TY - JOUR
T1 - Analyzing spatiotemporal species spread by three declustering methods utilizing monitoring data based on national programs and citizen science
AU - Goldshtein, Eitan
AU - Soroker, Victoria
AU - Mandelik, Yael
AU - Sadeh, Asaf
AU - Haberman, Ami
AU - Ezra, Nadav
AU - Cohen, Yafit
N1 - Publisher Copyright:
© 2022
PY - 2022/12
Y1 - 2022/12
N2 - Systematic and dynamic monitoring of species is often performed at preferential locations. Thus, while models of spatial distributions assume evenly distributed spatial sampling, sampling designs are often clustered, resulting in biased estimates. Therefore, spatial declustering methods should be used to reduce bias, for example when quantifying the spread of invasive species. However, little is known about the relative efficiency of such methods. In this study, the spatiotemporal spread of a species was estimated from synthetic data and from field-collected data, using three alternative declustering methods. We compared two grid-based methods that involve data depletion, and a polygon area weighting method (Multiplicatively Weighted Buffer-Thiessen Polygons; MWBTP), in which constrained weights account for spatial redundancy without any loss of data. We applied the effective range radius method to estimate spread rate from range areas in different time steps. The spread rate was determined as a linear regression slope between the effective range radius (range area/π) and time. Range areas were produced using 95% isopleth of kernel density estimation, with either fixed or adaptive bandwidths. Directional spread was quantified using the displacement of the population's weighted mean centers over time, and the direction of spread was determined using their standard deviational ellipse. We simulated spatially clustered sampling data, and compared the spread rates based on each of the three declustering methods, relative to that measured under an evenly distributed sampling. MWTBP best estimated the spread rate using adaptive-bandwidth KDE (−2.2% relative difference from the even distribution), whereas the clustered distribution produced a −34.4% relative difference. The grid-based methods produced relative differences of −19.6% and −28.5%. Our field data span two years from an Israeli national monitoring program of the invasive red palm weevil, Rhynchophorus ferrugineus, in which weevil traps were distributed in clusters. The grid-based methods showed small relative differences of up to 2% in spread rate, compared to uncorrected estimates, with MWBTP resulting in the greatest difference (−7% relative difference). The spread direction as evaluated by MWBTP deflected 18.3° from that of the uncorrected method and best reflected the actual spread direction. In conclusion, MWBTP appears to best overcome clustering, and yields superior or similar results to those of the other, more commonly-used methods. Databases from field monitoring programs are accumulating at accelerating rates, commonly with spatial clustering. Thus, we suggest further development and application of declustering methods on such spatial data to reduce biases in ecological and agricultural research and practice.
AB - Systematic and dynamic monitoring of species is often performed at preferential locations. Thus, while models of spatial distributions assume evenly distributed spatial sampling, sampling designs are often clustered, resulting in biased estimates. Therefore, spatial declustering methods should be used to reduce bias, for example when quantifying the spread of invasive species. However, little is known about the relative efficiency of such methods. In this study, the spatiotemporal spread of a species was estimated from synthetic data and from field-collected data, using three alternative declustering methods. We compared two grid-based methods that involve data depletion, and a polygon area weighting method (Multiplicatively Weighted Buffer-Thiessen Polygons; MWBTP), in which constrained weights account for spatial redundancy without any loss of data. We applied the effective range radius method to estimate spread rate from range areas in different time steps. The spread rate was determined as a linear regression slope between the effective range radius (range area/π) and time. Range areas were produced using 95% isopleth of kernel density estimation, with either fixed or adaptive bandwidths. Directional spread was quantified using the displacement of the population's weighted mean centers over time, and the direction of spread was determined using their standard deviational ellipse. We simulated spatially clustered sampling data, and compared the spread rates based on each of the three declustering methods, relative to that measured under an evenly distributed sampling. MWTBP best estimated the spread rate using adaptive-bandwidth KDE (−2.2% relative difference from the even distribution), whereas the clustered distribution produced a −34.4% relative difference. The grid-based methods produced relative differences of −19.6% and −28.5%. Our field data span two years from an Israeli national monitoring program of the invasive red palm weevil, Rhynchophorus ferrugineus, in which weevil traps were distributed in clusters. The grid-based methods showed small relative differences of up to 2% in spread rate, compared to uncorrected estimates, with MWBTP resulting in the greatest difference (−7% relative difference). The spread direction as evaluated by MWBTP deflected 18.3° from that of the uncorrected method and best reflected the actual spread direction. In conclusion, MWBTP appears to best overcome clustering, and yields superior or similar results to those of the other, more commonly-used methods. Databases from field monitoring programs are accumulating at accelerating rates, commonly with spatial clustering. Thus, we suggest further development and application of declustering methods on such spatial data to reduce biases in ecological and agricultural research and practice.
KW - Biodiversity monitoring
KW - Grid
KW - Invasion dynamics
KW - Kernel density estimation
KW - Multiplicatively Weighted Buffer-Thiessen Polygons
KW - Rhynchophorus ferrugineus
UR - http://www.scopus.com/inward/record.url?scp=85142138144&partnerID=8YFLogxK
U2 - 10.1016/j.ecoinf.2022.101916
DO - 10.1016/j.ecoinf.2022.101916
M3 - ???researchoutput.researchoutputtypes.contributiontojournal.article???
AN - SCOPUS:85142138144
SN - 1574-9541
VL - 72
JO - Ecological Informatics
JF - Ecological Informatics
M1 - 101916
ER -