Development Seed

Subscribe to Development Seed feed
Updated: 2 weeks 2 days ago

Alireza joins Development Seed

Wed, 07/16/2014 - 22:00



We welcome engineer Alireza J to Development Seed. Alireza will be helping us to build smarter and faster. He will help us to build more usable and useful APIs, develop robust open source tools for data processing, and improve our approach to infrastructure. Alireza will be making it easier for data to play with other data.

Alireza has a background in mining engineering and journalism. His most recent work is in international development, opening data and advancing transparency in some of the most repressive environments. Alireza is a tenacious problem solver and a smart storyteller.

We are delighted to have Alireza on the team. Follow Alireza on Twitter at @scisco7.

Visualizing Water Cut Backs in Las Vegas

Wed, 07/16/2014 - 17:00



Lake Mead, providing Las Vegas with 90 percent of its water needs, has reached its lowest water levels since its original filling and the city has implemented mandatory water cuts of 4.6 percent per person (from 130 gallons in 2012 to 124 gallons in 2013). Similar to identifying crop vulnerability related to North Korea's drought conditions, vegetation analysis can be used by city managers to evaluate effects of drought and to design, evaluate, and enforce water conservation policies in water limited environments. The map below shows the percent change in vegetation, per pixel, before and after the water cuts. Areas in red have experienced a decrease in vegetation, areas in yellow have remained the same, and areas in green have experienced an increase in vegetation.

Toggle OpenStreetMap labels for city locations. View full screen map.

To create this data we used Landsat imagery and NDVI to isolate and then analyze vegetation cover. We showed this last week on a much larger scale with our North Korea drought post. Both municipal managers and humanitarian relief workers can use vegetation analysis to evaluate effects of drought and to design and enforce water conservation or disaster preparedness policies that may help mitigate or avoid disasters.

Visualizing North Korea's Worst Drought in Decades

Tue, 07/08/2014 - 14:30



The Democratic Peoples Republic of Korea (North Korea) is experiencing its worst drought in over a decade. According to reports, some areas have experienced 70 days without rain as well as the lowest rainfall levels since 1961. As food shortages are already a problem for North Korea, the damage that this drought will likely cause for harvests is cause for serious alarm. Food shortages in the 1990s led to an estimated million deaths.

Toggle OpenStreetMap labels to pinpoint most vulnerable cities.

We used satellite data to measure what areas are most impacted by the drought. We examined MODIS Normalized Difference Vegetation Index (NDVI) data to evaluate current vegetation levels across the country as a proxy for drought. The map shows the difference between current vegetation levels (data comprising NDVI measurements from June 10 through June 25) and a baseline constructed from average vegetation levels, for the same June extent, over the past 5 years (2009-2013). Red areas have less vegetation than normal. Darker red shows significantly less vegetation. Green areas have higher levels of vegetation than average. Blank spots are the result of ground-obscuring clouds in the satellite imagery.

This drought map can be used to identify urban areas most vulnerable to food shortages right now, to plan irrigation and relief efforts, or to identify areas that are susceptible to widespread forest fires.

Afghanistan Runoff Elections from Space

Mon, 07/07/2014 - 19:30



On June 14 Afghans went to the polls to vote for a new President. As they voted, satellites overhead captured images of polling locations. High-resolution satellite images can be helpful in investigating election fraud.

This morning the IEC reported that over 8.1 million people went to the polls on June 14, a historically high turnout. At the same time, the IEC acknowledged that fraud occurred at polling locations around the country. Before releasing preliminary results, the IEC audited the results from 1,930 polling stations that reported 100% ballot submission.

We've pulled satellite imagery that captures the June 14 runoff elections. We mapped one set of imagery showing twelve polling centers at 11:26 AM local time on Election Day near Shindand in Herat Province.

Election day satellite imagery of twelve polling stations near Shindand

Four of these centers include stations that reported 599 or 600 ballots--more than one voter per minute. Processing, inking, and issuing ballots to 600 voters takes all day. One would expect to see lines and a steady flow of people at polling stations that returned the maximum number of ballots. Several of these locations include tents and buildings that will make it difficult to measure the number of people at the polling station. Still one would expect to see a steady flow of cars and foot traffic near polling stations that processed 600 voters. Our first look at the satellite imagery shows less activity than would be expected at polling centers with stations that returned the maximum number of votes.

June 14 satellite imagery of four polling stations that reported 599-600 votes

In comparison, at a building at an nearby air field we can see 38 vehicles and approximately 20 people around the facility.

