When using remotely sensed data, a cost-efficient sampling design for collecting ground data is needed to accurately map natural resources, environmental and ecological systems. The existing methods including traditional simple random sampling and kriging or cokriging variance based sampling designs can not lead to optimal sampling designs. In this study, a local variability based sampling design using a sequential Gaussian co-simulation by combining remotely sensed and ground data is developed. This method theoretically can lead to a sampling design with variable sampling distances, that is, grid spacings that are optimal at local and global levels. The method was assessed and compared with simple random sampling in a case study in which the soil erosion ground and vegetation cover factor was sampled and mapped using Landsat Thematic Mapper (TM) images and annual permanent ground measurements sampled from 1989 to 1995. The results show that the local variability based sampling greatly reduced the number of sampled plots and increased the cost-efficiency for sampling in comparison to simple random sampling. The difference in cost-efficiency between the two methods increased with increased global variation. This method can also be applied to analyse the sufficiency of a permanent plot sample and further provide information for additional sampling.
ASJC Scopus subject areas
- Earth and Planetary Sciences(all)