 Good morning and welcome to the final session of the Koala Coalab Conference series for 2021. My name is Jeff Lundy Jenkins and I'm the director of Southern Wildlife and Koala Operations within the Department of Environment and Science and I've had the privilege of acting as the emcee for each of the six theme sessions that have made up this year's Koala Coalab series. Before we commence On behalf of all the participants of today's Koala Coalab I'd first like to acknowledge the traditional owners of the land on which each of us attends today's virtual event. Pay my respects to their elders past, present and emerging and acknowledge their ongoing connection to country. In acknowledging the traditional owners, we also acknowledge their continuous living culture, their diverse languages, customs and traditions, knowledge and systems. We also acknowledge the deep relationship connection and responsibility to land, sea and sky country as an integral element of First Nations identity and culture. We recognize and acknowledge that country is sacred. Everything on the land has meaning and all people are one with it. We acknowledge First Nations people's sacred connection as central to culture and being. First Nations people speak to country, listen to country, sing up country, dance up country, understand country and long for country. We acknowledge and thank First Nations people for their enduring relationship connecting people, country and ancestors. An unbreakable bond that safely stewarded and protected the land, waters and sky for thousands of generations. Today's final event in our sixth theme series looks at koala survey and monitoring techniques and builds on the previous five weeks of presentations and builds also on the success of the inaugural koala collab event that was convened at Lone Pine Sanctuary back in 2018. The first presentation in today's series will be provided by a associate professor, Grant Hamilton, and will look at using drones and artificial intelligence to detect koalas. Grant Hamilton is a qualitative ecologist and associate professor in ecology at the Queensland University of Technology. With a focus on detection and risk in conservation, he has a passion for making real impacts in agriculture and the environment. He leads the Conservation AI Hub and his group has pioneered the use of artificial intelligence and drones for the detection of koalas and other species. He is currently an associate editor for PLOS Sustainability and Transformation, lead of the ecological monitoring program in the centre of the environment, centre for the environment, Queensland University of Technology, an academic lead, international engagement in the School of Biological and Environmental Sciences at the Queensland University of Technology. So I invite Grant to proceed with his presentation. Thanks very much for the invitation to come along and thanks very much for that introduction. I'll acknowledge first up Dr Simon Denman. I'm a quantitative ecologist as you said. Simon works on the machine learning side of things and we've worked very closely together for the past few years to get this methodology up and running. In fact, I'll start by acknowledging the whole team. We've done quite a bit of work in the past few years and that takes a lot of people. Evangeline Corcoran, now Dr Evangeline Corcoran, some of this work was hers during her PhD. Megan Winston, the current master's student and a number of other people who are really crucial in the work that we're doing at the moment. I'd like to acknowledge funding from the Queensland Government for past and present projects. Also, we've had support from the Neusershire Council and some federal funding from the Land Care Lead Bushfire Recovery Program. As you've no doubt heard, or excuse me, no, detection of wildlife is really fundamental. We need to find out where they are for translocation, health checks. But from my point of view, as a quantitative ecologist, it's fundamental data. We need to find out the abundance of threatened species. And in the long term, we need to be able to estimate trends and that's really essential. And if you can do that accurately, then that means you can get a good trend analysis and you can have some kind of idea of whether your management actions are actually having an effect. Traditionally, and quite reasonably, the way that koalas have been spotted has been by people on the ground, often with binoculars. So there are a range of other ways, scats and other kinds of approaches to it. But probably the most accurate method previously was people looking for them. Good on them. I can't do it. It's hard because koalas are cryptic. There's a range of experience out there. And the folks who work for government are very, very good. And there are other people who try hard and there would be people like me who might try hard and still wouldn't do very well. But it is true that there's a range of experience of people who are out there looking for threatened species, including koalas. And the success of our detections is partly dependent on the experience of those observers. It's also given that we have people walking through the bush looking for koalas. It's quite labor intensive. So part of what motivates me in this time of profound change and an absolute need to protect biodiversity is how to manage fauna and flora at scale. So while previously the ground-based spotters were the most accurate, studies suggest that at best of every four koalas out there, there were only spotting three of them. And so that suggests that sometimes the trend estimates may not have been wonderful just based on the data that was available. So really to try and improve those things, we have a challenge and that's to cover large areas. Koalas occur over large areas and often they're quite sparsely distributed and they tend to like places where it's difficult to get to. So we need to be able to cover those large areas quickly and efficiently. And we need to do this with a high rate of detection. Just to give you a little bit of the history of my group and in fact a little bit of what was going on with drones. Back in 2015 there were people who were using drones already and often simply as a platform. So a way to look at things from 20 or 30 or 50 meters up. And that was a really important way to start using this technology, inspecting power lines, looking at bridges for cracks, those kinds of things. People had been even back then using drones to count some very obvious and conspicuous kinds of species. So for example seals on snow or nesting birds on beaches. And it proved to be very useful for those kinds of things. That contrast made it quite easy. We figured at the time well you know koalas are in trees, drones fly, maybe the two should go together. And so we started working with some councils in South East Queensland to look for koalas. And at that time we you know we were reeling in a learning phase. We experimented with flight plans. We looked at different senses. And we manually searched through the images for koalas. So that presented a problem and I'll talk a little bit about why that is a problem. On the left we have a thermal image of a canopy and on the right we have an RGB image, red, green, blue, the typical spectrum that we see and that your iPhone or whatever camera you use takes. There is a koala in there and neither of these images really show it very clearly. So in fact there's a koala down there and we've subsequently detected it by a very intense manual scrutiny. So these are quite cluttered difficult scenes to look through. So it's not like looking for seals on a beach. In fact it's considerably more difficult. So even in those early stages we knew we were introducing new technology to an existing monitoring system and we expected that there would be some challenges. We were learning a lot of things. Thermal was a great way to detect koalas and in fact it was the only way that you could detect the koalas using drones from the kinds of heights that we were flying. So we were flying quite high, 60 meters above the ground, 30 meters above the canopy because that was safe and it also was far enough away from koalas that we felt that it was a reasonable way to be looking for them and not stressing them. It reiterated to us however as has been known for a long time that detection is not the same as abundance. So if you detect five koalas that doesn't mean that there's five koalas in there and we've known that for a long time with ground based spotters there are statistical models and from a range of other species as well but we needed to think about this differently because we were approaching it in a different way. Now it produced a lot of imagery, a lot. We were covering 30 to 50 hectares in a couple of hours which is fantastic. It means that it really is an efficient way to cover large areas but because we were manually reviewing all we had done was go from putting people in the field to putting people in front of a computer and that was difficult to see if that was actually advantageous. The manual evaluation as you might imagine having seen those slides was also quite difficult and so it was clear to us at the time that we needed to develop a more efficient standardized method and we figured well look let's have a look at artificial intelligence and machine learning. We were pretty naive it was a hard thing to do but we managed to crack it and that got a lot of attention worldwide. I'd encourage anybody who's interested in this this is the paper that's freely available. That was potentially seen these kinds of stories by up to 300 million people it was on BBC click and all over the place. So it was very challenging because while people were using artificial intelligence simply for example to recognize shapes as had been done with camera traps those seals on the snow for example or birds on the beach we were dealing with a very cluttered environment and that made for a very different process. Partly the reason that we were successful and Kudos to Simon for this is we started using a combination of a couple of algorithms so we would take the raw thermal image from the drone survey and we would apply two separate algorithms to it and these were off-the-shelf algorithms a convolutional neural network and a YOLO and we would look for the consensus between those and there was a logic to that each of them had been constructed for a different reason and looking at the consensus actually meant we were getting a much better more accurate output from this and as you will see in the slide there the output of that is a bounding box around a koala. I'll show you a little bit of what it looks like in the process on the right hand side we have the probability of any of those patches actually being a koala and when we get to a high enough probability then it will actually draw a bounding box around it and you can see that in the thermal imagery on the left hand side and so that's quite a bit slower than what you know that's simply done so people have an idea of what it looks like we run it much faster we run it on a super computer and we can really take that imagery from around 50 hectares and process that with the automated detection in a couple of rounds so the algorithms process data more quickly and accurately than people drones can cover a large area and really pretty quickly so for the koalas we can process the data quickly and we also demonstrated in that paper that we have a higher accuracy both than ground-based spotters based on previous studies but also it was more accurate than people going through the imagery and looking through it and that's probably due to viewer fatigue and a couple of other factors like that so we thought we were probably on to something with that but it actually it introduced a range of new questions there have been a lot of statistical modelling around ground-based observers and you know of course other kinds of occupancy or abundance-based methods for koalas but this was a new one there may have been new factors now that impacted on the modelling so that we could get a decent estimate of how many koalas were out there based on how many detections we're making