Auditing the Afghanistan Audits

Thu, 07/03/2014 - 21:30



The Afghanistan's Election Commission (IEC) released a list of the 1,930 polling stations to be audited for potential fraud. The stations in the audit returned 599-600 votes in the presidential runoff, 100% ballot submission since each polling station only has 600 ballots. The challenge for the IEC is to determine which ballot boxes represent legitimate high turnout and which demonstrate evidence of fraud.

The release of the audit list is a strong move by the IEC to demonstrate transparency during the audit process. We analyzed these polling stations to begin to determine the degree of possible fraud. We posted the raw data to github and encourage you to undertake your own analysis.

Where are the audit polling stations

The 1930 polling stations currently under investigation represent 1,157,470 votes for around 15% of total turnout. The locations of audited polling stations currently largely track to areas with high levels of ballot box stuffing in the 2009 Presidential Election. The audit list also covers most of the polling stations in Paktya where we saw evidence of fraudulent results in the first round of presidential election.

Locations of polling centers under IEC audit

Using April 5 results to identify deviation

Another way to analyze the audit polling stations is to examine how these stations performed during the first round of elections on April 5th. Of the 1,930 polling stations currently being audited, 1,372 returned results in the first round of elections. The remaining 558 either returned no votes or the results were rejected by the election commission.

Looking at these 1,372 polling stations whose votes were counted in the first round; these polling stations contributed 661,168 total votes in the April 5 election. For the June 14 election, these same polling stations contributed 822,790 votes, an increase of 161,622 votes (24.4%) or an additional 118 votes per polling station. In April these 1,372 polling stations returned 219,652 votes (33.2%) for Abdullah and 235,494 votes (35.6%) for Ashraf Ghani. The polling stations under question previously supported Ghani slightly more than average. Nationally Abdullah received 45% of the vote to Ghani's 31%.

Some polling stations in this group legitimately had 600 voters in both elections. We looked for potential fraud by isolating the polling stations that saw a substantial increase in votes between April 5 and June 14. Of the 1,372 polling stations retuning results in both elections, 829 also returned 599 or more votes in both elections. 422 polling stations saw a jump of of 100 votes or more. One polling center at Molla Joma Gul Mosque in Sar Rawza, Paktika reported a total of 31 votes between both polling stations during April 5 elections. Just one of these polling station returned 600 votes in June 14 elections.

Results fingerprinting

Let's look deeper at how audit polling stations performed during April 5 elections.

This scatterplot shows the total turnout (y-axis) of the 1372 polling station against the number of votes for Abdullah (x-axis) at those stations:

Across the top are all of the polling stations that reported around 600 in April, and whose results were accepted. Abdullah's performance at these stations was evenly distributed. He did poorly at some and well at others.

In the more concerning polling stations with significantly more votes, no notable pattern emerges. Abdullah preformed relatively poorly in polling stations with a significant jump in results.

This scatterplot shows the total turnout (y-axis) of the 1,372 polling stations against the number of votes for Ghani (x-axis):

Again we see a large number of polling stations that reported 600 votes in April. Again there is a relatively even distribution. Ghani performs well in some, poorly in others.

However, a look at the concerning polling stations that saw a significant jump between rounds shows that most of these polling stations voted almost entirely for Ghani in April.

Remember these are all polling stations that reported 599-600 ballots in the most recent round of elections. Many of these stations reported fewer than 100 votes in the previous round. The Ghani campaing asserts that they did a better job of campaigning and getting out the vote through local religious leaders. Independent analysts suggest that there was a surge in voter turnout in Paktia, Khost, and Paktika. Still, it is remarkable that so many small Ghani-leaning polling stations increased their turnout full capacity.

Polling stations with less than 100 votes in April and 600 votes in June are largely in Khost, Paktika, and Wardak.

This analysis is possible because the IEC is opening data about the process. We look forward to seeing other analysis of this data. We will complete an extensive audit once the full results are published.

OpenStreetMap, Disaster Preparedness, and Growing Cities

Tue, 07/01/2014 - 16:30



OpenStreetMap (OSM) is a powerful tool for disaster preparedness planning. OSM is most helpful where the data is complete (includes all buildings, road networks, and other points of interest) and detailed (building heights, materials, structural information). Good OSM data can be used by tools such as InaSAFE to generate better policy and operational insight.

