Geospatial Forum: Dr. Mirela Tulbure

Geospatial Forum: Dr. Mirela Tulbure


it’s a pleasure to be able to kick off the
series the fall series this year with actually two major events first the
arrival of Dr. Mirela Tulbure it’s really wonderful having Mirela here she was hired into
the Chancellor’s Faculty Excellence Program for Geospatial Analytics with
her academic appointment in the Department of Forestry and Environmental
Resources as an associate professor so we’re incredibly excited she’s here lots of
anticipation she moved all the way from Australia to join us and got here late
late late summer from the University of New South Wales and anyways I think we all know a lot about you so I’ll let
you take it away from here Thanks thank you so much thank you so much Ross for
the kind introduction and thank you all for coming with a hurricane looming
around the corner that’s really dedication so I really really appreciate
it so just um by show of hands how many of
you are grad students I’m just curious so a lot of you and faculty a few of you
and what else research scientists okay thank you thank you for that so I’ll try
to ask you a couple of questions to hopefully keep you engaged and somewhat
entertained and also for me to learn from you as I go through my talk so just
to start out I’ll be talking about remote sensing and spatial analysis of
surface water extent dynamics vegetation and landscape connectivity in a dry land
basin and the work that I’ve been doing with my team over the past couple
of years in in Australia so the work is on Australia so just to start out why do
we need geospatial analytics for surface water resources because it’s
changing so it’s dynamic sorry because of hurricanes and it’s
really dynamic we have a lot of hydro climatic variability we know that rivers
meander we know that lakes shrink and expand we know that farmers irrigate and
flood their fields we know that engineers build dams dams dry out so
it’s really important to capture these dynamics and of course having just
static layers might not capture all the processes that we might be interested in
in general surface water is obviously very important but in arid landscapes
surface water is critical not only for hydrological cycles but also for
biogeochemical cycles arid lands cover about one-third of the globe and they’re
shown here in different tones of of orange just out of curiosity who else is
working in arid lands we need to start a trend so aerial arid lands cover again
one third of the globe so we need to start a trend and they are highly
dynamic so this idea of looking at surface water and different processes in
dynamic landscapes is really critical and hopefully I’ll convince you by the
end of the talk that is really important to capture those dynamics
so the southeastern part of the US as you would all know faces water problems
and in water water supply shortages and of course on a day like this it’s kind
of hard for me to think of it as well this region has a problem with water in
the sense that it doesn’t have enough water but I guess we all know that it’s
the region has gone through hydro climatic variability it’s been through a
big drought and then of course the hurricanes who’s been here during the
drought the 2016 drought quite a few of you what have you seen on the ground
like what did you did you experience any changes any like what did you see your yard yeah that makes sense right yeah that makes perfect sense
were there any restrictions like how long your shower could be or any
restrictions on water usage to water your yard right so the region obviously
has water supply shortages and climate changes will impact the area in terms of
higher population as well as a drier climate as well as very high hydro
climatic variability and of course the elephant in the room is is the hurricane
so it’s important to look at these dynamics here in Australia we have a
pretty similar in some ways and very different in in lots of ways system so
what do you know about Australia what do you know about other than kangaroos and
Crocodile Dundee of course what do you know about the landscapes in Australia
there’s a lot of fire so if you study a fire that’s a good place to be if you
study snakes yeah there are a few things that happened there right up north yeah in
the tropics yeah yeah yeah there’s a lot yeah that’s
excellent yeah so you could do studies across gradients and it’s also it’s the
driest inhabited continent so water is really really crucial and it’s critical
and it’s really again water if you study water that’s another good place to be in
Australia is also impacted like the systems would be what we would call boom
and bust systems so they go from drought to floods so I guess very high hydro
climatic variability like here it’s also one of those places that has been
impacted by climate change really badly so the climate change commission has put
out a couple of reports pretty much every summer and that’s because climate
records have been broken every summer and just a couple of years ago the
Bureau of Meteorology had to come up with a new color scheme to add to its
colors scheme to accommodate temperatures soaring about 50 degrees Celsius so
about 125 Fahrenheit so that’s that’s happening and it’s been happening and
just a little earlier this year we’ve been through a major drought now farmers
are out in Canberra in the capital city protesting about the drought and when
the farmers are out protesting and asking the government to do something
about it that’s a really big deal farmers in Australia don’t typically do
that so that’s a really big deal and of course this made the cover of of the
Time magazine earlier this year talking about this big drought so in this
context the study area that we have been working on is the Murray-Darling Basin
