Henrik Madsen DHI - Hydrologidag

Big data og data assimilering i hydrologiskmodellering

Henrik MadsenDHI

Hydrologidag 2019, Odense, 22. oktober 2019

Transforming data into operational decisions

Only 5-10% of data collected in a typical utility are utilised

for actionable information

(Global Water Intelligence, 2016)

© DHI #2

Data is getting bigger

© DHI

And more diverse

© DHI #4

René et al. (2015)

Predictive models

© DHI #5

Use o

f p

hysic

s-b

ase

d kn

ow

led

ge

Use of data

Low

High

HighLow

Physic

s-b

ased

models

Data-based

models

Hybrid physics and

data-based models

Adapted from Karpatne et al. (2017)

Physic

s-b

ased

models

Accu

racy

Use of data

Low

High

HighLow

Adapted from Read and Kumar (2019)

Data assimilationMeasurements

ForcingsModel

predictions

Parameters State

Big Data challenges

• Data sources represent

o Different temporal dynamics

o Different spatial resolution

(supporting scale)

o Different measurement

uncertainties and representation

errors

© DHI

Challenges – SMOS soil moisture

© DHI

Coarse resolution (~44 km) Bias (0.02 – 0.23 m3/m3)

© DHI

Hydraulic Head

Cosmic Ray

Soil Moisture (3 depths)

River Discharge

Ahlergaarde, West Denmark

1055 km2

SMOS soil

moisture

High-resolution soil moisture product based on Sentinel-1

© DHI

10 % 90 %Relative Soil Moisture Content

1 km

10 m

© DHI

© DHI

#12

1. Describe

2. Diagnose

3. Predict

4. Prescribe

Physical System

Sensors

Automation &

control

System Loads Forecast

Model

Optimisation Model

Other data

Process Model(s)

Digital Twin

Thank you

Henrik Madsenhem@dhigroup.com