Detection of gradual changes in seasonal phenomena using time series analysis
programming skills (e.g. Python or R) required
The challenge of this topic is to detect a change or trend on top of seasonal fluctuations. For example: a crop or a seasonal forest has a characteristic phenological development over the year, which can be detected using a time series of satellite images and used for classification of crop type or forest type. Also seasonal flooding of river banks can have a characteristic pattern over the season, which has as a result different habitats depending on duration and time of flooding. Changes in these yearly patterns are of interest as indicators of climatic change or changes in management, such as increased logging in a forest. This topic is method oriented and the application can be chosen together with the student, provided suitable data can be obtained.
Methods using seasonal time series to classify crops exist, for example dynamic time warping (DTW). Methods to find sudden changes in time series with or without added seasonal patterns exist as well, for example B-FAST. Using these methods as points of departure, develop a method to detect gradual changes. Depending on interest of the student, the topic can be extended to include modelling of the change process using existing models, e.g. a deforestation process, a forest regeneration process, changes in flooding etc. Outcome of these models can be used to interpret the time series and find the relevant gradual changes.
Bagnall,
A., Lines, J., Bostrom, A., Large, J., Keogh, E., 2017. The great time series
classification bake off: a review and experimental evaluation of recent
algorithmic advances, in: Data Min Knowl Disc (2017) 31:606–660. https://doi.org/10.1007/s10618-016-0483-9
B-FAST:
https://bfast.r-forge.r-project.org;
For an M-SE student, the idea would be to combine remote sensing with process modelling, so it wil be a combination of SIS and TE fields.