the Murray-Darling Basin is located in the southeast of Australia it’s a pretty
large dryland basin it’s over 1 million square kilometers it’s about
the size of the four corner states it’s very important to Australia not only
ecologically it has the largest rivers in Australia so the Murray River and
the Darling River I keep forgetting this actually doesn’t work on the screen and
then it’s also very important to the country because that’s where most of the
agricultural production happens that’s also where lots of cities get
their water from and it’s also where irrigated agriculture happens and it’s
also been going through a pretty major drought a decade-long drought the
Millennium drought basically the worst that has happened in the southeast of
Australia which ended in 2010 with big floods they were primarily driven by La
Nina so in this context and also working together with the Murray-Darling Basin
Authority we wanted to integrate geospatial analytics and remote sensing and
looking at how do we actually quantify the surface water extent dynamics how do we
quantify some of the drivers and look at vegetation dynamics as well as
connectivity dynamics and how do we put it all together to come up with with a
broader view picture of the system so the first thing that we’ve done was to
develop a remotely sensed data set of surface water extent dynamics using the
Landsat archive we used Landsat 5 and Landsat 7 going back to 1986 up to 2011
we used the entire archive of data and we had about 200 core hours on the
national computing infrastructure we develop random forest models to map
every single pixel so it’s a 30-meter pixel every 16 days into water clouds or
land so the key thing here was that this seasonally continuous so among the
students that are taking Josh’s class what’s the big deal why is it important
to look at seasonally continuous time series over time so not just so typically with
with with time series of satellite data we might have had say a snapshot once a
year or a couple of times a year but it’s important to not only look at the
inter annual variability but also intra- annual variability and try to tease
apart seasonality who is using HPC among the grad students the CNR HPC what’s a
core hour so basically it’s about
one CPU on your typical desktop for one hour so you need two HPC to process the
entire Landsat archive I guess that was the main point of this so once we had
these almost three decades of seasonally continuous time series of surface water
maps that was pretty cool because you could do lots of lots of lots of nice
things with these data sets so just a couple of things to give you an idea of
what the data set look like I’ll just go through some frequencies so surface
water frequency in 2006 so one of the years during this time series so
basically what I’m showing you is how many times a pixel has been flagged as
flooded during the time that pixel has been mapped during that year so that’s
2006 you hardly see anything right so that was the driest of the millennium
drought years so that’s why there is very little water or close to no water
in these maps however in 2010 the picture is very very different we’ve had
lots of big floods primarily in the south eastern and northeastern part of
the basin and this rectangle right here is about the size of New Jersey so those
are massive floods that affected cities and they captured a lot of attention
just to show you an idea of the spatial detail that you can get with Landsat
which is I guess primarily why Landsat is is very useful for this I’ll just
quickly show you results of our surface water mapping with blue overlaid on
Landsat images I’ll just go through a time series starting at the end of the
millennium drought until the beginning of La Nina floods so you’ll see quite a
bit of dynamics that lake on your right hand side is Lake Victoria it’s about 15
kilometers east to west just for scale you’ll see a lot of dynamic along the
river so in the floodplains and then you’ll see a little more dynamic south
of the river and then with yellow you’ll start seeing results of our modeling of
clouds obviously we’re trying to mask out clouds were not interested in clouds so you can start seeing quite a bit of
dynamic so just out of curiosity what are these what do you think trigger this
water here there is actually localized rainfall but yes these are fields so you
can see some of the some of the dynamic that we can capture with the high with
with with the spatial resolution of Landsat we then did an accuracy
assessment so I guess how many of you are using remotely sensed products in
their work everyone pretty much what do you look for when you download the
remotely sensed product and what do you pick what kind of accuracy are you
looking for right so what’s your what’s your typical and so say you what would
you pick what would be your typical number that you might be looking for eighty percent so typically over eighty
percent so I guess I guess the thing that I’m trying to point out is that
whenever you’re downloading a remotely sensed product is really critical to
look at the accuracy assessment the rule of thumb is to be above eighty percent
it’s it’s actually a very important step in developing a data set and it takes
about as much as developing the data set and at the end you only have a couple of
numbers but it’s critical it’s important to know how much you can trust the
product it’s important to have the uncertainties associated with it so
basically to be able to have standard errors around your accuracy so in our
case our producers accuracy or omission error of water was around 85 percent so
we were happy with that and if you’re interested in reading a little more
about this work it was it was captured by NASA Landsat science team and yeah it
describes the work in more detail and with with more appeal for management
