Using spatially balanced vegetation mapping to improve aquatic animal biomass estimates

Nicole Strickland, Biology, Florida International University, North Miami Beach, FL, Joel Trexler and Evelyn Gaiser, Department of Biological Sciences, Florida International University, Miami, FL, Daniel Gann, GIS and Remote Sensing Center, Florida International University, Miami, FL

Background/Question/Methods
One of the many challenges for long-term, large-scale monitoring programs is selecting a sample design for estimation of population metrics and their assessment for diverse and dynamic habitats. This is a major challenge for monitoring and assessment of the Comprehensive Everglades Restoration Plan (CERP) because success will include the re-distribution of habitats across the landscape. Field surveys to sample aquatic animals and periphyton are allocated to areas with relatively low to medium emergent plant stem density, termed “sampleable” areas. This leaves about 50% of the ecosystem non-sampled either because it has vegetation that is too dense for travel or are non-aquatic habitats that are inhospitable for aquatic animals. The CERP Monitoring and Assessment Plan (MAP) for aquatic animals and periphyton was designed to accommodate the dichotomy of sampleable and non-sampleable habitats. Doing so, improves the accuracy of abundance estimates when scaled up to the landscape area. In this presentation, we report a new vegetation mapping effort of the CERP MAP aquatic animal and periphyton monitoring primary sampling units (PSUs) using Worldview 2 satellite imagery data. We will illustrate the implication for scaling up aquatic animal data with and without habitat stratification.
Results/Conclusions
Using a random forest classification algorithm, we mapped vegetation habitats in 146 PSUs (800m2) from multi-spectral satellite data. PSUs were selected by a generalized random tessellation stratified (GRTS) sampling design to provide a spatially balanced representation of the Everglades ecosystem. The vegetation maps allowed us to separate sampleable from non-sampleable aquatic habitat to then extrapolate fish biomass and density data sampled at the 1-m2 local scale to the PSU regional scale using the Horvitz-Thompson estimator. We scaled up surveyed fish density to the regional scale using the weighted probability for fish population of a PSU based on the size of sampleable habitats and two different weights for non-sampleable habitats. We assumed that non-sampleable habitats contain fish density that is equivalent to either 20% or 90% of sampleable habitats. We found that that when non-sampleable habitat had 20% fewer fish than sampleable habitat, scaled up abundance estimates were 52% less than estimates if both habitat classes were assumed to be equal. There was 14% fewer fish than estimated by equal density when non-sampleable habitat held 90% as many fish as sample habitat. This illustrates the need for targeted sampling in areas where samples are difficult to obtain, followed by scaling up data based on relative habitat coverage.