Urbanization in Astana, Kazakhstan (1994-2017)
Named the capital of a new nation, the city grew from about 303,000 people in 1994 to 1.7 million in 2017.
Named the capital of a new nation, the city grew from about 303,000 people in 1994 to 1.7 million in 2017.
Kazakhstan was declared independent from the Soviet Union in 1991. Since 1929, the capital of the republic had been Almaty. The capital was moved to Astana in 1997. A planned city, it was chosen due to strong economic growth potential, demographics, and its central location in the country. ("The History of Astana." Wayback Machine. January 19, 2013. Accessed December 01, 2020. https://web.archive.org/web/20141007033517/http://astana.gov.kz/en/modules/material/42 )
The city is at Latitude 51 09’ 37” N, Longitude 071 28’ 13” E, on the Central Eurasian steppe and is bisected by the Ishim River.
According to the Embassy of Kazakhstan in Washington, D.C., “In 1998, Nur-Sultan’s area was over 300 sq. km., but today, the city boundaries are rapidly expanding. Nur-Sultan’s population in 1998 was about 300,000. The population now exceeds 1.1 million and growing.” Further, “Between 2000 and 2004, the number of new legal entities registered in the capital more than doubled, and the local housing stock nearly tripled in size.” ("Nur-Sultan." Embassy of Kazakhstan in Washington, D.C. Accessed December 01, 2020. https://kazakhembus.com/about-kazakhstan/geography/nur-sultan.)
In this project, I used image differencing, post-classification change detection, and principal component analysis (PCA) to analyze remotely sensed satellite images showing the expansion of the city from 1994 to 2017. These are Landsat Ecosystem Disturbance Adaptive Processing System (LEDAPS) corrected images from Landsat 6 and 7 (LEDAPS is an algorithm that corrects for atmospheric anomalies). ("LEDAPS Calibration, Reflectance, Atmospheric Correction Preprocessing Code, Version 2." ORNL DAAC. October 11, 2013. Accessed December 01, 2020. https://daac.ornl.gov/MODELS/guides/LEDAPS_V2.html.)
Several methods of image change analysis are shown in order to empathize how the attributes of a study area influence the most useful analysis methods. Image differencing and PCA show a more accurate picture of the growth. Classification methods proved unable to differentiate well between vegetation and urban growth.
1994 color composite of Astana (Bands 1, 2, 3).
This first image shows the baseline urbanization of the study area. Bright white pixels represent human development surrounded by patches of brown and green (open land and agriculture).
2007 color composite of Astana (Bands 1, 2, 3).
This second images shows urban expansion to the southwest of the river. Note as well the body of water to the southwest.
2017 color composite of Astana (Bands 1, 2, 3).
Further marked expansion of development in is shown in this recent image, including to the northeast.
1994-2017 comparison.
The first image shows the extent of the city in 1994, while sliding the bar to the right depicts the city in 2017.
Screen grab of TerrSet ImageDiff output comparing 1994 and 2017.
Image Differencing 1994-2017 (exported to ArcGIS)
These images show the image differencing analysis performed on band 3 for 1994 and 2017 using the Image Differencing TerrSet module. This module provides a visualized model of change by highlighting the regions that are +- 1, 2, or more standard deviations from the mean. The regions in red show an increase in urbanization as well as some areas outside the city that appear to be cropland.
Post-classification change detection is another way to analyze change in a study area over time. It outputs, in pixel values, statistical data about the changes in land use/cover over time. The first step is creating false color composites.
False Color Composites of 1994 (left) and 2017 (right). (Bands 2, 3, 4).
The spectral classes are most simply converted to informational classes using Mahalanobis Typicality (a method of soft classification, which shows probabilities of pixels belonging to each class beyond the mean). The classes are vegetation, water, and urban development. Digitized polygons of each class were created as training sites for the images of both years. Signatures were then created with the module Makesig and placed into a signature group file. These were used to classify the images with the MahalClass module. These outputs were hardened with minimum values and a legend of the informational classes added to create thefinal map.
