When Crises Converge: Understanding the Impact of Floods on COVID-19 Infection Rates

Abstract

Climate change increases the likelihood and scale of environmental disasters, making the co-occurrence of disasters in the current context of COVID-19 highly probable. Large scale flood events, not only lead to socioeconomic losses but also cause people to congregate at evacuation centers, completely in contrast to the isolation requirements of the pandemic. While protecting vulnerable populations from catastrophic flood events requires the use of temporary shelters, they eventually contribute to increasing infection rates, since it is nearly impossible to maintain strict hygiene and social distancing regulations. Understanding the correlation between pandemic infection rates and flood inundation can help prepare the local health system in advance, thereby reducing fatalities caused by the lack of adequate healthcare during compound disasters.

Satellite-based Synthetic Aperture Radar (SAR) sensors provide weather and illumination independent images, with reasonable temporal frequency and spatial resolution, which are uniquely suited to mapping inundation. Studies have demonstrated the utility of SAR-images to map flood damage and connect these results with additional geodata, to understand the impact of floods on other areas of socioeconomic wellbeing.

This study investigates the correlation between people affected by floods and local COVID-19 cases in Bangladesh during Cyclone Amphan. Amphan was active in the North of the Bay of Bengal, between the 16th and 21st May in 2020, coinciding with the rising limb of the Covid-19 wave. The coastal area of Bangladesh was affected the most, leading to the evacuation of 2.6 million citizens into 12.000 temporary shelters.

Sentinel-1 (S1) SAR-data was used for a binary flood/non-flood classification, using a standard machine learning technique, Random Forest. Post-processing using topographic indicators based on Digital Elevation Models and Global Water Mask based on long-term Landsat optical water recurrence data was used to improve map accuracy. The S1-based flood maps resulted in overall accuracies of more than 90%, in comparison to optical Sentinel-2 based flood maps extracted using water indices. Multiple flood maps starting from one month before and until the end of the flood event were generated using all available data, to characterize the dynamic spatial evolution of the inundation extent over time. World population gridded datasets were subsequently used to estimate the affected citizens for each time step, as the spatial pattern of the flooding evolved.

The results show a lagged correlation between the affected people detected by the classification and the local COVID-19 cases, while infection rates increased by more than 70% on average. The trends of infections between coastal areas and inland regions, also display the impacts of the flood event, through differing trajectories. In future, predicted cyclone trajectories can be used in conjunction with shelter locations, to determine the expected rise in coincident viral infection numbers and better prepare the health systems for the upcoming onslaught, ultimately resulting in optimal resource utilization during emergencies.

Full text

When Crises Converge: Understanding the Impact of Floods on
COVID-19 Infection Rates
D2.11 Earth Observation for Health
Motivation
▪Climate change increases the likelihood and scale of
environmental disasters
▪In 2020, Cyclone Amphan affected over 2.6 Mio. people in
Bangladesh within the beginning of the COVID-19 pandemic
▪Crouding in evacuation center is counter-productive during
a pandemic
▪Combination of SAR-based flood detection and population
data provides information on COVID-19 infection rates
Tim Landwehr¹, Clara Lößl¹, Antara Dasgupta1,2, Björn Waske¹
References:
•Landwehr, T., Dasgupta, A., Lößl, C., and Waske, B. (2022): Understanding the Impact of Flood
Evacuations on Covid-19 Infection Rates. Natural Hazards and Earth System Sciences. (Expected
submission - July, 2022)
•IFRC (2021): Bangladesh: Cyclone Amphan –Final Report
Data Sources:
1 Ministry of Health and Family Welfare Bangladesh (2020):COVID-19 Dynamics Dashboard. Retrieved April 10, 2022 from https://dghs-dashboard.com/pages/covid19.php
2 Copernicus Emergency Management Service (n.d.): Flood in Bashan Char, Bangladesh. Retrieved April 15, 2022 from https://emergency.copernicus.eu/mapping/list-of-components/EMSR439
Data Summary
▪Sentinel-1 (VV, Descending) at 10 m for binary flood
classification validated using Copernicus emergency maps
▪Global Surface Water Mapping Layer (2020) for masking
out open water bodies with water > 9 months
▪HydroSHEDS (2000) topographic information for masking out
slope areas > 5°
▪WorldPop (2020) residential population at 100 m grid for
estimate exposed people by floods
▪COVID-19 infection rolling 7-day average by the Ministry of
Health and Family Welfare (2020)1
Conclusion
▪Highest correlation between exposed people and COVID-19
infection rates observed for a lag of 13 to 15 days
▪Flooded area and exposed people increase during the event and
persists thereafter
▪Stronger surge of COVID-19 cases caused by flood evacuations as
crowding encourages infections
▪Results generalizable to all natural disaster and infectious outbreaks
Sentinel-1
Mosaic
Binary Random
Forest Flood
Classification
Post-
Processing by
Slope &
Permanent
Water Bodies
Exposed
People by
Flood
Estimation
COVID-19
Correlation
Static No
Flood
Reference
Training &
Hyperparameter
Tuning Validation via
Emergency
Maps2
¹ Remote Sensing Working Group, Institute of Informatics, University of Osnabrück, Germany (timlandwehr@uos.de)
² Water Group, Department of Civil Engineering, Monash University, Australia
Fig. 5: COVID-19 Cases for Flood and No-Flood
Reference
Fig. 7: Impact of Day Shift for Correlation between
Exposed People and COVID-19 Cases
Fig. 4: Pre-Event (04- & 16-May), During-Event (22-May)
and Post-Event (28-May, 09- & 21-June) Estimation
Fig. 3: Flood Estimation for 22-May
Q2: Does a correlation between exposed people and
COVID-19 infection exist?
Fig. 6: 14-Day Shifts of COVID-19 Seven-Day Average
Workflow
Fig. 2: Performed Workflow for Flood Detection
Fig. 1: Study Area of Bangladesh including the Divisions Barisal and Sylhet
Study Area
Q1: Does SAR enable measuring flood impact over time?
For all dates:
F1 Score > 0.97