Outline of Presentation

An Introduction toAn Introduction toThe Localized Aviation MOS The Localized Aviation MOS

Program (LAMP)Program (LAMP)

National Weather ServiceNational Weather ServiceMeteorological Development LaboratoryMeteorological Development Laboratory

Mesoscale Prediction BranchMesoscale Prediction Branch

David E. Rudack, March 2008David E. Rudack, March 2008

Outline of Presentation

• What is LAMP? What is LAMP?

• What weather elements does LAMP forecast?What weather elements does LAMP forecast?

• What steps are taken to create LAMP forecasts? What steps are taken to create LAMP forecasts? • Data inputs Data inputs • Predictand and predictor typesPredictand and predictor types• Generating LAMP regression equationsGenerating LAMP regression equations• Post-processing forecastsPost-processing forecasts

• LAMP Verification and ProductsLAMP Verification and Products

• SummarySummary

Click the hyperlink headings to navigate to the desired portion of the presentation or

click to continue

Basic Properties of LAMP

• LAMP provides hourly updates of Global Forecast System (GFS) MOS LAMP provides hourly updates of Global Forecast System (GFS) MOS forecasts from 1 - 25 hours.forecasts from 1 - 25 hours.

• LAMP updates MOS by utilizing the latest observational data (METAR, LAMP updates MOS by utilizing the latest observational data (METAR, lightning, radar), GFS MOS forecasts, output from simple advective models, lightning, radar), GFS MOS forecasts, output from simple advective models, and geo-climatic data (hi-res topography and relative frequencies).and geo-climatic data (hi-res topography and relative frequencies).

• Many of the LAMP weather elements are important to aviation operations.Many of the LAMP weather elements are important to aviation operations.

• LAMP provides guidance for the contiguous United States (CONUS) and LAMP provides guidance for the contiguous United States (CONUS) and Alaska, Hawaii, and Puerto Rico (OCONUS), except the thunderstorm Alaska, Hawaii, and Puerto Rico (OCONUS), except the thunderstorm guidance is limited to the CONUS.guidance is limited to the CONUS.

• Forecasts are issued for METAR stations, except gridded thunderstorm Forecasts are issued for METAR stations, except gridded thunderstorm forecasts are issued on a 20-km grid.forecasts are issued on a 20-km grid.

Stations for which LAMP generates forecasts

Geographical Coverage of Thunderstorm Forecasts on 20-km Grid

Attributes of the MOS Approach

• The MOS approach is followed in developing LAMP guidance.

• This approach:

• relates observations of the weather element to be predicted relates observations of the weather element to be predicted ((predictandspredictands) to appropriate variables () to appropriate variables (predictorspredictors) through ) through multiple linear regressionmultiple linear regression

• is mathematically straightforward • yields forecasts which are accurate and/or skillful • generates non-probabilistic (e.g., temperature) and probabilistic generates non-probabilistic (e.g., temperature) and probabilistic

forecasts (e.g., probability of precipitation)forecasts (e.g., probability of precipitation)• improves upon direct model output forecasts and generates improves upon direct model output forecasts and generates

forecasts of elements not directly output from dynamical models forecasts of elements not directly output from dynamical models (e.g., ceiling and visibility)(e.g., ceiling and visibility)

Conceptual Approach of LAMP

• The objective of LAMP is to improve on MOS The objective of LAMP is to improve on MOS forecasts and persistence out to 25 hours forecasts and persistence out to 25 hours through rapid infusion of current observational through rapid infusion of current observational data.data.

• The predictive information in observational data The predictive information in observational data is extrapolated through the application of is extrapolated through the application of advective models.advective models.

Theoretical Model Forecast Performance of LAMP, MOS, and Persistence

0 6 12 18 24

Projection (hr)

Skill

LAMP

MOS

Persistence

LAMP outperforms persistence for all projections and handily outperforms

MOS in the 1-12 hour projections.

The skill level of LAMP forecasts begin to converge to the MOS skill level after the 12 hour projection

and become almost indistinguishable by the 20 hour

projection.

The decreased predictive value of the observations at the later projections causes the LAMP

skill level to diminish and converge to the skill level of

MOS forecasts.

LAMP Updates of GFS MOS

• LAMP produces hourly updates for LAMP produces hourly updates for the most recent GFS MOS cycle.the most recent GFS MOS cycle.

