The method used here to accurately map land/water boundaries is illustrated and discussed in the companion poster titled "Mapping shorelines to subpixel accuracy using Landsat imagery"
A 30 years long archive of Landsat images represents an asset to map the temporal evolution of shorelines during such period. However, the native resolution of Landsat data is not accurate enough for most applications. High resolution imagery (e.g., SPOT, WorldView2) is expensive, not available everywhere and has poor temporal resolution. In a companion poster titled “Mapping shorelines to subpixel accuracy using Landsat imagery” a subpixel method to accurately map land-water boundaries with the Landsat multi-spectral satellite is presented. Then the improved method is tested in Shasta Lake (California, USA), which is an excellent site for some reasons: (1) the availability of hour by hour ground truth; (2) the water level is highly variable; (3) the shoreline is a typical barren land (when the water level is down). Some case studies of applications using this method have been selected according to their peculiarities, i.e. shoreline shape, type of water body, nature of the surrounding land, water level variability, etc. In this poster we show some examples that illustrate the improvements and the potential exploitation of around 30 year history measurable changes at land/water interface.
Improved shoreline in small lake in North New York state
Small stretch of shoreline of George Lake
It is evident the apparent improvement in resolution from 30 m in raw imagery to 5 m in processed shoreline.
Note that Landsat coordinates are shifted -30m relative to Google image.
Best modelling of the Area-Level function of Shasta Lake
In the left panels are the A-L curves based on the cloud-free Landsat data set (about half of 612 Landsat images during 1984-2011). Water level is on the x-axis. Water area is on the y-axis. The subpixel version (right panel) shows a tighter, cleaner A-L relationship than the dark pixel method (left panel). Outliers are probably related to light clouds or haze that were not automatically detected. The relationship is tighter at lower water levels, less at higher water levels. The probable explanation is that at low water levels the shorelines is further from tree line and a 2-component mixture is appropriate. At higher water levels the shoreline is closer to trees with 3-components mixture of water plus bare ground and trees in adjacent pixels. The vertical line indicates the water level for the Google Earth image, which unfortunately was at a time of high water level when the method did not work as well.
The shoreline butts trees and the mixture of water and trees is well determined
Bend in a river
The biggest improvements is in the bend of the river. This permits an accurate estimation of river widths