we also wanted to make sure that the methodology was not biased and I don't mean that in any in a way that's negative as in trying to convey incorrect information we simply meant that there may have been some factors because we'd introduced the technology that we needed to look for to ensure that it was accurate and of course given the motivation of myself and my group is to do things at scale to make conservation more efficient the question of course was how to scale it the reason that just counting detections isn't enough is because of detectability and as most people will know the detectability of virtually everything is not 100% and certainly that's true for cryptic species like koalas so whenever we go out we're always going to be missing something and we need to find some way to be able to adjust our estimates and to get some error around that and mixture models had been very popular and still are quite justifiably their repeated survey method that became popular because there was no need to physically mark individuals and it hadn't been applied to automated detections from drone derived thermal imagery and so we were quite interested to find out whether that was going to be a relatively simple way for us to get some decent abundance estimates and I'll point you towards Evangeline's paper which is in ecology and evolution that's from 2020 so I won't bother talking about the music down here because it hurts people but we were very very lucky and a lot of this to have surveys from Petrie Mill in Queensland and the important thing about this is working with John Hangers team who are in there doing the spotting from the ground but they also had radio collar koalas in there and the absolutely essential part of this was that meant that we had true positives so we knew for for the first part of this we knew how to train the algorithm because that meant that we could find out the things that were actually koalas and then we could tell the algorithm what wasn't a koala even though it may have seemed to be one it was important here too because what it meant was that we had true counts and so whenever we were going to make an abundance estimation method we could start doing a comparison against what was actually there and that's a pretty rare thing certainly it's quite hard to find for many species with sufficient numbers so in mixture models consist of a couple of different bits there's a p formula which is has to do with the impact of factors that vary across the study site on the probability of detection and in this case the probability of detecting koalas there's a five formula which has to do with the impact of factors that vary between surveys on the probability of detecting koalas and there's a lambda formula which is the impact of environmental factors on the underlying abundance of koalas so eventually indeed an extensive literature search and I'm not going to go through all the modeling because it probably goes beyond the scope of the current presentation but there were a number of covariates investigator for each component and they would be things that are probably on a great surprise to people forest cover distance to observer things like ambient temperature is important for detections and wind speed so that impacts on the detections from drones and things like forest cover grass cover and the end mixture model really performed surprisingly badly and we were quite curious about that at first so we knew for example in the northern side of Petrie mill that there were 18 koalas present it was giving us an abundance estimate of 85 in the south there were 11 and it was giving us an abundance estimate of 124 so we needed to explore a little bit further and probably to come up with a different method now the reason upon exploration it turned out that we were having challenges with this is because with any automated detection method there are a couple of errors that are potentially can occur and one of those has to do with duplicate detections and one of them has to do with false detections so we looked around well that's the royal we have angeline looked around and did a fantastic job she looked at an estimator by Tilecki and Coons who developed this for aircraft flights so from for observers looking at large animals so when they're in light aircraft you'd have two observers hanging out of the aircraft and they developed this estimator and the important thing here is that there was a component which was the likelihood that the detection is a duplicate and they accounted for that so eventually took this a little bit further and incorporated something which is important for drone or you know artificial intelligence automated detection methods which is the likelihood that a detection is a false positive and so again there is that's a relatively simple formula but there's a lot of work behind that with some generalized linear modeling as well and for details i'm happy to talk to you if you want to get in touch or certainly feel free to read the paper so with some extra work to to find out what were the factors that we're going to feed into these different parameters there were things like the likelihood the detection is a duplicate depended heavily on the distance to the nearest detection likelihood of a detection being false was quite strongly dependent on the distance to habitat edge when Evangeline put all these together and went back we could see for example here that we might have a raw count of detection from drones of 41 but when fed through the formula the estimator we would come up with an estimated abundance of 15 and well within the confidence limits and in fact in this case it was the true count of koala's presence koala's present that was a bit of a lucky one but you will see here that all of these estimates the true count of koala's present is well within the error bounds and really pretty close to the estimated abundance of koala's so that we hope is a reasonable approach to take for using these kinds of methods using artificial intelligence and in fact there's no reason it just needs to be used for koala's so in summary for that small component the end mixture models probably aren't a great way to probably not a great approach for automated detection methods and even with when Evangeline compared her method against aerial survey data it was performing much better than that so there are a number of factors that are important ambient temperature is one of them because it will impact on false detections distance to habitat edge is an interesting one that we're still exploring it may have to do with canopy cover and the distance of the drone between detections has to do with simply the likelihood that you're picking up the same koala from a different angle so we would suggest this is probably a relatively general approach that could be used for automatic detection the current project that we have which is being funded by the state government is assessing predator risk zones for koala protection and we're all false the aim of this is to establish areas where dogs wild dogs and potentially domestic dogs that are let off lead where dogs and koalas may come into contact so one of the aims is to develop automated detection for wild dogs from camera traps looking at putting cameras on trails for domestic dogs and sending a ping off if we if we find one there and to do by now standard methodology using trains and artificial intelligence and to include an information and education program in this and you can probably tell this is the last slide that I made but there's a huge number of spelling mistakes on there I do apologize for now these are hot off the press and we do need to go over them one more time but this is the Ural state forest and this is an example of kind of output that we can get we covered approximately 50 hectares here and what you can see there is 12 koalas that were detected in that patch of bush so that was the flights were back in august and I didn't get a chance to put the camera groups in sorry I do want to talk briefly about it was question number three which is about how to scale things we were really lucky to get some federal government funding and what that has allowed us to do is to construct a conservation AI hub the real purpose of this is to scale through Queensland and in fact beyond Queensland and make it Australia wide I came from the land care led bushfire recovery funding and it's not really just about the model or the methodology it's about the model for data collection and analysis using AI drones and camera traps so the intention here is for land managers and they might be land care groups other groups and the state governments to collect the data we would help them to collect that in an appropriate way to feed that back to us to use artificial intelligence algorithms to do detections potentially other things as well so to translate that data into knowledge and feed it back to people so that they can use that for their land management so the construction of the data portal is well underway we're aiming to have it constructed by the middle of December and what that will allow for is remote access so for people wherever they're sitting to be able to upload their data and for us to be able to ingest that and turn it around I do want to touch on just some recent work from Kangaroo Island I'm not sure how I'm going in terms of time here but the Department of Environment and Water in South Australia contacted us post bushfire and asked us to do some work down there the really important thing about that the methodology is fundamentally the same but the really useful thing about this was we for the first time took this methodology and told someone else how to collect the data so previously we have a highly experienced drone team at QUT they're real experts but of course we can't scale across Australia using them we need to get people on the ground so what we did in this was we used a local drone pilot we told him how to collect the data he fed the data back to us we processed it and we fed that back to the department and they were really happy with that so what that meant was that we had a model and really that model is being translated now into this AI hub we're starting off with koalas but now that we have the workflow we know that we can train the artificial intelligence for many more species we have a really good workflow in place where we know how to capture the data train the algorithms and simply churn out one more algorithm that we can start using for threatened species or in fact for pest species we can also do vegetation analysis canopy cover and we can use LiDAR data as well for this I will just briefly touch on one more thing I mentioned previously but does bring up a question when you're using a new technology about what are the factors that influence detection we wanted to make sure that we had a good grasp of what was going on so that you know we could plan things a little bit better so Evangeline looked at some of that data with repeated surveys and the true positives that came from Hangar's group so we had 119 observations of radiocolor koalas from North and South Katrina what she did with that is looked at the notes that the ground-based observers took and from that developed a scale of visibility to ground observers so remembering that they knew where the koala was because the koala was radiocolid so they knew there was koala in the tree but they were also writing down whether they could actually see the koala so for example whether it was completely obscured from all angle angles or whether it was marginally obscured so we could develop a code of how well a ground-based spotter would have been able to detect that koala versus what we would be seeing with both automated image analysis from Adrienne and manual image analysis okay thanks so I'm not going to go through this in any great detail but with the construction of some ordinal and generalized linear models we found several factors that led to greater success effectively in both ground-based observers and drone flights observing koalas so for ground-based observers they were less likely to detect koalas when the koalas were female when they