Good morning everyone welcome to the September webinar for the Atlas of living Australia my name is Martin Westgate I run the science and decision support team at the ALA and I'm really thrilled to be here um we've got a great webinar for you today um but before I launch into that I'd like to acknowledge uh while normally I'd be calling from the lands of the Ngunnawal People I'm interstate today so I'm calling from uh from Sydney from the lands of the Gadigal People of the Eora Nation and I'd like to pay my respects to Elders past and present. And yeah just really thrilled to be here today with three speakers from all around the country so um look the topic of today's webinar you have seen um from joining in today whether you're logging in from the uh in the future on webinars or whether you're watching live um and it's about technology now that may not be entirely surprising we've been doing these webinars for coming up on four years now a lot of interesting technology talks but something that's particularly exciting to me about today is that it um impinges on an area that's very close to my heart which is about ecological monitoring um look to say that technological development is moving quickly is sort of an obvious statement these days um and yet in ecology um the ability to detect species in the environment is increasing rapidly and markedly and so I'm really thrilled to be able to hear from three experts today in that field and to talk about different technologies that all have that that shared goal and I think are really genuinely revolutionizing the field and environmental monitoring and a Australia and around the world so uh we've got three speakers today we've got uh Christine from Macquarie University give us a wave Christine we've got Paul calling in from QUT and we've got Matthew calling in from WildObs in at the University of Queensland now um for those who are watching live we have um a questions portal you'll be able to see it in on your screen we also have a poll at the end if you want to tell us something about yourselves um we'd love to hear about it always helps us to customise our future webinars to you and your needs um but with that in mind I'll get out of the way our first speaker is Christine um and so I'll hand over to you thank you so much no all right that's looking great perfect so good morning everyone as Martin said my name is Christine Chivas I'm a PhD candidate from Macquarie University and today just going to be giving you an insight into some of the work that we've been doing on monitoring birds and mammals in Kakadu using minimally invasive approaches I'd first like to begin with an acknowledgement of Country so today I'm tuning from the lands of the Wallumattagal Clan of the Darug Nation I'd also like to acknowledge the Mirarr people whose lands of Kakadu I am fortunate enough to be able to do research upon and extend that respect to any Aboriginal or Torres Strait Islander People present today So first and foremost why do we need novel and new approaches to biodiversity monitoring So currently we're seeing a global biodiversity decline and this has been particularly pronounced within Australia's loss of mammals a third of all modern mammal extinction events have occurred in Australia despite only being home to six% of the world's mammalian diversity this has resulted in 39 mammals going extinct since European settlement and listed a further 110 at a risk of Extinction despite Australian birds faring better we do have 163 Birds currently listed as at risk of extinction and we've seen a 60% decline in abundance of these threaened birds over the last 40 years this brings me to Next Gen biodiversity techniques so ultimately the first step in being out of temper this rate of biodiversity decline is being able to gain reliable and robust biodiversity monitoring data today I'm going to be talking about three of these techniques passive acoustic monitors camera traps and eDNA or iDNA based approaches so people are probably more so familiar with the first two on that last slide but not so much familiar with the eDNA or iDNA side of things some people may have heard of the term eDNA which stands for environmental DNA well essentially iDNA is a subdiscipline within this field and stands for invertebrate derived or ingested DNA it relies on the capture of invertebrate samplers so things such as carrion flies leeches and mosquitoes essentially to their day-to-day feeding are collecting that vertebrate DNA for us this offers us the opportunity to collect these invertebrates as a means of passively sampling the vertebrate diversity of an area so far the approach has shown great promise and its ability to detect multiple vertebrate groups with one method approach that can be challenging with traditional techniques as well as increased sensitivity in difficult to monitor groups including small bodied and arboreal mammals and studies offers demonstrated its ability to detect more mammals and established minimally invasive techniques such as camera traps as well as in some cases being able to differentiate between cryptic and morphologically similar species so this brings me to what we've been doing in Kakadu so we have sampled six 1 hectare reference plots on the northeastern Edge um of Kakadu National Park during two rounds of sampling undertaken during a wet and dry season we have employed these three minimally invasive techniques so acoustic monitors were only deployed during the dry season camera traps for a six to eight weight deployment during each field campaign and mosquito iDNA for two consecutive nights of sampling at the beginning of each of these field campaigns so this is just an idea of what our set up looks like so we have an acoustic monitor in the middle program to record for 2 hours in the morning 2 hours in the afternoon to capture that Dawn and dust chorus camera trap baited with some rolled oats honey and peanut butter and a CO2 beta mosquito trap that essentially pulls the mosquitoes in and they're sucked into that net that we come along and collect the next day so as you can see we definitely had no shortage of mosquitoes in Kakadu we were getting on average between 800 to 1,500 mosquitoes per trap and collected over 48,000 mosquitoes across the two field campaigns so what do you do with 48,000 mosquitoes well essentially iDNA follows the same sort of metabarcoding workflow you would in an eDNA based study to simplify it it starts with a DNA extraction process