AI for Data-Driven Decisions in Water Management

© 2017 IBM Corporation
AI for Data-Driven Decisions in Water Management
https://sites.google.com/site/aiwatertutorial/
Biplav Srivastava,
IBM Research
Sandeep Singh Sandha,
PhD Student, UCLA (Ex-IBM)
Acknowledgements: Our colleagues at IBM Research and collaborators at various agencies.
Tutorial at AAAI 2017, San Francisco, USA, Feb 04, 2017

© 2017 IBM Corporation2
What to Expect: Tutorial Objectives
§ Background
– The importance of water to life is unique. Humanity faces unprecedented
water challenges.
– AI techniques have high potential to impact management of water resources
– We want common citizens to make better decisions around water, and
through them, motivate government and businesses
§ The aim of the tutorial is to
– Make both early and experienced researchers aware of the water
management area, its issues and opportunities
– Discuss available open water pollution data, both quantitative and qualitative
in nature from a modality of sensors, that researchers can use to help a wide
spectrum of decision makers, e.g., farmers, tourists, environmentalists,
health officials, policy makers and businesses.
– Demonstrate usage of AI techniques to make real world impact that matters
– Help create ecosystem for water informatics innovations

Acknowledgements
All our collaborators & partners over the last 3 years:
§ Government agencies
– India: Center and States of Delhi and UP
§ Academia
– India: IIT Roorkee, IIT BHU, IIT Madras, IIT Kharagpur
– USA: Univ. of Chicago, University of Southern California, UCLA
– Canada: York University
§ Non-profits
– Center for Science and Technology (India)
– 2030 Water Group
– RiverKeeper, USA
§ Startup
– Platypus, USA
§ IBM: Sukanya Randhawa, Supratik Guha, Theodore G van Kessel, Hendrik Hamann, Bharat Kumar, Jaikrishnan Hari, Sachin Gupta, Karthik
Visweswariah, Anupam Saronwala, Rashmi Mittal, Deepak Turaga, Kush Varshney
For discussions, ideas and contributions. Apologies to anyone unintentionally missed.
Material gratefully taken from multiple sources. Apologies if any citation is unintentionally missed.

Outline
§ Motivating Examples
§ Water: Basics and Problem
§ Towards Solution
– Data & Infrastructure
• Data Sources
• Existing Mobile Apps
• BlueWater and GangaWatch
– Experience Working in the Field
– Usage Scenario
• Inspection
• Tourism Impact
§ Discussion
– AI Research Issues
– Water Data Standards
– Call for Action
4

Motivating Examples

Many Views of Water
Buenos Aires, Argentina, July 2015
Chennai, India, April 2015

[India] Ganga – Local Ground Situation @ Varanasi (Assi/ Tulsi Ghats) + Patna
Photos of/ at Assi/ Tulsi Ghat, Varanasi on 25 March 2015 during 1700-1800 Hrs
Assi Ghat post recent cleanup Bathing on Tulsi Ghat
A nullah draining into Ganga
A manual powered boat
Photos at Gandhi Ghat, Patna on 18 March 2015 during 1700-1800 Hrs
Common scene around Indian water bodies

[India] Ganga – Local Ground Situation Near Haridwar
Photos at different spots along Ganga at Haridwar-Rishikesh’s 70Km stretch during Feb-Apr 2016

Decision Example –River Water Pollution
§ Value – To individuals, businesses, government institutions
– Individual Use Example – Can I take a bath? Will it cause me dysentery? Can I
drink this water? Can I use this water for irrigating my kitchen garden? Can I eat the
fish I catch?
– Business Use Examples – How should govt spend money on sewage treatment for
maximum disease reduction? If an industry is being set-up, which site is best among
feasible alternatives with least impact and latest technology?
§ Data – Quantitative as well as qualitative
– Dissolved oxygen,
– pH,
– … 30+ measurable quantities of interest
§ Access –
– Today, little, and that too in water technical jargon
– In pdf documents, website
Key Idea: Can we make insights available when needed and help people make
better decisions?