2200 UTC LAMP

1800 UTC MOS

2300 UTC LAMP

0000 UTC LAMP

0100 UTC LAMP

0200 UTC LAMP

0300 UTC LAMP

0400 UTC LAMP

0000 UTC MOS

0500 UTC LAMP

0600 UTC LAMP

0700 UTC LAMP

0800 UTC LAMP

0900 UTC LAMP

1000 UTC LAMP

0600 UTC MOS

1100 UTC LAMP

1200 UTC LAMP

1300 UTC LAMP

1400 UTC LAMP

1500 UTC LAMP

1600 UTC LAMP

1200 UTC MOS

1700 UTC LAMP

1800 UTC LAMP

1900 UTC LAMP

2000 UTC LAMP

2100 UTC LAMP

The following LAMP cycles

update the 1800 UTC GFS MOS

cycle



cycle



cycle



cycle

LAMP updates the GFS MOS

• 2-m temperature and dewpoint (F)2-m temperature and dewpoint (F)

• Wind speed (kts), wind gusts (kts), and Wind speed (kts), wind gusts (kts), and wind direction (degrees)wind direction (degrees)

LAMP Forecasts of Continuous Weather Elements

• Categorical forecast (yes/no) indicating if precipitation, Categorical forecast (yes/no) indicating if precipitation, not necessarily measurable, will occur on the hournot necessarily measurable, will occur on the hour

• Categorical forecast of precipitation type (liquid, freezing, Categorical forecast of precipitation type (liquid, freezing, or frozen) conditioned on precipitation occurring or frozen) conditioned on precipitation occurring

• Categorical forecast of precipitation characteristics Categorical forecast of precipitation characteristics (drizzle, continuous, or showers) conditioned on (drizzle, continuous, or showers) conditioned on precipitation occurringprecipitation occurring

LAMP Forecasts of Categorical Weather Elements

• Total sky cover (clear, few, scattered, broken, or overcast)Total sky cover (clear, few, scattered, broken, or overcast)

• Ceiling heightCeiling height

• Conditional ceiling height (if it is precipitating)Conditional ceiling height (if it is precipitating)

• VisibilityVisibility

• Conditional visibility (if it is precipitating)Conditional visibility (if it is precipitating)

• Obstruction to vision (no obstruction, haze, mist, fog, or blowing Obstruction to vision (no obstruction, haze, mist, fog, or blowing phenomena)phenomena)

• Occurrence/non-occurrence of thunderstorm Occurrence/non-occurrence of thunderstorm • Thunderstorm is defined as one or more cloud-to-ground (CTG) lighting strikes in Thunderstorm is defined as one or more cloud-to-ground (CTG) lighting strikes in

a 20-km grid box in a 2-h period (CONUS only)a 20-km grid box in a 2-h period (CONUS only)

Elements of special interest to aviation forecasters

LAMP Categorical Forecasts

LAMP Probabilistic Forecasts

Probability of:Probability of: EventEvent

Liquid Equivalent Precip. ≥ 0.01 inch during past 6 hours/12 hours

Yes/No

Precipitation occurring on the hour (not necessarily measurable) *

Yes/No

Precipitation type

(Conditional on Precipitation)

Freezing

Frozen

Liquid

Precipitation Characteristics

(Conditional on Precipitation)

Drizzle

Continuous

Showers

*This is not the same as precipitation occurring during the hour.


Ceiling Height

< 200 feet

200 – 400 feet

500 – 900 feet

1000 – 1900 feet

2000 – 3000 feet

3100 – 6500 feet

6600 – 12,000 feet

> 12,000 feet

Conditional Ceiling Height (Conditional on Precipitation)

Same as above




Visibility

< ½ mile

< 1 mile

< 2 miles

< 3 miles

≤ 5 miles

≤ 6 miles

Conditional Visibility (Conditional on Precipitation)

Same as above

Thunderstorms within a 20-km grid box and within a 2-h period

Occurrence/non-occurrence of one or more CTG lightning strikes



Temporal Resolution of LAMP Forecasts • LAMP weather elements have an hourly temporal LAMP weather elements have an hourly temporal

forecast resolution except for probability of measurable forecast resolution except for probability of measurable precipitation in a 6 and 12 hour period (PoP6/PoP12) precipitation in a 6 and 12 hour period (PoP6/PoP12) and thunderstorms.and thunderstorms.

