No Thumbnail Available
Details
- Supervisors
- Faculty
- Degree label
- Abstract
- Central Belgium is one of the most muddy flood affected regions in Europe. One of the main causes of muddy floods is the presence of spring crops (e.g. maize, sugarbeet, potato) in the catchments. In Belgium, winter cover crops are implemented during the intercrop period preceding a spring crop and following a winter cereal. The Land Parcel Identification System provides the information on crop rotations every year in Belgium. Also, remote sensing acquires significant information on vegetation and plant behaviour that are becoming more and more useful in agricultural management and crop identification. With the increasing frequency of muddy flood events, the damage and its costs to municipalities, as well as the psychological stress on people, there is a need to identify hazardous areas in order to implement mitigation measures. To achieve the design of a such an early warning system against muddy floods, it is essential to discriminate winter cereals from winter cover crops during winter season. In order to meet this objective, different procedures were carried out to classify sample plots from a study zone into different classes (winter wheat, winter barley and winter cover crops): a rule-based classification method and a Maximum Likelihood classification method. By using different combinations of dates and remotely sensed features, what stands out in this master thesis is that using the Maximum Likelihood classifier with a combination of three features (NDVI, 10 band and SAVI images) and a combination of chronological dates (27/12/2016, 15/02/2017 and 14/03/2017) provide the most accurate results and allow the identification of winter cover crops with a high level of accuracy. With the aim of improving the classification methods, it would be interesting to add feature images taken during October and November because the differences between winter cereals and winter cover crops are expected to be really significant during these two months. Then, using a different classification algorithm, like Random Forest, could be a way to improve the robustness and replicability of the classification pattern. The challenge in the building of an early warning system against muddy floods using crop discrimination is that the information about the next crop on a given agricultural plot in real time will never be available. It will need some kind of predictions which will involve more errors and more uncertainty.