This thesis will focus on an inter-disciplinary (intersection of artificial intelligence and climate change) research topic. Lake ice is an important variable to understand the regional and global climate change and has been recently recognized as an Essential Climate Variable (ECV). Monitoring and analyzing the (decreasing) trends in lake freezing using artificial intelligence techniques [1, 2] provides important information for climate research. This work is part of the project “Integrated lake ice monitoring and generation of sustainable, reliable, long time series” initiated and financed by the Federal Office of Meteorology and Climatology (Meteo Swiss) in the framework of GCOS (Global Climate Observing System), Switzerland. Multi-temporal satellite images are a natural data source to survey ice on lakes. This work aims to integrate the measurements from two types of "medium-to-high" spatial resolution satellite images (Landsat 8 from NASA, Sentinel 2 from ESA). The four target lakes are: Sihl, Sils, Silvaplana and St. Moritz. The primary goal of this project is to monitor these target Swiss lakes and to detect the extent and duration of ice, and in particular the ice on/off dates.
The major tasks are as follows:
•Data collection and pre-processing (including cloud filtering) in Google Earth Engine  platform. This includes the data of four lakes (Sihl, Sils, Silvaplana and St. Moritz) from 2 winters (W16/17, W17/18).
•Develop few-shot learning  based algorithm for lake ice detection. Using the developed methodology, process the data (independently for both Landsat 8 and Sentinel 2) from the two winters. As a contingency plan, adapt the existing machine learning-based lake ice detection methodology  developed for Terra MODIS and Suomi NPP VIIRS images to Landsat 8 and Sentinel 2 images. Improve the existing methodology, especially for the transition period (freezing, thawing).
•Detailed experimentation that involves generalization across lakes and winters (independently for each sensor). This also involves qualitative and quantitative analysis.
•Combine the results from Landsat 8 and Sentinel 2. (Additionally, on the dates with no Landsat 8 or Sentinel 2 data, fill the gaps using the existing results of VIIRS, MODIS and Sentinel 1 and if needed Webcams. In addition, provide a reliability estimate for each daily result).Perform multi-temporal analysis to reduce errors, increase accuracy and reliability.
•Correlate the final results with auxiliary data (temperature, precipitation, wind, snow height etc.) from the Meteo stations near the lakes of interest. Use auxiliary data also for better interpretation of results + to bridge gaps due to clouds.
Manu Tom MSc. (firstname.lastname@example.org)
Lake Ice Detection
Satellite Image Analysis
NASA Landsat 8
ESA Sentinel 2
IDEA League Student Grant (IDL)
CLS Student Project (MPG ETH CLS)
Information, Computing and Communication Sciences
Engineering and Technology