• LAMP PoP6 forecasts have a temporal resolution of 6 LAMP PoP6 forecasts have a temporal resolution of 6 hours and are valid over the 6 hour forecast periods hours and are valid over the 6 hour forecast periods ending at 0000, 0600, 1200, and 1800 UTC.ending at 0000, 0600, 1200, and 1800 UTC.

• LAMP PoP12 forecasts have a temporal resolution of 6 LAMP PoP12 forecasts have a temporal resolution of 6 hours and are valid over the 12 hour forecast periods hours and are valid over the 12 hour forecast periods ending at 0000, 0600, 1200, and 1800 UTC.ending at 0000, 0600, 1200, and 1800 UTC.

• LAMP 2-h thunderstorm forecasts are issued at hourly LAMP 2-h thunderstorm forecasts are issued at hourly intervals to 7 or 8 hours (depending on the cycle time) intervals to 7 or 8 hours (depending on the cycle time) and at 2-h intervals thereafter (see subsequent slide). and at 2-h intervals thereafter (see subsequent slide).

Odd Cycles Example: 09 UTC

Even Cycles Example: 10 UTC

Schematic of LAMP Thunderstorm 2-h Valid Periods

11 12 13 14 15 16 17 18 20 22 00 02 04 06 08 10 11Projection (h)

UTC Time

Start time

1 2 3 4 5 6 7 8 10 12 14 16 18 20 22 24 25

10 UTC

White arrows indicate overlapping 2-h valid periods out to 6 hours from issuance

Green arrows indicate subsequent 2-h valid periods, which end on even UTC hours

10 11 12 13 14 15 16 18 20 22 00 02 04 06 08 10

1 2 3 4 5 6 7 9 11 13 15 17 19 21 23 25

UTC Time

Projection (h)

09 UTCStart time LAMP thunderstorm forecasts are issued every 2 hours after the 7th or 8th projection depending if the cycle issuance time is odd

or even..

LAMP thunderstorm forecasts are issued at an hourly temporal resolution beginning at

the 3-h projection until the 7th or 8th projection depending if the cycle issuance

time is odd or even.

The Development Process of Generating LAMP Forecasts

1)1) Collate the data from a variety of sources for the Collate the data from a variety of sources for the regression analysisregression analysis

2)2) Generate a regression equation for each element at Generate a regression equation for each element at each projection using a specific training period of dataeach projection using a specific training period of data

3)3) Post-process the forecasts to ensure consistency Post-process the forecasts to ensure consistency (e.g., ensure that the temperature is always equal to or (e.g., ensure that the temperature is always equal to or greater than the dewpoint)greater than the dewpoint)

4)4) Verify the weather element at each projection hour Verify the weather element at each projection hour using a station list (typically includes the entire station using a station list (typically includes the entire station list used to produce LAMP forecasts) over an list used to produce LAMP forecasts) over an independent sampleindependent sample

LAMP Predictand Data

• The predictand data (which are observations) are The predictand data (which are observations) are extracted from two possible sources:extracted from two possible sources:

• Hourly METAR observations (All LAMP elements Hourly METAR observations (All LAMP elements except thunderstorms)except thunderstorms)

• Lightning strike data from the National Lightning Lightning strike data from the National Lightning Detection Network (For LAMP thunderstorm Detection Network (For LAMP thunderstorm development only) development only)

• All predictands in the LAMP system are defined at All predictands in the LAMP system are defined at stations except for the probability of thunderstorms which stations except for the probability of thunderstorms which is defined on a 20-km grid.is defined on a 20-km grid.

• The data is quality checked for gross errors as well as The data is quality checked for gross errors as well as for temporal consistency.for temporal consistency.

LAMP Predictand Data

• Predictand data can take two forms:

• Continuous values such as temperature and dewpoint

• Binary values are used to define the predictand in terms of the occurrence (“1”) or non-occurrence (“0”) of an event.

Example of binary value:

Event: Wind speed > 14 kts If yes binary value = “1” If no binary value = “0”

LAMP Predictor Data Sources

• Predictors are data (e.g., temperature) that explain a portion of the Predictors are data (e.g., temperature) that explain a portion of the behavior exhibited by the predictand.behavior exhibited by the predictand.

