1 km Carbon maps project
- 1 Objective
- 2 Technical issues
- 3 Conceptual model of final (mosaic) map production using several input map products
.. to create a map of carbon stock according to the specs outlined in Media:Guidelines_GSOC17.pdf
Various pathways to acheive our goals
- Bottom-up (Province-territory)
- Top-down (National scales)
- Soil grids, SLC (covariables)
- Ensemble predictions (Merging all together)
The effect of Landcover
There is a need to isolate soil than non-soil area in the task of mapping carbon. Mineral soil should be infers separately from organic soil due to their particular properties and behaviour. Landuse as Rock, lakes, wetlands, urban area should also be extract to ensure that final products is not covering those areas. The use of landuse map derive from remote sensing technologies could provide an insight of the distribution of Landuse around Canada.
- Geobase (http://www.geobase.ca/) some datasets to map surface water, urban area and Landcover. Most of the information is relatively old and should be update.
- The potential of a recently global landcover at 30m should be explored (http://www.sciencedirect.com/science/article/pii/S0924271614002275)
- Potential of morphological data for carbon estimates : morphological data are more numerous than analytical data. Relationships between colors, granulometric fraction, depth and organic carbon contents could be assess to train models.
(RAM) Absolutely a good idea to use point data for which we have many point observations to improve predictions of soil analytical properties for which we have many fewer locations. Tom Hengl showed that predictions of soil analytical properties (in Africa at 250 m) were improved when we first used point data that gave soil classification information (10 x as many point locations as we had for analysed soil profiles) to first predict surfaces of likelihood of belonging to each soil class and then used these likelihood maps of soil class as covariates in predicting most likely soil property values. The improvement was substantial and was demonstrated to be statistically significant.
Data is available at the following locations:
The 90m gridded attribute dataset derived from SLC can be found here
The circumpolar carbon database can be found here
- Challenge of optimisation
What will be the procedures to update maps for FAO
- Work on harmonized dataset
- GSIF - spline funciton to homogenize dataset
- Malone et al. - spline equal area
Notes added on April 18...
- 1. a single purpose, national point dataset used to validate all products, and not used for modeling. The dataset should be ;
- be homogenize (JD)
- provide a full coverage of Canada (JD)
- be open access (JD)
- provide an equiprobable spatial accuracy for 1 km square area (JD)
- 2. point distribution considerations
- point density,
- geographic distribution,
- physiographic/climatic/land cover (eg latin hypercube)
Mapping method technical specs
- 1. global SOC stock 0 - 30cm (requires corrrections for bulk density & stones)
- 2. global SOC concentrations
- 3. uncertainty
- spatially explicit
- standard deviations
- quantiles (ensemble predictions
- others considerations
- 4. Questions
- a) soil/air sepration
- b) report SOC concentration over the whole (0-30cm) layer even if part of it is eg
- c) model forest floor FH separately form mineral
- d) application of masks for water, built up
- e) rock: model it as part of the predictions or use a
- f) organic: ...
- g) wetlands: ...
Data use guidelines.
- 1. there was discussion at the steering committee regarding our data use policy.
- 2. need to be sure we are using external appropriately
- 3. some version of free-use license?
- 4. need to include in our metadata for all inputs
- a) source,
- b) reference citation,
- c) license info if available.
Time is going on and I personally think that we should used an open access datasets to alleviate the difficulties that are already numerous(JD)