implications we’re currently using the harmonized Landsat Sentinel 2
product so the newer satellites to basically have a denser time series
because even though we have seasonally continuous time series of surface water
extent dynamics we’re missing lots and lots of floods we know that and with
Sentinel 2 data we’re capturing a little more of these floods but ideally
we would get even higher temporal density of data to capture floods such
as this right so once we have these time series that our fingertips it was pretty
cool to do some some analysis with this with this data set here is an animation
of flooding frequency per year going all the way from 1986 up to 2011 so you can
see that the place is really dynamic again dry lines are really dynamic to
start with and this is a particularly dynamic place and there are a few things
that jump out first of all that there is less water during the millennium drought
but also that even during the millennium drought there are still some areas that
are being flooded so so there is a little more to it than just there is
less water during the millennium drought here is another way of looking at the
data set so basically I’m taking the entire time series and then looking at
percentage water across this entire basin so again over 1 million square
kilometers first-season per year on your x-axis so I guess just looking at this
way of displaying the data what what do you see what are some of the key things
that you see the millennium drought was here so 99 – 2009 right so there is seasonality to it so
we’re seeing again the the broader patterns that that there is less water
during the millennium drought and more water during La Nina floods but there is
a change in seasonality and again having this seasonally continuous allows us to
capture and try to understand how did the drought affect water resources also
per season and this seasonality for example you see that before the 90s we
used to have lots of fall floods whereas after that those fall floods pretty much
disappeared you can also see that even during the millennium drought they still
had pretty large summer floods and and that’s that’s an interesting thing in
itself that characterizes the surface water dynamics and how that impacts
drought the other thing to point out here is the
amount the maximum amount of water that we’ve seen it’s only 2.5 percent now
this is a over 1 million 1 million square kilometres basin so it’s really
dry like even when it’s wet it’s really dry so water is is really precious in
this area of course having these data set you can slice it and dice it the
way you want not only temporarily but also spatially so these basically this
is looking at the northern basin versus the southern basin and these are two large
water management units that are quite different you can see here percentage
water on your middle top row percentage water per year and then on your
right-hand side top row per season per year again percentage in this northern
basin and again you can start seeing and disentangle some of these dynamics over
space as well looking at the northern Basin this is a very dry flat area which
has a lot of water coming in summer as a result of summer downpours whereas the
southern part of the basin shows a steady decline during the millennium
drought and most of the water comes up in spring primarily as a result of of
snowmelt in the Australian Alps which are located on the southeastern side of
this basin one key thing that most of the catchment
managers were interested in knowing was well this basin is great and it’s great
that you can do this across this entire basin but we’re really interested in our
catchment area primarily so again with these data set you can you can again
slice it and dice it the way you want and you can provide the local managers
with the data the way they wanted for the area that they wanted the next thing
that we wanted to do was to look at some of the drivers climatic drivers in in
surface water extent dynamics and this was work that was done by one of my
former PhD students Tino Heimhuber who’s now Dr. Heimhuber and he’s a
postdoc at the University of New South Wales and in Tino’s PhD we really wanted
to understand what are the major climatic drivers of surface water extent
dynamics now if you were to do this type of work what kind of drivers might you
use for this in your own work soils data vegetation also okay cool
impervious surface yeah cool yeah I’ll definitely add these to my next
iteration so in Tino’s case we we didn’t use impervious surface but we did look
at a soil moisture evapotranspiration rainfall and in river gauge data so I
guess we use some soil moisture data the way we approach this was top-down using
a data driven type of analysis so we split this entire basin into eco
hydrological zones they were done prior to us arriving to the scene and then
because we wanted to get a higher spatial resolution of these of these
models we’d split the basin into ten by ten kilometers square so basically
applying a grid on top and then as our dependent variable we had surface water
extent that was our y-variable surface water extent
dynamics over time so there was the time series and there’s our explanatory
variables we had rainfall data soil moisture data evapotranspiration and we
had river gauge data and the way we assigned a river gage per each of these
modeling cells was using the hydrologically reinforced river network
to make sure that the river gauge that we assigned corresponds to the river
area and the modeling unit if that makes sense and then we applied our models
using these data sets to three different areas and they picked these three
different areas to be quite different in the way they have their their water
resources allocated so first of all we picked one that’s less developed that’s
the one to your right hand side this is one that’s quite developed the Murray