• Possible predictor sources used in LAMP developments include:Possible predictor sources used in LAMP developments include:• Hourly METAR DataHourly METAR Data• GFS MOS forecastsGFS MOS forecasts• Simple models (such as advection of moisture) Simple models (such as advection of moisture) • Radar mosaic dataRadar mosaic data• Lightning strike data* from the National Lightning Detection Lightning strike data* from the National Lightning Detection

Network Network • GFS model output GFS model output

• Only those predictors that make physical sense are chosen from Only those predictors that make physical sense are chosen from these data sources as predictors. these data sources as predictors.

* * Archives obtained from Global Hydrology Resource Center (GHRC)Archives obtained from Global Hydrology Resource Center (GHRC)

LAMP Predictor Types and Transformations

• For station development, all gridded predictor values are interpolated For station development, all gridded predictor values are interpolated to stations by either:to stations by either:

• Bilinear interpolation for continuous values such as temperature, Bilinear interpolation for continuous values such as temperature, oror

• Nearest neighbor interpolation for discontinuous values such as Nearest neighbor interpolation for discontinuous values such as visibility visibility

• There are four types of predictors that may be used in the regression There are four types of predictors that may be used in the regression analysis:analysis:

• Predictors possessing the actual values of the forecast element Predictors possessing the actual values of the forecast element (e.g., temperature in degrees)(e.g., temperature in degrees)

• GFS MOS probability forecasts GFS MOS probability forecasts • Point binary predictors (Values of “0” or “1”)Point binary predictors (Values of “0” or “1”)• Grid binary predictors (Values range between “0” and “1”)Grid binary predictors (Values range between “0” and “1”)

Multiple Linear Regression Basics

• Multiple linear regression relates a dependent variable Y Multiple linear regression relates a dependent variable Y (predictand) to a set of “n” independent variables (predictand) to a set of “n” independent variables X1,,X2,,…Xn (predictors). The relationship is expressed through a (predictors). The relationship is expressed through a linear equation:linear equation:

Y = b + a1x1 + a2x2 + … + anxn

where the values of awhere the values of a11 through a through ann are the coefficient values are the coefficient values for each of the predictor terms for each of the predictor terms x1 through through xn, and “b” is the , and “b” is the equation constant.equation constant.

• Each term “aEach term “ai i xxii” represents the contribution of the predictor ” represents the contribution of the predictor xxi i to the estimate of the predictand.to the estimate of the predictand.

• The predictors, XThe predictors, X11…X…Xnn, can be nonlinear. That is , can be nonlinear. That is to say, the predictors can be derived from values to say, the predictors can be derived from values that were raised to a power other than one (e.g., that were raised to a power other than one (e.g., X=ZX=Z33).).

• The equation represents the best possible least The equation represents the best possible least squares fit to the data of any other possible squares fit to the data of any other possible linear equation with the same predictors. linear equation with the same predictors.

Multiple Linear Regression Basics

Sample Linear Regressionfor KATL, June 2005

Y = -16.568 + 1.2498*X60

65

70

75

80

85

90

60 65 70 75 80 85 90

0000 UTC Forecast of 2-m Temperature (°F) (Predictor)

Ob

serv

ed T

emp

erat

ure

(°F

) (P

red

icta

nd

)

The sum of squares of the vertical distances between

the line and the data points is minimized.

LAMP Equation Development Technique

• The regression analysis first determines the predictor from the set of The regression analysis first determines the predictor from the set of potential predictors that explains more of the predictand variability potential predictors that explains more of the predictand variability than any other potential predictor. than any other potential predictor.

• Each subsequent predictor is selected based on its ability, in Each subsequent predictor is selected based on its ability, in conjunction with the predictors already selected, to explain more of conjunction with the predictors already selected, to explain more of the remaining predictand variability than any other potential the remaining predictand variability than any other potential predictor.predictor.