GangaWatch Video Demo
Video at: https://youtu.be/MbVvVGsZoTo
GangaWatch App on Android store,
https://play.google.com/store/apps/details?id=com.ibm.research.ga
ngawatch
GangaWatch blog - https://www.linkedin.com/pulse/ganga-watch-
app-using-water-data-every-day-decisions-srivastava

Initiative: UK
§ Summary
– A web-based interface
specific to bathing at
major water-bodies
§ Reference
– https://environment.data.gov.uk/bwq/profiles/

Initiative: New South Wales, Australia
§ Summary
– Web and Mobile based tools to visualize water availability in river bodies in
New South Wales
§ Reference
– http://waterinfo.nsw.gov.au/
– http://realtimedata.water.nsw.gov.au/water.stm

Initiative: US Geological Survey
§ Summary
– Collects water data from different locations, makes available online along with
web services to access them programmatically
– Interpretation of goodness for a purpose left to user;; Clear Water Act and
state laws impact interpretation
§ Reference
– https://www2.usgs.gov/water/, https://waterwatch.usgs.gov/wqwatch/
https://www.epa.gov/laws-regulations/summary-clean-water-act

Initiative: Jefferson Project at Lake George
§ Summary
– Joint collaboration between academia and industry, project tries to
collect large-scale water data and do hydrological modeling;; help
promote conservation
– Example of outcome: Zooplankton Rapidly Evolve Tolerance to Road
Salt, https://www.rdmag.com/news/2017/01/zooplankton-rapidly-
evolve-tolerance-road-salt
§ Reference
– https://fundforlakegeorge.org/JeffersonProject

Initiative: RiverKeeper for Hudson River
§ Summary
– Impact on recreation activities due to human waste contamination
§ Reference
– http://www.riverkeeper.org/water-quality/hudson-river/

Initiative: Sewage Processing in Spain
§ Summary
– Optimizing operation of sewage processing regularly rather than
statically / seasonally.
– In city of 2 million people in Spain, an operation optimization system
showed a dramatic 13.5 percent general reduction in the plant’s
electricity consumption, a 14 percent reduction in the amount of
chemicals needed to remove phosphorus from the water, and a 17
percent reduction in sludge production.
§ References
– https://www.ibm.com/blogs/research/2016/08/used-iot-data-optimize-
wastewater-treatment/
– Operational optimization of wastewater treatment plants: a CMDP
based decomposition approach, Zadorojniy, A., Shwartz, A.,
Wasserkrug, S. et al. Ann Oper Res (2016). At:
http://link.springer.com/article/10.1007/s10479-016-2146-z

Analytics: Potential Use Cases
S.
No.
Stakehold
er
Use case Data Analytical
techniques
1 IT Identifying and removing outliers,
data validation
Sensor data Data mining (outlier
detection)
2 Individual Which bathing site to use? Sensor data, ghat data Rule-based decision
support
3 Individual/
Economy
What crops can I grow that will
flourish in available water?
Sensor data, crop data Distributed data
integration, co-relation
4 Institution Determine trends/anomalies in
pollution levels
Sensor data, weather
data
Time series analysis,
anomaly detection
5 Institution Attribute source of pollution at a
location
Sensor data,
demographics, industry
data
Physical modeling,
inversion
6 Institution Sewage treatment strategy and
operational planning
Sensor data,
demographics, STP data
Multi-objective
optimization
7 Institution Promoting wildlife/ dolphins Sensor data, wildlife data Rule-based decision
support
8 Institution Using waterways for commercial
shipping
Sensor data, commercial
shipping routes
Optimization, planning
17

Water Basics and Problem

Water Cycle (aka Hydrological Cycle)
Source: Economist, May 20, 2010

Fresh Water: Supply and Demand
Source: Economist, May 20, 2010
Supply Demand

Water Usage Trivia

Fresh Water Stress: Spatial Distribution
Freshwater stress will affect both developed and
developing nations. Billions will be affected.

Water Challenges Summary
§ Increasing demand due to
– Population
– Changing water-intensive lifestyle
– Industrial growth
§ Shrinking supplies
– Erratic rains due to climate change
– Sewage / effluent increase
§ Poor management
– Below cost, unsustainable, pricing
– Delayed or neglected maintenance
Water is the next flash point for wars

Better Information Flow is Critical for Better Water Flow
“One barrier to better management of water resources is simply
lack of data — where the water is, where it's going, how much is
being used and for what purposes, how much might be saved by
doing things differently. In this way, the water problem is largely
an information problem. The information we can assemble has a
huge bearing on how we cope with a world at peak water.”
Source: Wired Magazine, “Peak Water: Aquifers and Rivers Are Running Dry. How Three Regions Are Coping”, Matthew
Power, April 21st, 2008
The nature of water management must rapidly evolve
From To
Manual Data Collection Automated Sensing
Managing Collaboratively
Intermittent Measurement Real-Time Measurement
Multiple Data Sets Data Integration
“Guesstimation” Tools Modeled Decision Support
Commodity Pricing Value Pricing
Tactical Problem Solving Strategic Risk Management
Managing in Isolation