were higher above the ground which makes sense if they're further away from the observer in taller trees and it's probably part of the same thing and in a particular species in Perimbia which is an interesting one and we're not exactly sure why that is now the really fascinating thing for us we were expecting a whole host of different factors that were going to impact on detection using automated analysis and there were some with manual analysis and I'm not going to go into that it's just going to take up more time but we found that none of the covariates explored reduced the probability of detecting koalas when we used an automated analysis method so while there were a number of factors that were impacting on ground-based observers none of those same factors were impacting on the automated analysis so it seems like it really is quite a robust system and that was really important to us to find a method that is not only efficient and broad-scale but that we know is both accurate and robust so the key findings from that females are likely to be underrepresented we're likely to miss koalas in taller trees from ground-based observations and species composition makes a difference and that can change potentially change the estimates of koalas koalas accounts or detections from the automated analysis well the detection isn't impacted by koala sex or attributes of host trees and that's it for me now thanks very much for your attention and I'm really happy to take any questions thanks very much Grant thanks very much Grant we've got time for a couple of questions and you've certainly generated some interest through through your presentation the first question is from a practical sense what would the cost be to use thermal drone imagery when surveying koalas um well the the nice thing about technology is it's always going to get cheaper so at the moment the kind of kit we use costs the same as a small car it's um it's not cheap but it's it's not just about the capital investment it's about what you're saving in terms of labor costs and certainly if you're doing repeated surveys it's going to mount up so you're going to save money even though you need an initial investment having said that there are new drones that have just come out for around ten thousand dollars that have really good both RGB and a high resolution thermal sensor so you're probably looking at the moment um at 10k you need to um train and all those kinds of things but um that's probably where you're looking yep okay i've got another question here which is um the the person asking the question will be interested to see if there's any correlation between regional ecosystem types and koala density or koala detectability i guess as well um that you've found using using the drone research and yeah comments on that uh fascinating question and yeah obviously that's something we continue to look at there there are going to be factors um that impact on drone detections one of those are can is canopy and re types certainly feed into that as well and we're we're really keen to get more data um to be doing that to to be figuring that out but you know let's be realistic though those um so forest types canopy types they impact on all detections so it's not just going to be from drones yep and just a final question um what's the mechanism that leads to duplicates i think in during your presentation you mentioned that sometimes you get duplicates just interested in how that happens because what you're doing is flying strips so typically it's called a lawn mower pattern so like you do at home um you get the lawn mower and you um run up and down and there's an overlap so there's a chance that at the edges that you might detect koalas um twice once going that way and another time coming back up that way so we go through and um with the uh we have a quality assurance after we do the automated detection and with the qa we you know remove most of those um but there's always a chance that some are going to creep through yep so that's the mechanism okay um we're going to have to finish it off there there are a couple other questions that people have posed and um those people can try and connect with with grant and we'll certainly follow up with grant to get some responses back to you um thanks very much for your presentation this morning grant um if you want to join us for the next session please use the back to the timeline button and we'll see you again shortly thanks very much hey good morning and welcome back to the second presentation of this morning's koala co-lab conference series our second presentation this morning will be given by professor zhun zhao and it's in relation to koala identification using artificial intelligence zhun received his phd degree from the university of alberta in edmonton in canada in 2006 he joined griffith university in 2012 where he's now an associate professor in the school of information and communication technology at the nathan campus his previous appointments include research fellow in the research school of computer science at the strane national university and researcher in the camber research laboratory um of the nisc ta dr zhao has a was a recipient of an austrian research council discovery early career research award in 2012 his research interests include patent recognition computer vision and hyperspectral image processing with their applications to remote sensing and environmental informatics i look forward to professor zhao's presentation yeah thank you for the introduction hello everyone um today i'm going to give a brief introduction on an ongoing project that we are doing titled predicting koala road crossing behaviors using AI powered observation networks this is a project funded by the concerned department of environment sciences community sustainability action ground round four for koala applied research in southeast prinsland um the reason we are doing this is because vehicle collision is one of the main reasons for koala injuries and fatalities uh our colleague uh doglass curling made a statistics uh and formed from 1997 to 2018 an average of 356 koalas entered care due to the vehicle collisions therefore mitigating koala fatalities and injuries is one of the most important tasks for koala conservation um however to do this task it requires a very deep understanding and better prediction of koala road crossing behaviors which requires a more advanced koala monitoring and tracking technology uh so far quite a few technologies have been used by the our colleagues here in in prinsland which include a gps and vhf tracking koalas radio frequency identification uh drone technology which has just been introduced by grant and camera chaps all these technologies have their pros and cons and among them were particularly interested in the camera chaps the reason is it has some unique advantages first it is not intrusive we do not need to capture the koalas and install devices and secondly it can be deployed easily uh at different locations and it can provide continuously 24 seven monitoring capability and more importantly it can see the content in the field of view we can see the status of the koalas but um as many of you may have already experiences in using the cameras there are also some challenges in in deploying the camera chaps first there is no uh good data transfer and management uh technology or systems available so we often need to go to the site where the cameras are installed and retrieve the memory card and copy the data and take them back for analysis so we cannot do the analysis in real time and secondly if we want to identify animals in the images or videos that have been captured we'll have to inspect the videos and images visually which is a quite tedious and labor intensive process um so so far many researchers in artificial intelligence have been making efforts in developing automatic analysis video and image analysis technology so over the past several years there has been some breakthrough in machine learning in particular powered by a technology called deep learning which is derived from the traditional neural networks the technology has enabled more reliable detection and classification capability for from image and video analysis so this trend of booming in the artificial intelligence technology started from to some trial when deep-dune network was firstly applied to a image classification contest digital division which successfully reduced the error detection rate by almost a half and since then many advanced deep learning models have been developed which greatly increased the reliability of object detection and classification so for this reason we generated the idea of adopting the technology for quality monitoring and recognition so we would like to develop a deep learning based new technology tool that can integrate the recent advances in wireless networks, data analytics and deep learning basically intelligent intelligence and to provide the capability of doing automatic data collection processing and management so here is the proposed system structure for caller monitoring the system starts from a camera that can be triggered by motion for any movement in front of the camera then the trigger will allow the camera to capture images and videos these images and the videos will be transmitted to a backend server at previous university through the wireless 4g network the images and the videos sent through the email will be automatically pulled into a sharepoint server where we have developed the AI technology for caller recognition that can be applied to automatically recognize callers in the captured videos at the same time we also allow the general public to get involved to upload their images for the training of our AI technology so there are three types of detection tasks in the whole system first we want to detect whether there is an animal in the captured images or videos and secondly we want to identify whether this animal is a caller and the third one which is most challenging is we want to know whether which caller it is so it's basically an individual caller recognition task so far we have completed we have almost completed the first two tasks and we're going to start the third task very soon so this is the camera that we have used for data collection it is a compromise 4g channel camera it can be set to capture both images and videos at different resolutions or the length of the videos it can be set to be triggered once there is an animal movement in front of it or any types of movement in front of it and it can also capture images during both day and night in order to change the AI system we need a large amount of image data however it is quite difficult to capture callers in the wild therefore we have collaborated with the local caller centers and the centuries so that we can deploy cameras in their centers and capture the callers we have collaborated with long time with this hill caller center with Korean bean gym ward and local caller hospital at the same time we have also obtained get the help from many colleagues and the general publics for example you know South Queensland Queensland caller action group in red land and also from a ds who provided us a large amount of images they have captured in their project at the same time we are also collecting images from the Atlas of Lingwall Australia and also from the internet to expand our image pool so after we have collected a large amount of images from these caller centers in the second stage of the data capture we deployed cameras in several road crossing structures we have obtained the permit and got the help from Brisbane City Council from the Red Land City Council Queensland Department of Transportation and Main Road and also Queensland Rail so we deployed several cameras in the road crossing structures which I'm going to talk a little bit more later so here shows some examples of the deployed cameras the data capture system has several components the camera itself which has an antenna that allows it to connect to the mobile network the camera is powered by batteries but it is also connected to a solar panel so that the solar panel can charge an external battery to allow the