We looked at the OSM data for five of the fastest growing cities and evaluated what data they currently have available on OSM, and what gaps they should address to best identify their risks. Quickly growing cities are interesting for a few reasons:

  • Fast population growth often exceeds the ability of local authorities to maintain critical infrastructure, building codes, and smart resource allocation. This leads to at-risk structures and vulnerable populations.
  • Large future populations mean large exposure to potential hazards.
  • The fastest growing cities are currently small to mid-sized. They have received relatively little attention from the international mapping community and provide a fantastic opportunity to start building a community around mapping and planning.
Kabul, Afghanistan

Kabul, Afghanistan on OSM as of June 25, 2014.

The road framework in Kabul is well developed but building mapping is largely limited to non-residential buildings. Furthermore, of the 771 buildings in the selected area, only 44 have additional tags to indicate what the building is used for. This makes it difficult to distinguish a school from a hospital from a Mosque. We've worked extensively with polling station locations in Afghanistan, many of which are schools, mosques, and clinics. We are working with the OSM community to import this data to improve points of interest across Afghanistan. However this will not address the significant gap in residential buildings and construction data.

Kabul isn't at particularly high risk of most natural disasters. There are occasional earthquakes and flash flooding though, for which is critical to have accurate population and building information for preparedness, response, and evacuation.

Surat, India

Surat, India on the banks of the Tapti River.

Similar to Kabul, Surat's OSM map shows good road network mapping in the downtown area but very few buildings--there are only 40 buildings in all of Surat. There is also no data on building materials, levels, or structure.

Surat is in a precarious position in regards to flooding. In 1994, there was a high profile medical emergency when flooding led to an outbreak of a pneumonic plague epidemic. An effective health care response helped mitigate the damage but better flood planning can certainly help avoid these types of problems in the future.

Sana'a, Yemen

Sana'a, Yemen. Hosptials shown in red.

Like Kabul and Surat, the road network in Sana'a is quite good with very few buildings mapped. However, there are notably 47 hospitals and clinics mapped in OSM.

Ghaziabad, India

Ghaziabad, India just east of New Delhi.

At first glance, Ghaziabad appears to have a good amount of road and building coverage. Upon further investigation, it turns out that the real center of the city is towards the upper right of this map and has almost no mapping done.

Use the slider to toggle between the OSM data and satellite imagery.

As a neighboring city of Delhi, it's not surprising that the outskirts are well mapped but it is surprising to see how little is mapped in the central and eastern portions of the city. It would be a boon to disaster risk planning if the areas around the nearby Hindon river were mapped in case of persistent rains and flooding.

Beihai, China

Beihai, China on the north shore of the Gulf of Tonkin.

Beihai is the least well-mapped of these five cities. This is not surprising given legal restrictions on OSM in China. The Surveying and Mapping Law prohibits any citizen mapping or surveying in mainland China. Restrictive mapping policies in places such as China and in Pakistan will severely hamper the ability of local communities to smartly plan their growth and prepare for disasters.

Currently the population within the urban area is under one million but the city is predicted to experience the fastest growth of any urban area by 2020. The city is relatively low-lying, coastal, and experiences a significant monsoon season so planning ahead for floods and other associated hazards is increasingly important as the population grows.

Coordinating around OSM

The Open Data for Resilience Initiative (OpenDRI) is one group working to coordinate activities using OSM and open data "to reduce the impact of disasters by empowering decisions-makers with better information and the tools to support their decisions." The field guide released this spring provides a wealth of knowledge on this topic, particularly in regards to building inspection and tagging OSM data. As a part of the Understanding Risk Forum in London this week, OpenDRI will be holding a workshop on Friday on learning the tools used in OpenDRI projects. Check out the OpenDRI Field Guide for additional information.

In the long run, the success of projects like OpenDRI will depend on efforts to improve the OSM footprint in the most vulnerable places.

OpenStreetMap for disaster risk management

Mon, 06/30/2014 - 16:30



The 2014 Understanding Risk Forum kicks off today in London. One thing that I'll discuss while I'm here is using open data to make better decisions, specifically looking at how we are using OpenStreetMap (OSM). Disaster preparedness begins with access to information about population and infrastructure. This starts with open data and the ability to use timely and relevant data for disaster risk assessments and preparedness activities.