Decisions Within Water Life Cycle
1
1
1
Water Sources
• How much water is available?
• What is the quality of the water?
• How secure is the water?
• How is the water changing over time?
• How is/should the water be mixed?
• Do we comply with regulations/water rights?
1
2
2
2
Modeling and forecasts
•Feeds to modeling (e.g. SWIM)
• Meteorological models
• Hydrological Models
• Groundwater protection
• Feeds to storm water management
• Alerts to municipalities/first responders
2
2
2
2
2
2
3
3
3
3
3 3
3
3
3
Metering Analytics
• How is the water being used?
• When is the water being used?
• When patterns change dramatically,
what to do?
• When the sum of the parts is less than
the whole.. Where is the water going?
3
Combined Sewer Overflow
• Was this truly an event?
• How large was it?
• What was the composition?
• Can the existing capacity be used better?
• How do I notify downstream entities?
• How do I coordinate upstream agencies?
• How can we backtrack to owners?
4
4
4
4
4
4
4
4
Environmental Analytics
• How is the environment changing?
• How do we notify interested parties as it
changes?
• How does the entire system react to
changes?
• How can we predict changes?
5
5
5
5
5
5
5
5
5
Advanced Water Management Capabilities

Towards Solution: Data & Infrastructure

Water Pollution Sensing
§ Method 1: Sample collection and lab-testing
– Accurate when done well;; rich historical data which is under-utilized;; open data
technologies can make them accessible
– Time-consuming, costly and thus feasible for a few places only at a time
– Only quantitative
– Science: lab tests, sample collection
§ Method 2: Real-time sensing
– Timely, inexpensive
– Some important parameters are NOT feasible but can be inferred (e.g., BOD, FC)
– Only quantitative
– Science: how to deploy sensors and analyze data
§ Method 3: Crowd-sourcing
– Timely, inexpensive
– Only qualitative assessment
– Practical for India with people and mobiles
– Science: Combining qualitative and quantitative data

Quantitative Sensing Scope
Dimension {Yamuna | Hindon| Ganga}
Scenario focus General, Agriculture
Real-time measurement DO, pH, conductivity, turbidity
Lab / samples BOD, COD, FCC
Sensing COTS sensors, Machine
learning, In-lab test
Data ingestion Bluemix cloud, Cloudant
database
Primary
§ Temp
§ ORP
§ D.O
§ EC
§ Turbidity
§ Pressure
§ Nitrate
§ GPS Lat
§ GPS Long
Secondary
• Resistivity
• TDS
• Salinity
• SeaWater Sigma
Sensor
Measures
§ Microcontroller
§ Sensor probes
§ Communicator
(Shields)
§ Data Storage
(Server)

Water Qualitative Data Via Crowdsourcing –
NeerBandhu App
Data at http://nalanda.haifa.il.ibm.com/naturetrack/visualization.php

Creek Watch – Crowd Sourced Water Information Collection
As on 14 Oct 2014

Location: http://creekwatch.researchlabs.ibm.com/call_table.php
~3120 data points in 4 years from around the world
As on 14 Oct 2014

Gaps Filled by Our Approach
§ High spatial and temporal resolution (real-time)
– Current data are at low resolution of few places and limited time
points;; limits usage in applications
– Use floating platform and real-time sensor to collect GPS-enabled
data
– Use location to re-create water body condition
§ New source of data (qualitative;; crowd-sourcing)
§ Fusion of historic and new real-time data on single platform with
safety levels and purpose
§ Future: contextualize quantitative data with qualitative inputs for
data validation and stakeholders buy-in

Demo: GangaWatch (1/2)
Data Covering Ganga Basin
Fine-grained
Geo-tagged
Data from a
Real Time
Run on Yamuna

Demo: GangaWatch (2/2)

Blue Water Architecture: Water Data
Data (Multiple
Sources)
• Stationary Stations
• Labs
• Mobile Stations
• Historical Data
• ….
Applications
Web Service
Authenticated:
• Data Upload
(Excel, CSV)
• Meta Data
Upload
• Auto Meta Data
generation
NoSQL Database:
Cloudant
Stores:
Data (Semi-Structured Data)
Meta Data (Places, Limits)
IBM Cloud : Bluemix
Data Query Support
Spatio-temporal Queries
Meta Data Based Queries
https://bluewater.mybluemix.net/
Rest API
Authenticated
Multiple queries
are supported
1 2
3
4
5
6

BlueWater In-Depth: Web Service
Authenticated Data Upload
https://bluewater.mybluemix.net
Manual Places
Definitions
Limits on Water Quality
Water Quality Data of
Parameters in CSV
Water Quality Data of
Parameters in Excel

