Tools for open geospatial science

How to make your research reproducible

Vaclav (Vashek) Petras

NCSU GeoForAll Lab at the Center for Geospatial Analytics
North Carolina State University

US-IALE, Baltimore
April 9-13, 2017

Science

Science existed before journal papers.

Open Science

CC BY Stefan Janusz, Wikipedia

Open Science

registration so that scientists get credit
archiving so that we preserve knowledge for the future
dissemination so that people can use this knowledge
peer review so that we know it's worth it

Open Science

[Buckheit and Donoho 1995, Peng 2011, Rodríguez-Sánchez et al. 2016, Marwick 2016]

Image credit: CC BY-SA Comtebenoit, Wikimedia

Scientists rely on software

It's impossible to conduct research without software, say 7 out of 10 UK researchers

— Hettrick et al, UK Research Software Survey 2014

Software needs to be shared

Software [...] developed as part of novel methods is as important for the method's implementation [...] Such software [...] must be made available to readers upon publication.

—Nature Methods - 4, 189 (2007)
doi:10.1038/nmeth0307-189

Screenshots

Scripting

Bash

r.in.lidar input=points.las output=elevation -e

Python

from grass.script import run_command
run_command("r.in.lidar", input="points.las", output="elevation", flags="e")

library(rgrass7)
execGRASS("r.in.lidar", input="points.las", output="elevation", flags="e")

Paper: PDF

Paper: Code

File versions

Revision control

Git

git commit script.py -m "replaced part of the main equation"

There are GUIs as well.

Collaborative writing: Overleaf

Text and code: Jupyter Notebook

interactive document with text, code, and figures
languages: Python, R, Bash, C, C++, Octave, ...
alternatives: R Markdown (Notebook), Emacs Org-mode, ...

Publishing a model

Source code

Source code history

Documentation

Graphical user interface

Integration into a larger project

Do I want to create a new software project?

Preprocessing, visualization, and user interface (GUI, CLI, API)
Integration with existing analytical tools
Inputs, outputs, memory management

FUTURES model implemented as GRASS GIS modules
r.futures.pga, r.futures.demand, r.futures.parallelpga, ...

Alternatives: R package, Python package, QGIS plugin, ...

Running the code

dependencies
environment

Complete environment

	My laptop	Any computer	Docker
What is required to recreate the study and run the software?	nothing*	text	text file
How are the dependencies handled?	?	?	specified

*Except that I'm not going to lend you my laptop.

Dockerfile

FROM ubuntu:16.04
RUN apt-get update
RUN apt-get install -y \
        g++ \
        python \
        python-numpy \
        ...

Research publication

Paper	background, methods, results, discussion	PDF, HTML
Environment	details about all dependencies and the code	Git, Docker
Specific code	scripts to perform the analyses	Bash, Python
Reusable code	methods as GRASS GIS modules	Python, C

Petras, V., D. J. Newcomb, and H. Mitasova. Generalized 3D fragmentation index derived from lidar point clouds. In: Open Geospatial Data, Software and Standards [in print]

Motivation

Am I doing science to publish journal papers or to advance knowledge?
What happens when I need to rerun an analysis from last year?
How can my collaborators run and improve the code I'm running?

Next steps

Publish R scripts with the next paper.
Collaborate on code using Git.
Collaborate on text using Overleaf.
Use Jupyter Notebook for a report.

Like irreproducibility more?

Resources