Quanti-litative Revolution in GIS

Pre-NextLab

Licence in Maths, ULg

Licence in Computer Sciences, ULg

Geospatial specialization, ULg

Many years in GIS (Ionic Software and its mutations)

Always in data management and semantic

New technologies enthusiast

NextLab age

Project in GIS for the SPW

Project in IoT for Virdata

Wrote a book about Play! 2

Open data evangelist (OKFN, ...), Big Data speaker

Distributed Computing + Distributed Machine Learning expert

Wajug co-founder | Devoxx4Kids organizer

And the list goes on... and on...

@Noootsab

@NextLab_be

Qualitative GIS

Coined in 1963

Raster

vulgus: Big Fat MultiBand Images

• SPOT (see later)
• IKONOS (greek: image)
• LANDSAT
• ALOS
• QuickBird: resolution 64cm → 2m44cm

Vector

vulgus... nah, vector

• Feature extraction
• Landmeters (rare)
• Field surveying
• ~GPS (GNS)

hence, Producers were...

Govs!

thus, Users were

Govs!​

Oh yeah, researchers as well... huh!? wait!

Okay, very feeeew privates

in consequence, Analyses were

• Performed by experimented folks...
• ... or nerds

finally, Tools were...

You don't even know them!

ESRI, Oracle Spatial,
OGC, RedSpider (yeah), ...

Qualitative GIS

"Started" around 2000

Actually, not really before 2010

GIS Revolution

Induce theories from data

The river has flooded on 3km

The river will flood if it rains more than...

New producers

Google, 4², OSM, GPS

old: Mappy, Michelin, ...

New Users

Devs, Lambda

Billioooooons

Analyses

Routing, Accurate GWR¹, ...

Disaster prediction, flood prevention, crash probability, ...

¹Geospatial Weight Regression

Tools?

User friendly and/or Open Source

OSM

GeoServer

Google Maps

OpenLayers

Quantilitative GIS

Coined by @Noootsab^^

WTF?!

Questions

Data

Hypotheses

Rasters today (f.i. Spot7)

• Red Green Blue + Near Infra Red Bands
• One shot: 60x60km (3.600 sq km)
• Takes 3.600.000 sq km of geodata per DAY
• Resolution 2 satellites at 1.5m and 2 others at 50cm
• Revolution: 110 minutes
• 26 days to complete the geoid (all pieces of crap covered)
• 1 single f*****g file for a 60x60km tile is worth up to 

Vector (mostly position)

• Twitter, Facebook, ...
• Foursquare, Instagram
• Waze
• Google (in its whole)
• Connected Devices

everywhere and everytime

Presidente

Model-Driven

Deductive

Top-Down

Quantitative

Lagged Time

Commandante

Data-Driven

Inductive

Bottom-Up

Qualitative

Real Time

GIS Putsch

Marcelino

Data Lake

Machine Learning

Variety

Value

Velocity

VOLUME? It was there for ages!

Who, What, How

You thought I was joking, huh!

Socrata

Evan Chan using Spark

• Customers have streaming point data (many millions of rows)
• PostGIS: point-in-polygon and other does not scale

Soluce:

• Partitioning point data in Cassandra (tiling, Z-curve)
• Partitions into Spark for quick analysis
• Adding spatial indexes to Spark for speeding up

Azavea

Rob Emanuele on Geotrellis

• Run distributed model on 200 TB of climate data against daily temperature and precipitation data out to 2009
• Create suitability maps over high-res raster layers spanning the entire continental US for Urban Forestry Modeling.

Soluce:

• GeoTrellis is providing Spark with geospatial capabilities.
•  Ingest, mosaic and pyramid raster data into Accumulo or HDFS for fast (sub-500ms) tile fetching

Snips

Rand Hindi in his Labs

• Tranquilien: Predicting seating availability in public transport
• RiskContext: Determining the risk of bicycle and car accident from context

Soluce:

• Using blazing fast technologies that can scale linearly (Akka, Scala)
• Thanks to the bottom up reactivity of the architecture, the clusters keeps crunching data in a resilient manner