BlueWater In-Depth: REST API
License Key Generation:
https://bluewater.mybluemix.net/license.jsp
Add few basic details and get license online.
Supported queries
§Spatial query
§Temporal query
§Mixed query
§Meta data query

API URL: https://bluewater.mybluemix.net/query (All results are returned in json)
Returns the info about data: Parameters, Units, statistics of data
Spatial query
Inputs: Latitude, Longitude, Range and License
Returns: Data in Json
Example:
https://bluewater.mybluemix.net/query?Latitude=28.66501&Longitude=77.2393&Range=10&Lic=XXX
Data Returned:
{"source":"live_sensor_data","lat":28.66505,"lng":77.23923,"unix_timestamp_creation":1485297242,"userid":
"ibmadmin","date":"2015/12/16","time":"12:02:48","datetime":1450267368,"Water_Temperature":15.94,"pH":
7.82, ……}
(Lat,Lng)
R
Haversine

Temporal query
Inputs: Date_1, Date_2 and License
Time_1, Time_2 and License
Example:
https://bluewater.mybluemix.net/query?Date_1=2015/12/16&Date_2=2015/12/18&Lic=XXX
https://bluewater.mybluemix.net/query?Time_1=1450267414&Time_2=1450267426&Lic=XXX
Mixed queries
Examples:
https://bluewater.mybluemix.net/query?Latitude=28.66501&Longitude=77.2393&Range=1000&Time_1=1
450267414&Time_2=1450267426&Lic=XXX
https://bluewater.mybluemix.net/query?Latitude=28.66501&Longitude=77.2393&Range=10&Date_1=2015
/12/16&Date_2=2015/12/16&Lic=XXX
Date = YYYY/MM/DD
Time = Unix Timestamp(UTC)

Meta Data query
Inputs: Places parameters and License
Limits and License
Example:
https://gw-ser1.mybluemix.net/query?Places-By-User=UserID&Place-Type=type&Lic=XXX
https://gw-ser1.mybluemix.net/query?Limits-By-User=UserID&Limit-Type=drinking&Lic=XXX
Rest API Demo (Demo Usage Code in Java)
https://github.com/sandeep-iitr/BlueWater_REST_API_DEMO

BlueWater In-Depth: GangaWatch Using REST API
Rest API
Authenticated
Multiple queries
are supported

Experience Working in the Field
Water-bodies of focus:
(1) Hindon, sub-tributary Yamuna, tributary of Ganga
(2) Yamuna, tributary of Ganga
(3) Ganga

Experience Working in the Field - Hindon
River: Hindon (sub-tributary Yamuna, tributary of Ganga)

Hindon, Near Meerut, India (Sep 2015)
Places on Hindon River
1. Kinoni Village (K)
2. Barnawa (Ba)
3. Budhana Road (Bu)
4. Budhana (Bu-City)
5. Titavi, Muzaffarnagar (T)
6. Saharanpur
1
2
3
4
5

Hindon on the Ground
Kinoni Village (K)
Barnawa (Ba)
1
2
3
4
5

Data Around Hindon
4
6
Historical Qualitative Collected

Location Findings
K Ba Bu Bu-City T
Depth 3 feet 1-2 feet 5-6 feet 1 feet 3 feet
Boat
navigation
100 m 100 m 200 m 10-20 m 50 m
Water flow Medium Low Medium Low Fast
Visible
Trash
Low Low Medium High Low
Distance to
major
highway
Far Near Near Near Near

Insights from Hindon
§ Boat cannot go at many places
– Paddle boat may be more useful than powered boat
– At one stretch, boat can go for 100m max at most places visited
§ Mobile data collection was done and useful with NeerBandhu;; GPRS
signal strong during the whole trip
§ Diseases prevalent
– Humans: Cancer, gastro, infertility, skin diseases
– Animals: infertility, sudden death
• Current water usage: Bathing cattle, irrigating fields, drinking by buffalo
§ Impact of sensed data/ use cases: using river water data for
– Plantation of trees along the river bed
– Distribution of water filtration systems or setup of overhead tanks by
government in villages (Note: current data may itself be useful)
– Spreading awareness about river water usage for vegetables has lead to
change

River: Yamuna (tributary of Ganga)
Experience Working in the Field - Yamuna
Reference:
A multi-sensor process for in-situ monitoring of water pollution in rivers or lakes for high-resolution
quantitative and qualitative water quality data
Sukanya Randhawa, Sandeep S Sandha and Biplav Srivastava,
14th IEEE/IFIP International Conference on Embedded and Ubiquitous Computing (EUC 2016),
August 2016.

