multitemporal image analysis for monitoring fuzzy shorelines
Ratna Dewi is a PhD student in the department of Earth Observation Science. Her supervisor is prof.dr.ir. A. Stein from the faculty of Geo-information Science and Earth Observation.
Rapid development and population growth in coastal areas always bring a risk of coastal damage. In this situation, monitoring of shoreline position plays an important role in achieving a balanced condition between economic development and coastal protection. For this purpose, local authorities and coastal planners require information on shoreline changes for coastal land use planning and disaster risk management. Monitoring shoreline change for larger areas and longer time spans, however, is challenging due to limited data availability and high cost. Remote sensing and specific image processing methods for the identification and monitoring of shorelines are needed, especially methods that can handle the uncertainty in shoreline positions. This dissertation investigates and develops image analysis methods from remote sensing images to provide information for the sustainable coastal development. It focuses on using a fuzzy classification and a change detection technique to identify shorelines and monitor their changes. Emphasis is given on data quality and the estimation of uncertainty. The methods proposed in this dissertation are applied on a series of images to identify shoreline positions in the northern part of the Central Java Province, Indonesia which experienced a severe change of shoreline position over three decades.
First, an unsupervised fuzzy c-means (FCM) classification is presented to observe the shoreline positions by taking the gradual transition between water and land into account. The FCM is a clustering method that separates data clusters with class means and fuzzy boundaries allowing for partial membership. Two methods to generate shorelines are proposed. The first method derives the shoreline as a single line by applying a threshold of 0.5 on the water membership images. The second method derives shorelines as an area or a margin, presented as a crisp object with a boundary determined by threshold values resulting from parameter estimation. Crisp and fuzzy methods are combined for change detection. The post-classification comparison method is implemented to distinguish abrupt and gradual changes at the object level and provide the change uncertainty at the pixel level. Two perspectives of uncertainty are addressed: uncertainty that is inherent to shoreline positions as observed from remote sensing images, and the uncertainty that propagates from object extraction and implementation of shoreline change detection method. Shoreline and its changes are presented as crisp sub-areas. The changed areas are thus associated with the spatial distribution of change uncertainty.
Second, the possibility of using fuzzy-crisp objects to derive shoreline positions as the transition zone between the classes water and non-water is addressed. Pixels at which the membership value (μ) exceeds 0.99 are the core of a class, for example the water class, whereas pixels with 0.01< μ <0.99 belong to transition zones or shoreline class, and pixels with μ <0.01 do not belong to objects of water or shoreline. A change detection method for shorelines which accounts for their fuzzy character in remote sensing images is proposed and implemented. The change of shoreline is explained in terms of change magnitude and change direction using change vector analysis (CVA). Information provided by CVA allows us to see the trend of the fluctuating shoreline over time. The analysis of information provided by the change magnitude and direction reveals that each change combination represents one specific type of change process. It shows a multi-year pattern of water membership changes over the observation periods that could indicate certain coastal processes, for instance, erosion and accretion. Based on these results, it can be concluded that the proposed method can assess changes in a shoreline by taking into account that it is a fuzzy boundary.
Third, uncertainty modelling of shorelines by comparing fuzzy sets and random sets is presented. Both methods quantify extensional uncertainty of shorelines extracted from remote sensing images. Two datasets are tested: pan-sharpened Pleiades with four bands (Pleiades) and pan-sharpened Pleiades stacked with elevation data as the fifth band (Pleiades + DTM). Both fuzzy sets and random sets model the spatial extent of shoreline including its uncertainty. Fuzzy sets represent shorelines as a margin and their uncertainty as confusion indices. They do not consider randomness. Random sets fit a mixed Gaussian model to the image histogram. The random sets represent shorelines as a transition zone between water and non-water. Their extensional uncertainty is assessed by the covering function. The results show that fuzzy sets and random sets result in shorelines that are closely similar. Kappa values are slightly different and McNemar’s test shows high 𝑝-values indicating a similar accuracy. Inclusion of the DTM (digital terrain model) improves the classification results, especially for roofs, inundated houses and inundated land. The shoreline model using Pleiades + DTM performs better than that of using Pleiades only, when using either fuzzy sets or random sets. It achieves κ values above 80%.
Fourth, the transferability and upscaling of a fuzzy classification of shoreline changes to a different area and towards larger area is investigated. Three strategies are conducted: (i) optimizing two FCM parameters based on the predominant land use/cover of the reference subset; (ii) adopting the class mean and number of classes resulting from the classification of reference subset to perform FCM on target subsets; and iii) estimating the optimal level of fuzziness of target subsets. From the experimental results, 𝑚 values in the range from 1.3 to 1.9 are obtained for seven land use/cover classes that have been analysed. For the ten images used in this research, 𝑚=1.8 is obtained as optimal value. For a coast with similar characteristics, this 𝑚 value can be adopted and the relation between land use/cover and the two FCM parameters can help to shorten the time needed to optimize the parameters. The proposed method for upscaling and transferring the classification method to a larger and to different areas is promising, showing κ values >0.80 and agreement of water membership values >0.82 between the reference and target subsets.
To summarize, this dissertation focuses on modelling shoreline as an object with vague boundaries using multi-temporal remote sensing images. The associated uncertainties are estimated by means of possibility and necessity measures, and by confusion index. In this sense, this dissertation contributes to the monitoring of shorelines trough the development and the implementation of image analysis methods to quantify and monitor the changes of shorelines and related change uncertainty using remote sensing images.