camera to work for longer for the project and also to work at the night so we have also put a warning sign next to the deployed camera system to tell the general public that this is a research project and in particular we have put a QR code on the sign so for those people who are interested they can scan the QR code and upload their photos to our solar to expand our training pool so this is the sign so at the bottom here is the QR code to facilitate data management we have developed two applications the first application is a dashboard which collects all the information of the cameras and uploaded images and the videos so we can see from the center here center part here we have the location and a list of cameras which shows from each deployment site and from each camera how many images we have collected so far in total and how many images have been collected over the past day at the bottom it shows that we have collected more than 35 000 images and videos up to yesterday so this was the data i collected yesterday and at the top right it shows the more information about the captured images and the camera it shows the time when the images and video was captured and what's the status of the battery and the quantity of the signals and the bottom is a statistics about the number of images captured from each caller over the time so we can select a particular site or select a particular camera and at the bottom it will show the statistics so this interface allows us to quickly check the status of all cameras so we know whether a camera is working or whether the battery has to be replaced or it also tell us whether at a particular site there are any interested activities if not then we can if we move the camera to a different site this is the second app that we developed which allows us to have a live view of the captured images and videos so we can the site where interested and select a particular day then a particular camera it will show a list of videos we can click the videos and the paid content so before our automatic caller detection system can be linked to the whole monitoring system we can use this tool to check the content of the captured data so here I want to show more statistics about the deployed cameras the location and the data captured so here are two examples on the animal crossings at Compton Road and Illamina Street so at Compton Road we deployed five cameras at Illamina Street we deployed four cameras and so far we have captured about 20,000 images and videos from these two locations we deployed four cameras at Berkdale which is to monitor the behavior of Golas underneath a rail crossing so this is in collaboration with Queensland Rail and we also got the help from our colleagues in University of Southern Queensland they have deployed a GPS system to track a caller called Rainbow there and the Queensland Rail wants to know whether Rainbow has been using the actual railway or the underground pass so we deployed a camera there at the end of September and about three weeks ago we captured a caller twice using the pedestrian pass on the railway but surprisingly it's not Rainbow it's a different caller because Rainbow has a Joey it's a female caller and has Joey but this captured caller does not and at the beginning of this month we completed the camera deployment at April Park Creek and Cullin being Cullin Creek in Redland so far we have captured about more than 100 videos from these two sites so once the data have been captured we need to label them so that we can provide the AI system with the training data so we have used a tool called label IMG which allows us to select a jaw bounding box on the images and then we can specify which classes of the animal a target it is so here it shows some statistics on the labeled data so far these are all manually labeled we have labeled more than 7,800 images and with more than 16,000 instances or objects among them are most a caller we have more than 10,000 caller images or call us labeled and we also have labeled a lot of people who can grow person bird and so far in total at this stage we specify 17 classes of target for detection but we have not got the training images for for many of the classes so for the audience of this event if you have some data and if you can share them with us please contact me that will be very helpful for for the development of the AI system so there have been rapid the development in the deployment models since 2012 so every year there are large number of models being published so we need to select the most suitable one for the detection task so we select the model called YOLO which has also been mentioned in grass talk it is one of the best object detection models that can integrate both detection and classification into the same network structure this model is very well maintained so there has been updated every year this year the latest version five has been released it is very accurate and reliable so here is a bit more about the data structure of YOLO at the left hand side is the input images or videos then in the middle are the deep neural networks uh then on the right is the output basically the output shows the location and the the size of a burning box of the detected objects in the image there can be multiple objects detected and for each detect detected object a confidence value will be given and then and possible classes for this detect object so at the bottom I show the some performance parameters so we can see that the precision of detection is pretty high if we set the confidence to about 0.5 and and same for the recall and the F1 measure so here I want to show some detection results in videos these from the base field see that the and also the confidence here are the videos collected at a long time objects at the same time and even when the objects are heavily uploaded for example call behind the leaves we still can detect them reliably and here are some results of call captured in the road so you can see the same caller it's probably the same caller that has been captured in august september and surprisingly in this october video the caller is injured so you can see it it's creeping so we have also captured the types of animals like fox or possums so so so in the next two months we're going to detect our expand our detection system to other classes of animals so far we have already been able to detect crawlers and in a longer term we will start the work on individual caller recognition models so finally I would like to introduce our team so we have Dr. Wilenton who is an expert in wireless sensor network Dr. Kerling and Guy Custley they are animal ecologists and who are specialized in caller appellation research caching texture who is working in red and safe console and very experienced in caller monitoring and conservation first hell and oven they are my students who by developing the AI based recognition system and the TANLI is our research support officer who developed the apps I'd also like to acknowledge Queensland Government for funding research and also all partners for their support to the project so it's my talk thank you very much thanks very much John um really interesting talk and and really advanced technology we're looking at in terms of assisting us in the way in which we survey crawlers in the future you've generated a couple of questions that we're interested to follow up on so first one is when we are doing I am not a robot image recognition are we really training artificial intelligence to recognize what we see yeah yeah we are so AI needs to be chained in order to provide very accurate and reliable detection and recognition results so that's the reason we have produced this large amount of image pool and and get them labeled in order to provide this accurate detection capability um so far we have only labeled about seven seven thousand images and we're planning to label more with the data captured in a while so one plan is to get the general public involved so we have discussed with caller action group in red land who has continually agreed to to provide some help and we're also trying to engage our students to do the labeling so so basically the AI needs to be taught uh what the object is so yeah so that's the that labeling is useful okay just to follow up from that one um in in that labeling with those large image sets that you've got is that you're using students or specific experts to do that identification it's students students so that that's why we in the the classes uh list of classes that we have produced we have classes we did not provide two very detailed uh species so because we are worried that the general public does not have the domain knowledge to distinguish uh or different uh species yep um somebody's asking what software models are you using for for identification processing and um are you able to share links with regards to that software yes of course we're using a um a deep learning model called YOLO which uh the source code can be downloaded online i uh i can uh yeah please just email me if you are interested in using the link and also uh planning to release the data to the the public so especially our colleagues here in Queensland uh so if you're interested in using our system or if you're interested in uh monitor callouts or other animals at a particular location please contact me so we can walk together to get the system deployed and capture more data for analysis great um one final question is um will will the data capture image be able to detect things such as either whether animals have pouch young or back young and also whether animals might show signs of disease such as the the sort of staining you get associated with chlamydia infection yeah i i think my understanding is anything that is recognizable uh to human can be recognized by uh ai so as we change the ai to do it for example if we want to identify a particular then we have to provide the ai with images labeled that these are the symptoms for particular disease for example for the injured caller video that i just showed we can teach the ai that this way of working means the caller is injured so we can label the video as this injured caller and if we provide the machine learning algorithm with many of these training data uh quite a few injured callers and quite a few uh good and healthy callers then ai will be able to detect injured or healthy distinguish injured and healthy callers great that's good thank you very much john for for your presentation this morning um the next session is actually a wrap-up session which will be run as a as a panel session so if you want to join that please um return back to the timeline and join us then so thanks very much good morning and and welcome back to the final session of our uh sixth theme sessions under the koala colab series for 2021 um this final session will be an opportunity for myself and professor jonathan roads to share some information and reflections with regards to the koala colab series for this year to provide some feedback in relation to the information that's been provided by attendees uh and presenters um so we'll have some short presentations from jonathan and myself and then uh we'll open it up for us a question and answer session just to introduce and background jonathan for you jonathan is a landscape ecologist working at the interface between ecology and policy to improve and better understand decision making for biodiversity conservation and ecosystem services he's based in the school of earth and environmental sciences at the center for biodiversity and conservation science at the university of queensland he received his phd in ecology in 2005 and has been employed at university of queensland since 2007 jonathan has a particular interest in understanding how dynamic landscape and climate changes drive biodiversity and ecosystem service outcomes and the implications for this for effective environmental decision making and policy setting under uncertainty he's also interested in optimal monitoring and quantifying the benefits of investment in learning and research for environmental outcomes his work has application across a wide range of scales from local to global and in both urban and