Real-Time Sensor Deployment

Day
1
-‐ multiple
anchoring
approaches
for
real-‐time

sensor
on
another
day
(16
Dec)
in
2-‐3
km
stretch

16-‐Dec-‐15
Location
Name Description
Sample
-‐
collected Sample
-‐ testing Sensor
@site
Realtime

(Stretch) Neer
Bandhu
1Point
1
[A] Nigambodh,
in
waterY
Y
(ph,
DO,
Temp,

Turb,
Cond,
BOD,

FCC) Y Y
2Point
2
[B] Y Y
3Point
3
[C] ITO
bridge Y Y Y
4Point
4
[D] Y Y
5Pointe
5
[E] Y
Y
(ph,
DO,
Temp,

Turb,
Cond) Y
6Point
6 Moving
(7-‐8
Kmph) Y
7Point
7 Moving
(10
Kmph) Y Y
8Point
8
Drain Y
Y
(ph,
DO,
Temp,

Turb,
Cond) Y Y
9Point
9 With
Ted
buoy Y

Dec 16
Example Run
16/12/15 13:59:34
16/12/15 13:46:50
• ~12 minute downstream
travel
• 765 data points

Turbidity in
Yamuna
– measured
on
16th Dec,
2015
Data
min:
56.7
Data
max:
127
Gradient:

Default
56
91
127
TurbidityFNU

Day
2
-‐ Covered
~7-‐8
km
one-‐way
on
one
of
the

days(18
Dec)
roughly
covering
33
%
of
the
navigable

stretch
of
Yamuna
in
Delhi
(22
km
one-‐way).
18-‐Dec-‐15
Location
Name Description
Sample
-‐
collected Sample
-‐ testing Sensor
@site
Realtime

(Stretch) Neer
Bandhu
1Point
21
[AA] Nigambodh,
in
waterY
Y
(ph,
DO,
Temp,

Turb,
Cond,
BOD,

FCC) Y Y
2Point
22
[AB] Past
rope
(ISBT) Y Y
3Point
23
[AC] 2nd
rope Y Y
4Point
24
[AD] Drain Y
Y
(ph,
DO,
Temp,

Turb,
Cond) Y Y Y
5Pointe
25
[AE] Drain Y Y
6Point26
[AF] Drain,
gurudwara Y Y
7Point
27
[AG] Wazirabad
bridge Y
Y
(ph,
DO,
Temp,

Turb,
Cond) Y Y Y
8Point
28
[AH]

Majnu
ka
tila,

greenery Y Y
9Point
29
[AI] 1st
rope,
ISBT Y

Dec 18
Example Run
2015/12/18,12:13:45
2015/12/18,12:51:37
• ~38 minute upstream
travel
• 2273 data points

Turbidity in
Yamuna
– measured
on
18th Dec,
2015
Data
min:
50
Data
max:
144.4
Gradient:
Default
50
97
144
TurbidityFNU
56
91
127
TurbidityFNU
Reference:
Turbidity of 16-Dec

Turbidity and
Conductivity
in
Yamuna
– measured
on
18th Dec,
2015
50
97
144
TurbidityFNU
1268
1424
1582
ConductivityµS cm

Lab Samples and Traditional Testing

16/12/2016 18/12/2016
Temp(°C) 15.93 15.34
pH 7.82 7.81
ORP(mV) -182 -86.4
D.O(mg/L) 3.76 3.53
EC (µS/cm) 1604 1279
Turbidity (F.N.U) 84.25 66.9
BOD (mg/L) 46 28.2
Fecal Coliform
(No./100 mL) 430 210
Change
in
parameters
measured
for
two
different
days

SensorLab

NB
Qualitative
Data
http://nalanda.haifa.il.ibm.com/naturetrack/visualization.php

Correlating
RT Sensor and
Crowd Data to Get
Verifiable Data!

Towards Solution: Usage Scenario

Towards Solution:
Usage Scenario – Inspection
Reference:
Protecting the NECTAR of the Ganga River Through Game-Theoretic Factory Inspections,
Benjamin Ford, Matthew Brown, Amulya Yadav, Amandeep Singh, Arunesh Sinha, Biplav
Srivastava, Christopher Kiekintveld, Milind Tambe,
14th International Conference on Practical Applications of Agents and Multi-Agent Systems,
Sevilla, Spain, June 1-3, 2016

