2016-02-12

fireMIP

Infomation on benchmarking models participating in the fire Model Inter-comparison Project (fireMIP).

Note this project is still under development.

Project Repo

The projects code and version history is accessible from github: github.com/douglask3/fireMIPbenchmarking

To clone this to you own computer:

git clone https://github.com/douglask3/fireMIPbenchmarking.git

When I have finished initial development, I will provide a link for the latest 'stable' version of the project.

Benchmark data

Benchmark data is on the fireMIP server. Contact Stijn Hantson for info. Each model should be stored in the following way:

<<PATH>>/<<NAME>>/<<EXP>>/<<OUTPUTS>>

Where:

  • PATH is a common path to all models, which must be defined in the cfg file as outputs_dir.BenchmarkData
  • NAME is a unique name for the model. It doesn't not have to be the actual model name, but can be for example an abbriviation. So long as no other model shares that name
  • EXP is the experiment (i.e "SF1")
  • OUTPUTS: all the model outputs. These should be netcdf files. The can be individually zipped with gzip. And they are allowed to be in different directories.

Configuring

There are several points in the 'cfg.r' where you can set the comparisons to be made, add new models, add extra observatioanl comparisons, and define which models and comparisons are to be made

Set Parameters

Here, you can set the following parameter:

experiment   = 'SF1'

Sets the experiment directory. Here, its set to SF1

mask_type    = 'common'

What mask to apply to the spatial comparisons. The options are still comoing online, but will include:

  • 'common': Compares areas of where observations and all models have data
  • 'commonHi': All models are resampled to the highest resolution of models and data (typically 0.5 degree) before applying the common grid
  • 'commonLw': All models are resampled to the lowest resolution of models and data before applying the common grid
  • 'Observations': models resampled to observations grid and compare for common grid between observation and model on a model by model basis
  • 'Modelled': observations resampled to model grid and compare for common grid between observation and model on a model by model basis.

Paths

data_dir.ModelOuutputs = 'data/ModelOutputs/'

The path to model output data (<> in model data section)

data_dir.BenchmarkData = 'data/BenchData/'

The path to the benchmark datasets

outputs_dir.modelMasks = outputs/Masks'

The path and start of pattern of the mask from model and observations, to be applied before metric comparison.

Model Information

Model.RAW lists information about the model outputs and how ethy should be processed. Each model forms an item in a list, and each item has three entires:

  • Directory name, or <> in model data section
  • Processing function, the function used to process the particular model. Most models will get away with process.default. If a new model needs a new function, this can be defined in a separate file and placed in 'libs' dir or its sub-directories, and it will be automatically sourced when the system is run
  • Start date, in decimal years - the timing of the first day of the models run.

For example

Model.RAW = list(
                CLM  = c('CLM' , process.CLM , 1996),
                CTEM = c('CTEM', process.CTEM, 1859))

Model.Variable Description soon For example

Model.Variable = list( #Line 1  variable name; Line 2  scaling; Line 3 - timestep
    varname  = rbind(c("BurntArea", "gpp"    , "ModelMask"),
                     c(100        , 1/1000   , 1          ),
                     c('Monthly'  , 'Annual' , "Monthly"  ),
                     c('mean'     , 'sum'    , "sum"      )),
    CLM      = rbind(c("BAF"      , "gpp"    , "cSoilt"   ),
                     c(100        , 1        , 1          ),
                     c('Daily'    , 'Monthly', "Monthly"  )),
    CTEM     = rbind(c("burntArea", "gpp"    , "cSoil"    ),
                     c(100        , 1        , 1          ),
                     c('Monthly'  , 'Daily'  , "Monthly"  )))

Model.plotting Description soon For example

Model.plotting = rbind(#Title  #Colour
              CLM  = c('CLM' , 'red'     ),
              CTEM = c('CTEM', 'green'   ))

Comparison Info

Infomrmation for comparisons are in a list per comparison. There a number of common items each list must contain, along with some extra items specific to the comparison type:

  • ObsFile is the filename, relative to data_dir.BenchmarkData that the observation is contained in

  • ComparisonFun the function which will be used to make the comparison. Available functions include:

  • FullNME, ...

  • Others listed soon

Spatial

Spatial comparisons can be an NME comparisons of Annual Average or itemized MM comparisons.

For NME

BurntArea.Spacial = list(obsFile = "Fire_GFEDv4_Burnt_fraction_0.5grid9.nc", obsVarname = "mfire_frac", obsLayers = 8:127, obsStart = 1996, ComparisonFun = FullNME, plotArgs = plotArgs)

where plotArgs are defined as:

BurntArea.Spacial = list(cols = c('white', "#EE8811", "#FF0000", "#110000"), dcols = c('#0000AA', '#2093FF', '#C0D0FF','white', '#FFD793', "#F07700", "#AA0000"), limits = c(0.001,.01,.02,.05,.1,.2), dlimits = c(-0.2,-0.1,-0.5,-0.01,0.01,0.05,0.1, 0.2))

Model and Comparison selection

More info coming soon

Participating Models

More Info coming soon

Packages Used

Packages are downloaded by running initalise.r ./initalise.r in command line and source("initalise.r") from within R.

benchmarkmetrics

The actual metrics used are in an R-package, and are sourced and called by this project.

To use in this package separately in you own project

install.packages('devtools')
library(devtools)
install_github('douglask3/benchmarkmetrics/benchmarkMetrics')
library(benchmarkMetrics)

For info on how to use this package:

?benchmarkMetrics

Click here for more info.

TM2r

This package will be used to calculate atmospheric seasonal concentrations of CO2 from model outputs of fluxes from Net Primary Production, hetrotrophic respiration, and fire. Click here for code development. More info coming soon.

gitBasedProjects

gitBasedProjects is used to link comparison outputs to a specific point in the projects development, and to specific configurations the model will be run in. This is done by adding the git version number, url, date and time to:

  • figures, as a watermark
  • netcdf outputs and temporary files, as attributes
  • csv tables, as footers.

To use in this package separately in you own project

install.packages('devtools')
library(devtools)
install_github('douglask3/gitProjectExtras/gitBasedProjects')
library(gitBasedProjects)

For info on how to use this package:

?gitBasedProjects

Click here for more info.

Extra raster functions.

More info coming soon.

Andy questions, contact me