Effects of stratification and misspecification of covariates on species distribution models for abundance estimation from virtual line transect survey data
The aim of this study was to examine the effects of stratification of the survey region on the performance of species distribution models (SDMs) described by generalized linear models or generalized additive models when estimating school abundance by using a line transect survey. True covariates that define spatial school distribution are not always obtainable explanatory variables. When the true covariates differ from explanatory variables in the model, the explanatory variables are determined to be misspecified. We evaluated the performance of SDMs in abundance estimation with misspecified covariates by using dummy datasets for which the true abundance was known. Simulated replicates of spatial distributions of a whale school and sighting data were generated from possible scenarios motivated by the spatial school distribution of Antarctic minke whales Balaenoptera bonaerensis. This distribution was obtained from the Japanese Whale Research Program under Special Permit in the Antarctic. Our results showed that the relative bias of the abundance estimators was large when covariates were misspecified and a survey region was stratified. Although stratification of the survey region is intended to produce a conventional line transect estimator with a smaller variance than that of non-stratified survey region, it also acts to increase the bias of the abundance estimate obtained from SDMs.
%0 Journal Article
%1 shibata_effects_2013
%A Shibata, Yasutoki
%A Matsuishi, Takashi
%A Murase, Hiroto
%A Matsuoka, Koji
%A Hakamada, Takashi
%A Kitakado, Toshihide
%A Matsuda, Hiroyuki
%D 2013
%J Fisheries Science
%K Balaenoptera, Ecology, Food Freshwater Marine Science, Statistical \& design, indictators, management model, stratified transects, wildlife
%N 4
%P 559--568
%R 10.1007/s12562-013-0634-5
%T Effects of stratification and misspecification of covariates on species distribution models for abundance estimation from virtual line transect survey data
%U http://link.springer.com/article/10.1007/s12562-013-0634-5
%V 79
%X The aim of this study was to examine the effects of stratification of the survey region on the performance of species distribution models (SDMs) described by generalized linear models or generalized additive models when estimating school abundance by using a line transect survey. True covariates that define spatial school distribution are not always obtainable explanatory variables. When the true covariates differ from explanatory variables in the model, the explanatory variables are determined to be misspecified. We evaluated the performance of SDMs in abundance estimation with misspecified covariates by using dummy datasets for which the true abundance was known. Simulated replicates of spatial distributions of a whale school and sighting data were generated from possible scenarios motivated by the spatial school distribution of Antarctic minke whales Balaenoptera bonaerensis. This distribution was obtained from the Japanese Whale Research Program under Special Permit in the Antarctic. Our results showed that the relative bias of the abundance estimators was large when covariates were misspecified and a survey region was stratified. Although stratification of the survey region is intended to produce a conventional line transect estimator with a smaller variance than that of non-stratified survey region, it also acts to increase the bias of the abundance estimate obtained from SDMs.
@article{shibata_effects_2013,
abstract = {The aim of this study was to examine the effects of stratification of the survey region on the performance of species distribution models (SDMs) described by generalized linear models or generalized additive models when estimating school abundance by using a line transect survey. True covariates that define spatial school distribution are not always obtainable explanatory variables. When the true covariates differ from explanatory variables in the model, the explanatory variables are determined to be misspecified. We evaluated the performance of SDMs in abundance estimation with misspecified covariates by using dummy datasets for which the true abundance was known. Simulated replicates of spatial distributions of a whale school and sighting data were generated from possible scenarios motivated by the spatial school distribution of Antarctic minke whales Balaenoptera bonaerensis. This distribution was obtained from the Japanese Whale Research Program under Special Permit in the Antarctic. Our results showed that the relative bias of the abundance estimators was large when covariates were misspecified and a survey region was stratified. Although stratification of the survey region is intended to produce a conventional line transect estimator with a smaller variance than that of non-stratified survey region, it also acts to increase the bias of the abundance estimate obtained from SDMs.},
added-at = {2017-01-09T13:57:26.000+0100},
author = {Shibata, Yasutoki and Matsuishi, Takashi and Murase, Hiroto and Matsuoka, Koji and Hakamada, Takashi and Kitakado, Toshihide and Matsuda, Hiroyuki},
biburl = {https://www.bibsonomy.org/bibtex/215a65f636e9685de7f5638262aedbe13/yourwelcome},
doi = {10.1007/s12562-013-0634-5},
interhash = {fa51071c5907855822e97721645f5417},
intrahash = {15a65f636e9685de7f5638262aedbe13},
issn = {0919-9268, 1444-2906},
journal = {Fisheries Science},
keywords = {Balaenoptera, Ecology, Food Freshwater Marine Science, Statistical \& design, indictators, management model, stratified transects, wildlife},
language = {en},
month = jul,
number = 4,
pages = {559--568},
timestamp = {2017-01-09T14:01:11.000+0100},
title = {Effects of stratification and misspecification of covariates on species distribution models for abundance estimation from virtual line transect survey data},
url = {http://link.springer.com/article/10.1007/s12562-013-0634-5},
urldate = {2013-08-28},
volume = 79,
year = 2013
}