LiveE! Project weather sensor network Seiichi X. Kato (Hyogo University of Health Sciences) 1.

LiveE! Projectweather sensor network

Seiichi X. Kato(Hyogo University of Health Sciences)

1

Weather hazard information• Global Scale

– Global Warming – Large scale hurricane

• Local Scale– Urban heat island – Urban squalls– Flood

It is important to observe the weather information in detail in

order to predict these phenomena

Squalls

Flood Hurricane

2

Scale of meteorological phenomena

• Meteorological phenomena occurs in a variety of time/ spatial scale

time scale

spatial scale

Tornado

Heat Island

Typhoon

Low pressure

minute hour day week

100m

1km

10km

100km

1000kmWarm/cold

front

meso/micro-scale

Synoptic-scale

Squalls

Our target

3

Grass-roots Weather Observing System

• Background– Low-cost weather sensors are marketed– Broadband network is in widespread use

• Some companies and individuals have weather station.

• It will be possible to observe the weather information in detail if these weather station are connected each other by internet.

4

Live E! Project

• Founded by WIDE project & some industries in Japan (2005)– WIDE project is a research consortium on the internet

technology among industry and academia

• We’ll establish the platform to share all the digital weather information and devices by individuals and organizations in order to recognize the environment of the Earth.

• If you have some weather sensors and are interested in this project, please contact me.

5

Live E! sensor map(May. 2009)

http://www.map-asp.net/Spatial_Gateway/pl/Gate_100.html 6

TokyoKurashiki

～ 230 sensor sites

What kind of sensors we use.

• Weather sensors that can read ...– Temperature– Humidity– Pressure– RainFall– WindDir– WindSpeed

• Cost– US$200 ～ 3000

Vaisala WXT510 WM918 WMR968

VantagePRO2 One-Wire Weather Station

7

Live E! server architecture

Java VM

SOAP Web Service (Axis) Database(PostgreSQL)

Live E! Service

8

Live E! global system

.

jp. th.

wide.jp. hoge.jp. hoge2.jp. ku.th. ait.th.

• DNS like architecture• Control information for this system (profile, schema, query

etc.) are exchanged by link

Delegation of sub-authoritiesMetadata of the sensor dataare replicated in all server.

9

Live E! service architecture

LinkManager

DataManager

Resolver & Retriever

Archive

Profile

Schema

Sensor Data Upload

ProfileManagement

User

Link to Other Sites

Live E! service10

Current Web service API• Get Observation data

– GetCurrentDataAll – GetCurrentData – GetDataByTimespan – GetCurrentDataOf – GetCurrentDataByType – GetCurrentDataByAreaRect

• Ger detail profile of each sensor– GetProfileAll – GetProfile – GetProfileByType – GetProfileByAreaRect

11

Live E! data (xml)<SensorSet> <Sensor> <Profile> <sensor_id></sensor_id> <sensor_vendor></sensor_vendor> <sensor_type></sensor_type> <longitude></longitude> <latitude></latitude> <location></location> <address></address> </Profile> <Data> <value time=“”>10.2</value> </Data> </Sensor> <Sensor> 　　　…</SensorSet>

• Profile– Sensor_Id– Sensor_type

• Temperature, Humidity etc.

– Location information• Longitude,latitude etc.

• Data– Observation value

12

<< Live E! Application >>Provisions for natural disaster

• Kurashiki-city, Okayama, JAPAN– Rainfall has a locality; i.e., many sensors are

needed to correctly monitor the area.– About 30 sensors on schools– Weather sensor mesh

3km × 3km

– The local government uses these sensor data for flood prediction.

http://live-e.naist.jp/map/ 13

75km

Kurashiki

Web service -> Overlay network

• Current system– Server-client model– Single point of failure– Load of server will be enormous if the number of

sensor become enormous.• Next system

– P2P(We use PIAX developed by Osaka Univ.)– Load distribution system– Realtime alarm for disaster

14

AR Model for forecast

model oforder :m time,:n noise, white:

ueweight val:a series, time:y

• AR model is one the model of time series analysis, and can forecast future value by validating it from the past data

n

m

iinin vyay

1

Example of AR Model Validation results

AIC minimization AIC ( Akaike’s Information Criterion) isone of the index that selects the bestorder of the model, and the minimumAIC model is the best model.

PARCOR Method PARCOR means Partical Autocorrelation Cooefficient,and the following expression consists in PARCOR and model’s AR parameter.

From this expression, if PARCOR 1,..,m isobtained, all AR parameters can be calculated. 　

)parameters of2(number )likelihood log(2 AIC

1,,111

miaaaa mim

mm

mi

mi 　 　

PARCOR:a parameter, AR:a model oforder :,time: iimi 　

● observed values

○ forecasted values

15

Example of our application for weather forecast

16

Historical data

PredictionObs. data

Prediction by AR(auto-regressive) model

Prediction by AR model is independent of the data of the

other points.

Test application: contour map

Temperature

Barometric pressure

humidity

Contour map on Google Map

17

Current problem

• The number of sensor is small– The accuracy of our interpolation is incredible

• Not suitable for long-term forecast

• We’d like to combine the satellite data with our local sensor data.– Check the accuracy of interpolation and the value

of each sensor(discovery of failure)– Get the information of valuable phenomena

18

Future work

• Collaborate with GeoGRID– Now we are implementing web service in order to

convert Live E! xml data to SensorML by OGC (Open Geospatial Consortium).

– To use the satellite data in order to check the error data in Live E! data and predict more accurately.

19