Sneak preview of ongoing research work in WP#4
Spatial analyses of cities require an extensive knowledge of the prevailing land cover. The practice of remote sensing allows researchers to record up-to-date datasets of land cover by carrying out a semi-automated land cover classification. In March 2020, the WP4 local research team from the Royal University of Phnom Penh surveyed and verified more than 600 “ground-truth points” in Phnom Penh using a mobile phone mapping app called “Input APP” which are used as training and validation data for the very high resolution land cover classification of Phnom Penh. The map below shows the first preprint of our very high resolution classification results based on Rapid Eye satellite date provided by Planet.
WP4 applies a machine learning classification approach using a Support Vector Machine (SVM) algorithm on a satellite image with a 3x3m resolution (Basic Scene of February 2020) provided by PlanetScope and integrated ground truth training data into the classification algorithm to enhance the classification performance. We selected eight separate classes including built-up, sealed surfaces, dense urban classification, low urban classification, bare soil, burned crops/ dry vegetation and shadow.This pre-printed map is a work in progress and underlies further refinement iterations based on the classification results achieved in order to train the model further towards the reality on the ground.