rural landscapes a key focus of jonathan's work and one of the key reasons we wanted jonathan involved in this session is the application of our understanding of drivers of change in environmental systems to decision making and policy formulation he uses decision science approaches to achieve this and works closely with end users especially government to maximize impact and uptake this work has been has made important contributions to understanding appropriate management and policy for koalas particularly in austral in queensland and in coastal planning and linear infrastructure planning so i'm really pleased that jonathan has agreed to join us today and participate in this group session now first of all what i might do is i'm just going to share some reflections in relation to the koala colab during the course of the the conference series we've collected information from participants and we've also made observations in discussions with both attendees and with presenters so a quick review of the koala colab 2021 by numbers um as you'd all be aware we had six themed virtual sessions over a six-week period those sessions featured 22 presenters that included six from state government three from local government eight from universities and five from non-government organizations that resulted in 550 minutes or nine hours and nine minutes of presentations and q and q and a's and all of those are being stored to a youtube channel so they will be there for some time for people to review and to reference during the course of the colab series we had over a hundred over 1300 registrations across the six theme sessions which was over 200 people registered for each session whilst the actual attendance at those sessions wasn't that hot wasn't as high as the registrations we did achieve over 50 percent of registrations attending each session which the nectar creative team who are helping us run this virtual series say is is quite high for these types of virtual events we had really good responses to our online line session surveys and also to the koala colab series survey and i'll present a little bit information from those surveys as part of this presentation and further there's a commitment to host the next koala colab in two years time during the during the surveys that we set out during sessions and after the the fifth session we asked a couple of questions and one of the primary questions at the end of each of the sessions was did this topic provide you with learnings that you can apply to your work and pleased to say that overwhelmingly the response was that people agreed with that statement so over 70 percent of people agreed that the information that was provided in each of the sessions provided them with learnings and was relevant to their work 20 over 20 partially agreed with that so again we're looking at over 90 percent of people who found value in in the presentations and the information that was provided through the colabs as part of our major survey we question people is how did they find out about the colab and majority of people over 50 found out via email as we sent a vision six email to people who'd been engaged in the koala strategy consultation process it's pleasing also to see that people have obviously passed that on so word of mouth and website information was also used to access and identify the opportunity provided by the colab we had a couple of free format questions within our surveys and the initial analysis was to generate some some word clouds and that first question was how did you think we can improve the koala colab so there was really strong feedback to say we needed to provide more time for questions and I think as a virtual event we were limited by how we could do that but certainly a keen as we move forward with future colabs to look at opportunities to have increased time and increased opportunities to workshop and discuss the information a few people raised some difficulties with some of the technical aspects of having a virtual conference and again we'll be sitting down with the nectar team to reflect on how we worked through those and some of the things we might do in the future there was also some feedback with regards to some stronger focus on legislation on how people can get engaged and involved in in the programs that have been run so that's important feedback and we'll continue to work our way through that in the coming weeks we posed a couple of other questions one was those was how likely are you to recommend koala colab to a friend or colleague and again really strong support there for the sessions and interest in people providing recommendations about participation in colab so that's a really pleasing result for us we then asked people overall how would you rate the koala colab conference series and again by far and away the majority of people rated it four out of five some people rated it five out of five which again is really pleasing for us so really strongly positive support for the colab series and the information that was presented during the series we asked people about what they like most about the koala colab again the these responses are quite spread and that that again is an indication of the diversity of people who attended and the diversity of topics that we covered during the colab really wanted to highlight in particular the collaboration again it's a really strong theme that has run throughout the colab and was really important for us to to emphasize and demonstrate that there's significant cooperation going on there was also a significant focus on new research and information so information about research that was important to koala conservation but also updating people with regards to new initiatives and plans to support koala conservation the next question was around what topics interest you do most and this is probably the most even spread and again it reflects that people had interests in all of the sessions that we set up so there wasn't really a session that people didn't think was worthwhile all that didn't provide information that people were interested in so again I think that that reinforces for us that we probably didn't all right at getting the themes arranged and in the information that was presented the final question in the broader survey was what topics would you like covered at a future koala colab the word koala obviously stands out very strongly but that's because most people were using that in their responses some of the key stuff again was the focus on legislation and the need for changes in laws and policies around koala conservation I really highlighted the need for additional community engagement and involvement habitat again featured strongly both with regards to the protection of habitat but also the restoration of habitat and how we addressed such threats as clearing so again really strong feedback that we've received that we can again use to plan both how we respond in developing the future co labs but also how that influences some of the programs we're currently running now finally I just collected some some dot points on some personal reflections about the co lab and from my perspective I think the presentations highlighted the breadth of work that's being undertaken to support koala conservation in Queensland so we saw all things from quite applied research to on the ground planting trees protecting koala habitat managing threats in relation to dogs and cars modifying infrastructure to to prevent koala impacts on roads so a breadth of work that's being undertaken across Queensland the mix of presenters across state government local government universities and non-government organisations indicates there's significant interest and investment outside of government to support koala conservation so there's significant will and interest in these areas and it's really encouraging to see how many groups are involved and keen to support koala conservation a lot of the presentations also highlighted the extent of partnerships and collaboration and I think the the two sessions this morning again highlighted that the list of organisations that were working in conjunction with the universities to progress that research and how that research was then going to be applied and integrated into programs by government local government and non-government organisations all of the presenters highlighted gaps in knowledge they also highlighted a range of gaps or problems with legislation and policies that need to be addressed but in the majority of cases they've also offered solutions and ways forward so again I think there's quite a positive feedback that we receive through those presentations and in the feedback that attendees provided through their surveys and finally just the level of registrations and attendance at the koala co-lab really reflects the high level of interest and support for koala conservation and that in itself creates a challenge for us both as a state government but as a group of people who are interested in conservation of koalas is how do we effectively harness and deploy that support to both progress key initiatives and to lobby governments and other key groups to drive changes to support koala conservation so I think the final reflection is just really strongly positive feedback and experience from the koala co-lab some clear messages that we've received and now can progress in terms of both planning future events but also in how we approach the projects and initiatives that we have underway that's the end of my presentation I'll now invite Jonathan just to make some observations and provide some information from his own perspective. Excellent thanks thanks very much Jeff and I would just echo what Jeff said and that I think this koala co-lab series has been outstanding I haven't been to all the sessions but I've sort of read the reviews of what was in the presentation so and I'm really impressed about the breadth of work that's been going on and it's a really nice way to bring people together unfortunately we can't do it in person but in some ways there's some advantages of the online platform as well and more people can attend so that's fantastic so I think this is a really great idea and look forward to it continuing so look I was asked to really I guess reflect on knowledge gaps and some of the stuff we've heard about in the co-lab series is around knowledge generation and but some of it about implementation and actually doing action on the ground and so you know but rather than really just I guess you know go through a list of what I think of the sort of some of the knowledge gaps and clearly there are knowledge gaps across you know across sort of the koala conservation space I thought I'd just reflect a little bit on you know how you know the value of knowledge and monitoring and research and I guess reflect on how I think about knowledge generation and the importance of it and so I've been working for a number of years around using things like decision analysis to try and understand well when is knowledge valuable and when is it not valuable and when is research valuable so it's a real interest of mine so I thought what I do is really just you know to get us thinking a little bit I suppose is you know talk about you know how I think about knowledge and how we can perhaps think about it you know in a structured way in terms of you know what should we prioritize on so I'm not going to make any specific recommendations as such but I'm just going to going to to talk about it sort of I guess a framework of thinking about it and make some broad broad observations I suppose so so I sort of think of knowledge generation and research and monitoring as part of the same thing it's really about trying to generate new knowledge and and reduce sort of uncertainty so and and I think there's probably four things that we really try and you know with monitoring research probably four areas which we focus on and reasons for doing and one is so the first is really and this applies to monitoring is really tracking up progress so you know a quality population is going up and down really important reason for for monitoring and doing research the other is just engagement so do