Part A:
DASOS, my open source software, manages airborne laser scanning data (both full-waveform and discrete point clouds). It is fundamentally different from other available software as it employs a rasterisation process prior to feature extraction, thereby mitigating harmonisation issues of the data. In this presentation, I will give an overview of how the airborne full-waveform LiDAR literature has been evolved over the years and then focus on the functionalities of DASOS along with its applications. DASOS has three main functionalities: (1) Reconstruction of 3D polygonal models, (2) extraction of 2D metrics and alignment with hyperspectral imagery and (3) extraction of structural elements from 3D windows. Regarding the applications (1) Handling big laser scanning data is challenging, so I compared six different data structures for storing the data and assessed their efficiency while generating 3D polygonal models. I further proposed the new data structure named Integral Tree, (2) I used the 2D metrics along with the combination of multisensory data to improve tree coverage maps, (3) proposed using multi-scale 3D windows and a machine learning pipeline to detect dead Eucalypts without delineating the trees first native Australian forest for managing biodiversity. It further worth noting that in a study I co-supervised (led by Dr Martins-Neto) we showed that DASOS performed better in tree species classification than the widely used LidR. The improved performance is probably attributed to DASOS effectively addressing variations in point cloud density resulting from uneven scanning patterns.
Part B:
Forest ecologists traditionally collect detailed information in the field from predetermined locations known as plots. Although collecting plot data is a time-consuming process, Earth Observation (EO) technologies' potential to support these studies is largely unexploited. Many ecological studies use local-scale EO data and are limited to time-consuming image downloads and subsequent local processing. The availability of cloud platforms has facilitated the capability to address the challenge of scalability. In this presentation, I will introduce the PlotToSat framework. This innovative framework generates spectral-temporal signatures, which capture both the temporal and spectral dimensions of each plot, from multi-sensory EO data at thousands of scattered plot locations in various geographic regions for machine learning applications. Using Google Earth Engine's cloud processing, I streamlined the time-consuming task of downloading massive satellite imagery and processing them locally.
Dr Miltiadou is a Postdoctoral Research Associate at the University of Cambridge, with an EngD from the University of Bath and Plymouth Marine Laboratory and prior research experience at Cyprus University of Technology. She conducts research fusing Earth Observation (EO) data with thousands of plots, has worked on detecting dead standing Eucalypt trees from 3D-Windows data, managed LiDAR data efficiently for 3D polygonisation, and studied the SAR phenological cycle of Paphos forest in Cyprus. She secured 400,000EUR project funding, officially co-supervised a PhD, and gained industrial experience at Carbomap and Interpine Group Ltd. She's a reviewer for Remote Sensing of the Environment and MDPI Remote Sensing, featured in MDPI Remote Sensing Journal's Most Notable Articles, and her work was included in the LOL manuscripts by Ladies of Landsat. She's also an Arctic code Vault Contributor of the 2020 Github Archive Program for her open-source software (DASOS).