area and then the Murrumbidgee area is is one of the areas that have a lot of
irrigated agriculture and what we found was that our models explained about 70%
of the variability in surface water extent dynamics for two of these regions
which is pretty good to start with because we only had climate variables in
there and for Murrumbidgee we did pretty poorly and so we looked at the data and
the place and the reason why we didn’t do so well is simply because there is a
big water diversion scheme that happens here there’s a lot of water that’s being
taken out for irrigation and none of our variables capture that so that was
reassuring to us in a way that the models actually i guess did what they’re
supposed to do so there was so reassuring to us that we applied it
across the entire basin and we looked at climate drivers across the entire basin
and what I’m showing you here on your left-hand side is R-square values
and on your right hand side is reduction in RMSE
so basically areas that are shown in red are areas that capture more of this
variability across the basin so what you see here is that the northwestern part
of the basin is where climate has a more important role in explaining this
variability in surface water dynamics now why did we use a data-driven
approach there are I think some eco hydrological hydrologists in the room
could we have used a hydrological model what are some pros and cons of these two
approaches is this useful would you do this thank you it must be yeah so it
definitely has pros but it also has cons so it has limitations that you can’t
explore future scenarios unless you use I suppose you could use synthetic
river flow data projected into the future and then and then use that as a
model what else might you gain or might you lose when using this type of
data-driven versus physically based models yeah thank you so if you don’t care
about the actual process this might this might give you it definitely gives you
insights this is a large basin so it’s it’s typically easier to parameterize
hydrological models for a smaller area we’re always data poor when we when it
comes to parameterizing hydrological models for these areas so it’s it’s good
to have a data-driven approach it’s also one of these approaches that you can
start applying and then you can zoom in into certain areas once you identify
certain patterns that you might want to use hydrological models for and because
this is such a dry land basin with hydrological models you actually don’t
capture some of the evapotranspiration that happens in the basin as well as the
soil infiltration because water takes a long long time to get from point A to
point B of the basin and you can’t really capture that with hydrological
models so this was one of the main reasons or this was among the main
reasons why we picked a data driven approach but in work that I want to
start here I’m really keen to start incorporating these two approaches and
and marrying those up so I want to talk to you more and then besides looking at
hydro climatic variables we also wanted to look at land-use change so we added
land use change layers we had eight layers of historical land-use change and
what we found was again that climate was more important in the less developed
part of the basin the northwestern part whereas land-use change was more
important in the southeastern part of the basin as well as the central part
where he had a change in land use from irrigated or natural environments to
dryland agriculture why is it important to look at land-use change in a basin
that’s so hydro climatically variable yeah yeah that as well as I guess in in
in in in this part of the world the narrative has been for a long time there
well there is so much hydro climatic variability we have droughts we have
floods there isn’t a whole lot that we can do about it right so so the idea is
that well with land use I mean land-use change is really important and that’s
that’s something that I think in lots of lots of studies is probably not used
enough in my opinion and there are synergistic effects of
climate and land use and land use change and and of course this is one of the
things that we can actually do something about it’s harder to trigger lots of
actions for for changing hydro climatic patterns but there is some this is
something that we can actually do something about at the local level so so
we were quite quite happy to see that land-use changes was really important
and where there was important across this this entire basin the other thing
that we were interested in was to look at vegetation dynamic in space and time
as a function of flooding dynamics so the one of the one of the key things
with with vegetation dynamics is and then flooding in this area is that it
impacts the carbon cycle and there’s been a pretty I guess famous paper by
now that that looked at carbon sinks across the world and they came up with
with these numbers that there was a major carbon sink in 2011 that was
triggered by vegetation greening in the southern hemisphere and a lot of it was
actually due to the La Nina floods that triggered lots of vegetation
greening in this basin in the Murray- Darling Basin this is also important in
the context of the Murray-Darling Basin Authority who is trying to decide where
to allocate water as environmental flows so basically a lot of the riparian
vegetation shows dieback and they are trying to mimic natural
flow regimes to basically restore some of these riparian areas and they’re
trying to understand how much water to put where to get the effect that they
want to get in terms of vegetation greening and there’s been quite a few
studies looking at vegetation dynamics as a function of either rainfall as well
as flooding and there were studies that said well rainfall is really important
