<TITLE: Reserve Selection and Biodiversity Persistence
ACADEMIC DOMAIN: natural sciences
DISCIPLINE: biology
EVENT TYPE: lecture discussion
FILE ID: ULECD060
NOTES: lecture interspersed with discussion, USEMD200 and USEMD210 are part of the same course

RECORDING DURATION: 66 min 50 sec

RECORDING DATE: 1.11.2006

NUMBER OF PARTICIPANTS: 27

NUMBER OF SPEAKERS: 6

S1: NATIVE-SPEAKER STATUS: Finnish; ACADEMIC ROLE: research student; GENDER: female; AGE: 24-30

NS2: NATIVE-SPEAKER STATUS: English (Ireland); ACADEMIC ROLE: masters student; GENDER: male; AGE: 31-50

NS3: NATIVE-SPEAKER STATUS: Welsh, English (UK); ACADEMIC ROLE: masters student; GENDER: female; AGE: 17-23

NS4: NATIVE-SPEAKER STATUS: English; ACADEMIC ROLE: masters student; GENDER: female; AGE: 24-30

S5: NATIVE-SPEAKER STATUS: Russian; ACADEMIC ROLE: masters student; GENDER: male; AGE: 17-23

S6: NATIVE-SPEAKER STATUS: Dutch; ACADEMIC ROLE: masters student; GENDER: female; AGE: 17-23

SS: several simultaneous speakers>


<S1> <START MISSING> finished erm erm i don't have any copies left @anymore@ </S1>
<S5> mhm-hm (imagine) </S5>
<NS2> [yeah] </NS2>
<S1> [so] er as you see the title is slightly different than it was in the programme <NS4 KNOCKS DOWN A CHAIR> but er </S1>
<SS> [@@] </SS>
<S5> [@<NAME NS4> @] </S5>
<S1> [@don't worry@] </S1>
<NS2> you <S1> [erm] </S1> [should] (xx) </NS2>
<S1> @@ yeah okay (shall we sit down) should i actually shut down this </S1>
<NS2> i don't [know] </NS2>
<S1> [first] light (xx) </S1>
<NS2> no </NS2>
<S1> can you see that </S1>
<S5> oh [fantastic] </S5>
<S1> [that would] be much better </S1>
<SS> true yeah </SS>
<S1> <REFERS TO POWERPOINT SLIDES THROUGHOUT THE LECTURE> yeah er so i'll be talking about biodiversity persistence the original title was something about er spatial dynamics but basically it's pretty much the same thing because er the only way to assure biodiversity persistence is to take into account also spatial dynamics and spatial structure of the reserve network <NS2> mhm-hm </NS2> , and do you here we have again the questions that you should already have in your your handouts and again keep these in mind while going through the talk so how can we account for er biodiversity persistence in reserve selection and how can we incorporate spatial dynamics population dynamics in into quantitative reserve planning . mhm well what we did last last time we'll just go through briefly the basics of reserve selection here er so we might have a limited budget and we want to maximise with with what we can get with that budget which is the maximal coverage approach to reserve selection and the second alternative was er the minimum-set problem where we have er some target that we have to reach a representation target and we try to minimise the costs of doing that and then we had some important concepts of reserve selection you can base your selection basically on er species richness but as er er selection criteria in an algorithm that is not the most efficient possible way to do things it's better to look at complementarity where you look at all the sites put together how many i- species are represented by all sites , and not just the richness of individual sites then there was the concept of irreplaceability er which means that sites that contain unique endemic species that don't occur at other sites are irreplaceable so you have to select them if you want to have all species in your network you have to (xx) er select some of the sites and then we talked about use of surrogates and actually er something that people quite often don't realise is that reserve selection is always based on use of surrogates because we never have data on everything so you're forced to use the data that you have on for example some specific species groups and then you just have to presume that it works well for other species as well so er use of surrogates is another important issue here and then we need to consider what kind of data is available and the data quality for example if we have er only presence data or if we have also absences , and er also data quality how how reliable the data is . here's another example of a minimum-set selection on madagascar er this is a minimum-set selected to represent 321 different butterfly species in madagascar and er er it was done using this er greedy algorithm i'm not sure if i mentioned this name on <NS2> [mhm-hm] </NS2> [monday] greedy algorithm i did explain how it works but er the word greedy comes from because you base the selection on richness and er the richest sites might actually not be needed when you look at the complementarity of all the sites put together so it's greedy because it might end up selecting more sites than are actually needed to represent the the species , and here we have again er the irreplaceable sites with red and er replaceable sites in yellow . there have been quite a large number of this kind of very simplistic studies , but er as i said it's not enough to to look at the representation of these species because you might end up as in the example of a of the british birds last time er you might end up with having just for example one breeding pair of of one species and that doesn't really guarantee that the species will be there in the future as well , er actually when you , when you er have cost-efficiency as a selection criteria what happens is that er since areas usually cost more if they are larger <NS2> mhm </NS2> you end up selecting small sites and er possibly even a few of them , so basically you minimise the size of the reserve network which is not a very good thing considering species persistence this is something that i er er it's small and this phrase down here is in red because er i just added it there after i printed the er handouts so you might want to add something in your notes so er minimising the cost often also me- means minimising the size of the reserve networks and also the size of individual sites because if you select many small sites you probably get more species than by selecting er only a few large sites which is a problem regarding species persistence and er as you know small populations , and especially if you only have one population of each species they are prone to extinction more prone to extinction than large and several populations would be due to demographic stochasticity er and environmental