OpenStreetMap and disasters

OSM is an open, freely available global dataset of geographic and infrastructure data. We use OSM heavily in our work as do other disaster response organizations, ranging from the World Bank to university research centers and NGOs. Humanitarian OpenStreetMap Team's (HOT) is perhaps the most visible example of using OSM to provide timely data to disaster response efforts. The World Bank's InaSAFE hazard modeling tool uses infrastructure data mapped in OpenStreetMap to run preparedness analysis and impact modeling. The WorldPop project, a high-resolution human population distribution mapping project, uses OSM to increase accuracy of their population distribution data around the world. Using road and building data from OSM, WorldPop can produce higher resolution density mapping datasets for public use.

These groups all use OSM because it is the best data available. And through their efforts working with with OSM they continue to improve it. From hazard and exposure mapping, risk modeling and impact analysis, education and training, to disaster preparedness, open data sets are vital to understanding risk.

Precision and accuracy of population distribution models before and after using OSM. Photo: WorldPop

I'll also be discussing satellites, drones, data visualization and other essential tools for understanding risk. OpenStreetMap is a part of the critical infrastructure behind all these efforts. Find me on Twitter at @nas_smith if to meet up and chat!

Remote sensing conflict in Iraq

Thu, 06/26/2014 - 16:30



Yesterday ISIS captured the Baiji oil refinery. Using Landsat 8's Infrared Bands, we detected fires around the refinery from the first day of fighting. Analyzing satellite imagery to see beyond the visible light spectrum enables a better understanding of the situation on the ground.

Detecting Fires with Landsat-8 Imagery

Landsat-8 is an excellent source to observe events from space. Landsat-8 sensors capture visible light as well as infrared. To analyze the Baiji Fires we used three of Landsat-8's 11 bands: - Band 7 (Shortwave Infrared 2) which senses light at wavelengths between 2.11 and 2.29 micrometers - Band 5 (Near Infrared) which senses between 0.85 and 0.88 micrometers, and - Band 3 (visible, Green light) which senses between 0.53 and 0.59 micrometers.

To visualize the data from these three bands we created a 'false-color composite'. The 753 false-color composite combines data from all three sensors and reprojects the colors to clearly show infrared light.

The 753 false-color composite is useful for a range of applications from forest-fire management and post-fire evaluations; monitoring vegetation characteristics for environmental or agricultural applications; monitoring substrate to understand spatial distributions of soils, sands, and minerals; and the delineation of human landscapes. Here we use it to detect fires around the Baiji oil refinery.

Employing an Open Source Toolchain

To create and publish the false-color composite I used entirely open source tools:

I used GDAL to project the Landsat imagery into Web Mercator spatial reference system (SRS):

for BAND in {7,5,4,3,2}; do gdalwarp -t_srs EPSG:3857 $id"_B"$BAND.TIF RESULTS/$BAND-projected.tif; done

I used ImageMagick to create the false-color composite. Also, color, saturation, gamma, and contrast levels were altered to pull out certain features. Here I combined the 7, 5, and 3 bands of Landsat-8 and adjusted the contrast to highlight the fires:

convert -combine {7,5,3}-projected.tif 753.tif convert -sigmoidal-contrast 50x32% 753.tif 753-corrected_32.tif

Because ImageMagick is not geoaware I created a Tiff World File (.tfw), using libgeotiff. This reconnects the geo information that is lost when you use ImageMagick:

listgeo -tfw 3-projected.tif mv 3-projected.tfw 753-corrected-8bit.tfw

Finally, I used Tilemill to make the final map and pushed the image to Mapbox.

Pinpointing the Damage

On-site fires are a natural part of refineries. Gas flares (or flare stacks) will be visible at the Baiji refinery during normal operation. To better understand what fires were caused by fighting, we used imagery from before the conflict to establish a baseline. The map below compares Baiji imagery under normal operating conditions (May 14; on the left) and imagery from (June 18; the right image) after the first day of fighting.

Left: May 14; Right: June 18.

On the June 18 false-color composite we can easily detect a smaller fire southwest of the main blaze. When we compare to the May 14 image we see that this is a new fire. We quickly flag it for further investigation. Zooming into the satellite basemap shows what appears to be vegetation that is on fire.

Baseline image does not show fire activity in lower left of the facility. A fire is detected in this area on June 18.

Check out Mapbox's tutorial for an overview of processing satellite imagery and the code I wrote to produce the false-color.

Growing the team this summer

Wed, 06/25/2014 - 21:30



Culture is everything to us -- especially as we continue to grow. This summer, Eric Miller, Alex Kappel, and Yuriy Czoli are joining us to expand our team of subject matter experts, designers, developers, and strategists.

Eric hails from Pittsburgh where he graduated from Carnegie Mellon University recently. He has programmed for a wide range of projects from biomedical research to web development. He is diving into our technology stack and helping us build more impactful tools.