• The regression analysis for a particular equation stops when either:The regression analysis for a particular equation stops when either:• the maximum number of allowable terms is reached (pre-defined the maximum number of allowable terms is reached (pre-defined

by the developer – usually 10-15 predictors), orby the developer – usually 10-15 predictors), or• none of the remaining predictors further reduces the remaining none of the remaining predictors further reduces the remaining

variability observed in the predictand by a predefined amount variability observed in the predictand by a predefined amount (also stipulated by the developer)(also stipulated by the developer)

• Separate equations are developed for each season to address the wide range of intra-Separate equations are developed for each season to address the wide range of intra-seasonal weather variabilityseasonal weather variability

• All LAMP weather elements with the exception of precipitation type and thunderstormsAll LAMP weather elements with the exception of precipitation type and thunderstorms Cool season (October 1 – March 31)Cool season (October 1 – March 31) Warm season (April 1 – September 30) Warm season (April 1 – September 30)

• Precipitation TypePrecipitation Type Cool season (September 1 – May 31) Cool season (September 1 – May 31) Warm season (June 1 – August 31) (Alaska only) Warm season (June 1 – August 31) (Alaska only)

• ThunderstormsThunderstorms Spring season (March 16 – June 30)Spring season (March 16 – June 30) Summer season (July 1 – October 15)Summer season (July 1 – October 15) Winter season (October 16 – March 15) Winter season (October 16 – March 15)

• This seasonal stratification “fine-tunes” the equations to the appropriate season’s type of This seasonal stratification “fine-tunes” the equations to the appropriate season’s type of weather. weather.

• For a smoother transition of forecasts between seasons, an additional thirty days of For a smoother transition of forecasts between seasons, an additional thirty days of training data is included in the development sample for each year - fifteen days prior to training data is included in the development sample for each year - fifteen days prior to and subsequent to the development season.and subsequent to the development season.

- Seasonal Stratification -

LAMP Equation Development

• Some elements such as temperature are reported regularly at every station Some elements such as temperature are reported regularly at every station while hazardous events of visibility and/or ceiling height occur less frequently. while hazardous events of visibility and/or ceiling height occur less frequently.

• When a station reports a sufficient number of cases for a particular element, When a station reports a sufficient number of cases for a particular element, (usually no less than 200), a regression equation is generated for that element (usually no less than 200), a regression equation is generated for that element that applies to that specific station. that applies to that specific station.

• To increase equation stability for certain elements that are less common, an To increase equation stability for certain elements that are less common, an equation is developed by pooling other station datum into a region. This equation is developed by pooling other station datum into a region. This regionalizedregionalized regression equation can then be applied to any station in that regression equation can then be applied to any station in that particular region. particular region.

• Regions are determined based upon similar geoclimatic features (e.g., terrain or Regions are determined based upon similar geoclimatic features (e.g., terrain or relative frequencies of the event). relative frequencies of the event).

• Regionalization allows for the LAMP guidance to be produced at sites with poor, Regionalization allows for the LAMP guidance to be produced at sites with poor, unreliable, or non-existent observation systems. unreliable, or non-existent observation systems.

• All LAMP elements with the exceptions of temperature, dewpoint, wind speed, All LAMP elements with the exceptions of temperature, dewpoint, wind speed, and wind direction are developed regionally. and wind direction are developed regionally.


- Station vs. Regionalized Equations -

• The figure to the right is The figure to the right is an example of how the an example of how the CONUS might be CONUS might be subdivided into regions for subdivided into regions for equation development.equation development.

• It is important to find a It is important to find a balance between larger balance between larger region sizes which help to region sizes which help to create more stable create more stable equations and smaller equations and smaller region sizes which help to region sizes which help to better model the local better model the local effects. effects.

• Each element has its Each element has its own set of regions, which own set of regions, which usually differs by season.usually differs by season.


- Station vs. Regionalized Equations (Cont.) -

• Primary equations use observations, MOS forecasts, and other Primary equations use observations, MOS forecasts, and other variables as predictors.variables as predictors.

• Secondary equations do not use observations as predictors. Secondary equations do not use observations as predictors. • They use values interpolated from the LAMP analysis of available They use values interpolated from the LAMP analysis of available

observations, as well as MOS forecasts and other variables. observations, as well as MOS forecasts and other variables. • The analyzed value is used as a “surrogate” observation. The analyzed value is used as a “surrogate” observation. • Secondary equations are used in instances where a station does not Secondary equations are used in instances where a station does not

report an observation.report an observation.

• Primary equation forecasts are generally better than secondary Primary equation forecasts are generally better than secondary equation forecasts because the primary equations contain the equation forecasts because the primary equations contain the station’s actual observation and not its proxy. station’s actual observation and not its proxy.