stochasticity and genetic variability issues also and then there's also the fact that er we're not sure of the data so if you only have one presence of a species then what if if that's a false presence and the species isn't really there then you might not have any in reality <P:05> and er another issue is that species occurrence patterns are not static they change in time and if you have data for one year then the situation might be completely different the next year and if you base your selection or and you presume that the species that were there in one year are there er the next year as well you might actually er it might be better to have several oc- occurrences of each species also for this reason because er they change in time and there's some turnover in species composition in time . so er one way that i already mentioned on monday , er is to fiddle with the targets , er you might want to set for example higher targets for species representation you might instead of just one breeding pair or one population you might want to have for example five populations but better yet er instead of using these fixed targets you could use these continuous benefit functions where you always try to get more so er five is better than one and ten is better than five and so forth and what happens with these benefit functions is actually that you get trade-offs between different species so you might want to , select for example some sites that give a very high representation for some important species whereas some other species might be left completely unrepresented that might seem like a problem but that's why we should use species weight so that the endangered species they have a higher weight and er that ensures that we select the sites that contain those important species and we get a higher representation levels for those species whereas species that are common that don't really need to be protected they can have lower weight and then it doesn't really matter if you if you have them in the network or not <P:10> so er in i- in addition to having several populations there are some other characteristics of reserve net- networks that are important for biodiversity persistence size is the most obvious thing size of individual reserves because if you have very small sites that means that you have also very small populations and er the edge-effects are larger and also the size of the total network of course because if you have a larger network it means that you have er more populations and er you have larger metapopulations of of s- those species for which it's relevant and also the spatial configuration of the network is important it reduces the edge-effect that er of the whole whole reserve network and enhances the colonisation and migration processes between different sites and er another important issue is it also reduces management cost because it's easier to manage a compact reserve network compared to sites that are located really far from each other , and also this er quality of individual sites is important <P:05> so one way of measuring biodiversity persistence is to perform these so-called population viability analyses these are basically assessments of risk of population extinction and you can do it for either the current conditions or for something that is predicted under certain circumstances for example er the differences between a er if er a site is protected or if it's un- left unprotected so how large difference is there in er extinction risk er these are quite often done with species that are er interesting to the public er s- like flagship species and so forth because er these species have to be quite well-studied to be able to do these kind of analyses and er also because it's quite expensive so you need funding and funding is easier to get for interesting species er you can estimate vulnerabilities of populations to different factors and you can use these analyses to rank different kinds of conservation scenarios <P:06> and if you compare the results of a population viability analysis from species and a and the results of some basic reserve selection algorithms er you'll notice that different er results are actually quite different so if you do a PVA for a single species you probably end up <NS2> [mhm] </NS2> [recommending] selecting some single large area where the s- that species in question can persist whereas with reserve selection as i mentioned when you try to minimise cost and maximise the number of species there you end up selecting many small sites which is bad for the persistence of any individual species <P:05> but erm well as i said er PVAs actually they demand quite quite high data quantity and quality and this is not available for many species and it requires a lot of money for for starting from scratch doing this kind of ecological monitoring studies for a species and it takes time as well which is another problem in conservation applications because <SIGH> you you just don't have time you have to work quite quickly , but er PVAs could be used to improve the basic reserve selections done using these er optimising algorithms , so er PVAs can help us for example in defining threshold for size of selection units and for the size of the entire network so that if we know that er some species will not persist in a in an individual site that is smaller than some specific specific size then we can just directly exclude all sites smaller than that from the selection procedure or similarly we can set a minimum size for the network based on this information , and we can get some information regarding the spatial configuration of the network er on the if we have knowledge of for example dispersal distances of the species we know how how compact the reserve network should be and also about the quality of reserve sites which er which kind of sites are good quality for each species and we can use this information to do er produce actually probabilities of occurrence so we might have data er where we have presences and absences that have been confirmed but that's only a snapshot in time so er as you know species go extinct locally and er recolonise and so forth so each site will actually have a different probability of occurrence even though if we know that that species is there now we should also consider how probable it is that that site is occupied at any