COS 114-3 - Evaluating the Effects of Habitat Stratification on Sampling Bias for Estimations of Aquatic Animal Population Metrics

Abstract: Accurately estimating the size and characteristics of animal populations is important for determining the progress of restoration programs and managing vulnerable ecological communities. Many factors can influence the results from field data collection, but a carefully planned sample design can help manage bias in sampling data. Monitoring programs supporting the Comprehensive Everglades Restoration Plan (CERP) use a Generalized Random Tessellation Stratified (GRTS) sampling design to optimize sampling efforts by creating randomized spatially balanced sample plots across regions. Although an effective tool, the GRTS sample design alone does not account for the potential bias that is created by the heterogeneous landscape of the Florida Everglades.

As part of the CERP Monitoring and Assessment Plan (MAP), field surveys of aquatic animals and periphyton are directed to areas with relatively low to medium emergent plant stem density, termed “sampleable” areas. This leaves some of the ecosystem unsampled, either because it has vegetation that is too dense for sample collection methods or are non-aquatic habitats that are inhospitable for aquatic animals. The CERP MAP for aquatic animals and periphyton was designed to statistically accommodate the dichotomy of sampleable and non-sampleable habitats. However, sampling efforts currently only take place in favorable habitat conditions (sampleable habitat), leaving animals in unsampleable aquatic habitats neglected from ecosystem assessments. In this study, we evaluate the effects of using stratified habitat data to create a weighting scheme (inclusion probability) to adjust data for bias. We simulated a range of bias to existing CERP MAP samples to explore if it has ecologically important effects on population estimates.

We report calculations that correct for spatial bias in data collection based on a new vegetation mapping effort for the 146 primary sampling units (PSUs) of the CERP monitoring program using Worldview 2 multispectral satellite imagery data. From the vegetation maps that were generated, we found that among all sample sites the relative habitat composition was 55% sampleable, 36% unsampleable, and 9% non-aquatic habitat. The weighting scheme was applied to biomass & density estimates of different species from a 15-year fish sampling data set (2006-2021). Due to the wide range of landscape heterogeneity across our sample sites, we observed significant findings of spatial and temporal variation in the magnitude of bias in the scaled-up estimates of abundance metrics. These effects also varied among species. This study illustrates the benefits of identifying and accounting for sampling and spatial bias in estimations of population data.