in our case from bulk mosquito samples amplification of this DNA using mammal vertebrate and avian primers that then goes off for next Generation sequencing in return you receive millions of reads back that essentially go through a bioinformatic and taxonomic assignment process so what do these techniques tell us about the detection of birds so just to clarify what we mean by sampling effort on these um species recovery curves so acoustic monitors is the number of trapping periods per trap um due to time constraints we were only able to analyse the period that corresponded with the collection of iDNA corresponds to each of those bulk mosquito samples and camera traps is the number of days of deployment per camera trap so as you can see from this we're getting significantly more bird detections with both acoustic monitors and iDNA than with camera traps so in total during the dry season when we had all three methods deployed we were able to detect a total of 79 avian taxa this included 52 with acoustic monitors 49 with mosquito iDNA and 11 with camera traps only six of these were detected with both methods while 21 were detected by both acoustic monitors and camera traps despite this though there were still a considerable amount that were unique to each method essentially telling us that these two methods are additive and will be able to result in the detection of additional taxa than they would be on their own just to put into summary some of the examples of some of the birds that we were getting with different methods so with the acoustic monitors we found we were detecting additional honey eaters as well as transient species so things such as the magpie goose and red tail black cockatoo that may be passing over our sites and getting picked up by these monitors but aren't necessarily stopping to get fed on by mosquito while the taxa that were detected with both methods included relatively common birds so things such as the blue winged kookaburra tawny frog mouth and galah while with iDNA we found we were getting more ground dwelling birds so things such as the brown quail and the threatened white throated grass wren as well as semi- aquatic birds are things such as the Nankeen night heron and intermediate egret um generally aquatic birds can be challenging to monitor using approaches such as acoustic monitors so what about mammals so we found we were getting pretty much double the amount of mammal detections with iDNA as with camera trapping with pretty much on quarter of the sampling efforts so with camera traps we were able to detect a total of nine tax across both field campaigns and these included relatively common ground dwelling species so things such as dingos agile wallabies feral cats and the northern brown bandicoot despite this though the prolonged deployment of the camera traps over that six to eight week period did result in the detection of additional species that weren't picked up by iDNA most notably the threatened black footed tree rat while with iDNA we were detecting a total of 20 taxa this included additional arboreal species that aren't picked up by camera traps so things such as the sugar glider and ghost bat small mammal so things such as the pale field rat and common rock rat and as well as macropods most notably the short eared rock wallaby at one of our sites that is near the sandstone escarpment and the relatively elusive spectacled hare- wallaby so what about reptiles you might ask so we found in this study that although mosquito iDNA appears to be a good approach for mammals and birds not so much for reptiles in our case we did find that camera traps were able to detect the most reptiles however these were mostly large and diurnal species so you can see down on that left hand corner we've got a fill neck lizard and in the right hand side of that right hand picture we have a yellow spotted monitor and currently studies do demonstrate that currently a combination of camera traps to catch those larger species and live trapping for smaller and nocturnal species appears to be the best approach really hiding that there's still work to do and being able to develop many invasive sensitive techniques for the detection of reptiles so as with any approach all of these methods do have their own sets of challenges so a correct identification can be challenging with all these methods so acoustic monitors you can have challenges with incomplete calls crossover or species variation and calls camera traps you can have issues with blurred imagery especially for small fast moving species and morphologically similar species so to put this in perspective with iDNA we were able to detect three different rat species this included the introduced black rat as well as the common pale field rat and common rock rats however we were challenged be had to get any speeches level identifications for any of our rat images from camera traps iDNA can also have its challenges so essentially any iDNA or eDNA based study is really only as good as the reference library that you have to be able to match those sequences to furthermore all these challenges do have their own specialised skill sets this ranges from the molecular skills required for iDNA based approaches to the localised knowledge for the identification of acoustic calls however in saying that though both the use of AI as well as the commercial labs offering eDNA services are making these approaches more accessible for a wider audience additionally to this we do have challenges with the transportation of CO2 to be able to capture mosquitoes particularly when working in places such as Kakadu and you do always have challenges with equipment failure and destruction so over half of our bait stations were destroyed during our campaigns um I'll give you best guesses who may have been responsible we've got um suspect number eight down that right hand corner so just to summarise this all so we found that mosquito iDNA outperformed camera traps in the detection of mammals detecting additional macropods and arboreal species we found that acoustic monitors and iDNA appear to be complimentary approaches with acoustic monitors resulting in the detection of more honey eaters and transient species and iDNA more ground dwelling birds in semi aquatic Birds however camera traps appear to be at best for reptiles in this study however our current approach of live trapping and Camera trap appears to be the best approach with that really demonstrated need to continually to