An autocorrelated conditioned Latin hypercube method for data-driven spatio-temporal sampling and predictions

by Van Huong Le¹, Rodrigo Vargas¹

¹Department of Plant and Soil Sciences, University of Delaware, Newark, DE, 19716, USA

Corresponding author affiliation and e-mail:

Rodrigo Vargas

Department of Plant and Soil Sciences, University of Delaware, Newark, DE, 19716, USA

[email protected]

Description

This repository contains the source code to perform different sampling methods:Fixed sampling, conditioned Latin Hypercube Sampling (cLHS), and autocorrelated conditioned Latin Hypercube Sampling (acLHS). The predictions are then based on the samples using a Bernstein copula-based stochastic cosimulation (BCSCS) method. Here are two case studies using data of soil CO${2}$ efflux (i.e., the CO${2}$ efflux from soils to the atmosphere known as soil respiration) that are relevant for carbon cycle science. The first case study represents data from a time series (1D approach), and the second case represents spatial data (2D approach) across the conterminous United States (CONUS).

Content

RProject_acLHS_1D: The case study represents data from a time series

• Data: this folder contains the data
• Functions: this folder contains the useful functions
• Scripts: this folder contains the scripts
• Results: this folder contains the results
• RProject_acLHS_1D.Rproj: This file is the R Project

RProject_acLHS_2D: The case study represents data from spatial data

• Data: this folder contains the data
• Functions: this folder contains the useful functions
• Scripts: this folder contains the scripts
• Results: this folder contains the results
• RProject_acLHS_2D.Rproj: This file is the R Project

install

The code has been tested using packages of:

• R version 4.2.1

• RStudio 2022.07.1

How to run the code?

RProject_acLHS_1D

Opening the project RProject_acLHS_1D.Rproj with Rstudio. Then open all the scripts in the "scripts" folder. The scripts are run in the following order: 0_Getting_Started.R, 1_Exploratory_data_analysis.R, 2_Variogram_analysis.R, 3_Sampling_Design.R, 4_Simulations.R.

• 0_Getting_Started.R: this script is for installing and loading R packages, and also loading functions from the functions folder.
• 1_Exploratory_data_analysis.R: this script is for exploring and calculating univariate statistical properties and dependency relationships between variables.
• 2_Variogram_analysis.R: this script is to explore the temporal or spatial distribution of the variable of interest and calculate its autocorrelation function.
• 3_Sampling_Design.R: this script is for applying sampling methods based on the data.
• 4_Simulations.R: this script is to model the characteristic functions of the variables and perform the simulation.

RProject_acLHS_2D

Opening the project RProject_acLHS_2D.Rproj with Rstudio. Then open all the scripts in the "scripts" folder. The scripts are run in the following order: 0_Getting_Started.R, 1_Exploratory_data_analysis.R, 2_Variogram_analysis.R, 3_Sampling_Design.R, 4_Simulations.R.

• 0_Getting_Started.R: this script is for installing and loading R packages, and also loading functions from the functions folder.
• 1_Exploratory_data_analysis.R: this script is for exploring and calculating univariate statistical properties and dependency relationships between variables.
• 2_Variogram_analysis.R: this script is to explore the temporal or spatial distribution of the variable of interest and calculate its autocorrelation function.
• 3_Sampling_Design.R: this script is for applying sampling methods based on the data.
• 4_Simulations.R: this script is to model the characteristic functions of the variables and perform the simulation.

License

Source code to replicate figures in publication: "An autocorrelated conditioned Latin hypercube method for data-driven spatio-temporal sam- pling and predictions"

By Van Huong Le and Rodrigo Vargas (in review)