flooding isn’t that important others that said no flooding is really
important rainfall isn’t that important in this area but those were smaller
scale studies and of course it’s region specific so here we wanted to take this
more holistic approach again across the entire basin and look at both rainfall
as well as flooding so what we used was an index of vegetation greenness and
again those of you that are taking Josh’s class what’s a vegetation index NDVI yay so we used EVI in this case
but a measure of vegetation vegetation greenness derived from Landsat and we
used our flooding layers derived from the same data set as well as Bureau of
Meteorology rainfall data and then we used a break point regression which
basically tells you where you have a change in coefficient into your
regression so basically do you have a change at some point in time during your
entire time series so some of the major things that we found was that it’s not
just flooding or just rainfall is that the synergistic effect of those two
actually triggers the highest greening events we also found that the first year
of break point corresponded pretty much with the beginning of the millennium
drought in most of the basin with the exception of some agricultural areas
where we had lots of lots of irrigation schemes that started in these areas
right here the other thing that we found was that this relationship was spatially
and temporally dynamic so there are certain areas where vegetation where
rainfall was more important than flooding and flooding more important than
rainfall and that mattered where in the river system you were closer to the
river channel versus further away from the river channel and and there was a
lag effect in this whole relationship so there was again very insightful and I
guess for this we crunched through quite a bit of data and and again that was
quite insightful and if you’re interested in reading a little more
about it again it was captured in a NASA Landsat newsletter we then wanted to
look at vegetation health so not only vegetation greenness even
though sometimes we might be using vegetation greenness as as a surrogate
for health but in other times you might not so we wanted to look at vegetation
health and whether that has something to do with flooding and so for this we
focused on one specific area and this is work that was done by Yuri Shendryk Yuri
was a PhD student in my lab and he’s now a postdoctoral fellow at CSIRO and
basically the the whole background of his work was that we have this river red
gums it’s a species of eucalypt there are these really big majestic trees
riparian trees they’re quintessentially Australian like the koalas right so
they’ve been there they’ve adapted to these systems they need flooding to be
able to sprout when there is a drought they drop branches and they have these
amazing complex structures they grow 30 metres tall and they have these complex
structures the older the trees are the more complex those structures become and
the reason I’m dwelling on this is that when you try to identify those
individual trees it’s not very easy from the data that we had and there have been
lots of episodes of dieback of river red gums and lots of people said well it’s
it’s got something to do with the reduction in the flooding frequency and
there is anecdotal evidence to suggest that but there was no study to combine
the two and put the two together so basically that’s pretty much what
we set out to do here so to start out with we
collected airborne lidar and hyperspectral data this is the forested
area Barmah-Millewa forest that we worked with about 750 square kilometers
and of course we couldn’t capture the entire forest and we collected data
airborne data across these red strips that you’re seeing and then we wanted to
look at tree health and link that to flooding so I guess how many of you have
used lidar data before lots of you quite a few of you how would you go about
delineating an individual tree and then you look for it yeah that’s perfect yeah
so that’s what most people would use would use it like the canopy peak and
then use something like an inverse watershed algorithm and delineate the
canopy well in our case that failed miserably these trees have complex
shapes we couldn’t get any trees out of it so instead of taking this top-down
approach of looking at an apex we decided to try a bottom-up approach and
see if that worked so what Yuri did was basically he used
lidar intensity to take away the understory in 0 to 10 meters where we
suspected the trunks would be and then we used Euclidean distance algorithms
which basically linked those points based on their location and those became
the trunks and then we used the top of the trunks as seeds in a 3D graph and
then we connected all these points based on 3D graphs and the distance between
these points that was the weight and then once we had these 3D graphs we
wanted to split them to get to individual trees so what we did for this
was random walker segmentation to basically decide which crown each point
would belong to so basically what they gave us was going from an unorganized
point-cloud to these individual trees and so then if you look along one of
those lidar strips you would see these individual trees so the different
colors would represent different individual trees we then so we were
happy with this we got an accuracy of about 68% which
may not sound like much but again given the the forested area that we were
working with and the kind of trees and that we were trying to get individual
trees we were super happy we then wanted to look at tree health and so for this we
did a lot of field work and a lot of measurements in the field and we did a
lot of visual assessments to look at basically tree health so this would be a
typical healthy tree declining or dead trees and once we had that basically
