Does the Presence of Shade Trees and Distance to the Forest Affect Detection Rates of Terrestrial Vertebrates in Coffee Home Gardens?

Round 1

Reviewer 1 Report

A very serious and worked article that I think I would improve by adding the following comments: Not detecting the most interesting species for conservation such as Panthera pardus or Presbytis comata in the agroforestry matrix is ​​an important result that should be highlighted as such. The replacement of the forest by the agroforestry matrix should be pointed out as detrimental to most of the species except the most adaptable or not strictly forestry. The role of the agroforestry matrix as beneficial to the leopard as a buffer zone should be stated in this context

Author Response

A very serious and worked article that I think I would improve by adding the following comments: Not detecting the most interesting species for conservation such as Panthera pardus or Presbytis comata in the agroforestry matrix is ​​an important result that should be highlighted as such. The replacement of the forest by the agroforestry matrix should be pointed out as detrimental to most of the species except the most adaptable or not strictly forestry. The role of the agroforestry matrix as beneficial to the leopard as a buffer zone should be stated in this context.

We thank the reviewer for acknowledging the importance of our study. We agree we highlighted more the positive effects of the agroforest but we did not highlight enough the two missing species (very important ones) in the agroforest matrix. We have now rephrased key parts (abstract, beginning of the discussion, and conclusion) and expanded the dedicated paragraph in the discussion to highlight the missed detection of some of the key species.

Reviewer 2 Report

This study aims at assessing detection rates in forest and home gardens under the assumption that these detection rates would then be representative for habitat use. The study has however two shortcomings. First, there is no detailed description in the methods how cameras were installed and it seems that there was no attempt of any quality control. If cameras are installed in more open habitats, the area surveyed might be larger, so detection rates can be higher only because the area is larger and not because the habitat is selected. The authors therefore need to show that detection rates are actually proportional to abundance, e.g. if the area surveyed is standardised or by using camera trap distance sampling, which would correct for differences in the area surveyed. Second, detection in a habitat does not mean that this is a good habitat. For example, if mortality in a given habitat is high, attraction to this can have negative population effects. Therefore, pure detection rates do not help in assessing the quality of a habitat. The study design thus does not help in assessing the value of home gardens. I therefore think that at least some of the data are overanalysed and that the authors need first to clarify the two concerns mentioned above and then accordingly adjust the analyses.

A few minor comments:

Line 130. Birds are also “terrestrial vertebrates”. Change to “mammals”.

Table 1. Here is a confusion about sample size. N is not the number of detections but the number of different sites, thus n is the same for all species. Which n did you use for statistical testing? Why is there no SE for forest sites? Because these were only 2 sites? In this case how do you know the sample size is sufficient? I would anyways recommend using confidence intervals for both because you can immediately see if they are different or not and no further testing is necessary

Fig 3: These are very strange daily activity patterns, why only from 0 to 12 hours? Maybe you wrongly labelled the x axis?

Author Response

This study aims at assessing detection rates in forest and home gardens under the assumption that these detection rates would then be representative for habitat use. The study has however two shortcomings. First, there is no detailed description in the methods how cameras were installed and it seems that there was no attempt of any quality control. If cameras are installed in more open habitats, the area surveyed might be larger, so detection rates can be higher only because the area is larger and not because the habitat is selected. The authors therefore need to show that detection rates are actually proportional to abundance, e.g. if the area surveyed is standardised or by using camera trap distance sampling, which would correct for differences in the area surveyed.

We thank the reviewer for spotting this limitation with the data analysis. We did indeed tried to standardise the data collection as much as possible by placing the camera traps in the similar position and at the same height, and considering a similar open area between camera traps (we have now added more information in section 2.2). But theory is a thing, practice is another. We thus checked the data and there is some variation in the maximum detection distance between camera traps (mean: 5.3 ± SD 1.8). Since we were not able to identify the traits to be able to recognise different individuals, and we did not plan the study to meet the assumptions of the alternative methods to estimate density of unmarked populations, we did not go for a distance sampling method as that would have been biased. We thus kept the same analysis on detection rates via GAMs but we included the maximum detection distance for each camera in the model to control for this potential confounding factor. The results changed only by a little, and the significant results remained. The reviewer can find the edited results plus the highlighted addition in the methods explaining that we included the control variable.

Second, detection in a habitat does not mean that this is a good habitat. For example, if mortality in a given habitat is high, attraction to this can have negative population effects. Therefore, pure detection rates do not help in assessing the quality of a habitat.