The statistical output of comparison between 1994 and 2017.
The classification was difficult due to the high levels of vegetation within and outside the city. In addition, the water and urban classes were under-reported in the 2017 images and can be estimated to exist in areas with less vegetation (and with the assumption that the lake did not move). It is possible the increased vegetation reported in the cross-tabulation is accurate due to some unknown changes outside the city, but it is more likely to be due to classification error. More study of the region, perhaps in situ, would be necessary to determine if there is relevance to these results.
1994 (left) and 2017 classifications.
Another way to analyze changes over time is with principal component analysis (PCA). In the PCA forward t-mode, every band is a time variable. The output is a series of uncorrelated images with less and less of the variance as the band numbers get higher. The first PCA band reveals the most variance (and is oriented to do so). Later bands are perpendicular to the ones that came before and show remaining variance. PCA compresses information, as sometimes variables are mostly copies of one another. (Watkins, Thayer. Principal Component Analysis in Remote Sensing. Accessed December 01, 2020. https://www.sjsu.edu/faculty/watkins/princmp.htm.)
In TerrSet, PCA output consists of extensive module results, such as the variance/covariance matrix (relation and variability), the correlation matrix, eigenvalues (variance amount each component includes), eigenvalues as percentages, eigenvectors, and Loadings (connection between the input and output). (Warner, Timothy A., David J. Campagna, and Florencia Sangermano. Remote Sensing with TerrSet/IDRSI: A Beginner's Guide. Hong Kong: Geocarto International Centre Ltd., 2019.)
PCA 1994-2017
98.13% of the variance in the images are accounted for in components 1, 2 and 3. Later components are more noisy and less useful for visual analysis.
PCA false color composite (Bands 1, 2, 4)
The components of PCA analysis can be visualized in a false color composite. This is shown above. Per the advice of "Remote Sensing with TerrSet/IDRISI: A Beginner's Guide", component 3 is not used due to its being an image showing vegetation. Therefore, components 1, 2, and 4 are selected. (Component 4 was reversed using the Scalar module to show increased vegetation as bright.) The output shows growth in urbanization in the center surrounded by vegetation. Reddish purple represents vegetation, while green and cyan represents albedo (brightness), which increased. There was extensive development southeast of the urban core in this 23 year interval.
Through performing a variety of analysis methods on these satellite images, several things have become clear about the pattern of urbanization in Astana from 1994 to 2007 and 2007 to 2017. The urbanization spread in a typical ring pattern, but there was particular growth to the southeast of the urban core.
The cross-tabulation of the hardened Mahalanobis typicalities classifications for 1994 and 2017 output exact changes in pixels between the two dates. There was a decrease in vegetation, from 805,417 pixels in 1994 to 339,082 pixels in 2017. Despite the other methods of analysis showing urban growth, this method does not. I hypothesize this is due to classification error. Astana has an abundance of vegetation within and around it, which may have made differentiating between urban areas and vegetated areas difficult.
In closing, this project allowed me to explore several methods of land use change over time and examine the strengths and weaknesses. Post classification change detection is the most tricky and had the most errors, but it is a powerful option for exploring probabilities. Image differencing was very useful and illuminating because it showed direct change in the pixels. Principal component analysis is advantageous for examining the different variables in a study area's satellite imagery.
According to Macro Trends, Astana's population growth increased from 1.71% in 1993 to a high of 9.66% growth in 2003, dropped to 3.46% in 2005 and leveled off at about 6% from 2011 to 2017. (Macro Trends. Astana, Kazakhstan Metro Area Population 1950-2020. Retrieved December 01, 2020, from https://www.macrotrends.net/cities/21679/astana/population)
The fast population (and thus housing) growth at the turn of the century has continued to an extent into the present and is shown in a variety of ways through the remote sensing analysis done in this project.