https://github.com/andrewzm/atminv

Supplement package for papers on atmospheric trace-gas inversion by Zammit-Mangion et al.

https://github.com/andrewzm/atminv

Repository

Supplement package for papers on atmospheric trace-gas inversion by Zammit-Mangion et al.

https://github.com/andrewzm/atminv/blob/master/

atminv
=====

This page hosts a supplement package to reproduce the results in the papers 'Spatio-temporal bivariate statistical models for atmospheric trace-gas inversion' by Zammit-Mangion et al. (2015) and  'Trans-Gaussian bivariate modelling with application to atmospheric trace-gas inversion'. I will call these two papers **Paper 1** and **Paper 2**, respectively. The accompanying vignette in this package may be used to reproduce the results shown in Section 4.1 of Paper 1 while the two R scripts in the top-level folder may be used to reproduce the results in the UK/Ireland sections of both papers.

To download the vignette without reproducing it on your machine, please view it in the `vignettes` folder or by directly click on the following links:

[Vignette (Section 4.1 of Paper 1)](https://github.com/andrewzm/atminv/blob/master/vignettes/chemo_sim_study.pdf?raw=true)

For a summary of the results of Paper 1 please see the [research homepage](http://niasra.uow.edu.au/cei/research/UOW202715).

Presentations
-----

The presentation I gave for the 27th Goulburn Research Fellows Meeting (2015) that is mostly focused on bivariate conditional models but that also contains some slides on this topic is [here](https://github.com/andrewzm/bicon/blob/master/pres/2015_03_Goulburn.pdf?raw=true).

The presentation I gave in the Spatial Statistics Conference in Avignon (2015) that is centred around this work is [here](https://github.com/andrewzm/bicon/blob/master/pres/2015_06_Zammit.pdf?raw=true).

Reproducibility 
-------
    
If you wish to reproduce the results, you will need to install the package and its dependencies. Two of its dependencies, `hmc` and `linalg`, need to be installed manually. To do this please open an R consoles, make sure you have `devtools` installed, and then type

    library(devtools)
    install_github("andrewzm/hmc",,dependencies=T)
    install_github("andrewzm/linalg",,dependencies=T)

To install the `atminv` package and the other dependencies, please make sure you have `devtools` loaded and then type

    install_github("andrewzm/atminv",build_vignettes=T,dependencies=T)
  
If you do not wish to compile the vignette (which takes a while) please set `build_vignettes=F` above. Otherwise, to view the vignette please type

    library(atminv)
    vignette("chemo_sim_study")
    
and select the vignette under `atminv`.

**Important:** The package `maps` has been updated to version 3 since publication of Paper 1. The results change slightly with the new maps. For reproducibility's sake, I now have saved the world map I use in the `data/` folder, which is equivalent to that in the `maps` package version 2.3.11. Also since `ggplot2 v2.0.0`, the plotting no longer works with grouped data frames. To obtain the required plots, convert the grouped data frame to a data frame using `data.frame()`.

The `sessionInfo()` one month after releasing the results for Paper 1 (after having rolled back the `maps` package) that yield the same results was as follows:

      R version 3.2.0 (2015-04-16)
      Platform: x86_64-pc-linux-gnu (6.4e+01-bit)
      Running under: Ubuntu 14.04.2 LTS

      locale:
       [1] LC_CTYPE=en_AU.UTF-8       LC_NUMERIC=C               LC_TIME=en_AU.UTF-8        LC_COLLATE=en_AU.UTF-8    
       [5] LC_MONETARY=en_AU.UTF-8    LC_MESSAGES=en_AU.UTF-8    LC_PAPER=en_AU.UTF-8       LC_NAME=C                 
       [9] LC_ADDRESS=C               LC_TELEPHONE=C             LC_MEASUREMENT=en_AU.UTF-8 LC_IDENTIFICATION=C       

      attached base packages:
      [1] grid      stats     graphics  grDevices utils     datasets  methods   base     

      other attached packages:
       [1] mapproj_1.2-4      fitdistrplus_1.0-4 hmc_0.1.0          atminv_0.1.0       gstat_1.0-26       sp_1.2-1          
       [7] gridExtra_2.0.0    ggplot2_1.0.1      devtools_1.8.0     Matrix_1.2-0       tidyr_0.3.1        dplyr_0.4.3       
      [13] maps_2.3-11       