- Primary vs. Secondary Equations -


• To help ensure better consistency across related To help ensure better consistency across related forecast elements (e.g., temperature and forecast elements (e.g., temperature and dewpoint), it is beneficial to develop related dewpoint), it is beneficial to develop related predictands simultaneously. predictands simultaneously.

• When weather elements are developed When weather elements are developed simultaneously, each element’s equation shares simultaneously, each element’s equation shares the same set of variable predictors but differs in the same set of variable predictors but differs in its coefficient values and the value of its its coefficient values and the value of its equation constant. equation constant.


- Simultaneous Development -

Practical Example of Solving a LAMP Temperature Equation

Y = LAMP temperature forecastY = LAMP temperature forecast Equation Constant b = -6.99456Equation Constant b = -6.99456

Predictor x1 = observed temperature at cycle issuance time (value 66.0)

Predictor x2 = observed dewpoint at cycle issuance time (value 58.0)

Predictor x3 = GFS MOS temperature (value 64.4)

Predictor x4 = GFS MOS dewpoint (value 53.0)

Coefficient values: aCoefficient values: a1 1 = 0.15147, a= 0.15147, a2 2 = -0.041273, a= -0.041273, a3 3 = 0.84864, a= 0.84864, a4 4 = 0.18787= 0.18787

Y = -6.99456 + ( .15147 x 66.0) + (-.041273 x 58.0) + (.84864 x 64.4) + (.18787 x 53.0)

Y = 65.2 F

Y = b + a1x1 + a2x2 + a3x3 + a4x4

Note that the equation contains at least one observation predictor indicating that it is a

primary equation.

Post-Processing LAMP Forecasts

Reasons for post-processing forecasts include:Reasons for post-processing forecasts include:

• Reconciling meteorological inconsistencies between weather Reconciling meteorological inconsistencies between weather elements (e.g., ensuring that the dewpoint temperature remains less elements (e.g., ensuring that the dewpoint temperature remains less than or equal to the forecast temperature at a specific projection)than or equal to the forecast temperature at a specific projection)

• Ensuring that the probabilistic forecasts behave properly (e.g., the Ensuring that the probabilistic forecasts behave properly (e.g., the probabilities range between zero and one inclusively, or that select probabilities range between zero and one inclusively, or that select probability forecasts for certain elements sum to a value of one, probability forecasts for certain elements sum to a value of one, etc.)etc.)

• Generating threshold values for elements that are used in producing Generating threshold values for elements that are used in producing categorical forecasts for specific weather elements such as visibility categorical forecasts for specific weather elements such as visibility and ceiling heightand ceiling height

Generating LAMP Threshold Values for Categorical Forecasts

• Typically, threshold values are generated for each probability category Typically, threshold values are generated for each probability category except for the most common category. except for the most common category.

• For a specific weather element, regional threshold values are generated For a specific weather element, regional threshold values are generated using the same regions that were used to generate the equations for that using the same regions that were used to generate the equations for that element. element.

• All stations in a particular region possess the same threshold value but the All stations in a particular region possess the same threshold value but the values vary by projection.values vary by projection.

• A set of threshold values is developed from the primary forecasts, and A set of threshold values is developed from the primary forecasts, and

another set is developed from the secondary forecasts. another set is developed from the secondary forecasts.

• Primary probability forecasts are compared to the primary threshold values Primary probability forecasts are compared to the primary threshold values when a station reports an observation. when a station reports an observation.

• Secondary probability forecasts are compared to the secondary threshold Secondary probability forecasts are compared to the secondary threshold values when a station does not report an observation. values when a station does not report an observation.

Process of Generating LAMP Categorical Forecasts

• To illustrate how LAMP computes a categorical forecast, consider the 5 categories of sky To illustrate how LAMP computes a categorical forecast, consider the 5 categories of sky conditions: conditions:

(1) Clear(1) Clear (2) Few(2) Few (3) Scattered (3) Scattered (4) Broken (4) Broken (5) Overcast(5) Overcast

• For this element, threshold values would be generated for categories (2)-(5).For this element, threshold values would be generated for categories (2)-(5).

• For a particular station and projection, the probability of the greatest threat event (in this For a particular station and projection, the probability of the greatest threat event (in this case overcast) is compared to the threshold for overcast skies. case overcast) is compared to the threshold for overcast skies.

• If the threshold equals or exceeds the probability of overcast skies, the LAMP categorical If the threshold equals or exceeds the probability of overcast skies, the LAMP categorical forecast is “Overcast.” forecast is “Overcast.”

