Roadside Car Surveys: Methodological Constraints and Solutions for Estimating Parrot Abundances across the World

Round 1

Reviewer 1 Report

The premise for the article is useful. You are correct in stating that parrots could use some easier ways to assess their populations and rightly point out the limitations of strategies that have been previously used. 

Your sample size and years of data was impressive! That was also alot of parrot encounters (>15,000 encounters and 137 species).

Authors tested whether detectability is related to species size and gregariousness, and whether it varied among biomes. They found that smaller, less gregarious species were harder to visually record. Biomes had no effect.

I see a lot of value in terms of offering a possible way to survey parrots and emphasize the loudness of parrots for the value of this sampling methodology. However; what I see missing from this is a discussion of the importance of seasonality in assessment and comparison to other forms of assessment beyond line transects (e.g. point counts) (Casagrande and Beissinger 1997, Wright, Dahlin et al. 2008). In parrots that are rare and/or congregate in roosts, I think reliance on road censuses could result in undercounts, and efforts to complete total counts of a species may be necessary, with efforts to survey locals to determine roosting sites. In addition, during the breeding season, I think road censuses could result in undercounts. Could you add a brief discussion of the limitations of this methodology and some recommendation as to in which conditions/species roadside sampling may be most appropriate? It appears you can cite yourself (Dénes, Tella et al. 2018).

Also please make sure to acknowledge the limitations of your study. You did not compare your sampling methodology to any other in this study; thus, you do not actually know if your methodology was similar, better or worse than other sampling methodologies. I think essentially the value of your study was in showing that it REALLY does not work well for certain types of gregarious/rare parrots and MAY work okay for other species. Still; it really needs to be compared to some other sampling methodologies to be sure. 

57; please clarify this sentence; does this mean that it was essentially expert guesswork?

346-353; Can you please clarify the patterns here? I found this paragraph confusing. What exactly do you mean, proportion of aural encounters? The sentences as currently written did not make the relationships obvious. Do you mean that as body mass increased, birds had a lower percentage of aural encounters because they were being seen? This is the correct interpretation, right? Also, as flock size increased, aural encounters declined because birds were more likely to be seen as well?

515; correct to “country”

Casagrande, D. G. and S. R. Beissinger (1997). "Evaluation of Four Methods for Estimating Parrot Population Size." The Condor 99(2): 445-457.

Dénes, F. V., et al. (2018). "Revisiting methods for estimating parrot abundance and population size." Emu - Austral Ornithology 118(1): 67-79.

Wright, T. F., et al. (2008). "Stability and change in vocal dialects of the yellow-naped amazon." Anim Behav 76: 1017-1027.

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 2 Report

This is an interesting manuscript and a timely one, sadly, given the proliferation of roads through large areas of habitat for keystone species such as parrots. Thus, there is an important need for assessments of the use of roads as suitable survey routes for threatened species populations, particularly those based on Distance Sampling protocols. I have some comments on the manuscript.

Distance Sampling analyses (1): Bin width selection and detection probabilities: Generally this has been done with good attention to detail, using some of the peer-review literature, something which not all users of DS attempt. So the authors deserve some praise for this. But a number of issues should be clarified for the Diversity readership. On lines 220-22 the authors write “We binned distance data for each study case to facilitate the fitting of detection functions, using three sets of breaks (25, 50, and 100 m). For each binning setup, we fitted DS models with a half-normal key function as previously recommended after visual inspection of the histograms of distances.” Firstly, I am confused by what the authors mean in this statement. My initial reading is that they varied bin width for each species, running three models for each species using 25m width bins, the second using 50m and the third using 100m bin widths. Is this correct? Then they compared between each model? Or did they vary the bin width in each model i.e. have multiple different bin widths within the same detection probability histogram? I think this needs to be clarified, preferably with examples of the detection probability histograms of some of the most commonly encountered species from each region presented as supplementary materials. This would enable the Diversity readership to give the modelling procedures the scrutiny it merits.