any given moment of time . so er often we only have a very sparse sparse dataset of er the species occurrence occurrences we don't we m- might have only presences for example at some sites which is er quite often the case if you use for example museum records and er we might also have in addition to the confirmed presences we might have some confirmed absences , but then we still have a lot , lot of sites where we don't we have no idea whether the species is there or not so what can we do in this kind of a situation <P:06> now <SIGH> first of all there can be an observation error we might have detected wrong species so all of these green dots might not actually be green in reality and of course the other way around we might not have detected the species even though we have surveyed the site so some of these might actually be green in reality , er what we what we would like to know is that where is the species most likely to occur and we can estimate this based on site quality <P:07> erm , okay so what we try to estimate is the probability of occurrence of the species in each of these sites including the ones that for which we already have the data but it might not be relib- er reliable and also for the sites that we we don't have any data for and we can construct these so-called habitat suitability models for this purpose . er presuming of course that we have some environmental information for the whole study area we might have different kinds of er layers these are called layers that are usually used in some kind of G-I-S applications directly so we might have for example the habitat types different categories throughout the study area and er temperature data and er for example ground-water level and soil-type and what we can do with this information is that we link all these different variables to the occurrences of the species so we try to deduce what kind of habitats and what kind of sites are most suitable for the species and which are least suitable s- for a species and based on these predictions we can produce probabilities that go from zero to one obviously er which is in a way it's better than just presence-absence data because it it's <NS2> mhm-hm </NS2> <SIGH> these are concrete quantities so we have different values for for even for the sites where we know that or we think we know that the species occurs we get different values we know which sites are most suitable , so instead of just presence-absences we actually have a gradation in the site quality and er in the probabilities of occurrence </S1>
<NS2> sorry <S1> yeah </S1> are you basing the criteria for that on the data you already have for that particular site or environmental data from other sites or the </NS2>
<S1> er we should have environmental data for for the entire study <NS2> [mhm-hm] </NS2> [region] all the sites but er the modelling is done on on those sites where we have recorded or </S1>
<NS2> yeah i just wondered <S1> [yeah] </S1> [if] if the data that you already have is so prone to error </NS2>
<S1> yeah that's [yeah that's a possibility of course] </S1>
<NS2> [your you how how reliable can your] model be </NS2>
<S1> sorry </S1>
<NS2> how reliable can your model be [under (those) circumstances] </NS2>
<S1> [yeah that's that's something] that has to be done also the model estimation that's er actually what i have here so there are different methods for for <SIGH> er estimating how well the models actually work , because er in addition to the your original data might not be <NS2> mhm-hm </NS2> reliable then also you might not have the correct environmental variables for example they not m- might not be linked that well to the species occurrences and there might be just some <SIGH> too much er stochasticity there and so forth so there are some techniques that you can use to examine the ac- accuracy of the er habitat models first of all you have to look at how well does your model fit the original data that you used for modelling the occurrence data and er er presence-absence data that you have used and this is called model calibration and there are different kinds of er techniques for estimating the model fits er you don't have to remember these names these are just examples you don't have to remember these techniques erm and on the ha- other hand you have to look at how well does it predict independent data so what you do in practice most often is that er the er data that you have you divide it in half or something use the first half for the modelling and then use the second half for estimating predicting so er if you use for example half of the areas then you predict for the entire study area and then you look at how s- how well the er predicted occurrences correlate with the with the occurrences that you had in the other half of your dataset <P:08> but er habitat quality is not the only factor in determining the occurrences of the species the probabilities of presence in the long-term they will also depend on the reserve size because larger sites are more likely to <NS2> mhm-hm </NS2> sustain the populations in long-term and er , and similarly you should also consider the reserve connectivity because if you select sites far from each other the species are more likely to disappear from there <P:07> so er as you know the spatial pattern of reserves is important for species persistence and in general clustering of sites is preferred because it reduces the edge-effect of the entire network er it enhances the recolonisations and regional persistence , and er a decrease in the boundary length helps also er economically because the boundaries need to be maintained and er usually the boundary length is positively related to management costs <P:06> so how to obtain clustering in reserve selection methods the simplest possible strategy that has been suggested is that you just use a normal regular minimum-set selection algorithm or whatever and when a tie occurs which means that when you have er two sites that have for example the equal number of new species you select based on the distance so you select the one that is closest to the sites that have already been selected but the problem with this approach is that you might not have any ties during the whole selection procedure so there won't be any difference at all and even if you have ties then the site closer or closest to the already selected sites