• Otherwise, the algorithm continues and the same question is asked of the next greatest Otherwise, the algorithm continues and the same question is asked of the next greatest threat event, “Broken.” threat event, “Broken.”

• This process continues for the next two categories, “Scattered” and “Few.” If the This process continues for the next two categories, “Scattered” and “Few.” If the probability forecast for each one of these two categories does not equal or exceed its probability forecast for each one of these two categories does not equal or exceed its respective threshold value, the LAMP categorical forecast is “Clear.” respective threshold value, the LAMP categorical forecast is “Clear.”

0

20

40

60

80

Clear Few Scattered Broken Overcast

Forecast Probability

Threshold Value

LAMP Categorical Forecast Selection Process

Pro

bab

ility

(%

)

Does the forecast probability of

overcast equal or exceed the threshold

for overcast?

The probability of “few” exceeds the threshold value for “few” – LAMP

categorical forecast is “few”

Does the forecast probability of broken equal or exceed the

threshold for broken?

Category 1 Category 4Category 3Category 2 Category 5

Does the forecast probability of

scattered equal or exceed the threshold

for scattered?

Does the forecast probability of few

equal or exceed the threshold for few?

Performance Scores for LAMP Non-Probabilistic Forecasts

• Typical verification scores used for checkout of LAMP Typical verification scores used for checkout of LAMP forecasts:forecasts:

• Continuous weather elements (e.g., temperature and dewpoint) Continuous weather elements (e.g., temperature and dewpoint) are verified by:are verified by:

• mean absolute error (MAE) – lower is bettermean absolute error (MAE) – lower is better

• Categorical forecasts (e.g., visibility category) are verified by:Categorical forecasts (e.g., visibility category) are verified by:• Heidke skill score (HSS) – higher is betterHeidke skill score (HSS) – higher is better• threat score (also called Critical Success Index) – higher is betterthreat score (also called Critical Success Index) – higher is better• bias - values near one are goodbias - values near one are good

Verification of LAMP 2-m Verification of LAMP 2-m Temperature ForecastsTemperature Forecasts

0900 UTC mean absolute error (MAE) for temperature Cool season (October 2003 - March 2004); 1523 stations

0

2

4

6

8

10

12

14

0 3 6 9 12 15 18 21 24Projection (hours)

MA

E (

°F)

Persistence

GFS MOS

LAMP

Verification of LAMP Categorical Verification of LAMP Categorical Ceiling Height ForecastsCeiling Height Forecasts

0900 UTC threat for ceiling height < 1000 feet Cool season (October 2003 - March 2004); 1523 stations

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1


Th

reat

sco

re

Persistence

GFS MOS

LAMP

Verification of LAMP Categorical Verification of LAMP Categorical Visibility ForecastsVisibility Forecasts

0900 UTC threat for visibility < 3 miles Cool season (October 2003 - March 2004); 1523 stations

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1


Th

reat

sco

re

Persistence

GFS MOS

LAMP

Verification of LAMP IFR or Worse Verification of LAMP IFR or Worse ForecastsForecasts

0000 UTC threat for IFR conditions or worseCool season (October 2006 - March 2007); 1462 stations

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1


Th

reat

sco

re

Persistence

GFS MOS

LAMP

Verification of LAMP Categorical Thunderstorm Forecasts

1800 UTC threat for thunderstormsSpring season (April 1997 - June 2005); 27,373 grid points

0

0.05

0.1

0.15

0.2

0.25

0.3


Th

reat

sco

re

Persistence

CMOS*

LAMP

* CMOS stands for calibrated MOS, wherein the GFS MOS is calibrated from a 3-h valid period to a 2-h valid * CMOS stands for calibrated MOS, wherein the GFS MOS is calibrated from a 3-h valid period to a 2-h valid period.period.

Performance Scores for LAMP Probabilistic Forecasts

• Basic measure of accuracy is Brier score (lower is better).Basic measure of accuracy is Brier score (lower is better).

• Measure of skill is the improvement in Brier score over a benchmark Measure of skill is the improvement in Brier score over a benchmark standard, such as climatology. standard, such as climatology.

