Earth Observation & Damage Assessment

Damage Assessment

Damage Assessment is the process for determining the nature and extent of the loss, suffering, and/or harm to the community resulting from a natural, accidental or human-caused disaster.  - Clermont County, Ohio EMA 

Damage assessments provide information on:

  • Type, scope and severity of the event
  • Impact on individuals and communities
    • People injured, killed, or displaced
    • Degree and cost of structural loss and damage Additional resource needs
    • destruction of natural resources
  • Additional resource needs
  • Justification for disaster declaration
  • Future hazard mitigation projects

Who is involved in damage assessments:

  • Local officials
  • National/ supranational governing bodies
  • Damage assessment departments or emergency management agencies
  • NGOs
  • Research institutions

Spotlight: disaster tracking and damage assessment by the United Nations:

The United Nation's Office for the Coordination of Humanitarian Affairs (OCHA) maintains a database on current ongoing disasters and disaster alerts via reliefweb.

A conglomeration of multiple damage assessments from around the world, OCHA also published a Global Natural Disaster Report on reliefweb that analyzes the characteristics of global extreme weather disasters from 2000 to 2021.

The United Nations Office for Disaster Risk Reduction (UNDRR) also compiles information on disasters and damage assessments through their Sendai Framework Monitor which tracks the collection, documentation and analysis of data about losses caused by disasters associated with natural hazards around the globe.


Damage Assessment & Earth Observation

The most crucial aspect of damage assessments is that they need to be conducted as soon as possible after, or even during, a disaster. Thats is where earth observation comes in. Earth observation has increasingly been used in damage assessment in the past couple of decades in order to better address the time sensitive nature of damage assessments as well as to increase the coverage of these assessments in terms of both time and area.

Damage assessments can be called for in the event of many different types of natural disasters. Source:  http://areti-aroundtheworldinenglish.blogspot.com/2014/05/natural-disasters-crossword-puzzle.html 

Steps for conducting a damage assessment using earth Observation

  1. Pre-processing data (get rid of clouds, etc.)
  2. Computing vegetation indices and spectral response of non-photosynthetic vegetation
  3. Performing zonal statistics between NDVI changes, land cover types, and precipitation amounts
  4. Mapping changes
  5. Performing statistical examination of factors related to the damage
  6. Validation through accuracy assessments, independent damage assessments, ground surveys, visual assessment

Data & Platforms

  • Ground survey
  • Aerial & satellite imagery
  • SAR
  • Google Earth Engine
  • CLASlite
  • ArcGIS Damage Assessment solution
  • Machine learning (CNN, SVM, Random forests)


GEO-CAN

The Global Earth Observation Catastrophe Assessment Network (GEO-CAN) was formed in response to the 2010 earthquake in Haiti to assess building damage cause by the earthquake. After the earthquake occurred, various entities banded together to try and conduct a damage assessment. These included the Haitian government, World bank, European Commission, and United Nations Institute for Training and Research. Due to a lack of institutional capacity and available qualified professionals on the ground this group of institutions turned to the online community of experts.

GEO-CAN was created as a consortium of scientific and engineering volunteers from all over the world which at its height had over 600 individuals involved in the effort. To help with the damage assessment different companies and organizations including DigitalGlobal, GeoEye, Google, NOAA, and Pictometry provided high-resolution imagery free of charge.  Most of the volunteers had some background in geoinformatics but they were also specifically trained on how to identify damage caused by earthquakes from imagery as, in Haiti, damage patterns had specific characteristics depending on construction type, neighborhood, building density, and terrain. All of the actual damage assessment was done remotely by volunteers.

Post and pre-earthquake imagery of the National Palace. Source:  https://journals.sagepub.com/doi/10.1193/1.3636416

Grades for damage levels adapted from the 1998 European Macroseismic scale. source:  https://emergency.copernicus.eu/mapping/ems/damage-assessment 

The assessment was composed of a three-phase methodology which used high resolution satellite imagery at 50cm resolution and very high-resolution aerial imagery at 15cm resolution.  In the later stages of the assessment, some LiDAR data was also used. The volunteers used the 1998 version of the European Macroseismic Scale to assign different damage grades to buildings. The scale defines 5 damage grades of increasing devastation with 1 being negligible damage and 5 being complete destruction.