what you see here on your left-hand side is tree crowns with yellow are the ones
that are declining red are the ones that are dead and green are the healthy ones
overlaid on top of Landsat flooding frequency which I showed you before and
then on your right hand side you have this display differently on your x axis
you have flooding frequency and then on your y axis you have number of trees in
each of these health classes so I guess looking at these two graphs what do you
think is flooding having anything to do with tree health or not yeah it does seem
like it so it looks like a flooding frequency of say greater than five to
ten with would have areas with healthier trees than very low flooding frequency
and that was really important and really insightful primarily for the managers
because they wanted to understand how many times would this area need to be
flooded to understand how to get healthy trees and then if that area wasn’t
flooded that many times could they add environmental water as environmental
flows to this area to basically produce more healthy trees so we’ll
we’ll see how that goes when it’s actually applied in in practice and so
when you do field work in the Australian outback in the bush you get
to see lots of lots of really cool wildlife so we were super excited to
find the nest of um of emu birds and by the time I was away I had to go back
and teach and my team was was left seeing all these cool things and they
did pretty much once they saw this emu bird which is about two metre tall bird
they did pretty much the only logical thing that one can think of and they
took spectra of the eggs so there you go the last thing that I’ll be talking
about is looking at assessing dynamic connectivity based on surface water
networks and this was a big component of the work that we’ve been doing and we
will be doing here as well and the way I’m talking about connectivity is really
the way I guess landscape ecologists would use the term so I’m not talking
hydrological connectivity as in the way hydrologists would think of of
connectivity and why is that important primarily in dryland areas well
hopefully I convinced you showing you these animations that these are highly
dynamic systems so they go from very dry to very wet there are surface water
bodies that would that would merge into one big one and then they shrink to
smaller ones even temporary or ephemeral water bodies can be really important
particularly during drought conditions because they can act as stepping stones
in time or in space and I guess a lot of the previous studies were looking at
some localized areas arguably because it’s it’s it’s not easy to do large
areas when you look at connectivity so the way we we approach this again is by
using graph theory based on network analysis whereby each of these habitats would become nodes in a network or in a graph and then they would be linked
via edges which can be distance or resistance distances how have you seen
connectivity or networks can you think of networks that you might have
encountered in your day-to-day transportation your Facebook network so
each of your friends would be the nodes in the network
edges could be anything from how well connected you are so networks are
everywhere they’ve been everywhere I guess in
ecology we started using them in the past I don’t know 10 20 years I don’t
know not for not for as long as they have been used in other in other areas
of study the importance of using a graph theory based on network analysis is
because we can look at capturing some of the network topology and we can capture
this systematically consistent over space as well as time and we can
identify areas and hopefully habitats that are really important to conserve
when we have limited resources so what are some of the most important things
that habitats can be in a network so you have your connectivity hub which would
be the one that’s the most connected in a network so it’s like your most popular
friend where you might be the most popular one in your network so
ecologically this is important because they can act as refugia during
environmental disturbance and then you can also have nodes that are acting as
stepping stones unlike hubs they’re not necessarily very well connected but they
are really important why are they so important say again
yes exactly if you remove this node pretty much the network structure
collapses so they are critically important even though they are not as as
well connected compared to hubs and ecologically they are useful for
organisms to move long distances so basically when you have adaptations to
climate change you need to have these stepping stones for organisms to move
from point A to point B so this work was was done by by Robbi Bishop-Taylor
who’s now Dr. Robbi Bishop-Taylor he he did his PhD on surface water networks
and he is now an Earth Observation Scientist with Geoscience Australia and
for Robbi’s work we were interested in looking at certain I guess species that
might be useful for connectivity analysis so we settled on amphibians for
a couple of reasons for surface water networks amphibians are really good
because they depend on the water habitats but also on the land use
between these water habitats so there was it was kind of convenient to us the
other thing that was again convenient was that the southeastern part of
Australia is a hotspot of amphibian decline globally and we don’t really
know why they’re declining it could be a loss of connectivity or a loss of
habitat or a degradation of these habitats but we don’t really know what’s
going on so it it seemed like a really good idea to start investigating that in
amphibian studies looking at connectivity a lot of them would have
had say a static habitat and they might have had something like a Euclidian
distance so as the crow flies from habitat A to habitat B obviously there’s
