Performance Evaluation of Semantic Kriging: A Euclidean Vector Analysis Approach

Abstract

Prediction of spatial attributes in geospatial data repositories is indispensable in the field of remote sensing and geographic information system. The semantic kriging (SemK) approach semantically captures the domain knowledge of the terrain in terms of local spatial features for spatial attribute prediction. It produces better results than ordinary kriging and other prediction methods. This letter focuses on the theoretical and empirical analyses of the SemK. A Euclidean vector analysis approach is adopted to theoretically prove the efficacy of SemK in capturing semantic knowledge.Index Terms-Geographic information system (GIS), kriging, prediction, semantic kriging (SemK).

Spatio‐temporal prediction of land surface temperature using semantic kriging

“…For example, if there is a single LULC class which covers the entire region (such as snow cover, desert), the performance of ST‐SemK will be exactly equal to that of ST‐OK. In contrast, if the LULC distribution is highly diverse, ST‐SemK generally yields more accurate results than ST‐OK and others (Bhattacharjee & Ghosh, ). Therefore, an analysis of the terrain entropy can be carried out before the actual interpolation, by following the lemmas in Bhattacharjee and Ghosh (), to check whether ST‐SemK is suitable for the given study regions and the application.…”

“…In contrast, if the LULC distribution is highly diverse, ST‐SemK generally yields more accurate results than ST‐OK and others (Bhattacharjee & Ghosh, ). Therefore, an analysis of the terrain entropy can be carried out before the actual interpolation, by following the lemmas in Bhattacharjee and Ghosh (), to check whether ST‐SemK is suitable for the given study regions and the application. Further, it is also evident from the spatio‐temporal analysis literature that experimental spatio‐temporal semivariograms are difficult to construct.…”

“…The differences exist in these methods based on how they model the spatial autocorrelation/dependency among the sample points for LST analysis. Their performance also depends on the spatial region of interest and the diversity of the terrestrial distribution (Bhattacharjee & Ghosh, ).…”

“…The total semantic change (Δsem) is then evaluated over time interval t if both the scores are non‐zero (Bhattacharjee & Ghosh, ). Once the temporal change in semantics is evaluated through the temporal semantic similarity and the temporal importance metrics, a 3D temporal semantic semivariogram is modeled further.…”

“…However, the performance of ST‐SemK is highly dependent on the study region, and spatial autocorrelation, LULC–LST correlation, and the diversity of the LULC distribution (entropy) are the major contributing factors. Thus, while choosing the region of interest (RoI) for the empirical study, the LULC‐based entropy analysis (Bhattacharjee et al., ), LULC–LST correlation study, and the performance evaluation lemmas (Bhattacharjee & Ghosh, ) can be performed beforehand to check the suitability of ST‐SemK for that particular RoI. The objectives of the proposed work are as follows: to analyze the behavior of temporal semantic change of the LULC distribution and incorporate this behavior to model the residual LST change for each sample point; to extend the traditional spatio‐temporal autocorrelation model with the semantic LULC information and carry out the spatio‐temporal prediction and forecasting of LST; and to compare the proposed method in terms of accuracy with some popular spatio‐temporal interpolation methods. …”