we want to engage with the general public do we want to change behavior citizen science is a really good example of this so we you know maybe that there's there's there's benefit in collecting information just for engagement so people know what's going on people engage in the science for instance the third reason I I just like to think about is is just doing monitoring and research and collecting new information for surveillance purposes by this I mean really we don't really necessarily have a question in mind or a problem which I'm sold we're just trying to collect information because something might come up that we didn't sort of before and that's a really valid reason for doing for doing collecting data and developing research programs and the last one which is probably the one I'm most interested in is where we're really we're really targeting knowledge gain towards improved conservation outcomes so it might be to improve policy it might be to improve management decision-making on the ground and so on so we've got a really targeted approach and that's what I want to talk about today and sort of highlight a bit of a framework for thinking about that and how do we what's the value of of research and knowledge gain and the other three reasons for doing research and monitoring and getting knowledge in some ways you know we're always thinking about well how does this improve conservation so so I'm just going to go highlight a bit of a framework for thinking about this and I guess the first starting point is really that knowledge gain on its own for management or policy making has has no value right it only has value in a conservation sense if you know if it actually impacts decisions on the ground or outcomes for koalas and so from that perspective we can think of knowledge having no intrinsic value apart from it might be interesting but for koala conservation it's not valuable until it results in a change it's okay and so the other thing is that knowledge gain is not free okay so and we sort of have to weigh up the costs and benefits often you know you know trying to prioritize what what we actually work on and so we need to weigh up the costs and benefits and we've been using an approach over the last few years called value of information analysis that that tries to link through from knowledge gain to outcomes in decision making and trying to evaluate the costs and benefits and it's surprising actually how often we find that collecting more data or more information isn't worthwhile you might as well just go ahead and make a decision so that's pretty common at least in our analysis and so so I think you know one perspective is that you know sometimes we may just need to make a decision now rather than waiting for new information but sometimes we might be valuable to wait so the thinking through through that is really important and so I'm just going to look at that this is a framework we worked on with David Panolf who's an economist an environmental economist at the University of Western Australia so we worked on this a few years ago and I think highlights you know how we can start thinking about Kuala research and this is really trying to link research to outcomes on the ground and this is really focused on policy changing policy through research but it could be you know it could be really also about influencing management decisions on the ground or restoration activities whatever so you can you can switch out policy here for whatever you want whatever you're trying to influence with your research and so you know so this sort of steps through this and the first thing we need for our research really is it provides relevant information and you know so if it doesn't provide relevant information or management decision making or policy decision making then it's not like unlikely to influence change and so let's assume that we have relevant information if we do have relevant information then really there's another step right so so once we have relevant information what's the chance that it's going to influence policy in this case or influence our management decisions or influence where we're going to do restoration or and and so and sometimes you know so we need to think about the pathway so for instance you know if you're if you're doing some research on transportation success what's the chance that that's actually going to result in a change in policy around you know maybe regulation or guidelines around um transportation that's that's that's so that's the first step how does it influence will it influence policy or management decision making often as well once we've changed policy um that you know depending on what that policy is there's there's often a requirement that you know there's a change in behavior so so does does it result in a change in behavior such as you know landholder protection of vegetation on their properties for instance or in a translocation case does the change in policy actually influence how people undertaking transactions operate and sometimes there's a direct link between this it's regulatory it might be much much more direct but if it's it's more incentive mechanisms for instance then the there may not be necessarily changing behavior even though policy has changed so that's another thing to consider in terms of the link between research and outcomes and then the final step is is really does the change in behavior or change in policy actually result in improved environmental conditions on the ground okay so are actually implemented that actually result in benefits to koalas and you know in the translocation example it maybe does you know better transportation practice for instance result in actual improved outcomes for koala populations say across southeast so you can see that actually there's a whole chain of things that have to happen if our research is to be valuable and and I think you know these are some of these are barriers to research impact and and ultimately we want from our knowledge gain or research to have impact but so I think but these these barriers and the the barriers vary a lot from the type of problem we're trying to try to solve some in some cases there may be you know depending on where the major barriers are there may be no chance of success of actually having any impact and so and I think what this and I think what you know so I think when we're thinking about knowledge gain and trying to think about where the where the where the sort of you know the gaps and such I think it makes sense to me to start thinking well what's the what's the decision or the policy or the outcome we're trying to influence first and then design our research around that and prioritize our research to really put bets on the ones that are really going to have the biggest impact and then secondly you know the my sense is we don't do enough research around how do we reduce and and certainly the first two two barriers around influencing policy and influencing behavior they're largely you know political and and socioeconomic in nature and I think if I was to identify any gap it would be really well you know can we start thinking more about how do we do read what research do we really need to minimize these barriers and in a broad sense we sort of know what to do to protect koalas the the issue really is the the socioeconomic and political barriers that that prevent us doing it so I think you know you know if I was to identify any research gap it would be really how do we minimize these or reduce these barriers so when we do our koala research there is this this pathway to impact so so I'm going to leave it there I mean I really wanted just to you know highlight this way of thinking about knowledge gain and I'm happy to have some discussions and hopefully it sort of triggers some thinking around how we think about research and monitoring and knowledge gain okay thank you thanks Jonathan really really insightful in terms of the the stuff that we've looked at before in terms of the research and work that's going on and how we might best frame that or progress that to to address some of the key challenges that we're facing um there's a question here that somebody's posted that um this is a comment but also I guess big some reflection is they're saying applied research is valuable but so is pure research applied research builds on pure research so it shouldn't be undervalued so yep is that a question for me I suppose any any observations or comments yeah yeah I definitely agree and then we do need pure research and I sort of put you know when I when I was talking about the three types or four types of research or or monitoring I you know I would put the the pure research really in that surveillance bucket I suppose which is really you know we do need to do this stuff because it's going to be useful down down the track potentially then maybe a lot of you know pure research that isn't useful but that's okay and um and I think so I think yes there needs to be we need to balance both you know really directed applied research with the with the pure research and I don't have an answer about for where that balance is um and then the other thing to recognize is that especially sort of pure research sometimes takes a long time for for for outcomes to materialize so it can take decades or sometimes even you know many decades to actually get those um those uh outcomes so you know um so yes I agree and there needs to be um so I talked particularly today about this this idea of these the this directed research but we need to have them yep um I've I've got a question which I guess I'll make a bit of a statement then I'll get your response to that is I think in in a sense for the koalas in particular we've done a considerable amount of ecological research so we probably have a it's fair to say a fairly good understanding of the the critical aspects of the ecology of koalas and the types of things that are impacting and affecting their populations within Queensland and probably across Australia some of the key challenges however I think are affecting that change that's required to provide better protection or to drive improved rehab outcomes or to change the way in which governments or other organizations invest in research or conservation initiatives and I guess my question is um that's really about as you said changing behaviors or influencing a change in thought processes and the way that we approach things would you agree that there's probably a need for some basic sort of psychological socio-economic type research to look at how do change the way landholders or the community value koalas and how they interact with government and with other organizations to to affect change yeah um yes I think I mean I think so I think the I think the the challenge of conserving koalas is a job for all of us it's not you know obviously government player really major role for setting the policy frameworks and regulation around that and also facilitating community groups and so on to to achieve conservation goals but at the end of the day you know it needs to be it needs to be a whole of um you know whole of community approach I think and um and and some of that maybe behavior change and and changing you know trying to change people's values to value you know the environment more and we know not beyond koalas and I think it's broader than just koalas and so you know I think you know bringing you know trying to focus around I think I think that's you know it's clearly a barrier right and and and so you know and research can help I mean it's not as an ecologist I don't work in that space but I do work with social scientists and economists quite a lot and they have a lot to say on that and lots of lots of nice ideas um so so I think you know bringing you know if you can bring in people who work on those aspects I think we could make a lot of gains I think um and you know it is it is an important aspect um and you know I guess the other thing is that you know when people are making decisions and governments are making decisions of individuals are making decisions that they're always trading things off right so that they're trading off you know their livelihoods against