By Tim Salabim
This is a guest post by Florian Detsch
What it is all about
With the most recent update of the AVHRR GIMMS data collection to NDVI3g (Pinzon and Tucker, 2014), we decided to create a package from all functions we have written so far to download and process GIMMS binary files from the NASA ECOCAST server. The package is called gimms and features a collection of fundamental work steps required to get the data into R:
updateInventoryto list all GIMMS files available online and
rearrangeFilesto sort (online or local) files by date,
downloadGimmsto download selected files,
rasterizeGimmsto import the binary data as ‘Raster*’ objects into R and
monthlyCompositeto aggregate the bi-monthly datasets to monthly value
How to install
The gimms package (version 0.1.1) is now officially on CRAN and can be installed directly via
## install 'gimms' package install.packages("gimms") ## load 'gimms' package library(gimms)
In order to use the development version (no liability assumed), please refer to the ‘develop’ branch hosted at GitHub. There, you will also find the latest news and updates concerning the package.
library(devtools) install_github("environmentalinformatics-marburg/gimms", ref = "develop")
List available files
updateInventory imports the latest version of the online file inventory as ‘character’ vector into R. By setting
sort = TRUE, it is at the same time a wrapper around
rearrangeFiles as the output vector will be sorted by date rather than in alphabetical order. The latter feature proves particularly useful when considering the GIMMS file naming convention, where e.g. ‘geo13jul15a.n19-VI3g’ means the first half of July 2013. In case no active internet connection is available,
updateInventory automatically imports the latest offline version of the file inventory.
gimms_files <- updateInventory(sort = TRUE)
## Trying to update GIMMS inventory from server... ## Online update of the GIMMS file inventory successful!
##  "http://ecocast.arc.nasa.gov/data/pub/gimms/3g.v0/1980s_new/geo81jul15a.n07-VI3g" ##  "http://ecocast.arc.nasa.gov/data/pub/gimms/3g.v0/1980s_new/geo81jul15b.n07-VI3g" ##  "http://ecocast.arc.nasa.gov/data/pub/gimms/3g.v0/1980s_new/geo81aug15a.n07-VI3g" ##  "http://ecocast.arc.nasa.gov/data/pub/gimms/3g.v0/1980s_new/geo81aug15b.n07-VI3g" ##  "http://ecocast.arc.nasa.gov/data/pub/gimms/3g.v0/1980s_new/geo81sep15a.n07-VI3g"
The next logical step of the gimms processing chain is to download selected (if not all) bi-monthly datasets. This can be achieved by running
downloadGimms which accepts various types of input parameters.
- ‘missing’ input → download entire collection
Specifying no particular input is possibly the most straightforward way of data acquisition. The function will automatically start to download the entire collection of files (currently July 1981 to December 2013) and write the data to
## download entire gimms collection downloadGimms(dsn = paste0(getwd(), "/data"))
- ‘numeric’ input → download temporal range
It is also possibly to specify a start year (
x) and/or end year (
y) to limit the temporal coverage of the datasets to be downloaded. In case
y) is missing, data download will automatically start from the first (or finish with the last) year available.
## download gimms data from 1998-2000 downloadGimms(x = 1998, y = 2000, dsn = paste0(getwd(), "/data"))
- ‘character’ input → download particular files
As a third and final possibility to run
downloadGimms, it is also possible to supply a ‘character’ vector consisting of valid online filepaths. The latter can easily be retrieved from
updateInventory(as demonstrated above) and directly passed on to the input argument
## download manually selected files downloadGimms(x = gimms_files[769:780], dsn = paste0(getwd(), "/data"))
Rasterize downloaded data
rasterizeGimms transforms the retrieved GIMMS data from native binary into common ‘GeoTiff’ format and makes the single layers available in R as ordinary ‘Raster*’ objects. Thereby, it is up to the user to decide whether or not to discard ‘mask-water’ values (-10,000) and ‘mask-nodata’ values (-5,000) (see also the official NDVI3g README) and apply the scaling factor (1/10,000). Since rasterizing usually takes some time, we highly recommend to make use of the
filename argument that automatically invokes
## list available files gimms_files <- rearrangeFiles(dsn = paste0(getwd(), "/data"), pattern = "^geo13", full.names = TRUE) ## rasterize files gimms_raster <- rasterizeGimms(gimms_files, filename = paste0(gimms_files, ".tif"))
With a little bit of effort and the help of RColorBrewer and sp, here is what we have created so far.
Figure 1.Global bi-monthly GIMMS NDVI3g images from July to December 2013.
Generate monthly composites
Sometimes, the user is required to calculate monthly value composites from the bi-monthly GIMMS datasets, e.g. to ensure temporal overlap with some other remote sensing product. For that purpose, gimms also features a function called
monthlyComposite which works both on vectors of filenames and entire ‘RasterStack’ objects (ideally returned by
rasterizeGimms) and calculates monthly values based on a user-defined function (e.g.
fun = max to create monthly MVC layers). The function is heavily based on
stackApply from the raster package and the required coding work is quite straightforward.
## 'GeoTiff' files created during the previous step gimms_files_tif <- sapply(gimms_raster@layers, function(i) attr(i@file, "name")) ## create monthly maximum value composites gimms_raster_mvc <- monthlyComposite(gimms_files_tif)
Figure 2.Global monthly composite GIMMS NDVI3g images from July to December 2013.
A more comprehensive version of this short introduction to the gimms package including a collection of use cases (particularly in conjunction with R’s parallel capabilities) can be found online at https://github.com/environmentalinformatics-marburg/gimms. Any comments on how to improve the package, possible bug-reports etc. are highly appreciated!
R-bloggers.com offers daily e-mail updates about R news and tutorials on topics such as: visualization (ggplot2, Boxplots, maps, animation), programming (RStudio, Sweave, LaTeX, SQL, Eclipse, git, hadoop, Web Scraping) statistics (regression, PCA, time series, trading) and more…
Source:: R News