then can still be quite far it just happens to be closer than the other other site of equal value . and the second strategy would be to select all sites within some specified threshold for distance which would be based on knowledge of dispersal distances er the of the species and and so forth , er but the problem is that you you have several species and the dispersal distances vary so it's it's quite hard to determine which which threshold would be appropriate for for each case , and erm also er the e- end solution actually depends on where you happen to start your selection because if for example the richest if you start with richest site and it happens to be in a certain region then you're constrained to select all the sites close to that one even though some other area might actually contain as a as a whole a better network , and also the solutions differ in size depending on the starting point as well if you happen to start from a site that is very isolated from all the rest then the end-solution will be quite large <P:06> a bit more sophisticated alternative would be er to include boundary length in the objective fou- er objective formulation so er instead of a a simple minimum-set you actually try to minimise the sum of costs and this so-called boundary length penalty , so here what is essential is actually this part which is the boundary length penalty so here this just indicates whether or not a site is selected and that site's costs so you sum up the costs of all the sites as in the basic minimum-set approach and try to minimise that but in this case it's the cost of the sites plus the boundary boundary length of the network multiplied with this boundary length penalty , and if the boundary length penalty is large it means it means that er this side of the equation has larger effect so you end up with a more compact more clustered network with the cost of probably having to select more expensive sites or even mainly poor-quality sites in some cases but the basic idea is that you can you can use different values for the boundary boundary length penalty and get different kinds of ne- alternative network configurations more or less clustered and here er the other part is erm that we use actually probabilities of species occurrence and the idea is er that we look at the entire network and the combined probability of species occurrence so that we actually multiply multiply these probabilities so that we get in the end er a value for the species occurring some- at least somewhere in the network . and we set some kind of target for the probability for example 95 per cent or something so we have er have the species present in the network with a at least a probability of </S1>
<NS3> er [excuse me excuse me er] </NS3>
<S1> [0.95 yeah] </S1>
<NS3> can you give an example of what a penalty might be for the boundary </NS3>
<S1> penalty could be j- just a figure from mhm just any number for example if your penalty is just one then you just count the boundary length but if you have <NS3> [mhm] </NS3> [a penalty] of ten then you multiply the <NS3> [but] </NS3> [boundary] length by ten </S1>
<NS3> yeah not not the number what what is the penalty , er would it be death by predators by the edge [or] </NS3>
<S1> [oh] so what it what it's based on <NS3> [yeah] </NS3> [like] er biologically okay er well the idea is that er er just in general having longer boundaries and less clustering is is bad and more clustering is better so this is not er actually a species-specific value <NS3> [oh] </NS3> [it's just] something that you you try to er you you use different values and you look at the results and how compact they are and it's not something that you can like determine the correct value for it's just something that you have to experiment with and see what the what comes out </S1>
<NS2> so it's edge-effective </NS2>
<S1> yeah the idea is that you r- use reduce edge-effect and you get a more clustered network where the species are better able to disperse between the sites and so forth but i have er an example of this actually </S1>
<S6> erm er s- sorry <S1> yeah </S1> before you move er further on i also had a question about er the second equation what where does the P stand for </S6>
<S1> P er oh sorry </S1>
<S6> is it just the number P or no no sorry the the greek letter P </S6>
<S1> oh <NS4> pi </NS4> yeah right pi </S1>
<NS2> [pi] </NS2>
<NS3> [pi] </NS3>
<S6> pi <S1> [yeah] </S1> [oh you] say pi in [english] </S6>
<S1> [yeah] that just means that these are multiplied with each other it's like er sigma is sum and then P <NS2> [mhm-hm] </NS2> [means] that these are [multiplied] </S1>
<S5> [does] yeah </S5>
<S6> uh-huh </S6>
<S1> there has [to be a word for it but @@] </S1>
<S5> [and does it @@] </S5>
<S1> it <NS2> [mhm-hm] </NS2> [doesn't] come to my mind now . <NS2> mhm </NS2> okay <P:09> okay so this is a more practical example , er these are four different sites and er now we're looking at the probabilities for just one species here species J so these are the probabilities of occurrence at these different sites and that would be the probability of of a of <SIGH> absence but you can basically ignore this now er so we have three alternative er reserve network configurations here the blue ones would be the sites that are selected , and er so what we do is we multiply the probabilities of occurrence and er , er these would be the value value for for the solution of the solution for species J and here you have to keep in mind that we're trying to minimise this figure so large number , er , no sorry actually no no large number here is good <SS> [@@] </SS> [@@] @it's in the next slide@ so er we just multiply these these with e- er each other and we get the combined probability over all all these selected sites for the occurrence of the species and it should be larger than the than this threshold set here </S1>
<NS3> er for this example do we know what the threshold is [or yeah] </NS3>
<S1> [er no] no actually that it hasn't been set here but these these are the final values with these different network configurations and as you can see with three sites of course it's higher higher than with two sites </S1>
<NS4> so with the first equation that one is not species-specific but this one is and are those species then your surrogates or your target species or endemic species in it </NS4>
