Geostatistics are a variety of techniques used to analyze continuous data e.g., rainfall, elevation, air pollution. The fundamental structure of geostatistics is based on the concept of semi-variograms and their use for spatial prediction kriging. Sampling methods are also discussed in Unit GD9 Field data collection.
Geostatistics is a subdiscipline of spatial statistics developed to estimate the value of a continuous spatial process at unknown locations by using the information of the value of these process at known locations. Furthermore, it aims to quantify the uncertainty related to the prediction (Calder et al., 2009; Emmanouil, 2019). In order to do such predictions, geostatistics entails some statistical methods which use as starting point the assumption of a random component that can define the spatiotemporal variability. These methods are developed to infer the parameters that can describe the spatiotemporal patterns of the input variables (e.g. soil moisture) so that finally these variables at unsampled locations can be estimated (interpolated) (Emmanouil, 2019). Geostatistical methods are strongly related with classic interpolation methods but differ by its use of random variables that allow to given an uncertainty indication associated with the prediction of variables in space and time.
In environmental research geostatistical techniques are often applied to infer (interpolate) variables at such unobserved locations by using information from known locations. One of such geostatistical techniques is Kriging, which is a geostatistical method that predicts variables by using spatial interpolation. This spatial interpolation is done by establishing a semivariogram that defines the spatial relationship between the variables of interest in function of the distance. Because of this, the Kriging technique can also give an indication on the variance or accuracy of the prediction (Calder et al., 2009); Van der Meer, 2012). On the other hand, cokriging is another important geostatistical technique and differs from Kriging by using the cross-correlation between variables to generate local estimates (Van der Meer, 2012). In earth observation studies, cokriging can be applied to better predict sparsely based data on the ground (e.g. biomass) by using the cross-correlation of this variable with a more continuously sampled satellite metric like NDVI. Furthermore, these techniques can also be used to enhance satellite image information, filling missing pixels or even downscale the information to a higher resolution (Van der Meer, 2012).