Art
of
Possible
Tannery
Example:
Kanpur,
India

Background
of
Leather
Tanning
Problem
• > 700 tanneries in Kanpur
– Employing > 100,000 people
– Bringing > USD 1B revenue
• Discharge water after leather processing to river or Sewage treatment
plants (STPs)
– Requirement
• Must have their own treatment facility
• Or, have at least chrome recovery unit
– But don’t implement due to costs which is a burden to main operations
• Installation
• Operations : electricity, manpower, technology upgrade, …
– State pollution board is supposed to do inspections to enforce but doesn’t
perform effectively
• Government’s STPs do not process chrome, the main pollutant
• Knee-jerk reaction: 98 tanneries banned in Feb 2016 by National
Green Tribunal;; more threatened

Flow
Chart
of
Tanning
Source:

home.iitk.ac.in/~sgupta/tannery_re
port.pdf

Water Pollutant Standards for Tanning
Pollutant standards to maintain depends on whether effluent discharged directly to river or to
STP in-drain
•To river directly
– Chrome: < 1 mg/l
– Sulphide: < 2 mg/l
– Suspended solids: < 100 mg/l
– Ph: 6.5 – 9
– COD: < 250 mg/l
•To STP (after which, STP will process)
– Suspended solids: < 600 mg/l
– Ph: 6.5 – 9
• Current state
– Chrome: 45 mg / l
– Other heavy metals are also at alarming level: arsenic, mercury, nickel

Consequences of Contaminated Water in Kanpur
§ Poor public health in Kanpur and downstream
§ Affects poorest the most and government eventually
– Rural people in India (70%) spend at least Rs.100 each year for the
treatment of water/sanitation-related diseases. Which is approximately
same as Central Govt’s Health budget (Rs 6700 crore;; $US 1.1B);;
doesn’t factor the costs to urban India (30%)
– Similar implications likely for Kanpur area too
§ Stresses future water supply: lowering ground water table
§ Slows future industrial growth and thus economy

NECTAR: Nirikshana for Enforcing Compliance for Toxic wastewater
Abatement and Reduction
.
Setting
• Attackers
• M sites with N factory units each
• When inspection at a site happens, all units
know
• Defenders
• Inspectors base office is fixed
• Inspection team consists of
• Environment Inspectors
• Security personnel
• Transport provider / drivers
• Inspection team starts and ends at their
office
• Security and transport can vary daily
• Objective
• Create daily inspection plan which
minimizes violation over a time period
Joint work with USC, USA

Demo: NECTAR
http://teamcore.usc.edu/people/benjamin/ganga/gangaDemonstration.htm
"Very promising approach. Use of decoys and data-driven random were not
known in the inspection community where it was known that random could
help. Surprise elements of decoys and variable fines provide new
factors for compliance. The data from drone monitoring can help
improve the plans significantly as future work."
Dr. Venkatraman Rajagopalan, IAS
Ex-Secretary, Ministry of Environment, Forests and Climate Change, and
Ex-Chairman, Central Pollution Control Board, India

AI in the Scenario
§ Inspection generation a green security game - Stackelberg
Security Game
§ Solve as MDP to generate inspection plan and schedule
§ Explain plan and what-if questions using simulation and rules
§ Advanced
– Utility-driven planning - Use pollution data to update inspection
objective and replan
– Mechanism design - Determine optimal fine strategy

Towards Solution:
Usage Scenario – Tourism Impact
Reference:
An Open, Multi-Sensor, Dataset of Water Pollution of Ganga Basin and its Application
to Understand Impact of Large Religious Gathering,
B Srivastava, S Sandha, V Raychoudhury, S Randhawa, V Kapoor, A Agrawal
arXiv preprint arXiv:1612.05626

Use-Case: Understand Impact of a Large-Scale
Religious cum Tourism Event
§ Haridwar Ardh Khumbh Mela 2016
– January 1, 2016 to April 30, 2016
– Over 100 millions attended;; Many took dip in
river at select spots
– Major bath sub-events during the period
have high burst of visitors
§ Question
– How much does human activity impact river?
– Where is the impact highest? Of what kind?
Sources:
1. https://en.wikipedia.org/wiki/Kumbh_Mela
2. http://www.kumbhamela.net/kumbha-mela-haridwar.html
2. http://www.thegreatananda.com/ardh-kumbh-mela-2016-haridwar/
Date (2016) Day
Main Bathing
Event (Snan)
14th January Thursday Makar Sankranti
12th February Friday Vasant Panchami
22nd February Monday Magh Purnima
7th March Monday Mahashivratri
7th April Thursday Chaitra Amavasya
8th April Friday
Chaitra Shukla
Pratipada
14th April Thursday Mesha Sankranti
15th April Friday Ram Navami
22nd April Friday
Chaitra shukla
Purnima

