Optimal land cover sampling

Lauri

Oct 15, 2023 • 3 min read

Land cover is defined as the physical material at the surface of Earth. Different types of land cover classes include for example grass, asphalt, trees, bare ground and water. Information on land cover is usually captured using field surveys and analysis of remotely sensed imagery.

Land cover sampling is a process where land cover information is extracted from a land cover dataset around predefined geographical sample points. The size of the sampling neighborhood is determined by the sampling radius, which has an effect on the resulting land cover characterization of the points. In a recent article draft we describe a method for optimally sampling land cover data around the sample points (click below to download the article draft).

Optimal land cover sampling

Draft, March 2023

Optimal_land_cover_sampling.pdf

5 MB

The general idea is to maximize the information obtained by land cover sampling, by adjusting the size of the sampling radius around the sample points. The method is primarily motivated by characterizing the human habitats in medical cohort studies (HEDIMED, www.hedimed.eu). However, the proposed sampling method can be used in any sampling problem with n-dimensional spatial data and a countable number of classes.

Land cover sampling with three different sampling radii. Figure (a) shows three geographical points p1, p2 and p3 located in the Helsinki area and the CORINE land cover map, where each color represents a specific land cover class. All points are located in the same land cover class (red="Discontinuous urban fabric"). By using a small sampling radius (a), the land cover characterization of all three points is equal. By increasing the sampling radius as depicted in Figure (b), different land cover class distributions are obtained for the areas around the three points, resulting in larger spread in land cover distribution. Using a very large sampling radius for land cover sampling as in Figure (c), the spread of the land cover distribution vectors will diminish due to the overlap of the sampling areas.

Initial numerical experiments performed on CORINE land cover data and random locations in built-up areas suggest that the sampling radius has a significant effect on the land cover characterization of the locations. The optimal sampling radius is found by studying the spread of land cover in sampling areas of different sizes.

Land cover spread as a function of sampling radius, initial large scale experiment. The land cover spread, that is, the information obtained by land cover sampling, is maximized with r=1300 meters.

In the initial experiments, the optimal sampling radius for CORINE land cover data is found at around 1.3 kilometers for random points in built-up areas. However, both the land cover data as well as the sampling point data set have an effect on the size of the optimal sampling area. The method proposed in this paper is easily implemented for any data sets and enables the sampling radius to be determined in a well-justified manner in all cases.

Download article draft:

Optimal land cover sampling

Draft, March 2023

Optimal_land_cover_sampling.pdf

5 MB

Sign up for more like this.