work on new sensitive techniques for reptiles and furthermore also they continue need to improve DNA reference libraries to really uptake that use of eDNA and iDNA based approaches so just to wrap up today I'd like to acknowledge a few people so obviously this research couldn't have been possible without especially my supervisors Anthony Chariton Adam Stow and Andrew Harford the support from Macquarie University and my lab in the eDNA and Biomonitoring lab as well as our collaborators and project partners of of the supervising scientists who this project wouldn't have been able to happen without and Special M thanks sorry to Tom Mooney David Lowensteiner and Kate Montgomery that's wonderful thanks Christine look I've already got so many questions and uh there's already some coming through the chat as well now normally in these events questions at the end though so um I'll cue those up fascinating talk and thanks so much for sharing your research with us um now uh Matthew over to you thank you can you hear me yes great and the slides looking great too thank you so um I'm very happy to present uh the Wildlife Observatory of Australia which has recently been funded through ARDC QCIF TERN and the University of Queensland it's a big initiative um with a lot of people that I will acknowledge at the end so I also want to start by acknowledging the traditional owners of the land that I work here at UQ and I'm going to hop right in so what is WildObs and why do we need it we are going to create a National Wildlife database that will deliver information on robust monitoring science and management we need this information on wildlife to track threatened species we need it to track invasive species and management we need it to produce high quality scientific um innovations we need it for our public to be able to access information about where their amazing wildlife is and we need it to meet our uh UN biodiversity goals that we have committed to as a country we also need it because there are dramatic events and these require timely uh information so that we can uh manage them these include things like bushfires disease outbreaks climate change and we want to be able to sell biodiversity carbons biodiversity credits alongside our carbon credits which means we have to be able to robustly quantitatively track wildlife so we have some opportunities and some problems with cameras so cameras are relatively easy to set strap them to a tree leave them for a few months easy to set 50 cameras per survey but these create so much data hundreds of thousands of images over 100 gigabytes per survey and we have many surveys in Australia but these are completely uncoordinated often targeting targeting a single species but they capture so many species that if they were brought together we could use it as a monitoring tool we are collecting a lot of data but we are not publishing as much as we would expect based on this investment in our sampling this basically boils down to camera traps are easy to use in the field but generate unmanageably large data streams and these data streams are not being leveraged to produce high quality science and monitoring so who is using cameras well basically everybody they're very accessible to researchers at all stages in their careers they're used in a wide variety of biodiversity research management and monitoring people that use cameras in Australia include NGOs government consultancies academics and Indigenous groups so right now camera data is being boiled down to presence only this is not the richest format for the data but it is suitable for species distribution models something that Australia really excels in but species distribution models do have some limitations they're really good range maps essentially but they just tell us if the species is there how likely it is to be there well camera surveys actually include the detections and non- detections this enables a type of modelling called hierarchal modelling to derive occupancy and abundance while accounting for variable detection what that means is two people can set cameras one with baits and one without and we can account for the fact that species are detected at different rates utilizing different methods that's extremely powerful for robust monitoring it allows us also to look at trends through time and to infer species interactions both of which I will show you examples of we also want to note that while ALA focuses on presence only data to date we are in process of making it more accessible to download full survey data using the same website that we all know and trust to go to for our biodiversity information so right now only a third of Australian publications utilize an analyses that account for detectability this is relatively low and it could explain why are comparatively low publication rates compared to other places so what are the things WildObs is going to address we're going to enable coordination and collaboration in sampling fieldwork so that we can be more productive and efficient and save money and um answer our questions more efficiently we are also going to help with image management so this is the storage and efficient processing of images this needs to be free needs to be cloud-based so we can access images anywhere we want we can have many people sorting images and we need to be able to power AI identifications and I'll explain that in more detail we also need open access to data sets right now these are often siloed away at specific institutions or in someone's old dusty hard drive this especially important for publicly funded um efforts so anything from ARC federal government like DCCEEW state governments these are all required to be open access but they currently are rarely Open Access we also need training in advanced statistics for impactful research and monitoring and this is something that we are going to be um really focused on so in order to store organize identify millions of images we need access to a cloud-based image platform that has identifications using AI one example is wildlife insights but there are many and we are debating which one right now that we're going to promote and provide free for all Australians we also need tagged images in a repository so that our AI researchers here in Australia have what they need to do new AI research and improve our algorithms we also need an aggregation of all the spreadsheet data that we're currently deriving in all these separate places separate institutions