Automated and efficient workflow for large airborne remote sensing vegetation mapping and research of Natura 2000 habitats

Kania, Adam; Kopeć, Dominik; Niedzielko, Jan; Sławik, Łukasz

The goal of HabitARS (Habitats Airborne Remote Sensing) project is the implementation of remote sensing methods for Natura 2000 habitat monitoring in terms of environmental protection and sustainable agriculture. The methodology of identification of non-forest Natura 2000 habitats and their threats (desiccation, succession, invasive/expansive species) will be developed. Very extensive field campaign and remote sensing scanning make it probably one of the largest remote sensings project in Europe. Ground truths were collected on multiple sites (spring/summer/autumn) over two vegetation seasons (2016-2017). 200 field campaigns collected 31.500 reference samples. Aerial scanning was performed using multi-sensor platform, integrating Riegl full-wave LiDAR, 50Mpix RGB camera and HySpex hyperspectral scanner (SWIR-384, VNIR-1800). In 400 h flights hours 2300 km2 of data was acquired, achieving 7 ALS points/m2 and ground pixels sizes of 10 cm (RGB) and 1 m (hyperspectral). We automated classification workflow to free teams from repetitive work. A system based on Vegetation Classification Studio software was deployed in datacenter, integrating all datasets and facilitating remote access to research teams from 6 universities. Simple text language defines input data, goals and parameters of scientific experiments. Batches of 10-100s of related experiments can be defined. The basic process automates e.g. preparation of reference data, splitting into training/validation sets, rasterization, model learning & validation, quality assessment & reporting and produces a set of final vegetation maps in multiple formats. Often in minutes. Many procedures/algorithms performing multiple cycles of classification, prediction and accuracy assessment – like feature selection, optimizations in search for specific target, or search over multiple parameters – are automated. Modern approaches e.g. fuzzy prediction and multiple fuzzy visualizations, dimensionality reduction and feature engineering algorithms are supported.. That way of work revolutionized the daily routine of research, allowing research teams to shift focus from performing experiments to concentrate on analysis and understanding of results, and designing new approaches. It brought more confidence to our results – we now base our conclusions on thousands of experiments, not just a few as before. For many vegetation maps, results are delivered in hour(s). Larger experiment batches, spanning multiple flights and study sites, are often ready “next morning”. 9 months after introducing VCS system, all teams created over 300.000 classifications (compared to just 600 during 1st year of the project) and 17.500 vegetation maps. While being an obvious success, such a big shift in organization of research work lead to its own problems. The 100x-1000x increase in number of experiments and results was quite disruptive for whole project organization. Besides many benefits and obvious improvements, it uncovered unexpected bottlenecks, creating need for further automation of related activities like data management, error detection, organization of results and better visualization and analysis of results. Research has been carried out under the Biostrateg II Programme of the Polish National Centre for Research and Development, project DZP/BIOSTRATEG-II/390/2015: The innovative approach supporting monitoring of non-forest Natura 2000 habitats, using remote sensing methods.


Citation style:
Could not load citation form.


Use and reproduction: