Student information
MSc thesis topic: Multi-temporal log yard monitoring with PlanetScope & Sentinel-1/2
Advances in computer vision can help to track where, how and in what quantities timber products are processed. Such insights are essential for understanding how changes in demand for timber products will affect forests in the future and they can guide the development of forest policies which aim to ensure a continued provision of timber along with other vital ecosystem services.
The use of harvested wood products may contribute to climate change mitigation by storing carbon in built environments and substituting for high emission construction materials. Sawmills play an important role in the wood supply chain but there is a lack of data on their production capacity. To make accurate timber production projections, forest economic models rely on accurate information regarding the availability of raw materials for sawmills.
In this thesis you will expand on existing methods for quantifying the raw input of sawmills from images by applying computer vision techniques to time series of high-resolution remote sensing data.
Relevance to research at GRS or other groups
- This Thesis is supervised by Marc Rußwurm (Remote Sensing & Geo-Information Science), Jens van der Zee (Aquatic Ecology) and Nicola Bozzolan (Forest Ecology)
Objectives & Research Questions
- create a dataset of dense multi-temporal annotations of log stacks using Segment Anything.
- train and compare mono-temporal and multi-temporal log stack classification models
- understanding long-term dynamics of intra and inter-annual sawmill productivity
Requirements
- required: deep learning course
- optional: advanced earth observation course
- optional: interest in timber supply chains
- optional: interest in computer vision methodology
Expected reading list before starting the thesis research
- Buongiorno, J. (2018). On the accuracy of international forest product statistics. Forestry, 91(5), 541–551.
- Aligning Wood Resources and Industry Capacity: A Spatial Analysis of the Wood Chain in Czech Republic, Norway and Germany using the European Forest industry Database Journal: Forest Policy and Economics (under submission, write Nicola Bozzolan if you are interested)
Theme(s): Sensing & measuring, Integrated Land Monitoring