not very realistic lots of studies though have used a least cost path type
of algorithm what’s the least cost path example that you can think of in your
day-to-day or in connectivity studies the sidewalk yeah between
two buildings I heard that I’ve actually learned that the hard way I don’t have a
car yet so my first day on campus I’m trying to walk and my phone tells me go
on Western Boulevard so I did and so I had to cross two highway entrances
basically and there is no stop sign so I basically decided that’s not my
least-cost path I need to find a different way so I now walk by campus
any other example ecological trails in a park a greenway perfect yeah and
corridors so so basically least-cost paths are kind of the way we we we look
at connectivity however when you look at organisms least-cost paths might assume
that the organism actually knows what the least-cost path is between habitat A
and habitat B and it turns out they don’t necessarily have an inbuilt
least-cost path algorithm there are actually studies that looked at
amphibians saying that they don’t know what the least-cost path is so better
way a more realistic way of coming up with this is basically look at again
random walker using circuitscapes of the way the current would move through a
circuit and basically instead of just having a single path you would have
multiple paths from habitat A to habitat B and it would give you a little area
that you can that you can think of as as connecting these habitats so in Robbi’s
first paper we started out with a static surface water layer from Geoscience
Australia and then we wanted to incorporate some dynamics and at that
point we weren’t done with our Landsat time series so what we did was using
flooding scenarios one in two years five years ten twenty fifty and a
hundred years and then we wanted to map where were these areas that were
important as hubs and the stepping stones so basically what I’m showing you
here are two rows of maps on your top you would have the areas that were
important as hubs and bottom the ones that are important
the stepping-stones red areas are more important and then going from left to
right you have no flooding scenarios of the static water maps going to one in
100 years flooding so what you see going from left to right is that you have more
red on the maps so you have more important habitat for connectivity as
you have more water in the system that makes sense you have more habitat more
connectivity the other thing that was pretty cool for us to see was that areas
that were important as hubs seem to correspond with areas that were
important as stepping stones as well and that was really important to us
because if you are to conserve certain areas it’s important to have both kinds
in the same area because you typically have limited resources to use for
conservation the other thing that we wanted to look at was how well are the
protected area schemes doing in terms of capturing areas that are important for
connectivity so what you’re seeing here is on the y axis areas that were
important as hubs and then on your x-axis going from left to right for each
of these three graphs are the static layers going from one in two and one in
100-year floods and then we split them up into habitats that were protected
versus habitats that were not protected and we looked at three different
protection schemes protected areas directory of important wetlands in
Australia in the middle and the Ramsar Convention on your right hand side so
what we found was that actually the areas that are protected are good for
for connectivity as well and this was one of the I guess few good news
environmental studies and good results for us I guess lots of environmental
studies might have doom-and-gloom answers and we were quite happy with
this because none of these protection schemes included connectivity among
their criteria but they were actually doing quite well in terms of protecting
habitats that were good for connectivity we then wanted to incorporate the true
dynamic captured from our satellite data so we considered that as the true
dynamic so we took our time series of surface water maps we had about 99 per
season per year surface water maps and then we apply connectivity on top of it
and then we looked at where are the areas that are important as hubs and
stepping stones over space and in time according to us at that point in time in
2018 this was one of the largest ecological networks because it was going
again across a large space and also in time so what have we found on your
left-hand side are habitats that are important the stepping stones in the
middle as hubs and on your right-hand side you have the amount of habitat
across this entire area areas shown in yellow are the ones that are more
important as hubs or stepping stones so just looking at this what jumps at you
was the first thing that jumps at you mm-hmm what else yeah yeah no those are those are those
are great examples I guess what jumps at me is that this is really dynamic so I
guess if we were to only have snapshots in time we would obviously miss out on
all of these dynamics so again the point that I’m trying to make is that for
dynamic systems we should be using time series of satellite data and incorporate
this in our analysis to be able to get a an understanding of these dynamics and
quantify these dynamics just to show this in a less dynamic way we basically
took the entire time series and we group the time series the bottom 25 driest
time steps the top 25% wettest time steps so that’s our wet and then everything
else in the middle was average so what you’re seeing here is with orange
habitats that are important as stepping stones with blue as hubs and yellow both
and what you’re seeing here is that during dry periods we pretty much had
habitats that were important as hubs and stepping stones along the major river
