Speaker
Description
Repeated observations are crucial for understanding trends in biodiversity changes, particularly shifts in species composition. However, next to biodiversity, changes in environmental drivers are of equal importance to understand these. As most monitoring projects only recently started, archival data offer invaluable insights into long-term ecosystem dynamics and environmental changes, revealing perspectives crucial for shaping conservation strategies and enhancing ecological understanding.
This project explores advanced AI-based techniques to provide data on the historical development of forest cover, management and age structure, thus, some of the most important drivers for changes in forest biodiversity in time. The goal is to develop innovative, user-friendly methods to extract time-series data, specifically by assessing changes across different years within the same location.
We focus on the repeated forest inventory maps of Thuringia from 1850 to 2000, using about 100 maps (out of several thousands) for model development. These maps span diverse geographical and ecological conditions. The research involves several steps: digitizing paper maps through high-resolution scanning to preserve original data integrity, followed by detailed manual annotation to define forest boundaries and features like management type, predominant tree species and age structure. Convolutional Neural Networks are then applied for deep feature extraction and model training, aiming to recognize spatial patterns. Additionally, computer vision techniques enable precise matching between the map features and legends, facilitating comprehensive classification and data integration. This methodology combines deep learning and computer vision to tackle complex map analysis, providing greater accuracy than traditional methods.
Our results demonstrate AI's significant effectiveness in image segmentation, legend-feature matching, and text recognition in historical map analysis. This achievement holds substantial potential, offering an automated method for handling historical records of environmental conditions. Future work will enhance model accuracy and expand its applicability to biodiversity conservation efforts, e.g. by identifying areas of high habitat continuance.
| Status Group | Other |
|---|
