Kenya Deforestation Analysis

About this Project

Urban GIS was contracted by Northwestern University to assess the scale of deforestation in Kenya from Landsat Imagery for the years 1973 and 2019.  In the course of the project, the team processed nearly 20TB of satellite imagery to generate annual cloud-free, spectrally consistent image mosaics as well as supporting the annual prediction of forest cover at 30m resolution for the entire country.

Each Landsat sensors measure solar reflectance slightly differently due to differences in the sensitivity of wavelengths of light for each band. The differences in spectral sensitivity complicate the effort to create composites from imagery across platforms. A Tasseled Cap Transformation (TCT) was employed to re-project the spectral detail into a single common image. The basis for our methodological approach is the LandsatLinkr project. However, the project is mostly defunct and significant changes/modifications/enhancements were required to adapt the project for our purposes. Modifications were made to both address bugs and errors, as well as to adapt certain aspects of the process to our revised methodological approach (e.g., using published TCT coefficients as instead of estimating the parameters from a regression model.

We performed a two-step random tree classification, initially training the model using Hanson’s (2000) predictions of global forest cover in 2000.  Subsequently, we refine the selected training data such that the predicted forest cover is within 1% of that predicted.  The resultant classification is repeated for each image acquisition and aggregated over the dry season to compute a maximum annual forest cover.