Correlated Spatiotemporal Data Modeling Using Generalized Additive Mixed Model and Bivariate Smoothing Techniques

Abstract

Background: The present article tries to analyze a correlated spatiotemporal data using an advance regression modeling techniques. Spatiotemporal data contains the information of both space and time simultaneously. Naturally, it is very much complicated and not easy to model. This article focuses on some modeling techniques to analyze a correlated spatiotemporal agricultural dataset. This dataset contains information of soil parameters for five years across the twenty six different locations with their geographical status in term of longitude and latitude. Soil pH and fertility index are the two major limiting factors in agriculture. These two parameters are governed by many other factors viz. fertilizer use, cropping intensity, soil type, geographical location, soil health management etc. Objective: The present study has been set up to explore whether there is any spatial gradient in the average pH levels across the geographical locations while fertility index and cropping intensity are acting as possible confounder. Methods: Soil pH is the response variable which varies with respect to time and space generally has a correlated structure. Besides this, some random effects component with fixed effects having a nonlinear association with the response is observed here. Generalized additive mixed model (GAMM) regression and Bivariate Smoothing techniques have been exercised to arrive at a meaningful conclusion. Conclusions: It is found that the pH value varies with change in latitude. Besides this, year, fertility index of available potassium and phosphate are also significant cofactors of this study. Final model has been selected through minimum AIC value (204.9) and model checking plots.