Crowd Sourcing Open Street Map and the Namibia

Crowd Sourcing, Open. Street. Map and the Namibia Flood Sensor. Web Dan Mandl NASA/Goddard Space Flight Center

Open. Street. Map and Crowd Sourcing • Open. Street. Map – Collaborative project to create free editable map of world – Driving force • Restrictions on use or availability of map info across world • Availability of low cost GPS devices – Free map clients – Free editing tools (Java Open. Steet. Map (JOSM), Potlatch) – Free database software (Post. GRES, Post. GIS and Osmosis) • Crowd Sourcing [from Wikipedia] – Process that involves outsourcing tasks to a distributed group of people. – Process can occur both online and offline. – Different from an ordinary outsourcing since it is a task or problem that is outsourced to an undefined public rather than a specific body. – In this case, use people to gather Global Positioning System (GPS) locations of water and collate information 2

End Goal! • Use historical Moderate Resolution Imaging Spectroradiometer (MODIS), Radarsat, and Earth Observing-1 (EO-1) Water Level Maps to relate Hydrologic Model Streamflow to Spatial Extent of 30% Flooding. 70% Streamflow 90% 70% 30% Riv er time

2010 Initial Experiment to Correlate Remotely Sensed Rain Upsteam with Flood Wave Downstream Flood Dashboard Note blue bars (TRMM data) indicating a surge of rainfall upstream Then a flood wave appears downstream at Rundu river gauge days later (gauge data) Early CREST Model trying to predict flood wave (Green) Riverwatch Model trying to predict flood wave (Orange) Better than CREST based on AMSR-E Zambezi basin consisting of upper, middle and lower catchments 4

Early Look at Inundation Extent Related to River Height Zambia water lines from old database Lower Zambezi catchment Multiyear river gauge measurements Envisat swath Envisat Data March 2009 EO-1 Data March 2009 Radarsat Data March 25, 2009 5

Unosat Flood Time Sequence In March 2009 in Caprivi when Katima Mulilo River Gauge Above 7. 5 M Vision is to generate similar product automatically when floods predicted and pair them with river gauge measurements 6

Vision Based on Preliminary Experiments • Use TRMM or other satellites to estimate rainfall in Angola and correlate satellite detection with resulting river gauge height increase (typically 5 – 10 days later) • Use satellite rainfall forecasts to possibly provide more advanced warning • Correlate historical river gauge heights with resultant flood inundation – Find river gauge level that results in flood event – Use historical satellite imagery to determine where water will go once over river banks 7

Challenges • Accuracy of remotely sensed Tropical Rainfall Measuring Mission (TRMM) rainfall estimates • Calibrating the Coupled Routing and Excess Storage (CREST) distributed hydrological or equivalent model without extensive rain gauge data, rain fall gauge data, Digital Elevation Model (DEM) data and soil maps • Even less accuracy for predicted rainfall estimates • Accurate determination of water locations when remotely sensed by satellites – How to calibrate and validate – How to collect, store and correlate ground GPS data with satellite classified imagery – How to work with questionable areas like heavily grassy areas • Floods do not stop at the grass but satellites often don’t see water within grass and trees 8

Approach 9

Upstream: TRMM Rainfall Forecast 10

Upstream: TRMM Precipitation Accumulation Calculator (PAC) User Defined Shapefiles 11

Planet OSM Baseline Water Database NRT Satellit e Data Proposed Open. Street. Map Cal/Val Process Waterpedia Diagram Version 1 1/7/13 Postgres Database (Joyent) Enhanced Water Reference and Flood Maps Baseline Water Layer Flood Map Processor Radarsat. Geobliki. Com (Joyent) Regional Hydrology SME Server W/ JOSM Automated Time Stamp GPS Stamp for shoreline (Lat, Long, Altitude) User Name/ User Type Water extent and Flood Maps (KML tiles, MB tiles, OSM tiles) publish NRT map National Hydrology Agency Server W/ JOSM Mobile Platform Track Collector Field Data Annotated High water reference Low water reference National Postgres Database Flood Map Edit polygons: publish corrected map • Drought OR • Low Normal OR • High Normal OR • Not flooded: misclassified as flooded • Flooded: misclassification due to deep slope reflection OR • Flooded: misclassification due to dry flat surface OR • Flooded: misclassification due to vegetation cover OR • Flooded: misclassification due to other Enter Notes • Input data source (field measurement/ other: describe) • Land Cover Type • text

Conducted Preliminary Open. Street. Map (OSM) Crowd-Sourced 2 Day Exercise • Traveled to Rundu area – – – 12 people from Namibia Hydrological Services 3 people from NASA 1 person Univ. of Oklahoma 2 boats 1 helicopter GPS, camera with GPS • Measured riverbank with GPS in Rundu area corresponding to Radarsat images overlaid with EO-1 images • Ingested GPS ground points, Radarsat images for Rundu and Divindu area • Conducted exercise to gather data multiple ways, ground, boat and helicopter • Combined in OSM to make preliminary assessment 13

Open. Street. Map Ground Validation Exercise at Rundu With Radarsat Detected Water and GPS Ground Truth 1 -31 -13 14

Open. Street. Map With Radarsat Detected Water and Global Positioning System (GPS) Ground Truth 1 -31 -13 Zoomed with Corresponding Helicopter GPS Photo Validation result on 1 -31 -13: Note that Radarsat water detection off based on GPS and Bing. Need to fix Radarsat processing. Probably projection error. (Also need to mark grassy area in database) 15

Another Example Near Divindu with Radarsat Data Versus Planet. osm Data in Red In this example, Radarsat is more accurate then existing data in Planet. osm 16

Integrated Water Edge Detection Display with Walking GPS Measurements, GPS located photos, Radarsat/EO-1 water edge detections Radarsat Water Edge Detection (yellow polygon) EO-1 Water Edge Detection Team 1 walking bank to collect GPS point s (red X’s) One of 500 GPS photos from helicopter Team 2 walking bank to collect GPS point s (green X’s)

Integrated Water Edge Detection Display with Boat GPS Measurements, GPS located photos, Radarsat/EO-1 water edge detections Radarsat Water Edge Detection (yellow polygon) Boat Team 1 track walking bank to collect GPS point s (purple track) EO-1 Water Edge Detection Boat Team 2 track walking bank to collect GPS points (orange track with numbered waypoints)

GPS photos Overlaid on Boat Track Boat Team 1 track with overlaid GPS photos (green track)

GPS Photo Shows Terrain Type Overlaid on Boat Track (geotagged elephants!) Boat Team 1 track with overlaid GPS photos (green track) with photo overlay showing land terrain

GPS Photo Shows Terrain Type Overlaid on Boat Track and Radarsat/EO-1 Water Detection 3/15/2018

Vision • Maps of existing water and flood waters increase in accuracy and detail over time due to many crowd sourced data sources • Map is edited and reloaded to database • Data is formatted in structured manner so that it can be retrieved by tags (e. g source, high mark, questionable area • Use database to develop more accurate water locations which then can feed into models • Models evolve over time to increase accuracy • Don’t need to gather all data at once, proceed as manpower and resources available 22

Conclusion • Integrating Open. Street. Map, crowd sourcing and various satellite and ground based data provides a mean of organizing flood related data to more easily mine for salient information for decision support • The exercise conducted in January 2013 between Namibia Hydrological Services and NASA was used to familiarize both teams with the process and examine possible future directions • Preliminary results demonstrated that at a minimum, using this approach would provide a measure of how good the data is relative to reality 23