Skip to main content
ORNL DAAC HomeNASA Home
DAAC Home > Get Data > Science Themes > Vegetation and Forests > User guide

Urban Land Cover Maps for Mekelle, Ethiopia and Polokwane, South Africa, 2020

Get Data

Documentation Revision Date: 2025-06-20

Dataset Version: 1

Summary

This dataset consists of very high resolution urban land cover maps for two African cities, Mekelle, Ethiopia and Polokwane, South Africa for 2020. Maps were generated from Planet SuperDove satellite imagery at 3.125-m spatial resolution, and Worldview-3 satellite imagery (Maxar Techologies) at two spatial resolutions, 2 m for multispectral imagery and 0.5-m spatial resolution for pansharpened imagery. An object-based image classification approach was used to produce a multi-class land cover product for each image source. The aim of this work was to support fine scale urban land cover analyses and comparative assessments between different high resolution satellite imagery sources. The data are provided in shapefile format.

There are six shapefiles (.shp) with this dataset provided in six compressed .zip files.

Figure 1. Predicted land cover generated from three very high resolution maps for Mekelle, Ethiopia, in 2020. The inset shows a more detailed depiction of land cover produced from Planet SuperDove imagery with spatial resolution of 3.125 meters.

Citation

Cardenas-Ritzert, O.S.E., S. Shah Heydari, D.T. Rode, and J. Vogeler. 2025. Urban Land Cover Maps for Mekelle, Ethiopia and Polokwane, South Africa, 2020. ORNL DAAC, Oak Ridge, Tennessee, USA. https://doi.org/10.3334/ORNLDAAC/2413

Table of Contents

  1. Dataset Overview
  2. Data Characteristics
  3. Application and Derivation
  4. Quality Assessment
  5. Data Acquisition, Materials, and Methods
  6. Data Access
  7. References

Dataset Overview

This dataset consists of very high resolution urban land cover maps for two African cities, Mekelle, Ethiopia and Polokwane, South Africa for 2020. Maps were generated from Planet SuperDove satellite imagery at 3.125-m spatial resolution, and Worldview-3 satellite imagery (Maxar Techologies) at two spatial resolutions, 2 m for multispectral imagery and 0.5-m spatial resolution for pansharpened imagery. An object-based image classification approach was used to produce a multi-class land cover product for each image source. The aim of this work was to support fine scale urban land cover analyses and comparative assessments between different high resolution satellite imagery sources. 

Related Publications:

Cardenas-Ritzert, O.S.E., S. Shah Heydari, D. Rode, S.K., Filippelli, M. Laituri, M.R. McHale, and J.C. Vogeler. 2025. A comparison of urban green and open space land cover characterizations among high and very-high resolution satellite imagery: a case study in Mekelle, Ethiopia and Polokwane, South Africa. In review at Frontiers in Remote Sensing.

Cardenas-Ritzert, O.S.E., J.C. Vogeler, S. Shah Heydari, P.A. Fekety, M. Laituri, and M. McHale. 2024. Automated geospatial approach for assessing SDG Indicator 11.3. 1: a multi-level evaluation of urban land use expansion across Africa. ISPRS International Journal of Geo-Information 13:226. https://doi.org/10.3390/ijgi13070226

Acknowledgement:

This research was supported by NASA's Land Cover and Land Use Change Program (grant 80NSSC21K0313).

Data Characteristics

Spatial Coverage: Mekelle, Ethiopia and Polokwane, South Africa

Temporal Coverage: 2020-01-01 to 2020-12-31

Temporal Resolution: one time estimates

Study Areas (All latitudes and longitudes are given in decimal degrees) 

Focal Cities Northernmost Extent Westernmost Extent Easternmost Extent Southernmost Extent
Mekelle, Ethiopia 13.569 39.416 39.571 13.433
Polokwane, South Africa -23.768 29.315 29.522 -23.940

 Data File Information

There are six shapefiles (.shp) with this dataset provided in six compressed .zip files. The data were generated from Planet SuperDove satellite imagery at 3.125-m spatial resolution, Worldview-3 satellite imagery at two spatial resolutions, 2 m for multispectral imagery and 0.5-m spatial resolution for pansharpened imagery.

The file naming convention is <source>_<site>_2020.zip, where <source> is ...

  • <source> = site imagery source: "Planet" for Planet SuperDove,  "MaxarMulti" for multispectral imagery from Worldview-3 (Maxar), and "MaxarPan" for pansharpened imagery from Worldview-3 (Maxar)
  • <site> = focal city: "Mekelle" or "Polokwane"

File names:

  • Planet_Polokwane_2020.zip
  • Planet_Mekelle_2020.zip
  • MaxarMulti_Polokwane_2020.zip
  • MaxarMulti_Mekelle_2020.zip
  • MaxarPan_Polokwane_2020.zip
  • MaxarPan_Mekelle_2020.zip

The shapefiles hold polygons projected into the Africa Albers Equal Area Conic coordinate system (ESRI: 102022; proj4 = "+proj=aea +lat_0=0 +lon_0=25 +lat_1=20 +lat_2=-23 +x_0=0 +y_0=0 +datum=WGS84 +units=m +no_defs").

Table 1.Predicted land cover classes as attributes of polygons in shapefiles.