separate Labs so we need standards for our data and we need APIs to share that data and then we need it to be accessible and searchable such as via online on ALA then we need robust analytics to drive key inferences this requires meticulous data standardization and preparation robust statistics and we need to create intuitive results so that people can really understand what is being um derived from their sampling we want to create automated reports you provide us data we give you back an automated report that includes the best robust analyses possible we're going to feed those analyses and results into the government regulation like the EPBC Act for endangered species and our indices like the threatened species index which I'll talk about in a second we're also providing training for statistical skills and we're initiating many more collaborative of papers and reports and this really is a carrot to get people to be involved because they can get a lot of publications if they're part of this group as a proof of concept we worked in Asia and our lab to put together 239 camera trapping studies published a data paper then that paper has been used in over 20 publications within the last two years including a publication that made the cover of nature so this indicates that if we can do this there's really power powerful science that can be derived we really want to contribute to the threatened species index this shows um trends through time for birds mammals plants right now there's very little camera trap data in this because um a variety of reasons and we can solve that so we can make that orange line extremely confident in what it's um reflecting and we can do an orange line for every single mammal species I want to give you one example which is the Eyes on Recovery project this was thousands of cameras set in after the 2019-2020 black summer fires massive amounts of data we were able to analyse the images using Wildlife insights and the AI um helped make that efficient and then we able to process all that data in eight weeks provide specific robust for with hierarchal models for 93 species and today we can do that in one to two weeks this is what we need to provide timely management relevant information we can also use these uh richer data sets to look at species interactions here's an example from Tom Bruce who I'm going to give you a little more information about his work in our lab and what it shows is the green line is the dingo um activity throughout a day as captured by camera traps that blue line shows how cats behave in the absence of dingos that red line shows what happens when how cats behave when dingos are present showing a strong shift toward nocturnality so these sort of inferences can cannot be gleaned from presence only data we're expanding that set that type of research to look at entire food webs if you want to learn more about this Tom's actually giving a seminar next week um we're going to drop the information for that in the chat um and to work in the lab managing all this data has a few easy conclusions that setting cameras in the going out in the field is really fun managing 25 million photos is not fun but we there to help so one of our some of our progress today is that we've already had a review paper with 194 Australian authors representing all major universities federal and state governments and NGOs we've also been collecting data I should say collating existing data we're already up near 30 million images we have a very defined data standardisation process this is led by um Dr Zach Amir and we are um nearing a choice of platform to provide for all Australians that will provide free limitless image storage easy management and AI identifications exceeding 95% or targeting 98% so that we don't even have to review them once we get to that accuracy we don't even have to look at our images we just trust the computer is doing a great job we're also um working with DCCEEW to provide a metadata collection app this is going to be called Monitor you'll hear more about it in the next year this is going to allow everyone to enter data in the exact same format from their smartphone in the field and that will automatically uploaded to the cloud the government can see where you've sampled it provides everything you need for reporting and that is linked to the image outcomes later and finally we're hosting statistics seminars and we're providing statistical tools so if you have data need help analysing it we can help you I'm going to let you under the hood just for a second so first I'm going to explain our um pipeline for images so image are sent to us either as SD cards or uploaded in the cloud in a dropbox they are moved into what we call a WIMP a wildlife image management platform examples of this include wildlife insights and there's even one being developed here in um Tasmania by Barry Brook um called MEWC so this platform hosts all your images allows you to see which cameras collected each images allows you to sort per for different species and what it also does is it puts all identified image unidentified images through something called Mega detector that tells you what has an Wildlife versus anything else if it has Wildlife it goes to a computer vision um classifier which is an AI tool to identify species if that species is identified over 97% accuracy it goes straight as data if it's less than that it gets verified by human all that data then goes into our um data standardization process and our um camera database okay now I'm going to shift this up to show you some more if you already have data your institution already has data that's been sorted or you have your own image management platform we can pull that directly either from your platform or through ALA they have an API and that goes also into our data standardization International um database we also are pulling directly from biodiversity aggregators so Queensland New South Wales BioNet WildNet federal government um BDR Biodiversity Data Repository this is where a lot of the data is already exist but it's currently siloed in these places and right now we cannot pull full detection histories only uh presence only and so we're changing all that so we are providing external access through ALA and this is what is going to enable wildlife biologists and government officials to monitor biodiversity easily we're also allowing a tagged image repository where we keep all these images um collated in a nice format so that can be used