Distance Sampling analyses (2): Key function selection. In addition to this, the authors have relied on the default approach of half-normal key functions (as do many other authors). However, in reality, a range of key functions would need to be compared in order to estimate the most robust detection probability given variation in parrot detection and behaviour across sites within the same region, let alone between different biomes. From experience, the use of hazard rate functions is typical given the nature of the ‘tail’ in detection probability histograms for common Neotropical parrots. So I would encourage the authors to provide additional analyses, to clearly show that the half-normal key function was the most appropriate, by comparing these with uniform and hazard rate functions. To go back to my first point, presenting the detection probability histograms would enable the readership to give better scrutiny to their approach.

Distance Sampling analyses (3): Model comparisons and goodness of fit. Lines 223- 229 the authors write: We compared models with adjustment terms and with cosine, Hermite polynomial, and simple polynomial adjustments, up to order 5. For models where group size was correlated with detection distances, we also fitted a DS model with a half-normal key function and a group size covariate. Akaike’s Information Criterion (AIC) was used to compare models within a distance break set, but it cannot be used to compare models fit to data with different binning setups. Thus, we performed chi-square goodness-of-fit tests to compare the best models from each binning setup to identify the best model for each study case.

A few things to clarify here. Can the authors clarify why AIC cannot be used to compare between models with different binning set-ups? What is the evidence for that (i.e. citations)? Secondly, AIC is used to compare relative fit, whereas chi-square is used to test the fit of the key function. However, the authors do not present any QQ plots for the readership to again, give the necessary scrutiny to their approach and assess the chi-square tests. Thirdly, and more importantly (once the clarity on bin width selection is provided), given the debate around the use of roads and potential bias, the authors need to present Kolmogorov-Smirnov and Cramer-von-mises tests to assess the influence of observations close to the road/transect line (see Buckland et al., 2004) for some of the most commonly encountered species from each region presented as supplementary materials. Fourthly, in a study of this nature its important to consider whether stratifications of density parameters and estimates with pooled or global detection functions should be used to understand differences between not just the study sites but also regions for species with similar traits. Finally, was there a need (or at least a consideration) to use model weighted averaging e.g. using bootstrap resampling, as a means of improving confidence limits of probability detection functions and their density estimates (see Buckland et al., 1997).

Use of GLMs: I am a little confused by the use of GLMs here, and why relatively little is made of them in the Results section. Or are they simply been used for calculating deviance explained? In addition, linear regressions are mentioned in the Results (e.g. line 422) but little mentioned in the Methods (or have I misunderstood or overlooked something). Thus I am left wondering about how many models were constructed and whether or not the most parsimonious model has been identified. Wouldn’t GLMMs be more appropriate here, maybe even LMMs to account for Biomes and even perhaps road/transect as a nested factor (particularly if Biomes do not have a significant effect - see lines 445-448). Although I do see that the authors state that a lack of model convergence has influenced their decision to use GLMs, but rescaling some of the parameters around the fixed effects (i.e. dealing with the unique species values in identity or flock size) or trying different optimizers in some R packages (e.g. lme4) or using quasi-likelihood estimators in the R package nlme could help resolve issues around model convergence. If the lack of convergence was due to an excess of zero values in the dataset, then these could be resolved by using zero-inflated models. Either way, there are options to explore.

Minor comments: Throughout the manuscript, the authors use two numbers after the decimal place, but in many instances these will need to be rounded up to make it more ecologically meaningful. For example on lines 413-414, a density estimate of 0.05 is next to nothing and not ecologically meaningful so, round up the numbers to bring them in line with the already published DS literature.

A good manuscript, of interest to the readership. The authors need to clarify and further justify/expand their Distance Sampling data analyses and reanalyse these, but also simply demonstrate/clarify that the use of GLMs was appropriate. Thus recommend major revision. 

Author Response

See attachment

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

The authors have been very conscientious in dealing with my previous comments and I thank them for this. I think the authors have addressed all of my concerns, only some very minor spelling and grammar issues to check throughout.