By Brad Evans, PhD student in the Centre of Excellence for Climate Change Woodlands and Forest Health.
A preliminary stage of my PhD study has involved sourcing and processing satellite, aircraft and field data to be used in the assessment of forest health. Data has come in from across the country and the world and so we are building up a substantial library. My colleagues and our network of partners have been very generous in sharing data, ideas and knowledge on and around data – thank you all!
As I understand it, satellite data is a useful tool for the assessment of forests, not only by us but globally. The good news is that the data has now become more accessible than ever. Jacques Diouf, Director General of the UN Food and Agriculture Organisation had this to say about the recently launched Global Forest Resource Assessment Portal (GFRAP):
“This brings a revolution to the forest monitoring field. Never before have data of this kind been provided directly to users in developing countries. Monitoring will be cheaper, more accurate and transparent for countries that want to participate in reducing emissions from deforestation and forest degradation…”
Now, use your imagination and make the link between the acronym “GFRAP” and what they are doing on their web portal. I came up with gift wrapped… This portal is a gift! Over the coming years, I will use it in my attempts to understand vegetation responses to recent climate variability. I will then use it to predict how our vegetation might respond to future climate changes. See below to have a look at it yourself!
Time to do some analysis of your own and win a prize!
We now have lots of data so the processing phase of my work has begun. My first objective is to understand temporal and spatial patterns of vegetation change in Dryandra, southwest Western Australia. To help me in this I have developed a method to test your pattern recognition. First have a look at multi year images of the SPOT VEGETATION normalised difference vegetation index for Dyrandra. Figure 1 shows the monthly variability of this index for the period from January 2000 to December 2008. Red indicates active live green activity in vegetation, i.e. photosynthetic activity (on a scale from 0-255 where 0 is also used to distinguish woodlands from farmland). The lighter colours are a sign of less live green activity in the vegetation. So far my nearest colleagues have made some very interesting observations from this image – so can you!
Figure 1. SPOT VEGETATION 1st of the month 1km NDVI from Jan 2000 to Dec 2008 in rows of months left to right for years top to bottom.
Win a prize for your skill in visual analysis!
Firstly, make sure you use this PDF version of the time series so you can zoom in (hint).
I am offering a prize for giving me your insights:
1. Make three significant observations you make from this image (click it to enlarge)
2. Describe two errors
3. Comment on this before the next blog posting (see below to comment)
The winner’s prize will be decided on the basis of your sense of taste….
And herewith my gift for everybody: http://geonetwork4.fao.org/geonetwork/srv/en/fra.home
Hi Brad,
Good work. It is interesting to see that the vegetation changes that much across seasons and years.
1. The woodland is greener in winter than in summer;
2. The start and length of the ‘greener’ period vary from year to year (I’m sure you already correlated this with rainfall)
3. There is an edge effect in the “greening up and greening down”: inner parts get greener faster and edge parts stay greener longer. I don’t know the scale and resolution but it seems that this is not due to limitations of the technology.
The data for December 2001 and 2002 don’t seem quite right.
I suggest you analyze an image at a higher resolution and correlate the pixels with cover from aerial photographs. There are some folk in DOLA that I had contact with awhile ago who may be able to help with this.
I think its a good approach to use monthly sequences to develop understanding of growth fluctuations, however, I donot think it is established that the ‘greening up’ observed in these photographs are indicative of live wandoo canopy. On the contrary, I find it rather implausible that wandoo trees green up approximately linearly with winter rainfall as temperature would limit growth and photosythetic activity in trees for the majority of winter. The live green activity in these photographs might be skewed by ground cover in a sparse woodland like Dryandra. Instead tree growth is likely to occur in spring and summer ‘flushes’. I suggest a MR using rainfall and canopy/ground cover as independents on a 1:8000 as a start. Target wet summers for summer ‘growth’ and look for ‘flushes’ in spring following wet winters for a good calibration. Edison Mill road is a good control site for field work, as it has a low ground cover.
I also have an unpublished report on landsat on wandoo around the traps; if you are interested ask Giles.
Good luck Brad, great to see some work being started.
Ryan is correct, what you see is not just canopy so the greening is not necessarily that of trees, and the phenology of the trees (in fact a mix of species) would not seem to match it.
The mix of species is a good point; as far as I know it appears to be one of the main issues with Landsat’s capability for forest inventory, even in “uniform” coniferous forests. For example, the ability of Landsat to distinguish brown mallett from wandoo would be critical in dryandra because brown mallett has its own problems related to plantation viability (Pauline Grierson might be worth a conversation Brad). Also tree species would have different phenologies, especially if they are co-occurring in the same patch. Phenology of trees is critical, and poorly understood for eucs.
Hiya Ryan,
How is España?
