The ECMWF forecast visualiza on challenge: winning idea

TheECMWFforecastvisualiza6onchallenge:winningidea
AneeshSubramanian
DaveMacLeod
HannahChristensen
StephanJuricke
14June2017@UsingECMWF’sForecastsmee<ng2017
The challenge
•  ECMWFdevelopedtheEnsemble(ENS)Meteogramtocommunicateprobabilis<c
forecastinforma<on
•  Themeteogramsummarizestheensembledistribu<onataspecificloca<onforaset
ofparameters
•  AchallengetoimprovethemeteogramwasissuedatUEF2016
The challenge
Threeideasforinspira<on:
1.Innova<onindisplayingtheensembledistribu<onforecast
2.Addingcomponents
3.Designingauserinterfacetoallowmorecustomiza<on
The philosophy: “Progressive disclosure of information”
AplaWormtograduallyexploretheforecastalongdifferentavenuesfollowinga
user’spar<cularinterestwithoutbeingoverwhelmedwithallthedataatonce
The philosophy: “Progressive disclosure of information”
AplaWormtograduallyexploretheforecastalongdifferentavenuesfollowinga
user’spar<cularinterestwithoutbeingoverwhelmedwithallthedataatonce
The implementation
•  Acleanandvisuallyappealingweb-baseduserinterface
•  Abilitytocomparecurrentwithpreviousforecasts
•  Probabili<esofexceedinganadjustablethreshold
•  Informa<ononexpectedforecastskill&decisionsupport
Acleanandvisuallyappealingweb-baseduserinterface
Acleanandvisuallyappealingweb-baseduserinterface
Oxford
Precipitation
✔Climatology
Show previous forecasts
Set threshold and get
forecast probabilities
Decision support
Abilitytocomparecurrentwithpreviousforecasts
Oxford
Precipitation
✔Climatology
Show previous forecasts
Set threshold and get
forecast probabilities
Decision support
Abilitytocomparecurrentwithpreviousforecasts
Oxford
Precipitation
✔Climatology
✔Show previous forecasts
Set threshold and get
forecast probabilities
Decision support
Verification:
0.8
Probabili<esofexceedinganadjustablethreshold
Oxford
Precipitation
✔Climatology
Show previous forecasts
Set threshold and get
forecast probabilities
Decision support
Probabili<esofexceedinganadjustablethreshold
Oxford
Precipitation
✔Climatology
Chance of greater than
9mm
<2%
<2%
<2%
15%
<2%
<2%
<2%
<2%
6%
<2%
5%
<2%
3%
5%
3%
Show previous forecasts
threshold and get
✔Set
forecast probabilities
Decision support
Informa<ononexpectedforecastskillanddecisionsupport
Oxford
Precipitation
✔Climatology
Chance of greater than
9mm
<2%
<2%
<2%
15%
<2%
<2%
<2%
<2%
6%
<2%
5%
<2%
3%
5%
3%
Show previous forecasts
threshold and get
✔Set
forecast probabilities
Decision support
Informa<ononexpectedforecastskillanddecisionsupport
Oxford
Precipitation
✔Climatology
Chance of greater than
9mm
<2%
0.4
<2%
<2%
15%
<2%
<2%
20%
10%
0.1
0.4
6%
<2%
5%
<2%
3%
5%
3%
These curves indicate a theoretical economic value of the
forecast across a range of cost/loss ratios, based on past
performance at this lead-time. The economic value of a
forecast is calculated by comparing the expense across
reforecasts based on a simple cost/loss model, compared to
default ‘climatological’ action (that is, always acting or never
acting). Each curve indicates the value of the forecast if action
is taken only when a certain forecast probability is exceeded,
the decision threshold.
0.3
0.2
<2%
The figure on the left shows potential economic value curves.
Potential economic value
30%
0.2
<2%
0.6
Cost/loss ratio
0.8
The figure indicates that the forecast has value at this lead
time across a range of decision thresholds. The maximum
value will be extracted by choosing to only take action when
the forecast probability is above 30%. However, the forecast
will still shows value over time for decision thresholds 10% &
20%, though not for all individual values of cost & loss.
Show previous forecasts
threshold and get
✔Set
forecast probabilities
✔Decision support
Oxford
Precipitation
✔Climatology
Chance of greater than
9mm
<2%
0.4
<2%
<2%
15%
<2%
<2%
Potential economic value
0.3
20%
10%
0.1
0.2
0.4
0.6
Cost/loss ratio
<2%
6%
<2%
5%
<2%
3%
5%
3%
Show previous forecasts
threshold and get
✔Set
forecast probabilities
✔Decision support
30%
0.2
<2%
0.8
Oxford
Precipitation
✔Climatology
Chance of greater than
9mm
<2%
0.4
<2%
<2%
15%
<2%
Potential economic value
30%
<2%
<2%
<2%
6%
<2%
5%
<2%
3%
5%
3%
Show previous forecasts
threshold and get
✔Set
forecast probabilities
✔Decision support
0.3
Summary:“Progressivedisclosureofinforma6on”
20%
0.2
•  Acleanandvisuallyappealingweb-baseduserinterface
10%
•  Abilitytocomparecurrentwithpreviousforecasts
0.1
0.2
0.4
0.6
Cost/loss ratio
•  0.8 Probabili<esofexceedinganadjustablethreshold
•  Informa<ononexpectedforecastskill&decisionsupport
Thanksforlistening–anyques6ons?
Oxford
Precipitation
✔Climatology
Chance of greater than
9mm
<2%
0.4
<2%
<2%
15%
<2%
Potential economic value
30%
<2%
<2%
<2%
6%
<2%
5%
<2%
3%
5%
3%
Show previous forecasts
threshold and get
✔Set
forecast probabilities
✔Decision support
0.3
Summary:“Progressivedisclosureofinforma6on”
20%
0.2
•  Acleanandvisuallyappealingweb-baseduserinterface
10%
•  Abilitytocomparecurrentwithpreviousforecasts
0.1
0.2
0.4
0.6
Cost/loss ratio
•  0.8 Probabili<esofexceedinganadjustablethreshold
•  Informa<ononexpectedforecastskill&decisionsupport