Things from meeting with Fenner School of Environmental Science

• Lots of spatial data is available, but it often requires specialist expertise to summarise/interpret data. By the time you access the data, it can be years after when you really needed it.
  • Website they're launching is meant to provide automated environmental analysis/summary stats to get around that problem.
• Statistics (e.g. fractional vegetation cover) are both displayed on a grid _and_ summarised by region. You can download the gridded data from the NCI, or download @.csv@ files containing the summarised data from their websites.
• Types of regions they used for aggregating data:
  • States
  • Local catchments
  • BoM river regions
  • IBRA bioregions
  • National parks
  • RAMSA wetlands
• Easy to access time series and descriptive statistics for any given region. Sometimes there are also fancy pie charts available (e.g. for land cover type).
• They're looking for more data and more website features to implement in 2016 (scientist feedback?)
• Being able to see YoY changes is really nice, esp. if it's broken down by land cover type. They colour coded changes and put them on a map (aggregated by state) and it looked awesome---what would it look like at ~1km resolution? Can we do that?
• Website seems to be capable of doing all sorts of complicated analysis, but it's not clear what the driver is for the different features. Is it just stuff that's easy to implement? Stuff that's most useful to scientists?

Some info about most of the types of data they can display:

• *Land cover:* based on MODIS 250m data ("temporal greenness patterns"). GA releases this annually. There are ~22 types (water, different types of tree cover, etc.). Tree cover (>=20% canopy cover with potential to reach 2m+ height) is also provided---was using national carbon accounting system data (1972-2013), but now using GA landsat data processed with some mystery algorithm.
• *Bushfires:* derived from GA Sentinel Hotspots system. Regular spatial resolution, but some temporal gaps---likelihood of detecting fire is variable (e.g. if there are few satellites).
  • They did a fire frequency breakdown by land cover type ("Which forests are most prone to burning?"). That seems like exactly the kind of thing linked data would be good at.
• *Water coverage:* data from from OzWALD
  • Average national rainfall by year is apparently a useful metric (BoM tries to estimate it annually or something)
• *Soil moisture:* also from OzWALD.
• *River inflow:* from OsWALD as well. Regional differences were relevant (e.g. is river inflow higher in the north/south of Australia?).
• *Inundation:* what fraction of this area was covered by water in a given time period?
• *Exposed soil (fraction of soil not covered by living or dead plants)*: derived from MODIS by CSIRO.
  • They like doing time series of measured quantities, but it seems like the time derivative of the time series is also relevant (YoY change).
• *Vegetation leaf area:* taken from MODIS (processed by NASA).
  • Everything is at totally different spatial and temporal resolutions. Some stuff is regular, some stuff isn't, etc.
  • Website doesn't do correlative stuff which might be interesting. I guess that's really just a matter of downloading the time series and doing your own regressions, though.
• *Carbon storage:* ANU OzWALD-derived again. Measures amount of "carbon taken up by the vegetation through photosynthesis".
• *Fire carbon emissions*: carbon released through bushfire, as estimated by GFAS (European system).
  • Summarising YoY difference by region identified some major fires from considered time period (sometimes shown as increases in fire-produced carbon where there were fires in 2015, or decreases when there were fires in 2014).

Stuff from the brief environmental overview at the end of the (semi-technical) seminar:

• They've quickly hacked up an "environment condition score" to try to summarise how well Australia's environment is doing. Work is totally unpublished and will probably change. It might not be very useful, but knowing how to calculate it (and ensure that we can calculate it!) would be nice. Apparently it's also politically useful/exciting.

Relevant things from Q&A at the end:

• **Q:** Is there an API?
  • **A:** You can download CSVs for the summary data, and netCDFs (from THREDDS) for the gridded data.
• **Q:** Can you mix it with other statistical data (e.g. demographic data)?
  • **A:** Not yet. Dealing with different definitions of administrative regions is the bane of a scientist's existence. Enabling something like that would be helpful, but it's outside the scope of the project.
  • A guy from the ABS implied that having data summarised over lots of different types of region (local government area, electorate, biodiversity zone, etc.) was a really helpful feature of the website. Linked data seems like it could be a huge boon here (if we get SPARQL queries working), as you can potentially query for data in arbitrary regions and have the server just calculate it for you.