<S1> yeah <NS4> [uh-huh] </NS4> [yeah] well it depends on what kind of data you have available but in general that that would be the case yeah , [and] </S1>
<NS3> [er] sorry could <S1> [yeah] </S1> [you] just say again what the pi is if it's not 3.14 what's the </NS3>
<S1> it means that these er are multiplied because you have this <S5> [no] </S5> [P-value] for each </S1>
<S5> what what is the meaning of this pi [er] </S5>
<S1> [yeah yeah] it <NS2> [mhm] </NS2> [means] that these this is multiplied with er each other because you have P-value for er for each species and site and then you multiply er one minus P for each species so you have one minus P-I-J you have one minus <NS3> [oh] </NS3> [P] I plus <NS2> [yeah yeah] </NS2> [one J] they're all multiplied <NS3> mhm-hm </NS3> so it's it's er like equivalent of the sum <NS2> [mhm-hm] </NS2> [sign] you know whatever comes after that sign is all summed [together] </S1>
<NS3> [(they're matching)] </NS3>
<S1> and in this case they're just multiplied not <S5> [oh] </S5> [(xx)] <NS3> mhm </NS3> <NS2> mhm-hm </NS2> , er i can come back to this slide but i i'll just show you because it's it continues on the next one maybe it's easier to understand if i first go through the whole example , so er in addition to the probabilities then you have to look at the boundary quality er boundary length penalty and , for this first reserve network you have two sites so the cost that we have here er would be presuming that each each individual site has the same price here for simplicity the cost here would be two and here it would two and here it would be three because we have three sites but then the boundary length is what makes the difference here we have a a boundary length of six we just calculate one for each side of each site so a boundary length would be six here it would be eight slightly longer and here it would be also eight but we have three sites so now we have the proportional boundary length would be three four and 2.67 so the value what what we have here what we try to minimise in this case er er for this network would be five for this six and here 5.67 , so in this case smaller is better this has the largest largest value because it's er it has a very long boundary the boundary length is as long as here but we only have two sites so it has the highest highest boundary length multiplied with the penalty , but of course it depends on the objective what we want for each species , and er depends on the target that we didn't determine for this expa- example wh- what what the end-solution what we want er <SIGH> er will depend on on the targets for for each species so the target would be determined species-specifically and this this on the other hand is is for the entire network and it it doesn't include as you can see it doesn't include the species indices as well you either you just look at the numbers and costs of the sites and the boundary boundary length </S1>
<S6> er sorry is it in this one that you that the penalty of for boundary length is just one , in this one because you just are looking at <S1>  yeah yeah </S1> the boundary length with this one <S1> yeah [yeah] </S1> [okay] </S6>
<S1> so that could be anything it could be like ten and then you would multiply it with ten and <S6> [yes] </S6> [then] it would have a lot much larger effect </S1>
<S6> yeah but in this example it's just </S6>
<S1> just one <S6> [okay] </S6> [yeah] <P:05> okay is this more or less clear should i try to explain something better </S1>
<NS4> s- just wondering so if you're looking in selecting an area would you first do perhaps your erm this boundary penalty kind of working that out for the different reserve selection sites and then go to the species-specific probability of their occurrence and see then what your best option is <S1> yeah </S1> [could you just choose from] </NS4>
<S1> [er that's a good] point <NS4> [yeah] </NS4> [because] you always have to keep in mind that these are er all the methods that i'm going through they're only tools they're not like the true answer to @anything@ <NS4> mhm </NS4> so you probably want to try out all sorts of different strategies and methods and and er then when you get all all these different kinds of results and different degrees of clustering for example then you also have to go back to the species data and think about which species are there and which species would actually benefit from the clustering and so forth and it's kind of going back and forth <NS4>  yeah </NS4> between different stra- strategies and methods and techniques and and without forgetting the actual species that live there and their biology that's that's a good point . erm okay perhaps there there have been several studies that have actually used these kind of techniques and they have found er that er that using these measures the species actually do have better better chances of survival in the long-term , and as i mentioned a high penalty for boundary length er creates more more clustered reserves or more clustered networks erm but there's also the risk that you may end up er including bad-quality sites as well , <NS4> mhm </NS4> . er the ideal situation would be that you could use species-specific spatial reserve dign- design and you could consider their the the needs of each species individually because species for example they differ in their dispersal abilities and er the problem here is that the spatial composition of the network should actually be good for all the species and not just not just some of them , but this can actually be taken into account er in the probabilities of occurrence . so er , you can consider er the connectivity of each patch you can look at the surroundings of the patch for example er within a certain , certain er distance from your site ideally you would you would determine this nei- neighbourhood size er based on the species that occur there and their dispersal distances , but that can be gui- quite hard because you might not have that kind of data but anyway you have some some neighbourhood , measure of the ne- or some measure for the neighbourhood surrounding the site and then you can look at the amount of suitable habitats for example within that neighbourhood here you would have like three highly suitable sites in the neighbourhood or you can look at the occupancy the amount of occupied hi- sites in