Very high resolution Land Cover maps OAL-UK (Catterline)

Surface features are produced as a result of internal deformation of active landslides, and are continuously created and destroyed by the movement. The detailed mapping of the evolution of bare ground patches and vegetation cover pattern over time provides useful insights on the mass movements and the most vulnerable areas. Monitoring the evolution of the latter, in turn, could help to describe the benefits of NBS past their implementation by observing a reduction of vulnerable areas and the increase of the extent of stable vegetation as a result of the interventions.
The preliminary analysis of satellite data on OAL-UK was devoted to the analysis of a time series of very high resolution multispectral satellite data (<1 m) in the period prior to the implementation of NBS. The goal was to explore the potential of remote sensing to describe the pattern and extent of vegetation cover and plant cover regeneration, as well as to observe the self-organisation of landslide scars - i.e. re-distribution of bare ground patches over time.

The dataset includes Land Cover maps obtained by Worldview 2 satellite images acquired between 2011 and 2016 in different seasons, notably on 29 Jun 2011, 22 April 2014, 08 June 2014 and 22 Feb 2016.

After the pre-processing phase, Principal Component Analysis (PCA) was applied on the pansharpened multispectral bands to reduce the dimensionality of each dataset (pansharpened multispectral bands), while retaining as much as possible its information content.  (Richardson, 2009).

A machine learning unsupervised classification (K-means) was applied to the first three PC found by PCA. K-means is an unsupervised classification algorithm that groups objects into k groups based on their characteristics.  The mean spectral reflectance values of the samples assigned to each cluster was analyzed to identify typical spectral profiles of land features as well as eventual similarities between the classes and successively used to assign labels to the unsupervised classes. In cases of spectral similarity between two or three clusters, thy were merged in a unique class. 

The results of k-means classification were analyzed to define typical spectral profiles of land cover features found in OAL-UK. First, the mean and standard deviation of the spectral bands of each cluster and image were calculated and compared with spectral libraries of land features. This led to identify 11 land cover classes with typical and recurrent spectral profiles within the OAL: 

1. Bare soil

2. urban materials and sea foam (uniform spectra)

3. water saturated soil

4. water logged soil/vegetation

5. vegetation (grass)

6. dense vegetation (shrubs)

7. vegetation with exposed soil

8. soil with sparse vegetation

9. mixed wet soil/vegetation

10. Water

11. Turbid Water