Project name: Development of detection method for invasive plant species using Sentinel data
Client: The European Space Agency – ESA
Period: 11/2021 – 01/2023
Location: Croatia
The aim of the project was to investigate the possibility of using high-resolution Sentinel images to distinguish and map selected invasive plant species based on their morphological and physiological characteristics such as flower and leaf color, and the amount of water and chlorophyll in the leaves. The project activities were divided into four work packages:
- Project management
- Market assessment
- Theoretical and determinant framework
- Image preprocessing and model development
Based on a questionnaire conducted among relevant stakeholders (work package 2) and a review of the scientific literature (work package 3), common ragweed (Ambrosia artemisiifolia) and false indigo bush (Amorpha fruticosa) were selected, and spatial (ground truth) data were collected in defined research areas by fieldwork (work package 3). Field data served as the basis for developing a detection model, which was done by identifying differences between the spectral reflectance of the selected species and surrounding vegetation within the chosen habitat, as well as by comparing their key phenological stages such as budding, flowering, or drying. Machine learning algorithms were applied to analyze the obtained results with the aim of extracting the most significant variables, or predictors, which achieved the highest accuracy of detecting the presence of common ragweed and false indigo bush.
Steps and results of project implementation:
- A SWOT analysis of the Croatian market was conducted from the aspect of application of remote sensing in monitoring invasive plant species,
- Stakeholders in the fields of forestry, nature and environmental protection, agriculture, water management, etc., were contacted with a survey questionnaire to determine problematic invasive plant species,
- Survey questionnaires were analyzed, and problematic invasive plant species and management methods were defined,
- Recent scientific and professional literature on remote sensing application in invasive plant species monitoring was analyzed,
- The results of the survey questionnaire and literature analysis were interpreted, and the most suitable invasive plant species for detection model development, as well as their habitats, were defined,
- Field data on the occurrence of false indigo bush and common ragweed were collected using an unmanned aerial vehicle,
- Digital processing of field data provided input data for the development of a detection model based on Sentinel 2 satellite data,
- Sentinel 2 satellite images of selected areas were downloaded, and preprocessing was performed, which included cropping the area of interest, cloud removal, and masking of unvegetated areas,
- Vegetation indices for the research areas were calculated, and those most favorable for differentiating false indigo bush and common ragweed from other land cover categories were selected by statistical processing,
- Various machine learning methods were implemented to obtain the best detection model – Random forest, K-Nearest Neighbor, Support Vector Machine, artificial neural network, and boosting classification,
- Classification results were validated based on spatial data collected in the field,
- Confusion matrices and accuracy percentages of each detection model were calculated,
- Testing of the obtained model was carried out in another area to determine its robustness and transferability,
- The possibility of using the synergy method of Sentinel 1 and Sentinel 2 data for the detection of common ragweed was tested and, based on the results, discarded,
- Final reports and technical documents were prepared, describing the methodology and research results.
Implementation of this project contributes to the following Sustainable Development Goals:
