beeoda: a suite of open-source software and educational ...beeoda: a suite of open-source software...
TRANSCRIPT
B E E O D A
BEEODA: a suite of open-source software and
educational materials for processing Earth
Observation data
Pontus Olofsson, BU & SilvaCarbon
Eric Bullock, Boston University (BU)
Curtis Woodcock, BU, NASA, USGS, GOFC-GOLD
Chris Holden, Boston University
GOFC-GOLD Workshop
Cote d’Ivoire, February 8th 2017
B E E O D A
Why are we doing this?
▪ For our own research and teaching responsibilities! And for
capacity building initiatives (SilvaCarbon, GFOI, GOFC/GOLD)
▪ Remote sensing data quality/quantity and analyses advancing
rapidly -- proprietary software lagging and expensive -- open
source solution are free and dynamic and powerful!
▪ Lesson learned from teaching CP workshops: software to take
home and that are in line with new advancements needed
2
B E E O D A
3
CAPACITY BUILDING
Capacity building needs in tropical
countries; SilvaCarbon, NASA,
FAO, etc.
SOFTWARE
BEEODA: free open-source
software and educational materials
for EO analysis
EARTH OBS. DATA
Vast Landsat archive, lots of new
imagery recently added through
LGAC; + high resolution data
Remote sensing advancements:
time series analysis, OBIA,
statistical inference, etc.
ADVANCEMENTS
B E E O D A
What is BEEODA?
▪ BEEODA: Boston Education in Earth Observation Data Analysis
▪ Collection of: existing open-source tools +
our own implementations +
educational materials
▪ Philosophy: “try to use as much existing open source tools as
possible and only implement the missing pieces”
▪ Educational material divided into modules in line with
advancements in RS analyses such as time series analysis, OBIA
and accuracy/area estimation
4
B E E O D A
Technical aspects
▪ Runs as a virtual machine in Windows, OS X, Solaris and Linux
platforms
▪ Oracle VM VirtualBox (freeware) is the only software required
▪ Can also be run from a USB stick with no additional software
needed
▪ In addition to homemade tools, VM includes QGIS, GDAL,
Orfeo, R, Python, Git in a Linux Ubuntu operating system
5
B E E O D A BEEODA 1/12/2016
6
B E E O D A
BEEODA modules
7
B E E O D A
8
B E E O D A
9
B E E O D A BEEODA
10
1/12/2016
B E E O D A BEEODA
11
1/12/2016
B E E O D A
Time series analysis module
12
B E E O D A
13
B E E O D A
▪ One of numerous different approaches for multi-temporal change
detection using Landsat data
▪ Algorithm uses all spectral bands to predict future observations and
detect changes in observed data
▪ Model coefficients for all bands can be used for land cover
classification
▪ Land cover maps ‘synthetic’ images can be created for any date,
and land change maps for any time interval in the study period
14
CCDC/YATSM
B E E O D A
Time series analysis
▪ CCDC implemented
Used by
▪ USGS: Land Cover Monitoring, Assessment, and Projection
(LCMAP; CCDC globally operational)
▪ IDEAM, Colombia: testing in national forest monitoring system
(SilvaCarbon Research and NASA CMS 2016)
▪ Various research projects (NASA Terra/Aqua, NASA CMS,
NASA LCLUC, NASA IDS, USGS Landsat Science Team)
15
B E E O D A
16
Land Change Monitoring, Assessment, and Projection (LCMAP)
B E E O D A
17
B E E O D A
18
Estimated loss of secondary forest
Estimated primary forest Estimated conversion of primary forest to pasture
Estimated conversion of primary
forest to secondary forest
B E E O D A
19
B E E O D A
20
Outputs:
Predicted, cloud-free images
(Any date)
Continuous Change Detection and Classification (CCDC)
B E E O D A
21
Continuous Change Detection and Classification (CCDC)
Outputs:
Land-cover classification
(Any date)
B E E O D A
22
Outputs:
Change maps
(Any time interval)
Continuous Change Detection and Classification (CCDC)
B E E O D A
Estimation module
23
B E E O D A
Estimation of activity data
▪ IPCC criteria: (i) no bias and (ii) quantify uncertainty to reduce
~ This means that you need to ~
1. Construct unbiased estimators of area (and not count pixels
in maps!)
2. Construct confidence intervals of area estimates
24
B E E O D A
25
B E E O D A
BEEODA support for:
▪ Estimation activity data compliant with IPCC and GFOI MGD
▪ BEEODA aims at providing tools that allow practitioners to make
practical use these documents!
Sampling designs supported:
▪ Systematic and random sampling
▪ Stratified sampling
▪ Two-stage cluster sampling
B E E O D A
BEEODA, estimation approaches:
▪ Analysis by stratified estimation (Cochran, 1953) following
Olofsson et al. (2013)
▪ Analysis of two-stage cluster sample using ratio estimators
Särndal et al. (1992) following a few recent research papers
▪ Analysis of sample when map and stratification differs using
indicator functions following Stehman (2013) (e.g. construction of
annual estimators using one sample)
▪ Implementation underway of model-assisted regression
estimator (Särndal et al., 1992) as used by McRoberts (2011)
B E E O D A
Two-Stage Cluster Sampling
28
B E E O D A
29
B E E O D A
Object based image analysis module
30
B E E O D A
Object based image analysis
▪ Popular mapping used methodology but proprietary software for OBIA
often prohibitively expensive; BEEODA contains open source
solution
▪ Two steps: Image segmentation and object classification
▪ Orfeo Toolbox: Mean-shift clustering and support vector machine
classification, in QGIS
▪ Still a work in progress
31
B E E O D A
32
High-resolution imagery
1.
2.
3.
B E E O D A
33
Landsat
1.
2.
3.
B E E O D A
Additional Modules
▪ Basic image processing
▪ Change detection
▪ Direct classification, not post-classification
▪ Using a global map
34
B E E O D A
Status and future directions
35
B E E O D A
Advantages
▪ Fully open-source, available on GitHub
▪ Developed by researchers -- in line with advancements
▪ Customizable (add your own or other open-source tools)
▪ Dependencies solved (major issue in Windows)
▪ One single software installation (or none if booting from USB)
▪ Runs offline on basically any computer
1/12/2016
36
B E E O D A
Current status
▪ Working on new and existing modules
▪ Used in workshops and university courses and by researchers
▪ Not really funded
▪ Currently maintained by BU researchers and grad students...
37
B E E O D A
BEEODA; teaching
38
• Five three-week hands-on workshops at USGS/BU
• Used proprietary software until recently – not ideal
• USAID/FCMC workshop in Peru on estimation using open-
source but no VM – better but issues
• START/GOFC/NASA workshop at BU Jul/Aug 2015 using
BEEODA – most successful solution to date
• Used in BU remote sensing courses since fall 2015
• Ready for hands-on workshops in countries
-START/GOFC/NASA, Bangkok Oct 2016, 40 people
-MINAM, Lima Peru Oct 2016, 8 people
B E E O D A
Email: [email protected]
Web: http://beeoda.org
GitHub: http://github.com/beeoda
http://github.com/ceholden
http://github.com/parevalo
http://github.com/bullocke
B E E O D A
Example: Using a Global Map
40
B E E O D A
Comments
41
Comments
B E E O D A
Comments from participants (Jul/Aug 2015)
▪ “I learnt processing of data using open source software [...] which makes it easily
transferable to my networks and other developing country researchers due to the
free access [...]” Mercy South Africa
▪ “A crucial advantage of the training was that we performed all these advanced
methods using open-source software [...] By this training, I not only acquired a lot of
data and processing techniques and tools which will be shared at my home
institutions, I can rather be considered as a candidate for any land cover change
detection and estimation analysis. Furthermore, I can contribute to a good
implementation of National REDD program and to the capacity building in my region.”
Maleki, Togo
42
B E E O D A
43
Comments from participants (Jul/Aug 2015) “One important limitation […] I had experienced was that commercial software are often
prohibitively expensive. Often there is a need of more than one software package for
completing the analysis procedures. Further, the installation and operation of those systems
are often operating system dependent and have many issues when running on even on
machines with different versions of windows. Here, […] significant add-ons was the open
source set of tools in the form of Virtual Machine provided to us. This has the potential to
address most of the issues with conventional software systems. This can be installed as an
independent operating systems and has a set of tools ranging from basic spreadsheets
program to advanced image processing. Further, the tools are open source, thus we can use,
adapt as per our requirement to solve diverse data analysis problems. And also all the
tutorials are provided […] with free access to anybody.
[In] my work, I have been volunteering a significant amount of time trying to help
students and other fellow researchers with their data analysis issues. In doing this one
important limitation was the software as for example I could demonstrate the data analysis
process in a machine with a licensed software but they could not try those on their own
machine. This was mainly due to the lack of licensed software and often the operation
system itself or component requirements within the operating systems. Now, I am confident
that access to as well as skills on using this tools will be very beneficial.” Shiva, Nepal
B E E O D A
Comments from participants (Jul/Aug 2015)
▪ “[The] two week data analysis exercise at Boston University is really excellent. I
really liked the neat planning and execution of training in terms of objectives like data
collection, classification and accuracy assessment. [...] The use of open source tools
such as QGIS, Orfeo toolbox, GDAL, Python and other in house developed tools by
Boston University team provided effective and ever improving solution for satellite
data analysis.” Suryakant , India
▪ We know for sure that many participants and their colleagues are currently using
BEEODA and at their home institutions.
▪ In short: very well received. This is also true for the BU students who have are
excited to have the whole system for analysis on their laptops – for completing labs
but also for future use.
44
B E E O D A
▪ QGIS to call GDAL library;
command-line optional
▪ GUI-based mainly but some
functions from command-line
▪ All tools run in a Virtual
Machine
▪ Open-source (CC BY-SA 4.0,
MIT and GNU GPL)
▪ Runs in Windows, OS X,
Linux
BEEODA
▪ “GDAL command-line
utilities”
▪ “Collection of prototype
command-line utilities”
▪ “Stand-alone programs and
scripts”
▪ Open-source (“GNU GPLv3
license”)
▪ Runs in Windows (CygWin),
OS X, Linux
45
Quotes from UN-FAO (2013). Open Foris Geospatial Toolkit User Manual, Version 1.25.4
UN-FAO OFGT