Using Planet Data and Google Earth Engine to Automate Mapping of Informal Roads In The Amazon Borderlands.

Keywords: Amazonia, Google Earth Engine, Roads, Peru, Brazil, Machine Learning
Session Type: Virtual Poster Abstract
Day: Friday
Session Start / End Time: 2/25/2022 03:40 PM (Eastern Time (US & Canada)) - 2/25/2022 05:00 PM (Eastern Time (US & Canada))
Room: Virtual 39

Abstract

Formal and informal road building in the Amazon borderlands is increasing at a rapid rate. Here, we focus on informal exploratory roads - those that are being built in the remote rainforests of the southwestern Amazon to reach coveted resources and agricultural lands with weak government oversight. Previous research has shown that forest disturbances, including informal-road clearing, can be mapped through either hand-digitization or using medium resolution imagery (i.e. Landsat, Sentinel-2, 20-30m). Here, we highlight our efforts in using high-resolution Planet satellite imagery (5m) and cloud computing (Google Earth Engine) to facilitate the high-resolution, automated mapping of forest disturbances and roads. We employed a Random Forest machine learning algorithm to the novel freely available NICFI Planet dataset to map all forest disturbances. We are currently working with this classified image to extract road features. The results of this research will ideally provide a computational methodology for mapping the development of informal roads in remote forested areas in order to allow for future analysis of road evolution and the associated socio-economic consequences.