      loaded via a namespace (and not attached):
       [1] Rcpp_0.12.1        RColorBrewer_1.1-2 git2r_0.11.0       plyr_1.8.3         R.methodsS3_1.7.0  R.utils_2.1.0     
       [7] tools_3.2.0        xts_0.9-7          SDMTools_1.1-221   digest_0.6.8       memoise_0.2.1      gtable_0.1.2      
      [13] lattice_0.20-31    DBI_0.3.1          curl_0.9.3         parallel_3.2.0     proto_0.3-10       stringr_1.0.0     
      [19] xml2_0.1.2         gtools_3.5.0       rversions_1.0.2    spacetime_1.1-4    R6_2.1.1           survival_2.38-2   
      [25] gdata_2.17.0       reshape2_1.4.1     magrittr_1.5       splines_3.2.0      scales_0.3.0       intervals_0.15.1  
      [31] MASS_7.3-40        assertthat_0.1     colorspace_1.2-6   stringi_1.0-1      lazyeval_0.1.10    munsell_0.4.2     
      [37] FNN_1.1            R.oo_1.19.0        zoo_1.7-12        


Simulation experiment in Paper 1
-------

Section 4.1 in Paper 1 details three sub-experiments. The first is the vanilla case with no model misspecification. The second assumes model misspecification, specifically that the flux field is uncorrelated even though it is not. The third assumes no misspecification but uses 1000 observations and shows that the EM-Laplace algorithm can run even in this scenario. If you wish to run these three different scenarios please download the `.Rnw` file from the `vignettes`, load `RStudio` and set `Tools -> Global Options -> Sweave -> Weave Rnw files using knitr`, and then press `Compile PDF`. Before doing so:

- To run the first case make sure `model = "full"` and `misspecification = 0`.
- To run the second case make sure `model = "full"` and `misspecification = 1`.
- To run the third case make sure `model = "full_big"` and `misspecification = 0`.

Note that if `load_results <- 1` then the computationally intensive results are simply loaded from cache. To run from scratch please set `load_results <- 0`.

UK and Ireland case study in Paper 1
-------

The file `chemo_UK_study.R` contains the code for Section 4.2. Unfortunately the files are too large to upload on Github so they have been uploaded [here](http://hpc.niasra.uow.edu.au/ckan/dataset/example-dataset-for-atmospheric-trace-gas-inversion) instead. We only provide an `R` file for this example, which, however is self-contained. The `R` file is named `chemo_UK_study.R` and is found in the root folder on this page. It should run without any problems when all the datasets have been extracted and put in a `../data/` folder. Obviously this path can be changed within the code to suit one's needs.

Plots of cumulant functions in Paper 2
-------

The code to reproduce the plots of the cumulant functions in Appendix B of Paper 2 is in `spasta_cumulants_plots.R`.


UK and Ireland observation system simulation experiment in Paper 2
-------

The file `spasta_UK_study.R` contains the code for Section 5. Unfortunately the files are too large to upload on Github so they have been uploaded [here](http://hpc.niasra.uow.edu.au/ckan/dataset/example-dataset-for-atmospheric-trace-gas-inversion) instead. We only provide an `R` file for this example, which, however is self-contained. The `R` file is named `spasta_UK_study.R` and is found in the root folder on this page. It should run without any problems when all the datasets have been extracted and put in a `../data/` folder. Obviously this path can be changed within the code to suit one's needs.

The results of Paper 2 are not as straightforward to compute as those in Paper 1. At the beginning of the script one has to choose one of the listed models (e.g., `Box-Cox.uncorrelated`) and set `do_inference = 1`. This will run an MCMC algorithm for that model (with the specified number of parallel chains, burn-in etc.) and save the results in a `./tempresults/` folder which might need to be created manually first. This needs to be done for each model, and requires a high performance computing environment. The results are saved in `inst/extdata/spastaMCMC` for quick loading and analysis and therefore can be loaded using `system.file()` as in the script.


Versions
--------

The results in both papers were carried out with 

    R version 3.2.0 (2015-04-16)
    Platform: x86_64-pc-linux-gnu (64-bit)
    Running under: Ubuntu 14.04.2 LTS

The `Matrix` package is v1.2-0. Results may change slightly across different `R` and package versions and across different platforms. To see the vignettes obtained under different platforms for Paper 1, see the pdf files within the folder `reproducibility`.

References
-----

Zammit-Mangion, A., Cressie, N., Ganesan, A.L., O'Doherty, S., Manning, A.J. (2015). Spatio-temporal bivariate statistical models for atmospheric trace-gas inversion. Chemometrics and Intelligent Laboratory Systems, 149, 227--241, doi: 10.1016/j.chemolab.2015.09.006.

Zammit-Mangion, A., Cressie, N., Ganesan, A.L. Trans-Gaussian bivariate modelling with application to atmospheric trace-gas inversion, submitted to Spatial Statistics.

Owner

GitHub Events