Designing drone surveys for N-mixture models

• Description
• Simulation experiments
• Repository contents
• Running code

Performance and optimal design of N-mixture models for spatiotemporally replicated drone-based surveys

Description

Repository for the manuscript OPTIMALLY DESIGNING DRONE SURVEYS FOR WILDLIFE ABUNDANCE MODELING WITH N-MIXTURE MODELS (Brack et al. 2022; Methods in Ecology and Evolution).
This repository contains the simulation results and R code to explore, in a very wide scan study, the performance (based on root mean squared error) and optimal survey effort allocation for hierarchical N-mixture models, focusing on their application for drone-based surveys. We also investigate the use of a double-observer protocol in image reviewing to decompose the detection process in availability and perception.
We provide an interactive and “ready-to-consult” webpage hoping to assist ecologists and conservationists in panning spatiotemporally replicated drone-based surveys for abundance modeling. There we show in detail simulation results and run some examples.

Simulation experiments

The simulation study is divided in three parts:

• Part 1: optimal design of count surveys for N-mixture abundance estimation

  This first experiment assesses the performance of N-mixture models using double and single observer counts under different scenarios of population density and detection probability and address the optimal survey effort allocation in terms of spatial vs. temporal prioritization for each scenario.

• Part 2: exploring the benefit of the double-observer protocol

  In this experiment, we investigate how the use of double-observer protocol increases model performance and affects optimal survey effort allocation.

• Part 3: reducing fieldwork effort by employing a double-observer protocol

  Here, we evaluate if the use of double-observer protocol can reduce the effort needed in fieldwork to achieve the same model performance as in a single observer approach.

Repository contents

• data/: simulation results files
  • raw_simul_resu/: raw simulation results are not available in this repo (download info below)
  • processed_simul_resu/: extracted and cleaned simulation results
• docs/: files for the website
• ms/: manuscript files
• outputs/: figures, tables and intermediate outputs
  • figs/: figures built with R scripts. RMSE curves for scenarios and RMSE relationships
  • tabs/: tables built with R scripts. Optimal J tables and budget savings table
• R/: R scripts and functions. Script file names reference to each simulation experiment with Part or P
  • ex.sumulation*.R: script to run a single iteration simulating spatiotemporally replicated counts and analyzing with the respective N-mixture model. Examples with maximum likelihood estimation (*_likeli.R) and Bayesian approach (*_bayes.R).
  • func_*.R: functions to support R scripts
  • simulNmix_scenarios_parallel.R: script to run several simulation iterations. This script was used to produce raw simulation results using maximum likelihood estimation.
  • Part*/: code for each simulation experiment
    • C&E_.R*: clean and extract raw simulation results and save in the processed folder
    • calcRMSE_*.R: code to calculate the RMSE for each scenario (under a combination of local abundance and availability) and get the optimal number of visits for each scenario.
    • fig*.R: code to generate figures

Running code

Download this repo

download.file(url = "https://github.com/ismaelvbrack/designNmix4droneSurveys/archive/main.zip", destfile = "designNmix4droneSurveys.zip")

Download raw simulation results

To reproduce the results from raw simulation results, download them from zenodo (645 Mb).
Otherwise, reproduce starting from processed simulation results (cleaned and extracted, available here) directly from this repo.

Manual download here and paste in ~/data/raw_simul_resu

or

Download from RStudio:

library(zen4R)
download_zenodo(doi="10.5281/zenodo.5156592", path=here::here("data","raw_simul_resu"))