the neighbourhood it's always better to have er neighbouring sites that a- also contain that species er or you can use both of them ideally you look at the amount of occupied suitable habitat within a neighbourhood because there might be also occupied sites that are actually sink-habitats and the habitat is not that good but if you look at both habitat suitability and occupancy then you probably end up with the best best possible result <P:05> so erm including connectivity measures into species distribution models enhances the evaluation of reserve networks and the probabilities of occurrence are higher in areas that are surrounded by occupied high-quality sites , but there you also of course have to consider that if you don't protect the surrounding sites then species might might start disappearing from there and then it doesn't really help that much but it's still better to have occupied habitats surrounding your reserve rather than unoccupied already unoccupied habitats , and this is an example er for nuthatch in er netherlands , here you have a habitat suitability mat map these are the different probabilities the darker colours indicate a higher higher habitat suitability and therefore higher probability of occupancy but if you include er connectivity consideration , then you get these more uniform areas where you have like core areas with highest probabilities and then surrounding areas , with the medium medium probabilities of occurrence and this is a much better indication of of what a reserve network protecting nuthatch in this case should look like compared to these sparse sparse occurrences here <P:06> erm here's another example with seven different s- bird bird species , and we try to select a minimum-set of sites with these different four different scenarios , that er you'll see the same numbers in the next slide i just explain here what they what they used so in the first one you use presence-absence data without any boundary boundary length penalty in the second one presence-absence data with boundary length penalty er and in the third one you use probabilities for habitat suitability without boundary length penalty and in the last one you use probabilities plus a connectivity measure without boundary boundary length penalty and here you can see the differences in the network con- configurations so all these are now without boundary length penalty and this is with boundary length penalty and you can see that here you get more connected more er er larger larger and more connected better connected areas this was the one with er presence-absence data without any any penalty so this is the just the basic basic most simple minimum-set selection and you have a lot of individual single cell rich cell sites that are located far from each other and this one looks a little bit better this one was the one with oh okay it's getting there with er habitat suitability with the probabilities probabilistic data , but without any boundary boundary length penalty and er this one is the one that uses also in addition to the probabilistic data it also uses some connectivity measure in the selection procedure so there's these two look pretty much the same except that he- here you still have some some individual isolated sites here you have just one but most are are quite neatly clustered using the penalty for boundary length </S1>
<NS3> why is there a difference between two and four if four takes into account connectivity and two also (xx) connectivity (xx) </NS3>
<S1> erm , basically wha- er well it's even with the boundary length penalty you can get different lesu- results using just different slightly different <NS2> [mhm] </NS2> [values of] the boundary length penalty <NS3> [okay] </NS3> [so] it's a matter of luck you might actually be able to get exactly the same solution with some specific value of boundary length <NS3> [mhm] </NS3> [penalty] as well so it's just differences in the de- degree i guess <P:06> but as you can see you can get very different results just by m- making these very slight changes in the method so when you do some kind of reserve analysis a- and you end up with some specific result just keep in mind that that's not the correct answer there are always different alternatives <P:06> and er these are actually this is the number of sites that you need to select with these different methods and er and and and <SIGH> you can see that if you don't use any additional constraints you do just the basic minimum-set you get away with the smallest number of sites but as soon as you start taking into account some additional criteria you have to select more sites and in in this case actually method number three selected the largest largest number of sites so considering both connectivity and costs it looks kind of like number two two would be might be a good solution because you don't have that many sites added to the mi- basic minimum-set solution but you still get clustering but that also makes you think that maybe in this solution you have more bad-quality sites included , there has to be some compromise , and this is the boundary length if you look at it the other way around the boundary length for this solution where there were no penalties or anything for clustering is way higher than with the other techniques , and of course where you try to minimise the boundary length it's it's the lowest <P:07> here's another example with birds in britain er 78 species in this case and er they had er presence-absence data available for two different periods in this case so they had er T-one the- these two occur in the next slides as well so keep in mind that T-one means period one from 68 to 72 and T-two means from 88 to 91 so two different separate time-periods and er they modelled the probabilities of occurrence er by fitting logistic regressions and as the response variable they used occurrence records in in this first time-period this response variable means that er , well the data that they they used for the modelling , and er predictor was the connectivity among the records so basically using the connectivity among the these different records in the first time-periods they then tried to predict occurrences in the second time-period even though they had the data for the second time-period but idea it was to predict based on this for this time-period and then see how