CLIPPED MULTISCALE SPATIAL PROCESSES FOR MULTIVARIATE PLANT COVER DATA

Wilson J. Wright

doi:10.1214/25-AOAS2088

CLIPPED MULTISCALE SPATIAL PROCESSES FOR MULTIVARIATE PLANT COVER DATA

2025-12-01

0

PRE

AI

DOI：10.1214/25-AOAS2088

Original

Original request for help

Save

Abstract

En 中文

Monitoring the distribution and abundance of plants is important for assessing the environmental impact of anthropogenic activity. Plant cover data are commonly collected using the point intercept method which records species presence or absence at points on a grid overlaying sampled plots. Multiple species are surveyed simultaneously, resulting in binary data that are both multiscale and multivariate. Fine scale and multivariate characteristics are not accounted for with current spatial models for binary data. For instance, conventional analyses of point intercept data discard the subplot scale information by aggregating observations to the plot level. We develop an alternative model for plant cover using clipped multiscale spatial Gaussian processes. Our model allows us to leverage the spatial configuration of observed points within a plot to model small-scale spatial structure in plant cover as well as the large-scale spatial structure among plots. To analyze data from multiple species, we include interspecies correlations that describe community structure. We develop a computationally efficient algorithm for fitting our model using Bayesian methods. We apply our model to analyze lichen and plant cover data collected at Cape Krusenstern National Monument, Alaska, U.S.A., and assess how the vegetation in this region is impacted by heavy metal pollution.

Keywords:

Bayesian statistics

community ecology

hierarchical model

joint species distribution models

point intercept method

vegetation data

CLIPPED MULTISCALE SPATIAL PROCESSES FOR MULTIVARIATE PLANT COVER DATA

Abstract

Journal

Organization