systems I guess you don’t know the major river systems but these are the major
river systems during average times we had more habitats so more connectivity
and then during wet times we had lots of interesting areas that started popping
out some of the interesting ones are located here in the river Rina
this is an area that’s irrigated so it turns out that in our modeling irrigated
agriculture was actually good for amphibian connectivity now this is a
controversial result when we go to talk to the Murray-Darling Basin Authority
and when we talk to I guess people yeah so it can be a controversial result in
terms of water usage but you know our modeling these turned out as important
for connectivity the other area that was important is located right here and this
is pretty much in the … area the northwest is pretty much the only area
that’s not developed across this basin the other thing that was interesting to
us was that again if we were to give suggestions for conservation or for
prioritizing resources for conservation we would definitely suggest this part of
the basin the north western part most of the protected areas
are located right here in the south and in the center part of the basin whereas
there a very few areas that are protected right here so these would be
the area that that we would suggest for as a priority area for conservation and lastly we wanted to look at how these
hydro climatic variability so the big drought and the big wet impacted
connectivity so again across space and time this is just a quick zoom in in the
middle showing you a network during the dry season and on your right-hand side
during the wettest season the size of the bubble is basically proportional to
the size of the habitat and then the links are the edges so how connected
this network was and we looked at connectivity across the entire basin so
using what we would call a global topology metric to see how this changes
during hydro climatic variability so basically we looked at the total area
connected versus the total habitat area and what we found was that while the
total habitat area was decreasing during dry periods very fast the total area
that was connected wasn’t decreasing as fast so the way we interpreted this was
that there is some resilience in the system in terms of connectivity given
that the habitat was decreasing at a faster rate the summary is cut off so so
so I guess just to conclude we used a time series of satellite data of surface
water maps for 26 years seasonally continuous and then we use this data set
in down the track applications looking at vegetation dynamics and vegetation
response to flooding looking at vegetation health response to flooding
as well as landscape connectivity to start looking at this system from
various angles and try to get a better picture of the system and I guess with
that I’m really thankful to my former lab members and all the people that have
been involved in this work and to my funding sources so I’ll just end with
this a short time series of my previous lab in Australia and and my very young
lab here a three week old lab so Mollie and Vini whom you probably met by now so
I look forward to growing as a lab and thank you very much all for coming
especially given the rain outside excellent and if you don’t have any
questions I have a question this is my question from natural environments to you
dryland agriculture and from irrigated agriculture to dryland agriculture not
as much in this basin the basin is pretty agricultural most of the major
cities are kind of on the edge of the basin you mean like individual habitats or
just so we kind of did that with with the hubs and stepping stones you mean to
add to basically restore new habitats we we did not look at adding habitat into
the network but that’s yeah yeah that’s a great suggestion we we basically had
tree health experts like eucalypt tree health experts and then we went out in in the
field and visually assess the trees there is a typical assessment protocol
that’s used for eucalypt and you have two people I was one of them assessing
the the tree health no so we had on- ground data to use with our lidar data
to get our accuracy assessment yeah they would that’s an excellent
question and that was actually one of my questions for you but I forgot to ask
you so I’m asking you now so so I would say yes because these are very complex
trees so I think if they if these methods work for those trees I would I
guess I would be confident to say that they might work for other trees but I
wanted to ask you are there similar trees in this area and which ones yeah
that’s another great point so in our study area about eighty five eighty
seven percent of the whole place was was a single species so that worked out
pretty well I don’t know that’s something to look into
yeah yeah yeah we did it on a small area we actually
only had that much funding to acquire lidar so we so we actually did for
our health algorithms we actually did so the 68% was for identifying individual trees
and then our accuracy for health actually incorporated hyperspectral data however
for this particular species when when there is a drought what they do they
start dropping branches so the tree looks pretty much the same as it would
look when it’s healthy and then all of a sudden it just drops branches so lidar
was the one that captured this so basically when you see a crown with very
few branches that’s that’s when you know the tree is not doing so well
so you didn’t we didn’t see a lot in in terms of hyperspectral data for this
particular species but I would suspect that is very different in other places I
noticed that the trees were very segregated we did have that in some
areas and that’s where the algorithm didn’t do so well

Leave a Reply

Your email address will not be published. Required fields are marked *