model models/flying41.pytorch. The log of detections will be saved in the folder log. These log files are then converted into time series through the following command:
$ python3 create_timeseries.py
The time series are saved in the folder timeseries. Both log and timeseries folders have pre-generated results.
If you decide to train your own model, run:
$ python3 train.py
This will use the annotations available in this repository to train a new detection model. Make sure you rename the model name on inference.py to use your own model for inference.
To perform the airports' activity analysis by detecting the breaking points related to COVID-19 and the recovery rate of the monitored airports, please follow the instructions available in covid19-airports-activity-analysis/README.md
If you find the code and data in this repository useful in your research, please consider citing:
@ARTICLE{9472964,
author={Pamplonasegundo, Mauricio and Pinto, Allan and Minetto, Rodrigo and Torres, Ricardo da S. and Sarkar, Sudeep},
journal={IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing},
title={Measuring economic activity from space: a case study using flying airplanes and COVID-19},
year={2021},
volume={},
number={},
pages={1-1},
doi={10.1109/JSTARS.2021.3094053}
}
@data{3mbt-tb11-21,
doi={10.21227/3mbt-tb11},
url={https://dx.doi.org/10.21227/3mbt-tb11},
author={Pamplona Segundo, Mauricio and Pinto, Allan and Minetto, Rodrigo and da Silva Torres, Ricardo and Sarkar, Sudeep},
publisher={IEEE Dataport},
title={A dataset for detecting flying airplanes on satellite images},
year={2021}
}