A River Channel Extraction Method To Analyze Channel Networks

The author, Prof. Bovik and his colleagues try to show that quantitative analysis of channel networks plays a very important role in river studies. In the paper, they have proposed a new method to extract channels from remotely sensed images to properly estimate the channel-widths which provides a quantitative representation of channel networks. The automated process proposed by them is based on the multiscale singularity index that strongly responds to curvilinear structures but not so strongly to edges.

The multiscale singularity index is a very useful algorithm for detection of singular curvilinear structures on different multiple scales. The tool comes handy in locating channels from satellite images. However, with the presence of channels charted on a wide range of scales, leads to creation of some artifacts in the singularity index response. The authors have modified and extended the multiscale singularity index to meet the multiscale nature of channel networks.

A channel map is produced by the algorithm by images wherever a contrast between the water and non-water pixels are found. At each pixel, the algorithm estimates the direction which is orthogonal to the curvilinear mass using 2nd order derivatives of the input images along both the evenly spaced directions. Then, the singularity index is computed at each scale and the maximum response across all scales is found out at each pixel location which retains the polarity needed for differentiating between the channel and island responses as both have opposite polarities. The island response is removed by discarding the negative polarity.

Due to the necessity of debiasing the input image before computation and further processing, the modified algorithm debiases the input image by using a Gaussian filters to subtract the local mean from the original image at every scale present. Also, the channel width estimation is done through interpolation between the scale that has the highest singularity index response and its neighboring scales.

In order to attenuate the ripples that get created by the MSI near the banks of wide rivers on presence of large range of scales after finding the maximum response from all the scales at every spatial coordinate, an adaptive smoothing algorithm is employed to adjust the strength of smoothing based on the estimated scale for each pixel for smoothing of the coarse scales more than fine scales.

The smoothing is implemented by computing an integral image over the singularity index response that enables the fast computation of summations over regions of any arbitrary size. Then, a box filter with a variable window size is used to smooth the response which is determined by the estimated scale. The channel centerline is determined by computation of the maximum response across all the scales at each coordinate and the dominant centerline direction is taken to be the orientation value at the maximum-response scale.

To finally display the computed channel width at each spatial coordinate along a centerline, a map of channels is created by regrowth of the channels. The channel-regrowth is achieved by drawing a line of length and orientation at each spatial location.

Critique

The paper has a very straightforward and plain-spoken approach written in a crisp language which makes it very comprehensible. The presentation of ideas and the analysis given were easy to follow and understand. The theoretical specifics mentioned by the authors are very insightful and help in complementing the elegance of the ideas formulated in the paper.

A complete automatic extraction of channel networks from the images of the satellites could profoundly facilitate the monitoring of water resources by elimination of the laborious manual inspection process. By the proposed method, creation of quantitative representations of channel networks will become uncomplicated and also be useful for a wide variety of studies.

The method proposed in the paper provides a robust alternative to the present procedures currently being used in the remote sensing of fluvial geomorphology. The modified MSI algorithm also makes the classification and analysis of channel networks much easier as the new purely automatic process can help in accurate estimation of the channel centerline, width and orientation along with the creation of a map of a channel networks using remotely-sensed data only.

However, being a just 4-page research paper has its own drawbacks too. Clearly, it becomes apparent to the reader that the paper being very crisp in its formulations, lacks the in-depth details, theories and conjectures that could have given the paper more substance.

Ideas for Follow - On Work

Even though the modified MSI algorithm was tested on 3 different regions having varying characteristics, i.e. the Mississippi River at Memphis, the Mississippi-Missouri-Illinois Rivers at St. Louis and the Ganga-Brahmaputra-Yamuna Delta using the Landsat-8 images, further analysis still need to be done on more riverine channels and deltas using images from other satellites too.

Also, as mentioned in the paper, the proposed method can be elevated to capture the effects of natural and anthropogenic changes and environmental forcing on networked-channels. The present work can also be extended towards automatic creation of topological maps which will provide the graphic representations of riverine channels.