Alex adds his expertise to our growing satellite imagery work. He is helping us to improve analysis and build open source workflows with rich experience in environmental sciences and remote sensing.

Yuriy is a GIS analyst with a strong spatial research background from Toronto. His interest in urban-political issues has shaped his ability to offer analysis and solutions to real conflicts with GIS techniques.

ISIS damage to Baiji oil refinery

Wed, 06/25/2014 - 16:30



After 10 days of fighting, ISIS has taken control of the Baiji oil refinery. The Baiji refinery is Iraq's largest, producing a third of Iraq's oil output. On June 18, USGS Landsat-8 captured smoke billowing from the refinery. Using new Landsat-8 infrared bands we are able to see through the smoke and pinpoint the exact location of fires. Enhanced imagery shows that main fire in this image was contained to oil storage containers.

We are waiting for new imagery to assess the extent of any subsequent damage. At present there is no evidence to suggest that processing facilities have been significantly damaged.

Swipe for near-infrared processed image showing fires. Zoom to see images from before fighting.

Open Data to Design Better Elections in Lebanon

Mon, 06/23/2014 - 15:30



With elections scheduled for late this year in Lebanon, lawmakers are considering proposals to reform the election law. Current debate on reform is based on assertions that are politically motivated and not backed up by data. The Lebanese Association for Democratic Elections (LADE) and Lamba Labs launched a data platform to provide open election data to facilitate evidence-based discussions on election reform as well as empower civil society to effectively engage in the process. The platform marks a growing community of election data hackers building tools for better, more fair elections.

Photo Credit: Lebanese Association for Democratic Elections

In Afghanistan and Tunisia, open election data supported better planning and execution of elections -- helping election administrators to make more efficient use of human and financial resources, political campaigns to more effectively engage in voter targeting and GOTV efforts, and civil society to determine where to deploy election observers and place observation findings within a broader context.

In Lebanon, civil society is using open data far in advance of elections to facilitate evidence-based discussions on election reform. In advance of upcoming parliamentary polls, Lebanese lawmakers have been engaging in debate on how to reform the country's legal framework governing elections. One of the key questions being considered is how to design an electoral system that is representative and inclusive, ensuring that the will of the diverse electorate is reflected and respected. Until present, data has not been used to inform discussions on electoral reform. Lebanon has not conducted a census in eight decades and there exists a dearth of credible information on present-day ethnic composition, religious affiliations, or gender breakdown. LADE and Lamba Labs, a local hacker space in Beirut, launched the Lebanese Elections Data Platform to provide data and trends on voter registration and participation by ethnicity, religion and gender.

The tool processes data on voter registration between 2005 and 2014 to present a clear picture of the current population by district, demographic characteristics, and growth trends.

Shifts in overall voter registration numbers, registration trends by district, overall demographics, and demographic trends can be used by lawmakers to engage in data-driven reform efforts, civil society to develop evidence-based advocacy efforts, media to engage in data journalism, and citizens to gain a better understanding of their country.

Women form the majority of the electorate, but less than 2% of Parliament. Visualizations ignite debate and advance understanding of developing policies and platforms that increase participation by women voters.

The Community of Election Hackers is Growing

Around the world a growing community of civic hackers are turning their attention to election data. The site was conceived and built in Lebanon by Lamba Labs and LADE. In Tunisia, the OpenGovTN community, a loose collective of civic activists and hackers, laid the foundation for the Tunisia Election Data platform by collecting and opening election data from the country's first democratic elections in 2011. Mourakiboun, a civil society organization dedicated to ensuring credible and transparent elections, built on that foundation to analyze the data and develop the Tunisia Election Data platform. Lamba Labs, OpenGovTN, and civic hackers all over the world are increasingly playing lead roles in helping to advance democracy by using open data and applying technology to help contribute to better elections.

Development Seed and Democracy International helped by providing strategic guidance and connecting LADE and Mourakiboun with like-minded civic hackers. However, Lamba Labs also received integral support from the OpenGovTN community, which offered ongoing technical support. Given their experience partnering with LADE, and based on what they learned from OpenGovTN, Lamba Labs is now embarking on a new effort: starting an OpenGov community in Lebanon. The birth of the @openleb project community, predicated on an understanding of the important role that civic hackers can play in civic life, is an important and welcome development.

Development Seed is excited to be part of a dynamic and growing community of election hackers in Tunisia, Lebanon and all over the world.