Value Description
1 Barren
4 Short Vegetation
5 Tall Vegetation
6 Water
8 Shadow
23 Impervious Surface

Application and Derivation

The aim of this work was to develop very high resolution (VHR) urban land cover (LC) maps in two rapidly urbanizing secondary cities to support fine scale urban LC analyses and comparative assessments between different VHR satellite imagery sources. The varying spatial resolutions, availability, quality, and cost of VHR satellite imagery influences the potential LC products that are created to support LC related assessments; therefore, understanding the advantages and disadvantages of each source can aid more appropriate image selection and product creation. Additionally, the LC map outputs add to the available information to be used for monitoring land cover change, assessing the recent distribution and configuration of important urban LC features such as green spaces, and informing land cover monitoring, management, and planning efforts at multiple levels.

The map outputs and subsequent analyses demonstrate the utility of different VHR imagery sources for LC applications. Utilizing the presented maps and a separate coarser resolution map, Cardenas-Ritzert et al. (2025) determined that the urban landscapes of Mekelle and Polokwane in 2020 were comprised primarily of undeveloped space. A major portion of the undeveloped space was dedicated for public use, but the share of land that was allocated to streets was low compared to the suggested share recommended by the United Nations (UN-Habitat, 2018). Considering the importance of streets in urban areas (Wood and Esaian, 2020), the findings may support future management efforts to improve the development and management of streets, as well as public open spaces.

Additionally, Cardenas-Ritzert et al. (2025) compared differences in the land cover characteristics of the two study cities derived from high and VHR imagery sources. The main focus was to quantify differences in the classifications of undeveloped space versus developed space and tall vegetation, the SDG Indicator 11.7.1 metric, and land cover configuration within sample public spaces between the different high and VHR imagery sources. The analyses revealed that coarser evaluations, including the comparison of undeveloped versus developed land cover, remained less affected by spatial resolution, but the higher image spatial resolution improved the detection of tall vegetation land cover and better dissected LC at finer scales. Additionally, SDG Indicator 11.7.1 and related metric values were less impacted by the spatial resolution of data at very-high resolutions (<=3 m). Land cover configuration did change starkly between VHR LC products, indicating the importance of utilizing highest resolutions for LC configuration analyses at very fine scales.

Quality Assessment

Model accuracies were derived from an area-adjusted confusion matrix in which the calculated model accuracies were adjusted for by polygon area. The area-adjusted approach is more appropriate for object-based analyses (MacLean and Congalton, 2012; Radoux et al., 2011). The best model outputs for Planet Superdove, Worldview-3 multispectral, and Worldview-3 pansharpened imagery in Mekelle exhibited an overall map accuracy of approximately 76%, 92%, and 87%, respectively, and 75%, 86%, and 68% in Polokwane.

Data Acquisition, Materials, and Methods

Very-high resolution (VHR) urban land cover (LC) maps were generated for two African cities, Mekelle, Ethiopia and Polokwane, South Africa for 2020. Data were from  Planet SuperDove satellite imagery at 3.125-m spatial resolution, and Worldview-3 satellite imagery (Maxar Techologies) at two spatial resolutions, 2 m for multispectral imagery and 0.5-m spatial resolution for pansharpened imagery. An object-based image classification approach was used to produce a multi-class land cover product for each image source. 

The initial processing step entailed segmenting VHR imagery into polygons and calculating zonal statistics, including number of pixels, mean, median, maximum, majority, minority, variety, and variance for each spectral band within each polygon. Validation data were then randomly extracted from the segmented dataset. The remaining polygons were classified as training data and incrementally increased until no model improvements were detected.

Subsequently, manual interpretations were carried out to assign a land cover label to each training polygon. Multiple image variants, including natural color, false color, and NDVI imagery, supported manual interpretations. Through a series of simulations, the best feature set was selected based on average and variance of obtained map accuracies, with correlated features or bands dropped, and Random Forest classifier models were trained and applied to create the LC maps.

Data Access

These data are available through the Oak Ridge National Laboratory (ORNL) Distributed Active Archive Center (DAAC).

Urban Land Cover Maps for Mekelle, Ethiopia and Polokwane, South Africa, 2020

Contact for Data Center Access Information:

References

Cardenas-Ritzert, O.S.E., S. Shah Heydari, D. Rode, S.K., Filippelli, M. Laituri, M.R. McHale, and J.C. Vogeler. 2025. A comparison of urban green and open space land cover characterizations among high and very-high resolution satellite imagery: a case study in Mekelle, Ethiopia and Polokwane, South Africa. In review at Frontiers in Remote Sensing.

Cardenas-Ritzert, O.S.E., J.C. Vogeler, S. Shah Heydari, P.A. Fekety, M. Laituri, and M. McHale. 2024. Automated geospatial approach for assessing SDG Indicator 11.3. 1: a multi-level evaluation of urban land use expansion across Africa. ISPRS International Journal of Geo-Information 13:226. https://doi.org/10.3390/ijgi13070226

MacLean, M., and R. Congalton. 2012. Map accuracy assessment issues when using an object-oriented approach. American Society for Photogrammetry and Remote Sensing Annual Conference. 2012. ASPRS 2012. 369-373. https://www.asprs.org/a/publications/proceedings/Sacramento2012/files/MacLean.pdf

Maxar Technologies, WorldView-3. https://resources.maxar.com/data-sheets/worldview-3

Radoux, J., P. Bogaert, D. Fasbender, and P. Defourny. 2011. Thematic accuracy assessment of geographic object-based image classification. International Journal of Geographical Information Science 25: 895-911. https://doi.org/10.1080/13658816.2010.498378

UN-Habitat. 2018. Annual Progress Report 2018. United Nations Human Settlements Programme (UN-Habitat); Nairobi, Kenya. https://unhabitat.org/annual-progress-report-2018

Wood, E.M., and S. Esaian. 2020. The importance of street trees to urban avifauna. Ecological Applications 30:e02149. https://doi.org/10.1002/eap.2149