Geo-Spatial Cluster based Hybrid Spatio-Temporal Copula Interpolation

Abstract: In the absence of Gaussianity assumptions without disturbing spatial continuity interpolating along the whole spatial surface for different time lags is challenging. The past researchers pay enough attention to Spatio-temporal interpolation ignoring the dynamic behavior of a spatial mean function, threshold distance, and direction of maintaining spatial continuity. Therefore, we employ hierarchical spatial clustering (HSC) to preserve local spatial stationarity. This research work introduces a hybrid extreme valued copula-based Spatio-temporal interpolation algorithm. Spatial dependence is captured by a blended extreme valued probability distribution (BEVD). Temporal dependency is modeled by the Bi-directional long short-time memory (BLSTM) at different temporal granularities, 1 month, 2 months, and 3 months. Spatio-temporal dependence is modeled by the Gumbel-Hougaard copula (GH). We apply the proposed Spatio-temporal interpolation approach to the air pollution data (Outdoor Particulate Matter (PM) concentration) of Delhi, collected from the website of the Central Pollution Control Board, India as a crucial circumstantial study. This article describes a probabilistic-recurrent neural networking algorithm for Spatio-temporal interpolation. This Spatio-temporal hybrid copula interpolation algorithm outperforms and is efficient enough to detect spatial trends and temporal influence. From the entire research, we notice that PM concentration in a year reaches a maximum, generally in November and December. The northern and central part of Del-hi is the most sensitive regarding air pollution.