exPrior: An R Package for the Formulation of Ex-Situ Priors

Abstract

The exPrior package implements a procedure for formulating informative priors of geostatistical properties for a target field site, called ex-situ priors and introduced in Cucchi et al. (2019). The procedure uses a Bayesian hierarchical model to assimilate multiple types of data coming from multiple sites considered as similar to the target site. This prior summarizes the information contained in the data in the form of a probability density function that can be used to better inform further geostatistical investigations at the site. The formulation of the prior uses ex-situ data, where the data set can either be gathered by the user or come in the form of a structured database. The package is designed to be flexible in that regard. For illustration purposes and for easiness of use, the package is ready to be used with the worldwide hydrogeological parameter database (WWHYPDA) Comunian and Renard (2009).

Estimating Hydrothermal Properties and High-Frequency Fluxes From Geophysical Measurements in the Hyporheic Zone

“…Amongst methods adopting the stochastic point of view, the Bayesian inference presented in section 2-in which probabilities are updated as more information or measurements become available-has been used in a wide range of settings in hydrological and hydrogeological studies (e.g., Woodbury, 2007;Freni and Mannina, 2010;Rubin et al, 2010;Harrison et al, 2012;Viglione et al, 2013;Zhang et al, 2014;Linde et al, 2017). Bayesian inference is flexible in the amount of information to account for prior to starting the assimilation of in situ measurements (Rojas et al, 2009;Kitanidis, 2012;Cucchi et al, 2019;Heße et al, 2021). To the author's knowledge, this study is the first to adopt a stochastic or Bayesian approach for estimating properties and fluxes from measurements in the HZ and it demonstrated how estimates become more reliable as the amount of prior information increases.…”

“…The prior pdf f Y (y) represents the knowledge about the target parameters Y before accounting for data measured in situ, in other words the data measured at the site under investigation (Rubin, 2003). As further detailed in section 2.3, Bayesian inference is flexible in the amount of information to account for prior to starting the assimilation of in situ measurements (Rojas et al, 2009;Kitanidis, 2012;Li et al, 2018;Cucchi et al, 2019;Heße et al, 2021).…”

Defining Hydrogeological Site Similarity with Hierarchical Agglomerative Clustering

“…The ex situ prior is the posterior predictive distribution for Y derived by marginalization over updated distributions of hyperparameters (Gelman et al 2014). Methods for fitting the model to provided data and for deriving $$$ p\left(Y|\mathcal{D}\right) $$$ are provided within the exPrior R package, available on CRAN as well as GitHub (Heße et al 2019b). In this paper, we introduce the use of clustering to determine for a target site a set of similar sites whose measurements are used as ex situ data $$$ \mathcal{D} $$$.…”

“…They are as follows:To import the data into R, we used the R package geostatDB being developed at https://github.com/GeoStat‐Bayesian/geostatDB. For the computation of the ex situ prior distribution, we used the R package exPrior being developed at https://github.com/GeoStat‐Bayesian/exPrior. The used software version was 1.0.1 (Heße et al 2021). The R code used for the analysis can be found at https://github.com/GeoStat‐Bayesian/siteSimilarity. …”

“…In data-scarce scenarios, a user can obtain data from an open-source database: the World-Wide HYdrogeological Parameters DAtabase (WWHYPDA) (Comunian and Renard 2009). Both the exPrior algorithm and functions to load and query the WWHYPDA can be accessed in the R programming language: The algorithm is implemented in package exPrior (Heße et al 2021), and the database can be accessed using geostatDB (Heße et al 2019c), an R interface for accessing hydrogeological information from the WWHYPDA. Cucchi et al (2019) recommends to derive informed priors with exPrior algorithm using data from sites that are similar to the site of interest, which we call the target site.…”