Phase 1

Volunteers used high-resolution satellite imagery to pinpoint completely destroyed buildings

Phase 2

Volunteers used very high-resolution aerial imagery to identify buildings with damage grades 4 (heavily damaged) and 5 (completely destroyed) and delineated the footprints of these buildings before and after the earthquake in order to quantify the amount of building area that needed repairs. The also employed statistical models to extrapolate the results of the damage surveys to lower levels

Phase 3

Volunteers used

very high-resolution aerial imagery to identify areas of liquefaction along coastal areas

Source:  https://journals.sagepub.com/doi/10.1193/1.3636416

After the initial phases of identifying damage, GEO-CAN had teams from different organizations validate their results through independent damage assessments and field surveys. These independent assessments found that the GEO-CAN volunteers were pretty accurate in terms of finding completely destroyed buildings but were only able to identify about half of the grade 4 (heavily damaged) buildings that other teams found as part of their assessments and ground surveys.

In the end, while the volunteers were able to accurately identify completely destroyed buildings, they often struggled to identify buildings that suffered less damage because the presence of shadows and the resolution of the imagery made it hard to detect damage. It was also hard for the volunteers to identify damage to buildings that were close to each other or had complex configurations. Even though their work clearly had some limitation, GEO-CAN clearly illustrates the value of remote sensing to damage assessment. Through crowdsourcing and the use of earth observation data and methods, a large group of analysts were able to assist in the earthquake damage assessment effort without putting any strain on Haiti's resources.

Since the GEO-CAN project there have been many improvements in the methods and data sources used to conduct damage assessments, but we see the legacy of GEO-CAN's use of crowdsourcing as a means to address natural disasters alive and well in the various challenges and hackathons proliferating across the world today. Increasingly, with improved data quality and quantity researchers have been employing indices as a means to conduct damage assessments for natural disaster events.


Indices

Haiti NDVI and EVI pre-earthquake from 2009 - 2010 using MODIS Terra Vegetation Indices 16-Day Global 1km

Damage to the natural environment from natural disasters can be assessed through comparing different index values from before and after disaster events. In general, there are two types of indices, those that are directly derived from the bands of remotely sensed images and those that are not directly derived from the bands (these indices usually compare historic data with current index values).

Normalized difference indices, like NDVI, EVI, VCI, NDWI, are often used in damage assessments to assess the impact of natural disasters such as floods or earthquakes on the natural environment. These types of indices take the difference between two selected bands and normalize them by their sum to minimize the effects of illumination (like clouds or shadows) and enhance spectral features that are not visible initially. Recently, a new index has been developed to aid in the creation of more accurate damage assessments through better detection of the specific impact of natural disasters on vegetation.

DVDI

In 2018 a group of researchers developed a new index meant to better assess the impact of natural disasters on vegetation through the calculation of the vegetation loss between a narrow pre-event window and a narrow post-event window called the Disaster Vegetation Damage Index or DVDI.

Source: https://link.springer.com/article/10.1007/s13753-020-00305-7

The DVDI calculates the difference between the median Vegetation Condition Index (VCI) from before and from after a disaster event. VCI reflects the current vegetation condition in comparison with the historical condition of vegetation in a given area. The median VCI relies on calculations based on NDVI values.

Here NDVI is the NDVI value of a specific day of interest, NDVI med is the historical multi-year median value of the specific day, NDVI min is the historical minimum value of the specific day, and NDVI max is the historic maximum value of the specific day. Source: https://link.springer.com/article/10.1007/s13753-020-00305-7

A median VCI value greater than 0 indicates the vegetation growth condition is better than the historical normal whereas a median VCI value less than 0 indicates the vegetation growth condition is worse than the normal. The researcher chose to use the median value in order to reduce the impact of noise in the data. The complete process for calculating DVDI is as follows:

DVDI calculation process. Source: https://link.springer.com/article/10.1007/s13753-020-00305-7

To calculate the DVDI it is necessary to calculate the mVCI from before the event as well as the mVCI after the event. The choice of pre- and post-disaster window can have a big impact on the results of the calculations, so it needs to be chosen carefully based on the situation.

DVDI is essentially the difference between pre and post event mVCI which relies on NDVI values. As a result, DVDI inherits the NDVI value range of values from -1 t0 1. A positive DVDI value indicates there is no damage whereas a negative value indicates considerable damage.

 Instead of comparing the current NDVI value of an area affected by a disaster to the temporal curve of NDVI (data which is not always available at the time of disaster) of said area as is traditionally done to assess vegetation health, DVDI only considers data from a narrow window before and after a disaster event which is meant to allow users to more accurately identify the specific impact of a disaster on vegetation. Moreover, because DVDI only considers vegetation change immediately before and after an event this reduces the impact of other natural causes of changes in the vegetation profile which helps to isolate and pinpoint the effects of a disaster.

