Graduate Project


Local Climate Zone Classification Using Random Forests Public Deposited

Downloadable Content

Download PDF


Attribute NameValues
  • Objective: The goal of this project is to recreate aspects of the article “Comparison between convolutional neural networks and random forest for local climate zone classification in mega urban areas using Landsat images” (Yoo et al., 2019), where methods for predicting LCZ classes for four large cities throughout the world were compared. To do so, a small training dataset from the 2017 Institute of Electrical and Electronics Engineers (IEEE) Geoscience and Remote Sensing Society (GRSS) Data Fusion Contest (Tuia et al., 2017) was used as ground truth for LCZ classes. This was combined with Landsat 8 satellite input data to create a series of models, which were then compared to a larger, full LCZ layer for each city and assessed for accuracy. The primary types of models considered in the Yoo et al. (2019) work were random forests and convolutional neural networks. However, for this project the focus will be only on random forests. In addition, this investigation targets just Hong Kong. This city was chosen because each LCZ that is classified has at least 4 polygons. Finally, here, a classification scheme like the one used by the World Urban Database and Access Portal Tools (WUDAPT) project, denoted as Scheme 1 in Yoo et al. (2019) will be the focus, with comparisons between accuracy at different values of a tuning parameter. All code and higher resolution images for this project can be found on GitHub at
Resource Type
Date Issued
Degree Level
Degree Name
Degree Field
Degree Grantor
Commencement Year
Committee Member
Academic Affiliation
Rights Statement
Related Items
Peer Reviewed



This work has no parents.