For this point we cannot do much in terms of data analysis as we do not have the possibility to include mortality risk in the different gardens surveyed. As we argue in the discussion, however, the hunting pressure in the area is very limited, so we have no reason to believe that animals are attracted to areas where they can be hunted by humans (the main potential threat for most of the species). We do not have any reason to believe that higher detection rates can be related to a higher attraction to areas where the mortality is high. We did however include a statement in the discussion highlighting the possible limitations of our study, but also the common use of detection rates to compare animal use of different habitats or general spatial and temporal trends. This is reviewed in Broadley et al. (2019), where they suggest that detection rates can be still suitable for studies with a similar approach that we use.

The study design thus does not help in assessing the value of home gardens. I therefore think that at least some of the data are overanalysed and that the authors need first to clarify the two concerns mentioned above and then accordingly adjust the analyses.

We hope the reviewer will appreciate our efforts in trying to fix the issues addressed both in terms of data analysis and discussion.

 

A few minor comments:

 

Line 130. Birds are also “terrestrial vertebrates”. Change to “mammals”.

Since we included the terrestrial bird Barred buttonquail, we changed the sentence to “In this analysis we focused on terrestrial mammals and birds.“

Table 1. Here is a confusion about sample size. N is not the number of detections but the number of different sites, thus n is the same for all species. Which n did you use for statistical testing? Why is there no SE for forest sites? Because these were only 2 sites? In this case how do you know the sample size is sufficient? I would anyways recommend using confidence intervals for both because you can immediately see if they are different or not and no further testing is necessary

The camera traps in the forest were two and we used one-sample t-tests to compare the detection rates in coffee gardens with the detection rate in the forest. As reference value we used the mean value of the two cameras (the value that we show in table 1). We did not add a SE as we report that we are doing a one-sample t-test, so for consistency we report the reference value. In the one-sample t-test, in fact, a population (in our case the cameras in coffee gardens) is compared to a single reference value that represent the mean of a distribution.  We did not do a t-test for independent samples as they require at least 5 values to calculate the variance. We thus used the only test available for our data.

Fig 3: These are very strange daily activity patterns, why only from 0 to 12 hours? Maybe you wrongly labelled the x axis?

We thank the reviewer for spotting the mislabelled axis. Now changed.

Round 2

Reviewer 2 Report

The revision has dealt with some of my concerns in the first review but in my opinion the analyses need some more work to make the results reliable and transparent. You now report maximal detection distances for camera traps that vary from 3 to 10 m (in line 157 you need to add m to the values). As this is a more than threefold range, please provide averages for the forest sites, home garden sites in shade and home garden sites in the sun (with confidence intervals) to show that there is no bias. If you can assess maximal distances for each camera site, why can’t you use this to estimate as well distances to individuals so that you can use point count distance sampling to estimate abundances? I additionally propose modifications of the analyses presented in the tables and figures:

Table 1. The number of detections is irrelevant here as this number is included in the calculation of the average detection rate. Also the information “gardens (%)” is not useful here as this is only an indirect info. I propose you change the whole table to make it more informative and more transparent by providing average detection rates with 95% confidence intervals (you can also calculate these for just 2 samples) for forest sites, home garden sites in shade and home garden sites in the sun. This is much more useful than significance testing, given the small sample size in forest. If you do this, you can also delete table 2 as already all information is available in table 1. It is anyway not necessary to use individual tests here as the averages with confidence intervals already provide all necessary information.

Figure 1. A and B. There is no reason to believe that the detection rate would actually follow such a pattern, why would the detection be high at the forest edge, then decrease, then increase again at mid distances and then decrease again? It is more likely that this is a just a coincidental pattern caused by small sample size. For plots C and D (and B if you want to keep it), you need to add individual data points (one point for each site) so that the reader can see variation. You need also to specify what the variation bands are, you should use 95% confidence bands here. You should put all species for which you detected an effect of distance to forest edge in this figure.

Figure 2. This figure does not provide much useful information and should be deleted.

Figure 3. The number of detections is not useful here, you need to standardise the data as the more encountered species have higher peaks, which is not actually representative for the intensity of activity. The best way to standardise your data would be to use the proportion of detections of a given species for each hour (detections in each hour divided by all detections of the given species). Add here also the number of detections for each species here, so that the reader gets an idea of the sample size.

Author Response

Please see attached file.

Author Response File: Author Response.pdf

Round 3

Reviewer 2 Report

The authors did a good job dealing with all my comments, just a few more suggestions:

Line 163. You should provide a confidence interval if you use plus/minus, this is anyways more informative here

Table 1. As the confidence intervals are for means they are symmetrical, so you can write average plus/minus interval. It would be more informative to separate garden in the shade and in the sun

I would replace the passive voice by the active voice throughout the manuscript.

Author Response

We thank the reviewer for the useful suggestions and for acknowledging our efforts in trying to deal with his comments. We think the paper really improved with his suggestions, and we accepted the final suggestions asked.