Beyond the Random Forest: How Deep Learning for Remote Sensing Monitors Trees

Abstract

Data to track our planet's health has increased to hundreds of petabytes in volume through the ever increasing number and ever decreasing costs to deploy drones and satellites. National programs by USGS, USDA, NASA, ESA, and the European Commission make freely available multi- and hyperspectral imagery. However, intelligent manual inspection of these geospatial data and maps by human eye is impossible. Recent advances in deep neural networks dramatically boosted the accuracy and diversity of computer vision applications to the point that those achievements become practical tools at our fingertip. My presentation invites you onto a journey of my research exploring deep learning for aerial photos and laser scans in order to generate an inventory of urban forests in New York City [1], discover ancient agriculture in the Negev desert of Israel [2], and an attempt to estimate global biomass [3]. I hope to spark fruitful discussions with peers from ecology on how to utilize AI for environmental good to protect and treasure our green spaces. [1] https://elib.dlr.de/187233 [2] https://elib.dlr.de/190710 [3] https://elib.dlr.de/191502