Ardh Kumbh 2016, Haridwar
Territorial Bird’s Eye View: ~76 KM (Road
Distance)

Turbidity Variations Feb 27-28, 2016
Turbidity values at different places (places marked red have turbidity value above the drinking
range, places marked blues ha turbidity value in range of drinking water)

Data Collection Points around Har-ki-pauri, Haridwar
Feb 27-28, 2016
Carrying sensor on a buoy for long stretch was not possible due to water speed.
45+ places from Rishikesh to Ganga Canal (Roorkee) (75+ KM)

Pollution on Major Bath Day around Har-ki-pauri, Haridwar
March 7, 2016
Turbidity values at different places (places marked red have turbidity value above the drinking range, places
marked blues ha turbidity value in range of drinking water)

AI in the Scenario
§ Data cleaning, normalization, missing values
§ Quantitative to qualitative data conversation
§ Water Quality Index
§ Aggregate Binary Clustering using parameters with opposite polarities
(e.g., Dissolved Oxygen, Turbidity), interval functions (e.g., pH)
§ Advanced
– Generating consolidated qualitative assessment across multiple
parameters
– Explaining and validating assessment

Discussion

Recap: Deliver Value From Water (Pollution) Data
§ Government for business decisions
– Source attribution, inspection
– Sewage treatment
– Public Health
§ Individuals for personal decisions
– Bathing (Religious, Lifestyle)
– Recreation
– Community practices

AI Research Issues
§ Sensing
– Deciding sensors to use (multi-objective optimization)
– How to sense cost-effectively? (Quantitative sensing)
• Install sensors
• Ensure sensor up-keep, inspections
• Decide sampling rate for sensors
– How to involve people-as-sensors? (Qualitative sensing)
• Use people as inspectors (increase resources for defense)
• Mobilization when needed on short notice
• Devising incentives for contribution

AI Research Issues
§ Interconnection
– Within water: quantitative and qualitative estimates;; relation between
fresh and sewage water
– Across domains: energy implications on water management, physical
safety, waste water treatment
§ Analytics
– Decision-support (optimization, planning, scheduling) for organizing
large-scale human activities
– Optimizing short-term and long-term investments (capex/ opex) for
maximizing overall-value from invested water assets
– Pricing to incentivize water conservation and behavioral change

Water Data Standards
§ A dis-organized collection of regulations varying by purpose (e.g., drinking,
agriculture) and regions (country, state).
§ Conflicts may occur;; some data may have privacy considerations (e.g., flow
volume);; standards often mixed with testing methods
§ WHO: water quality standards (from health perspective). At:
http://www.who.int/water_sanitation_health/publications/whoiwa/en/
§ US: Environment Protection Agency’s water quality standards
– https://www.epa.gov/wqs-tech
– http://water.epa.gov/scitech/swguidance/standards/
– https://www.epa.gov/wqs-tech/state-specific-water-quality-standards-effective-under-
clean-water-act-cwa
§ Europe: http://ec.europa.eu/environment/water/index_en.htm
§ India: Central Pollution Control Board, State Pollution Boards
– http://www.cpcb.nic.in/Water_Quality_Criteria.php
– CPCB guidelines for real-time data, http://www.cpcb.nic.in/FinalGuidelinse.pdf
– CPCB's list of pollutants, http://cpcb.nic.in/list_of_parameters.pdf

Call for Action
§ Use water data from BlueWater and find new insights
– Use APIs
– Build apps and/ or reuse GangaWatch app code
§ Collect and contribute your own data
– APIs exist to upload;; registration need
§ Focus on a water use-cases and look at how you can formulate
a basic problem;; solve them
– Fishing
– Water-borne public health
– …

Conclusion
§ We highlighted the importance of water and gave a snapshot of
potential for water informatics
§ Presented the BlueWater platform and GangaWatch app
§ Shared field-experience collecting and using data
§ Demonstrated use-cases of providing decision-support using AI
techniques in water context
§ Tutorial can serve as a resource for others to contribute

