Use of Remote Sensing methods in monitoring mussel beds in the Wadden Sea

As part of the Trilateral Monitoring and Assessment Program of the Wadden Sea, IMARES monitors the area and structure of intertidal mussel beds. This is an important issue for environmental policies and management concerning mussel seed fishery. Mussels are essential components of the Wadden sea fauna because of their influence on sediment dynamics and reef structure. They also represent hot spots of biodiversity with a high diversity of associated organism, as well as food for the migrating birds. Mussel beds show quite stable structures, but in dynamic environments such as the Wadden Sea, changes on shape and coverage occur frequently. Then regular inventories are necessary to monitor them. They are currently performed twice a year by direct field work measurements. This method is labour-intensive, time consuming and potentially dangerous. Moreover the work is restricted by meteorological and site accessibility conditions. Therefore, the use of automated remote sensing methods to improve the mussels detection and mapping (classification) is investigated. In this study, an object-based classification method is compared to a human eye interpretation method performed on the same aerial photographs and field data (ground truth).

Airborne photographs with a pixel size of 0.5 x 0.5 meter taken at low tide in spring are used for the analysis. Due to the fact that mussel beds are often mixed with other elements like sand and algae, it is not possible to base the classification rules on spectral parameters only. Because of their specific textural characteristics, mussel beds appear visually rougher on the image in comparison to the surrounding. This observation highlighted the necessity to apply an object-based approach combined with pixel-based approach in order to build the classification rule set. During the analysis, texture and shape parameters are used to identify mussel beds element and spectral characteristics in order to discriminate disturbing features. The obtained classification is assessed with the field survey polygon data (ground truth) and compared with the data produced manually from human eye detection. The latter comparison relies on accuracy, reliability, consistency, repeatability and time efficiency criteria.

The classification result compared to the ground truth shows that the main mussel beds are detected. The elements reminding not classified are mostly small beds, under 2 hectares. The results show also that some elements are wrongly identified as mussel bed, mainly located along canals and land areas that are transition zones between two land cover types and then present high heterogeneity in structure which is the basic characteristic used to detect mussel beds. This problem can partially be overcome by the use of masks to exclude all the surfaces where the mussel beds can not occurred. The remaining mistakes correspond to elements closely similar to mussel bed by there structure but that are potentially recognisable by an expert. Concerning the comparison with manual classification, more mussel beds are detected. However, the detection software is more accurate and consistent in drawing the actual contours. It also presents advantages concerning the time efficiency, the objectiveness and the repeatability.

Even if the automated detection of the mussel beds is possible, a difference in terms of spectral and even textural parameters among the beds is highlighted, showing the limitations of this method. To compensate, the automated object-based classification should be combined with manual correction to improve the final result. From the comparison with ground truth data, it results that the field investigation remains necessary to validate the analysis from aerial photographs. However, the later can take place in a much smaller scale as only uncertainties have to be checked.

To conclude, it can be stated that the object-based approach might be consider as a powerful method to help the human observer with the interpretation of the aerial photographs in complement to ground truth. It offers an efficient way to detect and monitor the mussel beds by saving time and labour as well as generating a more detailed and repeatable monitoring program.

The results and the observations made during this study lead to the investigation of further use of object-based approaches for the monitoring of mussel beds. Besides the quoted advantages of using an object-based classification for the detection of mussel bed structures from aerial photographs, it is also possible to give a quantitative representation of the patterns which can be observed within mussel beds. This technique might be helpful in classification of mussel bed patterns in studies on stability of structures, as patch sizes and patch size distribution can be easily calculated.