environmental um you know gains and and so governments right they're they're trading off you know economic factors against political factors against um environmental factors and so you know can we find ways that that minimize those trade-offs can we find smart ways of you know so so there aren't these strong trade-offs between say developments and um and outcomes for koalas which you know we're clearly there at the moment but you know and that you know could be around um smart development you know smarter ways of developing and so on so you know so that's another sort of thing I think about a bit is how do we how do we you know minimize these trade strong trade-offs that that allow us to not necessarily get win-wins but because I think win-wins are rare but at least um so we don't have to trade things off a lot for for environmental gains yeah because I tend to agree and that we end up in very heavily contested space where in some cases we just have winners and losers we don't have opportunities in some of those senses to have a negotiated outcome with that best benefits the conservation of koalas so in some cases there are clearly winners and losers in some of those interactions well I think just an observation I think that the behaviour change stuff it's it's pretty well understood that just more information or more knowledge is not enough to change people's behaviours you need more than that and so I think this idea that you just provide people information is sort of an old idea and at least that's my understanding from talking to the social scientists psychologists and so on that actually you have to be much more proactive than that and it's about communication and changing social norms and things like that so and that's hard yes so yeah I think we need to move beyond this let's just do let's just present the information and things will be okay because it's it's unlikely that's going to work on its end and that's that's certainly something we've been learning um one of the projects that was presented during the co-lab was the work we're doing with Griffith marketing school and that that again for us was quite a quite a step out of the norm in terms of um a project we would do and that that group is working very much with the community to identify how they value koalas um what action they would be prepared to take what um would encourage them to take that action and then designing programs around um understanding how the community thinks and values koalas so it's it's again I think we need some more of that in terms of some of the other work that we're doing to help us understand as you say the best way in which our knowledge can then influence positive change um I've got a question here which I think you may already have sort of touched on but I'll I'll pose it anyways um and it's relevant even your role on the koala advisory council and other government sort of committees associated with koalas it's what do you think is the best way of getting government uptake of applied research and what are researchers currently doing to get government to listen to and apply their research well yeah I mean this might be best answered by you Jeff but I can give my view from sort of the outside um look from from my experience I think it's it's it's working closely with government and um and co-developing research um and so I think from my perspective I think that's being probably the most successful strategy is actually you know when you start when you're thinking about research rather than just think thinking about think you're interested in is actually talking to government or talking to your partners who have they made the NGOs or you know and it's actually you know working out or what are you know listening and trying to work out or what is it that that's really challenging what why can't they get conservation out of koalas for instance and and so what are the things that are really hard to do and on what do they need and I think that's and trying to co-develop research I think is really is really is really important um you know and um you know some that's with various success in terms of having you know getting government to listen um and as many of you know but some you know there are and um and so you know but I think it's and I think it's um trying to develop that co-development of research um over a long time period as well so if you can you know research builds on each other on on itself right so it's actually trying to build a program of research around um the key challenges that governments face in getting outcomes on the ground and um you know and I know that's not necessarily easy to do is it you know talking if you're a researcher and but I think that's you know my experience is that's probably been the most successful way to proceed with you know in trying to get research visible and um and working with with end users more generally I think so but I don't know Jeff you you probably have some views on this from the inside as well and and what what what you find useful you know I certainly agree with the idea that um co-designing research and and strong partnerships with research organizations is a key to getting research that's fit for purpose and is tackling those key questions that we face in government with regards to implementing programs to to support the conservation of koalas and biodiversity more generally so it is critical to have those relationships but also to have I think enough knowledge within our organization our government organization to be able to engage in that robust debate with universities and researchers around what is important and what needs to happen so I think um from a government agency we need to maintain enough capability internally to understand what our research needs are but also to question and challenge the researchers in research organizations to to I guess convince us of the merits of what they might be proposing but also to think more deeply around um what what the challenges are and how they might best be addressed so I think that's an important part of it is that um yeah to have have that that sufficient knowledge to understand and engage with the research community in a productive way um got another question here which is um it's a bit of a statement than a question at the end so I'll just I'll read that one out and it says we already know that habitat loss and fragmentation is the main problem when koalas come to ground they are then at risk from motor vehicles and predators why is it so hard to get all levels of government to extend this basic discovery to actual implementing actions that retain mature habitat right now replanting rehabilitation is closing the barn door after the horse has gone or do all levels of government buy into the population growth at all costs and the issues that emerge from that premise so did you you can have some comments I'm happy to again it's yeah um yeah I mean it's it's um you're right we as I said before I think we know what we need to do for koalas in a general sense I guess you know if you look at Southeast Queensland for instance then we we know what the problems are we know what we need to fix and and obviously in local situations vary and and sometimes we don't in particular local case we don't necessarily know what to do but but you're right it's it's it's have it at loss and it's it's cars and things like that and and so um and again I think it is just this um you know this this trade-off between economic growth and um and the environment and um you you see it everywhere and um I don't really have an answer to fix it um you know but but I think you know we we had the talk about behavior change and so on um we were talking about behavior change I think that's crucial as well so you know how do we get people to value the environment um and you know that's not an easy task and I think it's um it's you know it but but I think focusing on that sort of behavior change it has to be part of the part of the solution I think um and um you know clearly there are these choices to be made between between economic growth and um the environment but you know you know but it's also short-term thinking as well because clearly in the long term it's not sustainable so we can't keep developing and developing and developing forever and at some point there has to be a stop so you know I think it's partly these trade-offs but partly you know how do we get people to think long term and um and I think humans are generally bad at thinking long term and thinking about long-term risks and um that's the problem too and again it's a chat it's a question about how do we how do we communicate or or or you know so that people realize about the long-term risks of this um ongoing um development and the unsustainable nature of it so yeah you know it's our solution but it's I guess it's just um that's how I see it anyway yeah and I think the the way in which the um climate change convention and things have played out over the recent days has certainly demonstrated that across the world there's um there's a will to make change but then there's clearly barriers to that change which has required compromise the challenge in in minimizing the degree to which we compromise on an environment is the way in which we can communicate and influence um people so I guess that as I said the challenge is both in how do we communicate in a way that is influential in changing the way people think and value koalas to the extent that um we we drive that sort of um thing closer to a to a center um position rather than a position where koalas are constantly lose as a consequence of population growth and development um the other the other challenge obviously in that space is also then um how do we how do we do that whilst recognizing that um we have people that are very passionate on both ends of this scale and and it's a matter of trying to bring those two ends together um to to find the middle ground that delivers the best outcome for conservation whilst not compromising that the other things that that society clearly values quite highly so um we've got another couple of questions um here uh one one is um again it's probably more of a statement that you might respond to so researchers need to partner with those who actually make the on-ground management actions happen such as local government rather than state government so I guess um I'm mildly offended that um people reflect that state government's not doing anything on the ground but um I'd certainly be interested in in your reflections with regards to the importance of partnering with local government yeah um yes I I agree I think um it's really important um that uh that that researchers partner with um with all levels of government um but also partner with community groups and um and you know and and NGOs as well I think that's really important and um you know um but you know I think so but but I see that as a you know different researchers will partner with different people and um you know some researchers work a lot with state government some work with NGOs and so on so I think across the board there needs to be that partnership obviously as a researcher with only so much you know time we have and um you know physically impossible to partner with everyone as an individual but I think as a community we need to have that partnership right across all um all levels of community um and I think it's recognizing that and as I said earlier I think you know it is it is the responsibility for koala conservation rests on all our shoulders and I think we all have to work together to do that so I think you know there's influence across in terms of outcomes all the way from state government down to individuals um and commonwealth government of course as well um so you know there's a role for for engagement with all those but hopefully you know I mean it would be good to identify are there gaps you know across the researcher community you know are there particular gaps with you know are there community groups for instance that really don't have the opportunity opportunity to engage with researchers and you know maybe it's something worth looking at is where are the where are