Spatial Changes of Pre- and Post-war City: Mariupol
Using Supervised classification - GEE-ArcGIS pro mixed method
Introduction
The Russian invasion in Ukraine started around February 2022. Kyiv city was the first city invaded, followed by Mariupol, Sumy, Kharkiv, and Kherson cities as of March 2022 (BBC News, 2023).
Three major periods and areas affected by the Russian invasion are represented in the map: [1] Stage 1: as of Feb. 2022; [2] Stage 2: as of October 2022; and [3] Stage 3: as of March 2023 (Source: BBC News, 2023).
As of February 2023, the Russian army stroke facilities including energy facilities and infrastructure stations in more than 24 administrative regions out of 27 in total. As of early December 2022, 50% of Ukraine’s energy infrastructure has been damaged or demolished (CNN News, 2023).
Recent data show that Mariupol city is one of the cities that is the most affected, with 90% of damaged areas consisting of residential area and industrial area (The economist, 2022).
In this story, the city of Mariupol is selected to identify the spatial changes between two distinct periods: [1] pre-war (July, 01-30th, 2021); and [2] post-war (July, 01-30th, 2022) periods. The classification of spatial changes includes seven categories, namely, urban, industrial, undeveloped lands, green/forest area, war-affected zones, agricultural fields (see figure 1 for details).
I. Research question Where are the areas affected by war and what are the major spatial changes?
II. Data used and methodology Satellite imagery data used is an API format provided by Google Earth Engine Platform. The satellite imageries provided are taken by the COPERNICUS satellite, and its surface reflectance data (level-2 data) is named as "Harmonized Sentinel-2 MSI: Multispectral Instrument, Level-2A" (the data provided are 10-30m resolution). The data is accessible " here ".
The overall methodology applied is represented in the graph below:
Figure 1: Methodology applied to identify spatial changes of Mariupol city (green/forest area, visible areas affected by war, etc.) using GEE supervised classification (learning) method ; Source: Author formulated
The supervised classification tool embedded in the GEE platform trains the classifier (using extracted sample) using classifier.train(). It classifies an Image or a "FeatureCollection" using "classify()" function. The algorithm embedded uses the "Classification and Regression Trees (CART) classifier" method (Breiman et al. 1984) to predict simple classes and areas identified under the "same" class (or category) (Google Earth Engine, n.d.).
III. Supervised classification Results
Figure 2: Result of Pre- (July 2021) and post- (July 2022) war spatial change identification using GEE-ArcGIS Pro mixed method with an accuracy rate of 88.25%. Source: Author formulated, 2023
Analysis of classification results: Zooming into the classiciation result of physical damaged zones (war-affected zones)
Figure 3: War affected zone: Traning sample #1
The first zoom-in region is illustrated in the image at the right frame. The area consists of an industrial region within a district named "Kalmiuskyi", which is categorized as an "urban district". This area is used to extract multiple training samples. The geometric boundaries of training samples are drawn in the GEE platform. Examples of geometric boundaries drawn are illustrated by red lines.
Figure 4: War affected zone: Traning sample #2
The second zoom-in region is an area located in "Livoberezhnyi" District, which is also categorized as an urban district. Similar to Kalmiuskyi district, the selected district also consists of industrial facilities (speculatively factories or energy facilities). This area is also used to extract small and large training samples.
Figure 5: War affected zone: Automated result #1
Area identified in figure 5 illustrated in the images on the right frame is an automated result generated from the training samples. The region is also an area located in the "Livoberezhnyi" District identified as an industrial area (speculatively identified as warehouses, factories, and facilities).
Figure 6: War affected zone: Automated result #2
Located in "Zhovtnevyi" district, the zoomed-in area in figure 6 is also an automated classification result from the GEE supervised classification method. Based on the image captured, one can speculate that the parcel is used for industrial purposes, such as logistics warehouse.
IV. Conclusion and discussion
The Google Earth Engine (GEE) provides useful tools to classify and identify spatial changes such as land use changes and war-affected zones in cities under risks. GEE is all the more useful in the sense that it permits to process surface reflectance data at any time from 2017 to the most recent period (such as the previous month). In addition, the GEE-ArcGIS pro mixed methodology helps to acquire data in remote and inaccessible regions such as Ukraine and process the data for further analysis such as spatial statistics and so forth.
The result of GEE classification enabled the identification of war-affected zones in Mariupol city, with an overall accuracy of 83%. However, due to limitation such as spatial resolution, subtle images such as war-affected single-detached housings are not identifiable from the provided (open) data, the Harmonized Sentinel-2 MSI: Multispectral Instrument, Level-2A data.
It has been also identified that after the war, green spaces (including forestry region) have substantially decreased, as illustrated in Figure 7 and 8.
Figure 7: Green spaces as of July 2021 and July 2022, excluding overlapping green spaces.
Figure 8. Aerial image capture of Mariupol city as of July 2021 and July 2022 using Harmonized Sentinel-2 MSI: Multispectral Instrument, Level-2A.
In sum, although limitation persists such as low spatial resolution that omits war-affected areas in residential areas, the GEE-ArcGIS pro mixed usage permits the analysis of spatial changes of cities under risks such as war. It enables professionals (such as planners and designers) to conduct the analysis of cities that were previously inaccessible and permit to ponder over the reconstruction and renewal of damaged cities. It also allows the consideration socio-economic policies such as investment plans in cities that were previously impacted by severe threats.
A. Technical notes - data sources
Note that the below list is the combination of data used for the analysis: [1] Surface reflectance data - Harmonized Sentinel-2 MSI: Multispectral Instrument, Level-2A [2] Urkaine administration boundary data - Diva-GIS (Accessible here ) [3] Spatial classification result data - Exported from GEE supervised classification result (cross validated with a result of 83% accuracy) [4] Screen capture data satellite imageries data - Images from Harmonized Sentinel-2 MSI: Multispectral Instrument, Level-2A and/or from Google Earth Pro
B. Reference
Breiman, L., Friedman, J., Stone, C. J., & Olshen, R. A. (1984). Classification and regression trees/Leo Breiman...[et al.]. Belmont, Calif: Wadsworth International Group, Belmont, Calif.
BBC News. (2023). Urkaine in maps: tracking the war with Russia. Accessible at: Ukraine in maps: Tracking the war with Russia - BBC News
CNN News. (2023). Russia has tried to rob the light from homes across Ukraine. Accessible at https://www.cnn.com/interactive/2023/02/europe/putin-ukraine-energy-infrastructure-attack/index.html
Google Earth Engine. (n.d.). Supervised Classification. Accessible at: Supervised Classification | Google Earth Engine | Google Developers
The economist. (2022). Nearly half of Mariupol has suffered grave damage. Accessible at: https://www.economist.com/graphic-detail/2022/04/23/nearly-half-of-mariupol-has-suffered-grave-damage
