Single tree based, satellite supported Forest ecosystem monitoring with auto-adaptive hyper-dimensional geodata analysis

Project period: 01.04.2021 – 31.03.2023
Funding No.: 033D014A

Prof. Dr. Barbara Koch, Albert-Ludwigs-University Freiburg, Chair of Remote Sensing and Landscape Information Systems (FeLIS)
GISCON Systems GmbH, Specialist for Forest and Geoinformation Systems (GISCON)
con terra GmbH, Specialist for Geoinformation Systems, esp. AI (CONTERRA)
Society of Scientific Data Processing mbH Göttingen, Processing- und IT-Competence center (GWDG)


The aim of the project, which is funded by the federal ministry of education and science (BMBF) by the funding agency Jülich, is the combination of high-time-resolution satellite data with high-spatial-resolution drone data. This approach should enable correlations of satellite data with single tree conditions and, by this, create a high spatial and temporal resolution single tree monitoring. This method can be the new gold standard for monitoring the forest regeneration on regional scale.

The available satellite and drone based high tech sensors will 1) be upgraded with a drone-based electronic nose sensor, 2) used at maximum resolution (BigData approach), 3) be analyzed referring to their data processing with HPC / HPDA (High Performance Computing / High Performance Data Analytics) with a hyperdimensional approach and 4) be used referring to the created data basis as training data for HPC-based deep convolutional neural networks.

As a result, correlation pathways will combine satellite and drone data to allow a high spatial and temporal resolution analysis of forest damages and forest growth.

Sub-goals in ForestCare are:
1) Drone-Based Electronic Nose
Goal is the development of a highly flexible autonomously flying early detection system of calamity onsets, e.g. bark beetle infestation or beech degradation, on the basis of chemical cues, which are specific for the selected calamities.

2) Big Data Approach
Goal is the acquisition of historical and recent high-resolution satellite data (SAR (1m), multispectral (1.24m²), NIR (2.16m²), SWIR (3.7m²) and drone data (Enose (<1m²), LiDAR (<1cm²), hyper- and multispectral (<1cm²), as well as forest inventory data (ConFoBi, own acquisition, Federal Forest Service Agencies).

3) High Performance Computing
Goal is a) to check the data from 2) on parametric correlations with stand data from analyzed forest stands, e.g. the DFG-research training group ConFoBi. Goal is b) the identification of not directly visible relations by a non-target approach. This should enable to find correlations, which will be validated afterwards with the ground truth data.

4) AI-Approach
Goal is the training of deep neuronal networks (ANN / deepCNN) with Big Data from 2) under consideration of the multidimensional correlations found in 3). The resulting algorithms should allow a high level of generalization in order to analyze the vitality of single trees based on high spatial resolution satellite data.

BMBFProjektträger Karlsruhe