Studies that use radio-tracking to reveal social structure and habitat use in populations of small and medium-sized mammals face a trade-off between number of location data (n) and monitoring of many individuals, to maximize efficiency.We simulated these conditions using location data from 30 radio-collared red squirrels, subsampled at different percentages of total number of locations and tested the performance of four home range estimators. Two linkage estimators, the minimum convex polygon (MCP), and the incremental clusteranalysis polygon (ICP) and two probability density estimators, the fixed kernel density estimation with reference smoothing factor (KDE with href), and with least squares crossvalidation to calculate smoothing factor (KDE LSCV with hlscv). KDE produced the largest home range estimates, MCP and KDE LSCV intermediate estimates, and ICP the smallest ones. Differences between estimators were larger at smaller n, but consistent throughout the entire range of locations (16–74) in our data set. Although KDE is widely used and LSCV is widely recommended to calculate bandwidth, our results confirmed that the value of h has a considerable influence on the home range estimate and varied more strongly when sample size (n) decreased. Our models showed that overestimation with KDE could be avoided by applying the average ratio of hlscv/href (in our case 0.75) as a multiplier of href and use this recalculated bandwidth to produce more reliable home range and core area estimates (KDEadj). MCP and KDE had lower variability than KDE LSCV and ICP. Stability improved with sample size and tended towards an asymptote at more than 60 locations for MCP and KDE. We conclude that high variation in ICP and KDE LSCV at small n limits their applicability to few situations (n > 70, landscapes with distinct habitat patches where ranges consists of several, separated cores). We recommend use of both MCP and KDEadj for home range size and KDEadj for core area size and propose to estimate a ‘best core area’ based on 85% MCP when a home range is mononuclear and 85% ICP when it is multinuclear.
Radio-tracking squirrels: performance of home range density and linkage estimators with small range and sample size
WAUTERS, LUCAS ARMAND;PREATONI, DAMIANO;TOSI, GUIDO
2007-01-01
Abstract
Studies that use radio-tracking to reveal social structure and habitat use in populations of small and medium-sized mammals face a trade-off between number of location data (n) and monitoring of many individuals, to maximize efficiency.We simulated these conditions using location data from 30 radio-collared red squirrels, subsampled at different percentages of total number of locations and tested the performance of four home range estimators. Two linkage estimators, the minimum convex polygon (MCP), and the incremental clusteranalysis polygon (ICP) and two probability density estimators, the fixed kernel density estimation with reference smoothing factor (KDE with href), and with least squares crossvalidation to calculate smoothing factor (KDE LSCV with hlscv). KDE produced the largest home range estimates, MCP and KDE LSCV intermediate estimates, and ICP the smallest ones. Differences between estimators were larger at smaller n, but consistent throughout the entire range of locations (16–74) in our data set. Although KDE is widely used and LSCV is widely recommended to calculate bandwidth, our results confirmed that the value of h has a considerable influence on the home range estimate and varied more strongly when sample size (n) decreased. Our models showed that overestimation with KDE could be avoided by applying the average ratio of hlscv/href (in our case 0.75) as a multiplier of href and use this recalculated bandwidth to produce more reliable home range and core area estimates (KDEadj). MCP and KDE had lower variability than KDE LSCV and ICP. Stability improved with sample size and tended towards an asymptote at more than 60 locations for MCP and KDE. We conclude that high variation in ICP and KDE LSCV at small n limits their applicability to few situations (n > 70, landscapes with distinct habitat patches where ranges consists of several, separated cores). We recommend use of both MCP and KDEadj for home range size and KDEadj for core area size and propose to estimate a ‘best core area’ based on 85% MCP when a home range is mononuclear and 85% ICP when it is multinuclear.
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