Totally agree with you on the understory and spring flushing. I almost made the Edison Mill road my site on my last trip. I ended up choosing a plot on the historical tour trail, listening post 2.
Dying to get my hands on a detailed map of Wandoo distribution for the area. Any ideas there?
We have some higher resolution data coming and my approach is to make links between to two since my regional scale modelling ingests the 1km data I want to be sure that what we see in it makes sense.
Keep in touch.
Brad
Agree on using multi-scale approach, but a problem remains in using high res landsat in that you still don’t know what your looking at.
Unfortunately a detailed map for wandoo distribution does not exist as far as I know – it should be a priority honours project in my view. However, you may need to do some veg mapping at your sites especially for the high res stuff. And aerial maps may be suited to low res, but this would still require ground-based surveys as vegetation types may be difficult to distinguish without an eye for it (roughness, greeness). I’m sure the old DEC dwellingup/manjimup offices would have some folk experienced in aerial mapping, and you may get lucky and borrow someones eye for wandoo. Try Joe Kinal @ dwellingup. To get a rough guide on distribution you can have a look a the old forestry map series (it has topographically interpolated woodland distribution) and also an old ecologia report has some quadrat surveys (only 1x1m though). Also Havel’s mapping series in the Darling range might be worth a look. Contact bryan shearer (Bryan.Shearer@dec.wa.gov.au) for the old maps and stuff, although it might be best to go into dec science division (before 12 pm most weekdays) to see him as he is best to catch in person. Visit the Cons library and speak to Deborah – they have regional maps. Also I’ll dig up those DOLA contacts for you, its worth having a chat as they know a lot and have alot of time for students.
I had some thoughts about calibration that may help.
1. Using natural cover. You will need multiple sites that vary in cover (understorey/overstorey) if you plan to use regression i.e. using the natural variation in cover among sites to validate pixels of landsat.
2. Site history. Target sites with decline history around Talbot and/or my PhD locations. Dryandra has not been investigated apart from some cursory study in the 80′s, however information is notoriously difficult to get on these DEC-housed studies. Remember all sites “have history but not all have a past” (Pickett 1985) and without some prior knowledge of decline/management history you can’t know what has influenced the trajectory of sites. My advice is, do your homework before selecting sites as you could be walking into a minefield of complex interactions that mask things you see. We all know landsat picks up broad-scale disturbance (e.g. logging and fire), but for subtle, spatiotemporally-complex disturbance such as tree decline in a sparse woodland, the jury is still out. In short, control for other sources of variation may be critical in the calibration phase of your work.
3. Temporal scale – this is where the interesting questions are and it looks like you a well aware of this. Climate between months/years (c. Eriks comment above) would need to be factored into the temporal component of a calibration model.
Hope this helps,
Ryan
Hey Ryan, Yes, definately all helps. Thanks!
I have been trying to arrange some meetings with DEC- no luck.
Eric has some ideas on a Wandoo map for Dryandra also. We are meeting next week to discuss, wish you could come. I am keen to read your PhD.
On the “Climate” side of things I am downscaling meteorology to 1km and possibly further, using regional climate models and planning to combine this, as an index, with the satellite imagery.
I have also done some seasonal trend decomposition analysis of this data. It correlates with LandMonitor and decomposed, this NDVI has a close relationship to fluxes of soil moisture and VPD. I am presenting this at UWA next week. I will send it to you via email.
Gotta go and follow up these great leads. Thanks again.
Brad
Happy to help Brad,
Thesis not passed yet, but I hope to get an answer soon. Will keep you posted.
Good luck!
BTW. Doesn’t the NDVI need a little more ground-work before it establishes wandoo’s canopy and hence soil moisture?
Thanks Ryan,
Yes, the NDVI (and all remotely sensed products) definately need more ground work. For now I am looking at all the off-the-shelf products, will tune and truth later.
Other team members are working on site selection, for my work, I will make do with what they come up with.
Cheers,
Brad
Hello Brad,
I love seeing how these things appear visually – i am no mathematician or modeller, but this imagery makes the data accessible to those of us with ‘different’ expertise. I would be very interested to see what others in the community felt about this summary image and what they think it means about the health of this woodland/forest area.
Lovely work
I love seeing how the patterns represent changes and also similarities in the seasons over the year.
I found the little error:
the image for 2004-12-01 appears twice, and so does 2004-04-01. can you post the corrected version? I would like to see if there is any difference
Keep up the great work!
PS – While i am here i might as well suggest that i would like the centre (us) to consider developing a glossary from this blog and other reporting activities, so that we can have links to language some of us may not be familiar with. There are parts of the above discussion that were a little mysterious to me… and when Marleen starts blogging about discursive practises, actors and governance I am sure there will be people in other disciplines who will have a similar experience.