well it matches . and er yeah well groups of occupied cells indicate spatial correlation of habitat variables or habitat quality <P:05> and the the idea here was to look at if probabilities of occurrence relate to probabilities of exti- extinction so if occurrences at at time one related to time er probabilities of occurrence at time two so whether or not the species had gone extinct . so er yeah if the species was not recorded the species that had been recorded in the first time-period was not recorded in the second time-period that was interpreted as an extinction , and the probabilities for these er extinction events were modelled using the same approach as for modelling occurrences , and it turned out that for 81 per cent of the species extinction events occurred significantly more often in areas with low probability of occurrence than in areas with higher probability of occurrence so there was a correlation between occurrence probabilities and extinctions , and these were just two species example species from the data and these would be high er red would be high probability of occurrence and er here red would be a high probability of extinction so you can see that er here the light-blue colours are here in dark-red meaning that the lower the probability of occurrence then the higher the probability of extinction and the same same for this second species <P:07> so the species' probabilities of occurrence were negatively related to local probabilities of extinction so can er could we possibly use er this kind of information for reserve planning as well can local extinction be reduced if reserves are selected to maximise current species probabilities of occurrence which would be very helpful . er the objective in reserve selection using probabilities is to find a minimum combination of sites required to represent all species with at least a target probability <P:08> here there's a comparison of some different reserve selection strategies , here's er er s- a reserve network selected with minimum-set er using probabilities , er here's a network selected using maximum coverage and presence-absence data and in this case this er budget limit for the maximum coverage selection is based on the results of this so when they er they did the minimum-set selection and they they got s- 64 areas then they decided to do with the same number of areas a maximum coverage selection and see how many species you can get represented with the same budget and this one is a minimum-set using presence-absence data , so this is still the same study that i was talking before this T-one is er time-period one and T-two is time-period two so using just the basic minimum-set and presence-absence method they could actually get away with selecting just six sites where all species were represented in the first time-period but er eight per cent of them went extinct before the second time-period when they did the maximum coverage using 64 sites , er they got 100 species of course in the first time-period but two of them went extinct in the second second time-period but when they used the probabilities and the threshold for the probability of occurrence of the species then actually none of the 100 or that's actually not the number it's percentage of the <NS2> [mhm-hm] </NS2> [species] none of the species went extinct , so it seems that using these probabilities based on habitat suitability is is better regarding local extinctions it better estimates <NS2> mhm-hm </NS2> whether or not the species is there also in the future </S1>
<NS2> but between those two methods the one resulting in the 100 per cent er retention and the 98 is the difference significant </NS2>
<S1> i don't know [i] </S1>
<NS2> [statistically] i mean to make to make a judgement favouring the probability method </NS2>
<S1> er for i actually have no idea because @@ i have <NS2> [mhm-hm] </NS2> [to admit] i haven't made these @slides i@ <NS2> okay </NS2> in- inherited them from the lecturer from last year so i i i don't remember this study so well , yeah it it's not that that big difference [if you have two species and mhm it could be just a coincidence] </S1>
<NS2> [mhm-hm er it it could be just (xx)] </NS2>
<S1> but i guess this is not also this is not the only <NS2> [yeah] (xx) </NS2> [study] <P:08> okay er i guess i already explained these mhm yeah @@ <P:07> there okay well this is something that i forgot to mention so as you can see of course the probabilistic approach requires much more area so the difference between six and si- 64 is it's quite substantial , but er also you should keep in mind that it's also the configuration not just the area that matters for for persistence , mhm okay , so there is a trade-off between reserve size and compactness if you want large large reserves they cost more , and er if you want compactness then you'll probably have to select some more expensive sites and then you have to give up on the total total size of the reserve network and er use of probabilities seems to increase reserve size or the mhm size of the total network at least and er it's er when you use only presence-absence data it's in a way it's over-optimistic concerning the amount of populations that are protected because they might not be there the next year or the year after that so er when you use connectivity measures the reserves become automatically more clustered and quantitative clustering can always be increased by using a boundary length penalty but mhm just keep in mind that there's there are always when you add more constraints to the objective formulation then there there are always trade-offs with with some other factors either the price or the quality quality of the network <P:05> erm yeah so connectiv- with connectivity you have to keep in mind that it's not enough to just look at in the initial selection stage er just to look at how many sites are near the area but you actually have to remember that er in the future these sites that are not protected might not actually be there anymore so you have to look at connectivity between the protected areas , and even though here it seems that when this this site actually exists then you have nice nice connections between these two and these two so these are all connected but if this one is not protected then you lose the connection between spe- between these two . so er the the