EESS PhD Student talk on "Unveiling the dynamics of upper forest shift in the Swiss Alps with deep learning and historical aerial images"
Abstract:
In the Swiss Alps, with a simple look at the mountains, any observer can trace a line separating the highest forest patches from alpine grasslands. Until the mid-20th century, this upper forest line has been largely influenced by grazing activity. Due to agricultural land abandonment taking place in the last decades, forests can now slowly expand uphill towards their potential climatic limit, the treeline. Climate change is also expected to strongly influence these dynamics. Mapping and understanding these interactions is crucial for climate adaptation and protection of alpine grassland biodiversity.Since accurate-enough forest covers maps from the 20th century are missing, we leverage time series of historical aerial imagery captured from 1946 and deep learning methods to map upper forest dynamics in the Swiss Alps. We overcome the lack of historical ground truth to train the deep learning model by injecting prior knowledge about forest cover dynamics into the training process. We then assess the influence of various factors including local climate, geomorphology and land use on the observed dynamics. By doing so, we hope to obtain a country-wide, spatially dense analysis of upper forest dynamics spanning the last 80 years.
Biography:
Thien-Anh is a 4th year PhD student at the ECEO lab. Her research interests lie at the intersection of machine learning, image processing and geospatial analysis for environmental applications. Before joining the ECEO lab, she graduated in Electrical Engineering at ENSEEIHT (Toulouse, France) and Georgia Tech (Atlanta GA, USA), and interned at CNES (Toulouse, France).
In the Swiss Alps, with a simple look at the mountains, any observer can trace a line separating the highest forest patches from alpine grasslands. Until the mid-20th century, this upper forest line has been largely influenced by grazing activity. Due to agricultural land abandonment taking place in the last decades, forests can now slowly expand uphill towards their potential climatic limit, the treeline. Climate change is also expected to strongly influence these dynamics. Mapping and understanding these interactions is crucial for climate adaptation and protection of alpine grassland biodiversity.Since accurate-enough forest covers maps from the 20th century are missing, we leverage time series of historical aerial imagery captured from 1946 and deep learning methods to map upper forest dynamics in the Swiss Alps. We overcome the lack of historical ground truth to train the deep learning model by injecting prior knowledge about forest cover dynamics into the training process. We then assess the influence of various factors including local climate, geomorphology and land use on the observed dynamics. By doing so, we hope to obtain a country-wide, spatially dense analysis of upper forest dynamics spanning the last 80 years.
Biography:
Thien-Anh is a 4th year PhD student at the ECEO lab. Her research interests lie at the intersection of machine learning, image processing and geospatial analysis for environmental applications. Before joining the ECEO lab, she graduated in Electrical Engineering at ENSEEIHT (Toulouse, France) and Georgia Tech (Atlanta GA, USA), and interned at CNES (Toulouse, France).
Practical information
- General public
- Free
- This event is internal
Organizer
- EESS - IIE
Contact
- Prof. Devis Tuia, Laboratory ECEO