directly um by AI researchers so we can be on the forefront of developing the um these tools so with that I'd like to thank many people listed on this um slide um and thank the funders and uh the websites learn more about work that we already have done for example on Asia is that our website or my website ecological cascades.com and if you want to learn more about WildObs the website is tern.org.au/wildobs thank you thanks Matthew fascinating to say and um I see that other people have problems with too much data at times as well it's something that hits us a lot at the ALA um so uh Paul, you're our last speaker today would you mind um sharing your screen that looks great thank you excellent okay thanks Martin thanks uh to the ALA for inviting me to give a talk and it's great to talk after um Christine and Matt because a lot of what they've already talked about in terms of the motivations and things are exactly why I'm excited about ecoacoustics uh so my name is Paul Roe I'm an academic at Queensland University of Technology and for the past 10 years I've been on really this journey into ecoacoustics and what I want to do uh today briefly is to talk about a couple of um projects um I'm involved with I would like to start off by acknowledging the traditional owners of the land uh where I now am or where QUT is that's the Turrbal and Yuggara as the First Nations um owners the traditional custodians of the land which was never ceded I'd like to pay my respect to the elders laws customs and creation spirits and recognise that these have always been places of um teaching research and science and the Aboriginal and Torres Strait Islander People play an incredible role in um in all of our communities okay so I'd like to start by just talking a little bit about ecoacoustics um and what it is so really eco acoustics is a new way to understand our world through continuous sound recording um it provides a sort of scalable way of monitoring biodiversity um through continuous sound recording and provides a direct environmental record of that so it's a little bit like um sort of remote sensing drones and satellite imagery uh which collect sort of images of um you know of the continent of the vegetation um and from those images we're able to deduce what's happening to the sort of forest we can look at kind of flowering event and understand the sort of vegetation but really up until now the problem has been that we haven't had scalable methods for understanding uh floral biodiversity um and that's why we're so interested in things like camera traps the ecoacoustics and the um iDNA ecoacoustics is a rather young discipline there are a number of unique issues Matt sort of alluded to some of those some of the issues of the same uh that we face in Eco Acoustics is with camera traps we've got sort of a lot of data uh it's collected all over Australia we need to sort of bring it together and also it's quite difficult to analyse um there's been some quite rapid progress over the last 10 years and I want to talk about a couple of those projects I've been involved with and Australia's been taking a um a leading role um before I do that though let me just play you um some sort of sound it's sort of somewhere collect um collected um uh on a property uh near Brisbane and this was sort of a beautiful morning chorus there [birdsong] so one of the um great things about working with sound is you get to listen to these beautiful um choruses to the beautiful birds not just birds but also frogs and some mammals um vocalising um like that of course the issue is how do we how do we make sense of that so one of the um projects which I've sort of created is called the Australian Acoustic Observatory uh this was a collaboration with some colleagues from other universities around the country um and what we set out to do was really to um realize that vision of the kind of continental scale biodiversity monitoring to really understand how can we monitor uh floral biodiversity at scale how can we actually understand what's happening um so we set out um and created a a project uh we were successful with getting some funding through the Australian Research Council um and we set up a large network of recorders and one of the things about ecoacoustics is we regard it as being very much a data first approach to science so in that sense it's a little bit like um astronomy um or bioinformatics where you might actually collect data and you might start with that data and mine that data to actually sort of come out with to understand what's happening to find questions um so it's a data first approach um and hence the name Observatory because we're observing or listening if you like to all of the sounds across Australia we are also very driven by an open science um uh methodology in that where believe we believe that the data should be it's been um funded through an Australian government research grant we believe that the data should be made uh sort of freely available just like um the observations through the ALA are um and we're using the data to support obviously sort of ecology um and computer science and citizen science but they're also creative uses of the data um as you can imagine with such beautiful recordings there's um a lot of sort of artists and other people that are interested in using that data um we started deploying recorders in 2019 which was of course a great time to start so we had a few delays due to bush fires and covid and things um but we now have over 365 sensors uh deployed all around Australia uh we're continuously recording sound so there's no sort of sampling involved this is continuous sort of sound recording each of the recorders there's a little sort of solar powered box that records to a set of SD cards and we're very much leveraging the good will of land owners to um collect the data for us to collect the SD cards dust off the solar panels and send the SD cards um to us so we have this this huge amount of data um rolling in the um the difficulty of course as Matt's already said is how do we manage all of this data we obviously want to get the data into the Atlas um but how do we sort of get it there from you know remote areas like the sort of Sturt Desert you know bunch of um sort of SD cards that are sent to us we need to get that into some database where it can kind of be analysed and then the observations um actually put into the um the Atlas so we can kind of generate those um environmental insights uh so to do this we um