References / Resources
§ Peter Gleick et al, The World's Water, Volume 8, The Biennial Report on Freshwater
Resources, 2014
§ IEEE Spectrum Special Report: Water vs Energy (June 2010),
http://spectrum.ieee.org/static/special-report-water-vs-energy
§ Economist Special Report (May 22, 2010), For Want of a Drink: Special Report on Water
§ World Bank
– InvestIng In Water Infrastructure:Capital, Operations and Maintenance, Diego J. Rodriguez, Caroline van den Berg and Amanda
McMahon, 2012, At: http://water.worldbank.org/sites/water.worldbank.org/files/publication/water-investing-water-infrastructure-capital-
operations-maintenance.pdf
– High and Dry: Climate Change, Water, and the Economy, http://www.worldbank.org/en/topic/water/publication/high-and-dry-climate-
change-water-and-the-economy
– World Bank India Ground Water Report, Mar 2010
§ Binayak Ray: Water – The Looming Crisis in India, 2008, ISBN-13: 978-0739126028
§ IBM
– Smarter Water Management Thought Leadership , http://www.ibm.com/smarterplanet/water
– Smarter Water Management Solutions Home Page, http://www.ibm.com/green/water
– IBM GIO Report on Oceans and Water, http://www.ibm.com/ibm/gio/water.html
– IBM Water Management Pains Summary Report, http://www-935.ibm.com/services/us/gbs/bus/pdf/ibm-water-pains-report-jan09.pdf

Blue Water Resources
Apps / tools
§ GangaWatch app for water pollution information on
Google playstore. See description here in the blog.(Jan
2016
§ Neer Bandhu (Water Friend) app on Google playstore.
See description here in the blog.(Nov 2015)
Data
§ From Blue Water/ GangaWatch, REST APIs are at:
http://researcher.watson.ibm.com/researcher/view_group
_subpage.php?id=7142 .
§ From Neer Bandhu (Water Friend), also available from
the mobile app. Find at:
http://nalanda.haifa.il.ibm.com/naturetrack/visualization.p
hp
§ From Creek Watch, also available from web site. Find
http://creekwatch.researchlabs.ibm.com/call_table.php
Blogs
§ Ganga Watch and Blue Water Revisited – Collected
Data is Now Externally Available, As-is,Aug 2016.
§ Ganga Watch – An App for Using Water Data in Every
Day Decisions, Jan 2016.
§ Two Days on Yamuna – Collecting Water Pollution Data
is Both Simple and Complex, Dec 2015.
§ Neer Bandhu, for Water Friends, Nov 2015.
§ Can Game Theory Help Clean the Ganga? Randomized
Inspection is a Case in Point, Benjamin Ford, Biplav
Srivastava*, Milind Tambe, Sep 2015. At data.gov.in's
envivonmentcommunity.
Papers
§ An Open, Multi-Sensor, Dataset of Water Pollution of Ganga Basin and its
Application to Understand Impact of Large Religious Gathering,
Biplav Srivastava, Sandeep Sandha, Vaskar Raychoudhury, Sukanya
Randhawa, Viral Kapoor, Anmol Agrawal
On Arxiv at: http://arxiv.org/abs/1612.05626, December, 2016.
§ The GangaWatch Mobile App to Enable Usage of Water Data in Every Day
Decisions Integrating Historical and Real-time Sensing Data,
Sandeep Sandha, Biplav Srivastava, Sukanya Randhawa,
Demo paper on Arxiv at: http://arxiv.org/abs/1701.08212, January 2017.
§ A multi-sensor process for in-situ monitoring of water pollution in rivers or
lakes for high-resolution quantitative and qualitative water quality data
Sukanya Randhawa, Sandeep S Sandha and Biplav Srivastava,
14th IEEE/IFIP International Conference on Embedded and Ubiquitous
Computing (EUC 2016), August 2016.
§ The GangaWatch App and BlueWater Platform to Enable Usage of Water
Data in India in Everyday Decisions Integrating Historical and Real-time
Sensing Data,
Sandeep S. Sandha, Sukanya Randhawa, and Biplav Srivastava,
Data Flow: Grand Challenges in Water Systems Modeling, Data
Management, and Integration, Louisiana State University, Baton Rouge,
May 9-10, 2016.
§ Protecting the Nectar of the Ganga River through Game-Theoretic Factory
Inspections,
B. Ford, A. Yadav, A. Singh, M. Brown, A. Sinha, B. Srivastava, C.
Kiekintveld, M. Tambe,
14th International Conference on Practical Applications of Agents and Multi-
Agent Systems, Sevilla, Spain, June 1-3, 2016.
§ Blue Water: A Common Platform to Put Water Quality Data in India to
Productive Use by Integrating Historical and Real-time Sensing Data,
Sandeep S Sandha, Sukanya Randhawa and Biplav Srivastava,
IBM Research Report RI15002, 2015.

AI for Data-Driven Decisions in Water Management

Recommended

Recommended

More Related Content

What's hot

What's hot (20)

Similar to AI for Data-Driven Decisions in Water Management

Similar to AI for Data-Driven Decisions in Water Management (20)

More from Biplav Srivastava

More from Biplav Srivastava (18)

Recently uploaded

Recently uploaded (20)