the gaps in terms of engagement you could almost do a maybe a sort of a network analysis to see where whether there's some gaps that could be formed and new collaborations formed that don't exist at the moment that could be really beneficial and that could be something worth considering yeah but I agree yeah um it's not a it's a thought that I've had rather than one of the questions that's come up and and I know the Queensland government for a period of time did actually operate a an annual program that was science in parliament in which they had scientists team with politicians for a day to talk around real science problems and challenges and it was an opportunity both for both to understand each other with respect to how a scientist communicates his or his or her thoughts with regards to what's important and what their research findings mean or need to translate into action but also helps the scientists understand that how what the politician deals with so um from from a change perspective I certainly think that was an important step in sort of demystifying both environments for the respective groups I just wonder if you had any thoughts on other things that might be worthwhile in breaching that gap between the people who hold the knowledge and the people who make the decisions yeah but yeah look I think it's really important and I think as researchers we should do a better job of well at least as applied researchers then we need to I think we need to do a better job of having our research driven by the question you know so actually as I said earlier I think talking and working with users of research to really identify the problem and then then building your research around that and so I think that can help so you know you know if there's for instance there could be you know there's funding opportunities that really force that to happen I think would be good that incentivize that would be maybe a good idea so that you know state government of funding research that you know there is you know a key component is making sure that it's end user driven and so at whatever level would be you know and trying to incentivize that would be useful and so yeah I think it's providing and also providing this the space for end users and researchers to get together and you know and so certainly that's something interests me and I think you know found that to be useful just the opportunity to be able to get together is really valuable and you know and as researchers I suppose we're also driven by interesting questions as well and some of those interesting questions are not useful but that's sort of okay but and different researchers depend on you know varying their extent to which which they tackle really applied problems versus more fundamental research and as we said earlier I think that's important so you know I'm not saying that every researcher should be applied but if you really are interested in the applied stuff then then working together from the start of projects is really important that's good we've got it again it's rather than the questions it's another statement so I'll read out the statement and I guess seek a response there seems to be a lot of one-way traffic here government are not the only responsible parties in this management objectives have been made clear researchers NGOs etc can tailor their research and actions to complement those objectives and keyers and research groups should be encouraged to collaborate and pool limited funding there are too many small groups competing for funding and debating ideology collaboration is the only way forward and government is not the only player in the game I'll probably comment first and so I guess there's a very strong focus in the sq koala conservation strategy around collaboration and there's a there's a clear statement in the in the strategy that says government recognizes and acknowledges that they can't make a difference in this space on their own and I was certainly encouraged by the co-lab presentations which focused on the extent to which we're partnering and collaborating in certain areas I think we certainly still have a way to go in that space but I certainly agree that in some cases we tend to be unproductive when we debate ideologies around things when clearly we have nine-tenths of agreement about what the key things are we need to do and I think we certainly as organizations need to get better at progressing the things that we agree on and not necessarily focusing on the minor things that we might disagree on and I think that's certainly again a rich area for research in terms of how we engage with each other and and to develop and collaborate better I don't know if you had any further comments Jonathan no I mean I I agree and I think the funding isn't is an issue that there is enough funding and if there's limited funding too many people need funding it's going to create competition I mean then so I think we do need more funding I don't know how we go about getting that but and also you know maybe there needs to be funding to to as I said before incentivize collaboration as well so so funding that's actually tied to collaboration because I agree that we have to work together if we're all competing then it's not how helpful it's inefficient and ineffective so so maybe funding needs to be you know there needs to be funding to drive collaboration as well and provide the space to do that collaboration in a fair and equitable way yep there's a question here that says education on real financial benefits to the community of living sustainably with nature is the key and not just in sq who is doing the research on these financial benefits so can I'm not sure if you're aware of yeah look I don't know I mean it I mean it's sort of the realm of probably economists so again it's an it's one of these things where I think you know this is ecologists we're not if you you know as an ecologist I'm not really that qualified to do that stuff but I think you know there are I mean there's various work going that economists are doing around financial benefits of sustainability and so on so yeah but again I think I agree with you I think we need the research on that but as I said before I think it's just you know doing the work and saying well here it is it's actually it's if people have a belief that there isn't a financial benefit that it's really hard to change or can be hard to change so behavior change programs will be really important as well and certainly yeah I agree completely with that with the sentiment and some of the work that we are doing I guess is touching into that sustainability and financial benefit stuff because I know some of the work that was presented by Chris Evenson from our department and also from Ian Wheatland from Queensland Trust for Nature was around using carbon outcomes to drive habitat restoration for koalas and the idea being that if we can demonstrate to landholders that there's a clear economic benefit to their participation in restoration then we are more likely to get people interested and willing to participate in it in the long run so it's having that base research but also then piloting programs that actually demonstrate clearly that you can derive those economic benefits from involvement in those types of activities I think is important because there's obviously a reluctance to drive in to dive in on those types of things initially until such time as the any bugs in how they operate are sorted so I think pilot programs are really important in demonstrating and giving people confidence that those types of activities can generate a financial benefit and hence are worth people engaging in. The other thing to recognize and our work we've been doing in New South Wales on private land conservation is sort of showing this that it's sometimes people respond to financial benefits but often they don't and for some people it's not about the financial benefit it's about you know having the capacity to protect koalas on their properties or it's you know or it's social norms that you know the people they know that prevent them it's not so and I think one really key question is when is it when is it that financial benefits are going to change people's behaviour and when is it not and if it's not what other strategies can we use to me that seems like a really important question because I think the trend at the moment is we just need to throw money at this pay pay people and they'll change behaviour I'm not convinced that's going to work everywhere. Yep there's a question here which is about what's what's our opinion on conservation covenants for koalas do they work um certainly from from the department's perspective covenants are a key component in protecting koala habitat but also in securing areas which have been subject to restoration activity so the programs that we're doing with Queensland Trust for Nature and that we're engaged in with healthy land and water a strong focus in those areas is that we're we're putting investment into a restore habitat that we want to secure that restoration for perpetuity so covenants are a really important part of that covenants are also a really important part of where we have development approvals in areas that contain koala habitat so again in providing ongoing protection within areas that are being developed to ensure that there is still connecting and sort of stepping stone type habitat provided in those fragmented environments to allow koalas to continue to persist in those environments um we're just about out of time in terms of what the the session um to run um somebody's asked a question about is their work being done on genetically distinct koala populations and how this affects management I don't know um I can certainly certainly um there's work across Australia that's being done led by the Commonwealth Government which is looking at establishing an overall genetic database for the diversity in koalas across Australia and that was driven very much by the bushfire impacts in the southern states there's also been some specific work done in Queensland under the auspices of a project called the Living Koala Genome Bank and that was um by UQ and University of Technology and that work has again been sampling koalas in populations in Queensland to establish the genetic links and to provide assistance for what type of management might happen in the future to manage the genetics of fragmented populations so there's certainly a body of work there this type of work we're doing in our translocation policies is referencing that work because we need to understand that as an important component of managing fragmented koala populations in the future um I think we're probably going to have to leave it there I really want to thank Jonathan for um his time this morning and for the work that he put into um providing those reflections around um the knowledge and how knowledge influences how we make decisions and progress effective conservation work so thank you very much for your time this morning Jonathan in in closing I'd also just like to thank and acknowledge all the presenters who have presented during the course of this co-lab 2021 series that's over the 22 presenters that have given their time and provided questions to address the feedback that we've got from attendees also want to thank and acknowledge the attendees for I guess making the forums what they are in terms of that interactive element we certainly look forward to the opportunity to do this face to face again in the future where we can I think engage in some more productive and some some robust conversations around some of the key things that we need to tackle as we move forward also just like to thank the DES team who did the planning and and implementation in relation to this conference series and also Nectar Creative who have planned and hosted this virtual event and allowed us to reach out to such a wide audience of people over the past six weeks so just thank you everyone and we'll see you in two years time but for some of you we'll certainly be engaging in relation to key initiatives and programs that we're rolling out under the conservation strategy so thanks everyone for your time