along with um some of my colleagues created a um a project open source platform this was supported through the Australian uh Research Data Commons um to essentially kind of complete that pipeline that workflow for processing um the data and to get the data into the ALA um it's been very important for us to um work with um sort of communities so we are co-designing the um tools and systems that we're producing all of the tools and systems are sort of open source for everyone to use so it's very much a program to a project to produce both a platform but also to um bring the community along with us because it's a young discipline so that we can understand um how best to use the data how people want to use the data the tools that they need to um deliver on this vision of being able to understand what's happening to our um beautiful country once we've got all this data of course there's still this issue of how to find calls um so you know somehow we needed to search through all the data so we thought who are we going to who we going to call when we need to do a search so we've got a um a project with Google um where we can actually uh undertake a kind of a search so if you've got an example call that you want to find we can actually scan through all of the data through hundreds of megabytes of data hundreds of um gigabytes rather of data to actually find those particular calls and if you would if you want to um perhaps take a photo with your phone or something or just um grab um that image or um if your phone allows you go to that link for the QR code and you can actually give the um the similarity search ago and this is all driven by um Google AI so just to sort of wrap up um a couple of projects we've um we're collecting data um with the Australian Acoustic Observatory for from all over Australia we've got other specific projects which sort of target um threatened species um or other monitoring regimes which very sort of NGOs and governments um are running uh we've got sort of literally thousands of users we've got over 200 projects um over half a petabyte of data now almost 500 years of um continuous um data obviously is too much to listen to which is why we need the Open Ecoacoustics platform to um to enable us to sort of store analyse the data um and also with such big dating you get a lot of errors so we've got a lot of sort of very sophisticated tools to um to fix the errors um but as much as this is about data um and about platforms and things it's also about people so I would like to uh thank all of my colleagues and collaborators that um have made this happen uh and I'd like to thank the ALA for inviting me to give this talk uh so thank you lovely thanks for that Paul um I'm slightly unsure where to go we now have about 20 minutes which is a wonderful amount of time for a discussion so thank you all for your presentations people have been queuing up questions in in the chat and I'll try to get to what I can and summarize what I can't um so I guess uh to get started then um look it seems like each in each of those three talks you ve benefited from some pretty massive leaps in technologies in the last few years um you know uh Matt you talked about processing things with machine learning and AI and Paul same I guess the uh eDNA um boom has been well documented as well like the sort of technologies for genomics um are we at the is everything else keeping up you're now generating all this data is do we need new development in our statistics to manage this kind of data or have we really just been waiting for that um sorry just open question there if anyone wants to jump in I think uh this could be a good opportunity for me to jump in because the statistics for wildlife ecology have largely been developed for cameras that came on the scene a little bit earlier than Acoustics and eDNA but can be used for those methods for example I mentioned these hierarchal occupancy and abundance models those take detection histories so we need detection versus non- detection in each amount of time at each sensor that can be created through multiple surveys of eDNA or through multiple surveys of acoustic data the statistics are there they're a bit complicated but there's packages being developed in R and other tools that make them accessible and one of the things that we're doing is producing um automated reports that include all those complicated analyses that we run on our supercomputers here at UQ and QCIF so that users who don't have a degree in statistics can still have access to robust outcomes from their data no thanks Matt Paul do you want to add to that at all well I'll probably defer to Matt I'm not a statistician but yeah certainly I mean with the new developments in sort of um hardware and software so as you saying Martin the sort of the AI developments which is both the sort of compute and also the techniques to um implement the AI those deep learning networks and the decreasing price of sort of storage that's enabled us to collect and analyse more data and as we get more data we are going to you know potentially got to um find out and understand new things but it is going to require um I think new kinds of statistics or certainly new um kinds of um models so I think things like you know the occupancy modelling and hierarchical models that Matt sort of talked to I mean those things perhaps have been around for a little while but we haven't really been able to use those because we just haven't had the data you know the great thing about you know putting out you know um iDNA or camera traps or eco acoustic um recorded is that you're not reliant on a person being out there in the field making all of those observations um and so yeah it is going to enable a whole new kind of science and that's obviously really important for um conservation and understanding um our environment as well as you know producing food and all of the other ecosystem services that we're reliant upon yeah no it is as someone who did a PhD in a in a long-term monitoring lab where every observation was someone standing in the field it's uh it's amazing to see the sort of wealth of data is coming through um Christine if I could um pick you up on this question of um dealing with data I mean it was a really stark example of a of a long-standing problem I felt in ecology that you demonstrated there which is that different methods detect different things um how do you see this playing out in future do you think that we're going to have to presumably