Case Study

As DVDI is a relatively new index, there are only a handful of studies using the index and most employ the same processes and dataset. However, one recent study did challenge the usefulness of DVDI. To test the validity of DVDI, a group of researchers conducted an assessment that compared different vegetation indices (DNDVI, ΔEVI, and DVDI) as well as machine learning models to conduct a post-typhoon forest damage assessment using the 2016 typhoon that struck Hokkaido, Japan as a case study.

Use of different indices to estimate typhoon damage in Hokkaido, Japan. Source:  https://www.sciencedirect.com/science/article/pii/S2212094722000731 

They found that DNDVI was the most accurate index for determining forest damage, though all of the indices tested produced comparable results. In their case, DVDI overestimated damaged areas and showed a lot of pixel noise on the classified images. As opposed to other studies which mostly relied on MODIS data at 250m resolution, this study used higher resolution Landsat imagery at 30m resolution. Moreover, MODIS data covers a greater number of years and has a quicker revisit schedule than Landsat which increases the likelihood of obtaining cloud-free images within the desired window. All of this indicates that the data sources for DVDI must be chosen carefully in order to maximize its affects.

Drawbacks of DVDI

  • Cloud cover can have adverse effects, sensitive to noise
  • Pre/post event window has to be chosen carefully to account for natural changes in vegetation
  • Overestimates damage with higher resolution data
  • Needs a lot of historic data for comparison
  • Needs short revisit schedule

Future Applications

While the use of ground surveys to assess building damage and the use of indices for conducting damage assessments related to the natural earth remain popular, recently there has been increased interest in automation and deep learning methods for conducting damage assessments using earth observation methods and data. For more information explore the papers below.

Ecuador

Ecuador. Click to expand.

A validated geospatial model approach for monitoring progress of the Sendai Framework: The example of people affected in agriculture due to flooding in Ecuador - ScienceDirect

Indonesia

Indonesia. Click to expand.

Sensors | Free Full-Text | Damage Assessment in Rural Environments Following Natural Disasters Using Multi-Sensor Remote Sensing Data (mdpi.com)

Italy

Italy. Click to expand.

Earthquake damage mapping: An overall assessment of ground surveys and VHR image change detection after L'Aquila 2009 earthquake - ScienceDirect

Japan

Japan. Click to expand.

(PDF) Applications of remote sensing and GIS for damage assessment (researchgate.net)

Japan

Japan. Click to expand.

Post-typhoon forest damage estimation using multiple vegetation indices and machine learning models - ScienceDirect

Mexico

Mexico. Click to expand.

Assessment of tropical cyclone damage on dry forests using multispectral remote sensing: The case of Baja California Sur, Mexico - ScienceDirect

Nigeria

Nigeria. Click to expand.

(PDF) Earth Observation-based Damage Assessment of 2018 Flood in Parts of Hadejia-Jama' are River Basin, Nigeria (researchgate.net)

United States

United States. Click to expand.

Evaluating the potential of LiDAR data for fire damage assessment: A radiative transfer model approach - ScienceDirect

United States of America

United States of America. Click to expand.

Remote Sensing Based Rapid Assessment of Flood Crop Damage Using Novel Disaster Vegetation Damage Index (DVDI) | SpringerLink

United States of America

United States of America. Click to expand.

Remote sensing-based comparative damage assessment of historical storms and hurricanes in Northwestern Florida - ScienceDirect

Another great resource for exploring earth observation for damage assessment is NASA's ARSET program which provides free trainings on earth observation for disaster management.

Damage assessments can be called for in the event of many different types of natural disasters. Source:  http://areti-aroundtheworldinenglish.blogspot.com/2014/05/natural-disasters-crossword-puzzle.html 

Post and pre-earthquake imagery of the National Palace. Source:  https://journals.sagepub.com/doi/10.1193/1.3636416

Grades for damage levels adapted from the 1998 European Macroseismic scale. source:  https://emergency.copernicus.eu/mapping/ems/damage-assessment 

Source:  https://journals.sagepub.com/doi/10.1193/1.3636416

Haiti NDVI and EVI pre-earthquake from 2009 - 2010 using MODIS Terra Vegetation Indices 16-Day Global 1km

Source: https://link.springer.com/article/10.1007/s13753-020-00305-7

Here NDVI is the NDVI value of a specific day of interest, NDVI med is the historical multi-year median value of the specific day, NDVI min is the historical minimum value of the specific day, and NDVI max is the historic maximum value of the specific day. Source: https://link.springer.com/article/10.1007/s13753-020-00305-7

DVDI calculation process. Source: https://link.springer.com/article/10.1007/s13753-020-00305-7