landscapes are dynamic species occurrences are not static and er for example in this case you have a let's see , you have a mhm species with certain disp- er dispersal capacities and in this case these er red ones would be reserves black ones are not not protected and for example er this species would have a dispersal capacity that extends to this one then it's connected to this one as well but if this unprotected site is lost then you lose the connection between these two , and on the other hand connectivity between these two sites regarding this species doesn't really matter because it's unsuitable habitat to hare the grey ones are unsuitable habitat for the hare whereas for this species it does matter because it occurs on both but then again the dispe- -persal distance might be different , so there are a lot of different issues to consider here and if you lose the unprotected sites then this is what happens you lose the connectivity <NS2> mhm-hm </NS2> between the protected sites as well <P:06> er so er , probabilities of occurrence that are based on a static landscape may be too optimistic , and if you include er considering cons- er connectivity measures in probability models you can evaluate the effects of loss of unreserved sites , so if you want to keep the probabilities high the reserve sites need to be close together and as a consequence reserve networks are more clustered so you don't actually have to include any boundary length penalty or something like that just by looking at the probabilities that depend on the connectivity you get more clustered networks . erm so er the probabilities at the edge of a reserve are are of course lower , so you actually need to add more sites to the network to achieve targets , the end-result is that you get a nicely clustered network but there are also some weaknesses to clustering , for example highly clustered and connected reserve networks they are more vulnerable to spreading of diseases and er fire for example the entire network can be destroyed in a large forest fire for example and you lose everything if you don't have like several alternative networks that are enough separated from each other so on the other hand clustering is good but maybe it's also better to have like several copies of the of the er network a couple of separate but within each oth- er within itself clustered networks and reserves are also not static there's natural succession that has should be considered the species er composition can change in time naturally and er management of course affects what happens in the networks and another important thing that i haven't really considered here specifically is climate change because climate change is going to change species distributions quite dramatically and if you're selecting reserves now based on the current situation <NS2> mhm </NS2> it might well be that in er 50 years the species are no longer there and er again regarding climate change connectivity is important but in a different sense because now you should actually consider connectivity between sites where the species occur in different time-periods so if the species distributions are for example moving uphill or towards north then you should have reserves that are connected in chains in a way in that direction so that <NS2> [mhm] </NS2> [the species] can move from one reserve to another which following the effects of climate change , this is er another quite complicated issue because it's it's quite hard to for i- er for starters the the climate change scenarios in in it in themselves are quite unreliable they vary quite a lot so it's hard to predict what the climate will be exactly after some time and on the other hand er predicting species distributions in the future will be even more difficult , so where exactly to locate the sites that's that's quite hard to predict where e- each species will actually occur in the future but er the basic general idea is that you have to maintain the connectivity between between the areas in time , so taking all these uncertainties into account will further decrease the probabilities of occurrence of of species so you become more and more uncertain where the species will be at each time-period <P:05> so er of course real persistence cannot be twes- tested we can't know which species in in advance which species will persist and which will not and er doing that afterwards is not very useful er population viability analyses and metapopulation models can predict occurrences but they require large amounts of good data to be reliable enough , and that's often not available er general concepts of metapopulation theory can be applied to reserve selection so er multiple connected reserve sites increase local persistence , and er distance between clusters of reserves enhances regional persistence because er as i just mentioned all the reserves are not subject to identical risks at the same time so if some natural catastrophe for example destroys or nearly destroys some reserve network there will be still some others others left and er probability models are a valuable tool in predicting the probability of occurrence of species these take connectivity and extinction risk into account but always one has to be aware of all the uncertainties associated with the data and with the predictions . okay so in the end just would like to remind you that all these methods and all these different techniques and strategies are only tools so er none of them give correct answers to any problems but they are intended to help people in decision-making and you can always come up with alternative configurations and er all kind of different alternatives and in the end it depends on what your specific targets are what kinds of species are you interested in in protecting and er and of course the funding is an es- essential issue , and er , in quite quite often actually er the end-result what what will be protected in fact is it depends on so many different things that it may be quite far from from any of your solutions that you have obtained with these techniques but er still the basic idea is that at least you get some indication of what the reserve network should look like what should be there what should not be there and and so forth <P:05> okay because that what it was it for today <SS> yeah mhm </SS> do you have any more questions </S1>