there's a lot of um a lot of testing needed to compare these methods yeah I definitely think there's more work required to really understand the bias especially associated with iDNA is sort of a newer approach uh so definitely understanding that and using that in a decision making framework and when planning biodiversity monitoring working out sort of what focal groups you're wanting to focus on and how to go about that um yeah obviously there's still a lot more work especially with iDNA that needs to be done there's a lot more questions than answers at the moment but yeah it's definitely showing great promise at the moment no it's fascinating and it's really good to see that work being done as well um there was a question oh sorry Martin I want to um just bring up some uh a frontier which is that these hierarchical models because they allow you to include a detection covariant for each survey they also allow you to merge eDNA acoustics and cameras with each of them having a different detection probability so for example Christine mentioned how there was a vastly different detection of mammals reptiles and birds using these methods you know cameras excel at vertebrates on the ground mammals predominantly ground dwelling birds acoustics excels in Birds particularly and insects and eDNA covers a lot of the most cryptic species that are hard to see on a camera don't make a lot of noise putting them together in a single model especially these multi-species multi-site models allows you to learn about the entire community in ways that we've never been able to do before this is the frontier there are models already available um it's not easy but um it's exciting and it is the future and presumably the uh the scale that you've got the amount of data that is putting into these models and the degree of standardization as well as helping with that I'd imagine absolutely um what I think the most important like innovation there is in this space is the ability to include species specific detector specific so whether or not your DNA acoustics or camera specific um co-variants for each of these sampling approaches and then that allows you to bring them all together under a single modelling framework no it's fascinating to see it be good to see how it how it um flows into things like the TSX for example that you mentioned Matt um look I'm going to try and get through some of the questions that have come in from uh people who are online and one of them sorry to go back to you Christine so soon there was uh one that was asking about co- benefits so for example whether it's possible to detect um like pathogens from mosquitoes simultaneously with wildlife is that something that's on the horizon um so that is definitely possible so you can do DNA and RNA co-extractions if you're wanting to look at that as well so in our case we were just focusing on the DNA but that's definitely a probability that you can also do the RNA extraction to be able to isolate any mosquito born pathogens as well I dread to think the amount of data you'd get from something like that but I imagine it would answer some very exciting questions um there was a similar one Paul for you about whether Eco Acoustics is feasible in the marine environment which um I wondered if you have any questions any comments on that oh yes there's a yeah I mean there's a sort of a long history of um using acoustics for um citations and things um I was also going to jump back to the previous question if I may and say that yeah I mean there are other things that you can use Acoustics for because it does detect other information so potentially you can get information about you know wind and rain and things but also you can detect um you know potentially you know if you've got some sort of human disturbance you know sort of gunshots and things like that or um you know motor vehicles are actually sort of noise pollution which of course may itself be affecting the environment so yeah these um these methods you can do a lot with them and detect more than just a sort of a single target species you really can get a bit of a handle on what's happening to the whole um environment and how far are we along on that journey so there was a specific question in the in the chat that was someone was asking about uh mimics so um birds that copy other birds presumably we're still at a point where um some of that stuff is challenging for these models is that fair to say um yeah that's a great question we have done some work with mimics particular with Albert's lyrebirds and uh generally if you know I think we're at the stage where if a if a human can um perceive the difference then usually providing we've got enough you know test and training data we and then actually train an AI model to distinguish the two so certainly we had success with sort of Albert's lyrebird but yeah um obviously there are there are limits to um you know what you can do and sometimes um uh you know I think as Christine was saying it's important to understand the bias of the of the technique I mean all techniques are biased even if you've got a sort of a human observer in the field um sometimes we can't distinguish um particular um species so sometimes we can't um distinguish for example between different raptors and things like that that make the same kind of noises other times we may actually be able to distinguish individuals so potentially with some birds and some um koalas and things like that we can actually get a handle perhaps even on the individual that's calling that's an interesting point actually because um we've had a couple of questions in the chat about um comparison against citizen science data or um I guess the and with birds of course in particular pick on those just because we get a lot of data on birds and I see it a lot many of the people um who are who are submitting those observations are extremely expert um you know to the extent that they could be doing this professionally but don't um do you have a sense of uh the extent to which um uh automated methods obviously they scale much better they're in a sense cheaper per unit um do you have a sense of their comparability against expert observers in the field um I think they're pretty good I mean I mean I think in a sense the advantage of a lot of these methods is so they can be deployed um by non-experts and then you can have experts review the data if you keep the data I mean even if you've got an expert birder in the field you know if you hear a