A comparison of search strategies to design the cokriging neighborhood for predicting coregionalized variables

Nasser Madani, Xavier Emery

Research output: Contribution to journalArticle

  • 1 Citations

Abstract

Cokriging allows predicting coregionalized variables from sampling information, by considering their spatial joint dependence structure. When secondary covariates are available exhaustively, solving the cokriging equations may become prohibitive, which motivates the use of a moving search neighborhood to select a subset of data, based on their closeness to the target location and the screen effect approximation. This paper investigates the efficiency of different strategies for designing a sub-optimal neighborhood wherein the simplification of the cokriging equations is challenging. To do so, five alternatives (single search, multiple search, strictly collocated search, multi-collocated search and isotopic search) are tested and compared with the reference unique neighborhood, through synthetic examples with different data configurations and spatial joint correlation models. The results indicate that the multi-collocated and multiple searches bear the highest resemblance to the reference case under the analyzed spatial structure models, while the single and the isotopic searches, which do not differentiate the primary and secondary sampling designs, yield the poorest results in terms of cokriging error variance.

LanguageEnglish
Pages1-17
Number of pages17
JournalStochastic Environmental Research and Risk Assessment
DOIs
Publication statusAccepted/In press - Jul 6 2018

Fingerprint

Sampling
Model structures
sampling
comparison
effect

Keywords

  • Cokriging neighborhood
  • Heterotopic sampling
  • Intrinsic correlation
  • Markov-type models
  • Multi-collocated cokriging
  • Screening effect
  • Strictly collocated cokriging

ASJC Scopus subject areas

  • Environmental Engineering
  • Environmental Chemistry
  • Water Science and Technology
  • Safety, Risk, Reliability and Quality
  • Environmental Science(all)

Cite this

@article{a92e3559fbce4f69ad34f5edf709ab3e,
title = "A comparison of search strategies to design the cokriging neighborhood for predicting coregionalized variables",
abstract = "Cokriging allows predicting coregionalized variables from sampling information, by considering their spatial joint dependence structure. When secondary covariates are available exhaustively, solving the cokriging equations may become prohibitive, which motivates the use of a moving search neighborhood to select a subset of data, based on their closeness to the target location and the screen effect approximation. This paper investigates the efficiency of different strategies for designing a sub-optimal neighborhood wherein the simplification of the cokriging equations is challenging. To do so, five alternatives (single search, multiple search, strictly collocated search, multi-collocated search and isotopic search) are tested and compared with the reference unique neighborhood, through synthetic examples with different data configurations and spatial joint correlation models. The results indicate that the multi-collocated and multiple searches bear the highest resemblance to the reference case under the analyzed spatial structure models, while the single and the isotopic searches, which do not differentiate the primary and secondary sampling designs, yield the poorest results in terms of cokriging error variance.",
keywords = "Cokriging neighborhood, Heterotopic sampling, Intrinsic correlation, Markov-type models, Multi-collocated cokriging, Screening effect, Strictly collocated cokriging",
author = "Nasser Madani and Xavier Emery",
year = "2018",
month = "7",
day = "6",
doi = "10.1007/s00477-018-1578-1",
language = "English",
pages = "1--17",
journal = "Stochastic Environmental Research and Risk Assessment",
issn = "1436-3240",
publisher = "Springer New York",

}

TY - JOUR

T1 - A comparison of search strategies to design the cokriging neighborhood for predicting coregionalized variables

AU - Madani, Nasser

AU - Emery, Xavier

PY - 2018/7/6

Y1 - 2018/7/6

N2 - Cokriging allows predicting coregionalized variables from sampling information, by considering their spatial joint dependence structure. When secondary covariates are available exhaustively, solving the cokriging equations may become prohibitive, which motivates the use of a moving search neighborhood to select a subset of data, based on their closeness to the target location and the screen effect approximation. This paper investigates the efficiency of different strategies for designing a sub-optimal neighborhood wherein the simplification of the cokriging equations is challenging. To do so, five alternatives (single search, multiple search, strictly collocated search, multi-collocated search and isotopic search) are tested and compared with the reference unique neighborhood, through synthetic examples with different data configurations and spatial joint correlation models. The results indicate that the multi-collocated and multiple searches bear the highest resemblance to the reference case under the analyzed spatial structure models, while the single and the isotopic searches, which do not differentiate the primary and secondary sampling designs, yield the poorest results in terms of cokriging error variance.

AB - Cokriging allows predicting coregionalized variables from sampling information, by considering their spatial joint dependence structure. When secondary covariates are available exhaustively, solving the cokriging equations may become prohibitive, which motivates the use of a moving search neighborhood to select a subset of data, based on their closeness to the target location and the screen effect approximation. This paper investigates the efficiency of different strategies for designing a sub-optimal neighborhood wherein the simplification of the cokriging equations is challenging. To do so, five alternatives (single search, multiple search, strictly collocated search, multi-collocated search and isotopic search) are tested and compared with the reference unique neighborhood, through synthetic examples with different data configurations and spatial joint correlation models. The results indicate that the multi-collocated and multiple searches bear the highest resemblance to the reference case under the analyzed spatial structure models, while the single and the isotopic searches, which do not differentiate the primary and secondary sampling designs, yield the poorest results in terms of cokriging error variance.

KW - Cokriging neighborhood

KW - Heterotopic sampling

KW - Intrinsic correlation

KW - Markov-type models

KW - Multi-collocated cokriging

KW - Screening effect

KW - Strictly collocated cokriging

UR - http://www.scopus.com/inward/record.url?scp=85049590303&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85049590303&partnerID=8YFLogxK

U2 - 10.1007/s00477-018-1578-1

DO - 10.1007/s00477-018-1578-1

M3 - Article

SP - 1

EP - 17

JO - Stochastic Environmental Research and Risk Assessment

T2 - Stochastic Environmental Research and Risk Assessment

JF - Stochastic Environmental Research and Risk Assessment

SN - 1436-3240

ER -