• The reliability of probability forecasts describes the degree to which The reliability of probability forecasts describes the degree to which the forecast relative frequency of the weather event has an the forecast relative frequency of the weather event has an overforecasting or underforecasting bias. overforecasting or underforecasting bias.

• When the average forecast probability within a bounded range:When the average forecast probability within a bounded range:• exceeds the observed relative frequency exceeds the observed relative frequency overforecasting bias overforecasting bias• equals the observed relative frequency equals the observed relative frequency perfect reliability perfect reliability• is lower than the observed relative frequency is lower than the observed relative frequency underforecasting bias underforecasting bias

LAMP vs. CMOS* Thunderstorm Brier Score Improvement on Climatology

1800 UTC Brier Score Improvement on Climatology for thunderstorms Spring season (April 1997 - June 2005); 27,373 grid points

0

5

10

15

20

25

3 4 5 6 7 8 10 12 14 16 18 20 22 24Projection (hours)

Bri

er S

core

Impr

ovem

ent

(%) CMOS

LAMP

* CMOS stands for calibrated MOS, wherein the GFS MOS is calibrated from a 3-h valid period to a 2-h valid * CMOS stands for calibrated MOS, wherein the GFS MOS is calibrated from a 3-h valid period to a 2-h valid period.period.

Reliability of 0300 UTC 3-h Ceiling < 1000 feet 2006 August - 2007 May, 1522 sites

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%

Forecast

Ob

serv

ed R

elat

ive

Fre

qu

ency

827

2642

4348

5189

4171

3625

3684

5574

14848

56727

145522

165669

Reliability diagram for LAMP3-h ceiling height forecasts of < 1000 feet. Note that the LAMP probability forecasts lie along the diagonal.

0

20000

40000

60000

80000

100000

120000

140000

160000

180000

0% 5% 15%

25%

35%

45%

55%

65%

75%

85%

95%

100%

Number of probability forecasts in each bin

Reliability of 0300 UTC 3-h Visibility < 3 miles 2006 August - 2007 May, 1522 sites

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%

Forecast

Ob

serv

ed R

elat

ive

Fre

qu

ency

127

3501268

2145

3175

3815

3233

7822

15090

48684

161115167357

0

20000

40000

60000

80000

100000

120000

140000

160000

180000

0% 5% 15%

25%

35%

45%

55%

65%

75%

85%

95%

100%

Reliability diagram for LAMP3-h visibility forecasts of < 3 miles. Note that the LAMP probability forecasts lie along the diagonal.


Reliability of 0300 UTC 3-h Thunderstorms 2006 August - September, 2007 April - May, 27373 grid points

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%

Forecast

Ob

serv

ed R

elat

ive

Fre

qu

ency

974

1117

1805

2839

4501

6956

11427

19702

40770

150318

1515938

1418921

0

200000

400000

600000

800000

1000000

1200000

1400000

1600000

1800000

2000000

100%

Reliability diagram for LAMP 3-h thunderstorm forecasts. Note that the LAMP probability forecasts lie along the diagonal except for the highest bins where the number of forecasts are relatively small compared to the total number of forecasts.


Overview of Available LAMP Overview of Available LAMP Forecast ProductsForecast Products

• Sent out on SBN/NOAAPort and available on NWS FTP Server• ASCII text bulletins• BUFR data• GRIB2 thunderstorm data

• AWIPS• Displayable in D2D • Guidance available for display and Terminal Aerodrome Forecast (TAF)

preparation via the Aviation Forecast Preparation System (AvnFPS)

• LAMP Website• Forecast products available:

• ASCII text bulletins• Station plot forecast images• Gridded thunderstorm forecast images• Meteograms

• Products Page: http://www.nws.noaa.gov/mdl/gfslamp/gfslamp.shtml

SummarySummary

• LAMP provides hourly updates of four-daily GFS MOS forecasts to 25 hours for the CONUS and OCONUS.

• LAMP forecasts are in probabilistic and non-probabilistic forms.

• LAMP weather elements include those important to aviation operations, such as surface wind, ceiling height, visibility, and thunderstorms.

• LAMP forecasts improve most on GFS MOS in the early forecast projections and on persistence at all projections.

• Please consult http://www.weather.gov/mdl/lamp/ for further information, inquires, and comprehensive performance scores.

Date post:	30-Dec-2015
Category:	Documents
Upload:	alvin-riley
View:	21 times
Download:	0 times

Outline of Presentation

Documents