weird sound that's not a bird you know you hear a frog or something and if you don't have a you know herpetologist with you what do you do but if you've got the recording you can then sort of take the data to the expert um so I think um keeping the data is you know it's about the sort of scalability that the techniques provide but it's also about the record having the data which can then potentially be kind of audited and assessed and used for you know the sort of green accounting schemes but also if it's a call that perhaps not a lot of people know way you might really need an expert you can take it to the expert um I think the I see the AI as being an assistant really I don't think it's ever going to replace people it's more a um an assistant to help people An entirely fair point um Matt if I could come back to you on this so um I'll just um what Paul said is it um about the AI being assistant I agree with that traditionally that's how we've used it but for cameras for a majority of species but not every species the AI is significantly better than a human and the accuracy is much better than a human so we're getting to a point where we never have to see the images and we can have a workflow once we have let's just say universal access to a signal the cameras can upload images the AI will process them and the data will flow into the spreadsheet so we can analyse it completely removing the human from the link and that will allow scalability on a whole another scale um so I suspect that will happen for many species of birds in the future maybe not all I'm curious what Paul thinks on that one well I guess some bird calls evolve over time so um it's not a sort of a static system that you can completely analyse um I you know I think that the AI is going to greatly help with dealing with the massive amounts of data that we've got but I think there's always going to be a role for a certain amount of um validation and checking I think we will also need to look at other um techniques um you know the sort of the finding bird calls in a long recording is a very reductionist approach to understanding the sound um I think and all you find out is what the recognizers have been trained to do I think it's also useful to have other perhaps unsupervised approaches where you also try to kind of characterize the overall sort of soundscape I mean crudely you know if a um if a microphone gets eaten or damaged as Christine sort of knows all too well with some of the problems that she's sort of had um the recognizers aren't going to tell you that they're simply going to say that we haven't found anything they're not going to tell you that well actually there's no sound data there yeah we've all had that happen I had a bushfire go through some of my sites when I was trying to record some things and that um that was in the tape days but let's not go back to that um speak while we're talking about this sort of scale um Christine we haven't really talked about um you've mentioned eDNA and iDNA and they're in some ways newer technologies but also becoming quite established do you think they have the potential to scale to the same sort of things we're seeing in Eco Acoustics and camera trap data uh definitely I think especially um with the uptake of also like things like citizen science projects employing eDNA and that we definitely have the application to extend the use of it uh so it's definitely getting more accessible and definitely increasing in its scalability especially as well as I was saying earlier with more commercial labs also offering eDNA services and that sort of workflow becoming more streamlined it's definitely increasing in that ability as well and you think with a with a body of users in the bioinformatics community who are used to large data already sort of sort of straight off the bat that that would be a benefit definitely also with super computing and like the increasing bioinformatic capability it's definitely going to be easier to add to process that data as well there's definitely people that are working on ways to try and essentially take that data and have it come out in a format that's accessible by you know government agencies or people who might not be familiar you know what a OTU table anything like that looks like so there's definitely things in the pipeline to try and make that as a data source that is more easily able to be understood by sort of general users as well definitely some challenges there though I'm sure some opportunities as well look we're coming close to time but I will try and sneak in one sneaky question which um and I'll direct it to you Matt first if that's all right which is that um now some time ago a year or two perhaps the um Australian Academy of Science put out a report suggesting that Australia needs a Bureau of Biodiversity in the same way as we have a Bureau of Meteorology now some of the things that that the presenters today have put forward are what looked like steps towards something like that do you feel that in future there's the potential for an integrated monitoring system for biodiversity in Australia that could be productionised in that way yeah absolutely that's in the cards um there's a variety of initiatives around biodiversity right now that are playing out so the landscape's evolving really quickly um one thing that has been great for at least Paul and I is um the Australian Data Research Commons ARDC has a new arm called ARDC Planet devoted to large scale data commons for biodiversity and environmental data um they've supported both me and Paul and are looking at expanding that to drones potentially eDNA in the future so the ball is rolling but the path it will take is unclear at this stage that makes sense and obviously we're speculating at this point you know it's uh with the technology mov moving so fast it's always hard to be sure but it is interesting to think that these things would have been impossible a decade ago for example um lovely okay well with that in mind um thanks everyone for your time uh I'll remind everyone uh who's uh watching along at home um we have a poll that um we put out if you want to tell us something about yourselves we're always glad to hear about it please uh click on that on your screen if you'd like um but with that uh thanks so much to all our speakers today and for those insights this will be available on YouTube in due course um and uh enjoy the rest of your day we'll see you at the next webinar thanks so much
2024-09-11