Predictability, forecastability, and observability

Jim Hansenjim.hansen@nrlmry.navy.mil

NRL Probabilistic-prediction Research Office

1

Predictability• The rate (or factor) of divergence of nearby

trajectories (norm dependent) • An intrinsic, yet unknowable, system property• The aim is to learn about the system dynamics

by asymptoting towards the system’s intrinsic predictability.

• Ultimate aim is to attempt to increase forecast skill

• Inherently probabilistic

A spectrum of -abilities

• Predictability– How trajectories of the true system diverge

• Model predictability– How trajectories of a given model diverge

• Forecastability– How a model trajectory diverges from a true

system trajectory

ForecastError

Time

Predictability

Forecastability

Back story• Aerosols are an important parameter for Navy

operations.• Aerosol estimation and prediction is a hard problem

fraught with uncertainties.• The NRL Probabilistic-prediction Research Office aims

to advance the estimation, communication, and use of METOC uncertainty for improved science and improved decision making.

• Got to talking with Jeff and Doug about predictability and uncertainty, but quickly learned that there are extreme observational challenges as well.

Observability• The ability to estimate the system state through

indirect observations– Predictability studies can tell us how close our initial

conditions need to be to the “true” state in order to produce useful forecasts

– Observability speaks to constraining the initial state to the necessary level

• what we need to observe • with what accuracy• with what spatial distribution • with what frequency

Indirect: sparse observationsAverage AOD correlation length scales (km) for a single 12hr forecast

Indirect: correlated variables

COAMPS: Correlation between AOD at a point and 10m dust concentration

Battlespace on Demand

Tier 3 – the Decision Layer• Options / Courses of Action• Quantify Risk• Asset Allocation / Timing

Tier 2 – the Performance Layer

Tier 1 – the (forecast)EnvironmentLayer

Initial and Boundary ConditionsSatellitesSatellites

Fleet DataFleet Data

40nm

100 nm

SSN AreaUPDATEDCONOPs

1800 sq nm

MPA Station #1UPDATEDCONOPs

4200 sq nm

MPA Station #2UPDATEDCONOPs

4000 sq nm

100 nm

28-00N

29-40N

60nm

30 nm

074-45W 072-50W

Province “B”

Province “A”“A” “A”

Updated CONOPs

Cumulative Probability of Detecting Both Threat Subs

0.0%

10.0%

20.0%

30.0%

40.0%

50.0%

60.0%

70.0%

80.0%

90.0%

100.0%

0 5 10 15 20 25 30 35 40

Hour

Cum

ulat

ive

Prob

abili

ty

Updated CONOPsUpdated EnvironmentOriginal CONOPsUpdated EnvironmentOriginal CONOPsHistorical Environment

Opportunities

Battlespace on Demand

Tier 3 – the Decision Layer• Options / Courses of Action• Quantify Risk• Asset Allocation / Timing

Tier 2 – the Performance Layer

Tier 1 – the (forecast)EnvironmentLayer

Initial and Boundary ConditionsSatellitesSatellites

Fleet DataFleet Data

40nm

100 nm

SSN AreaUPDATEDCONOPs

1800 sq nm

MPA Station #1UPDATEDCONOPs

4200 sq nm

MPA Station #2UPDATEDCONOPs

4000 sq nm

100 nm

28-00N

29-40N

60nm

30 nm

074-45W 072-50W

Province “B”

Province “A”“A” “A”

Updated CONOPs

Cumulative Probability of Detecting Both Threat Subs

0.0%

10.0%

20.0%

30.0%

40.0%

50.0%

60.0%

70.0%

80.0%

90.0%

100.0%

0 5 10 15 20 25 30 35 40

Hour

Cum

ulat

ive

Prob

abili

ty

Updated CONOPsUpdated EnvironmentOriginal CONOPsUpdated EnvironmentOriginal CONOPsHistorical Environment

• What about T2 and T3 in the context of civilian aerosols?

• T2 examples• Air quality• Volcanic ash impacts

• T3 examples• Advisories• Policy

Global Aerosol Community?

Field Experiments

Climate Satellite Products

Ground Networks

Operational Satellite Products

Tier 3:Safety of navigation/operationsAir quality mitigation/permittingClimate change mitigation/adaptation

Tier 2:PM2.5/PM10Plume locationsRadiative forcingTier 3:Operational forecasts models Long term re-analysesClimate model predictions

Same as DoD construct: many data providers, models, required performance metrics. Ultimately, need to aid many customers

Questions• Science

– What types of observing systems or networks should be put in place to best advance the science and/or forecast products?

– What are the relevant prediction problems and norms?

• Culture– Does the aerosol community really care about “broader impacts”,

or are the basic science questions motivation enough?

• Strategy– How far should the aerosol community move in the T2/T3

direction? Is throwing your product over the fence good enough, or do you need stronger T2/T3 links to justify and motivate your T0/T1 efforts?