Contributions from experimental geostatistical analyses for solving the cloud-cover problem in remote sensing data

Nowadays, in many fields and applications, such as mining engineering, environmental monitoring, soil science, natural resources and environmental topics, the potential of Remote Sensing have been exploited. Large amount of data, easy and fast accessibility, and time–space availability are the reasons of attractiveness of satellite data in many fields of geosciences. However, satellite images face the problems due to presence of shadows and clouds, which is a general and real challenge since the surface features are masked. In many cases, a dense cloud cover simply prevents any detailed study of the target area through Earth Observation techniques. Geostatistics is a science field properly developed to estimate the unknown values in two-dimensional or three-dimensional space; therefore, it can be a potential approach to solve cloud-cover issues in Remote Sensing investigations. This work applies the geostatistical tools over a sentinel-2 satellite image targeting land cover in Emilia Romagna (Italy). The three main spectrum bands (RGB-values) with 10-m spatial resolution have been selected as target variables. The objective is to estimate pixel values within an area which is covered by clouds. The spatial variability of pixels with available land cover information have been studied through the use of variogram tools. Different estimation neighborhoods have been tested for application of the Kriging interpolation method, in order to estimate the RGB values below the cloud-covered area. The estimation has been performed, by controlling the properties of the estimator, the varying sizes and shapes of the neighborhood, the available number of RGB data used for estimation and the spatial distribution of pixels in the image. An image with close time period, without clouds, has been used to validate the results. Moreover, the estimation variance of RGB values for each pixel has been mapped. Results have shown the advantages and limitations of the proposed geostatistical method for the specific application of cloud-covered areas.