Fleur Visser

Depth Estimation of Submerged Aquatic Vegetation in Clear Water Streams Using Low-Altitude Optical Remote Sensing

UAVs and other low-altitude remote sensing platforms are proving very useful tools for remote sensing of river systems. Currently consumer grade cameras are still the most commonly used sensors for this purpose. In particular, progress is being made to obtain river bathymetry from the optical image data collected with such cameras, using the strong attenuation of light in water. No studies have yet applied this method to map submergence depth of aquatic vegetation, which has rather different reflectance characteristics from river bed substrate. This study therefore looked at the possibilities to use the optical image data to map submerged aquatic vegetation (SAV) depth in shallow clear water streams. We first applied the Optimal Band Ratio Analysis method (OBRA) of Legleiter et al. (2009) to a dataset of spectral signatures from three macrophyte species in a clear water stream. The results showed that for each species the ratio of certain wavelengths were strongly associated with depth. A combined assessment of all species resulted in equally strong associations, indicating that the effect of spectral variation in vegetation is subsidiary to spectral variation due to depth changes. Strongest associations (R2-values ranging from 0.67 to 0.90 for different species) were found for combinations including one band in the near infrared (NIR) region between 825 and 925 nm and one band in the visible light region. Currently data of both high spatial and spectral resolution is not commonly available to apply the OBRA results directly to image data for SAV depth mapping. Instead a novel, low-cost data acquisition method was used to obtain six-band high spatial resolution image composites using a NIR sensitive DSLR camera. A field dataset of SAV submergence depths was used to develop regression models for the mapping of submergence depth from image pixel values. Band (combinations) providing the best performing models (R2-values up to 0.77) corresponded with the OBRA findings. A 10% error was achieved under sub-optimal data collection conditions, which indicates that the method could be suitable for many SAV mapping applications.

Mapping of Submerged Aquatic Vegetation in Rivers From Very High Resolution Image Data, Using Object Based Image Analysis Combined with Expert Knowledge

The use of remote sensing for monitoring of submerged aquatic vegetation (SAV) in fluvial environments has been limited by the spatial and spectral resolution of available image data. The absorption of light in water also complicates the use of common image analysis methods. This paper presents the results of a study that uses very high-resolution image data, collected with a Near Infrared sensitive DSLR camera, to map the distribution of SAV species for three sites along the Desselse Nete, a lowland river in Flanders, Belgium. Plant species, including Ranunculus peltatus, Callitriche obtusangula, Potamogeton natans L., Sparganium emersum R. and Potamogeton crispus L., were classified from the data using object-based image analysis and expert knowledge. A classification rule set based on a combination of both spectral and structural image variation (e.g. texture and shape) was developed for images from two sites. A comparison of the classifications with manually delineated ground truth maps resulted for both sites in 61% overall accuracy. Application of the rule set to a third validation image resulted in 53% overall accuracy. These consistent results not only show promise for species-level mapping in such biodiverse environments but also prompt a discussion on assessment of classification accuracy.

Rapid mapping of urban development from historic Ordnance Survey maps: An application for pluvial flood risk in Worcester

Between 2004 and 2008 the city of Worcester, UK experienced a number of pluvial flood events. The causes of this kind of flooding are the topic of ongoing research. This paper describes a study that aimed to investigate the urban development of Worcester over time in relation to the location of recorded pluvial flood incidents. A novel rapid mapping method has been developed to derive urban development over five time periods between 1886 and 1995 from scanned Ordnance Survey historical maps. The technique compared well with manual digitisation results with k-hat values ranging from 0.67 to 0.87 for the land use maps created for different time periods. The technique performed least well for the oldest map series, due to misclassification of the abundant symbols and annotation. The method will be particularly beneficial for investigation of town/city development in time over large areas. The resulting map of urban development in Worcester (scale 1:20,000) shows that almost half of the recorded pluvial flood incidents occurred in areas developed between 1956 and 1975, which contradicts local belief that an outdated (Victorian) drainage system causes most of the problems. The quality of the post World War II developments is more likely to be a source for concern.

Development of a Knowledge Driven Rule Set for Classification of Submerged Aquatic Vegetation (SAV) in a Clear Water Stream: Where Do You Draw the Boundaries...?

A recent attempt at mapping submerged aquatic vegetation (SAV) species composition of a clear water stream in Belgium from ultra-high resolution, multispectral photographs, using object based image analysis (OBIA), resulted in a low, but consistent overall classification accuracy (53-61%). Since the results were obtained with a single rule set they show promise for the development of an automated tool to map SAV despite the challenges of its submerged environment. This extended abstract investigates to what extent difficulties with species delineation in the validation data may have influenced the results. We compare class boundaries, as drawn by experts along image segmentation outlines, with the results from the expert knowledge driven classification rules. A comparison for ‘pure’ objects, where the expert is certain about the assigned object class, resulted in a moderately good overall similarity (68%), while inclusion of ambiguous objects reduces the results to 59%. Under ideal circumstances the rule set seems capable of 74% similarity with expert validation data.