Also, just a general comment: some less epxensive cameras peform very well but may be more prone to 'glitches' over the duration of an extended study. I think the expectation of long term reliability is part of the reason some people choose the expenisve brand. Systematic tests of long duration reliability in field conditions would be really interesting, albeit probably too difficult/expensive to achieve.

Thanks for the reply Chris!

When you state:

"Camelot has two modes for importing data: a bulk import, and a per-session import.  It sounds like you have many images up-front, and so bulk import may be the way to go.  In this case Camelot can create the camera trap stations based on the location of the images, and their metadata."

When you mention 'location' you refer to the location of the images from where they are imported from? And not the gps location from the EXIF data if it's available, right? In my case no gps data would be available. But if I've downloaded the images to a hard drive in the field, and then back in the office import them into Camelot I can tell Camelot that the pictures from folder X belong to camera station 1?

You're right about the two points for optimizing. If photos can be ordered by station, date, time, and reference images ordered by 'distance to station', it greatly reduces the number of reference images to look through as the vast majority of individuals will be recorded at only a few stations, effectively trimming down the number of individuals to look through from 100's to a few dozen.

Cascading the drop down lists Species > individuals (or even area > species > individuals, or species > area > individuals, or in some cases species > group > individuals) would be beneficial too in the process of identifying individuals. In a SQL database this isn't hard to code into the fields on a form, but might not be that easy in Camelot as you would have to chose if the field you add depends on another field on the form.

Hi Chloe,

I've had promising results with the Scoutguard MG983G-30M, and I believe a 4G version has just been released. The 3G version has some useful features, like two-way communication to change settings/get images and a audio call feature (which I haven't used). I'd be very pleased to hear how your tests go.

Cheers,

Rob

@@Chloe+Aust -  pls google “ltl acorn uk”， you will get a lot info.  No, no limited, just the company's sale rules. You could use it anywhere, but for example, if you wants to use it in kenya, you need a local sim card, such as safaricom... when you check photos from china, you will find most of them photoed by are ITI acorn. Panda,tiger, snow leopord, cloud leopard, leopard, golden cat.... 

Scoutguard is also a good china camrea traps. oldest company. only company have espano menu. but they are focus on US hunting market. not special for wildlife researching use.

The LOREDA is also good, the only brand focus on wildlife-bio researching camra traps in china. Might be do not have dealers/stock in uk.

An update to the automated species identification debate: 

A paper has recently come out which used deep learning ("very deep convolutional networks") and managed 89% accuracy for the Snapshot Serengeti Zooniverse dataset, IF the image was first manually cropped around the animal. Seriously, who has time to do that? If the image remained uncropped they managed a woeful 35% accuracy. 

Perhaps we have a long wait ahead of us for this to become a practical reality?

A number of good points have been made. In terms of remote-enable camera traps, you will mainly find one that use cellular data signals to transmit images. Typically these are thumbnails rather full resolution, so you will likely still need to retrieve teh emory cards for analysis. Also, traps that transmit images tend to have a lionger time lag between shots, which can be a problem.  Hunting web sites tend to have the most complete reviews and discussions of the various models.

If you want to use citizen science in the data analysis process, I would suggest looking at Zooniverse (www.zooniverse.org). It has a pretty well-thought-out platform. 

Hi Kate, It all depends on the specie and data set accuracy you are trying to collect. If you are purely documenting frequency of visits, this is straight forward. However, if you are recording numbers then further analysis of photos may be needed. Some animals like foxes have distinctive marks or shapes, whilst smaller mammals can be a lot harder. If this is impossible, then a time limit should be defined across all camera traps. From all the papers I have read and projects we have worked with, there doesn’t seem to be clear standard. Hope this helps. Mike - handykam

Hi Kate, this may be a bit late for your analysis but I will put here for future reference. The case you mention would be to detemine independent events, not necesarily different individuals. Unless there is conspicuous colour, pattern, size, sex or other scars/markings then you shouldn't label them as different individuals. I haven't found much analysis or discussion on this point so it seems prior experience or a "best-guess" is used to determine the time between independent events. Within the R package "camtrapR" vignette it is mentioned (https://cran.r-project.org/web/packages/camtrapR/vignettes/DataExploration.html) "The criterion for temporal independence between records ... will affect the results of the activity plots" and you may have other data to correlate activity with abundance or density.

If you are using and occupancy analysis this may be trivial as you will likely be lumping multiple nights of survey into a single detection period for each camera anyway (we have used six 10-day detection periods for cats in central Aus). 

SECR methods may also be more useful for calculating density. Timing of events BETWEEN cameras can be used to show that photos are of different indivduals as they cannont exist in two places simultaneously or travel at fast enough rate to be captured in that time frame (clocks must be synchronised). 

I hope you have had as good a season in Bon Bon as we have in Alice Springs. Cheers, Al

The facts are simple. If you violate the assumption of independence of sampling events you will bias the result. In the event of multiple observations by overestimating occupancy due to counting the same animal twice. There is therefore no answer to your problem. People try to overcome it by chosing a particular random number to try and standardise the method across sampling sessions. Let us say we decide to use one hour between new samples in all our sampling sessions. If abundance changes between sessions our ability to detect an animal may change and our result may therefore be biased up or down depending on the trend. So we allow detectability to vary between samples to try and reduce this bias. This is done through assessing the results of the replicate samples i.e. n days.
However, the session time determines how many counts are reduced to presence only and as session time increases the variance of the occupancy estimate may grow. But we don't know if this increasing variance/ loss of accuracy is based on a real loss of information or not because we don't know if we are correctly throwing away information portraying the same individual or information on the presence of multiple individuals. Therefore selection of session time may effect our ability to detect a change in occupancy even with accounting for varying detection. i.e. detection may be equal if our session time is too short. In other words if in both one hour AND two hours we detect multiple animals on the same number of plots, convert this to the binomial (presence/absence), then detectability will be equal for both sessions. But if in one hour we detect half the number of animals as in two hours what does this say about occupancy?
Spatially explict (SE) models will also not help because as far as I know they are based on allowing multiple detections across plots, not within plots. Or cameras in this case.
Having said this there is apparently some hope in marking a portion of individuals to increase the accuracy of SECR or SE unmarked in this case. So my only advice would be to run an experiment with marking some foxes and use a SE model with predominantly unmarked indiviudals.
On a more jovial note you could simply mine the data by assessing the power to detect a change between pre/post baiting using a variety of different session times. The reality of such a result is highly dubious but it might inspire more thought on the behaviour of this particular situation. Failing all this youy could run away and join the circus.

Hi Heidi,

You might also consider becoming involved in the eMammal project (emammal.si.edu) that the Smithsonian has recently developed. It has tools for both uploading and coding camera trap images, with the bonus that they are ultimately archived and made available by the Smithsonian for future researchers. My opinion is that moving toward open-access data will make all of our efforts more valuable into the future. I'm using eMammal, but not one of the developers or associated with the Smithsonian.

Cheers,

Robert