Components

Using the meteorological definition of direction: * i.e. where the wind is coming from * From north is 0, from east is 90

Checks that the value is between the min and max defined by the variable

Move air parcel dry-adiabatically up/down to desired location, from nearest neighbours.

Uses the mean of gridpoints within a neighbourhood.

Spread parameters based on inverse distance

Spread parameters based on inverse distance

Assign parameters based on which parameter pool the location is in

Pools all observations locations together

Create one pooling index for each elevation band.

Create one pooling index for each obs-locations. Uses nearest neighbour.

Store parameters in memory

Select dates/offsets in the past where the forecast issued is similar to today's forecast

Creates an ensemble by selecting past observations corresponding to a similar time of day and date of year

Creates a super ensemble by using members from multiple EPSes

Reducte the size of the ensemble such that it only contains the best members. Based on the mean absolute error of each member.

Moderates ensemble members towards the climatological mean

Uses linear regression between forecast and climatology

Bias-correction based no wind-direction. Bias is computed for different wind direction.

Bias-correction using degree of mass balance

Either add or set each ensemble member to a fixed value

Creates an ensemble with 'K' members, using the Kmeans algorithm Means that have no closest ensemble members are set to missing

Replace full ensemble with a single value created from an ensemble measure

Corrects forecasts by ensuring that the observations and forecasts have the same marginal distribution

Date: 20130923 Author: Thomas Nipen Use the most recent observation to bias-correct forecast NOTE: If multiple correctors are chained, this must be first

Rounds forecast value down or up to some threshold if the forecast is less than what the observations historically are when the forecast is on the threshold. Useful for precipitation ensembles to help discrimiate between occurrance and non-occurance of precip

Sorts the ensemble members in ascending order

Date: 20130305 Author: Dominique Bourdin Bayesian Model Averaging Adaptive parameter estimation based on http://www.dtcenter.org/events/workshops12/nuopc_2012/Presentations/Veenhuis%20-%20Ensemble%20Workshop.pdf NOTE: Assuming individual first moments, and pooled second moments

Binned probability ensemble, where the cumulative probability is given by the fraction of ensemble members below. The probability mass between two consecutive ensemble members is fixed and equals 1/(numMembers+1). Currently, the CDF is set to 0 and 1 below and above the ensemble, respectively.

Probability is the fraction of ensemble members forecasting the boundary. Assumptions: the obs is equally likely to fall between any consecutive ensemble members When one of two consecutive members fall on the boundary, half the mass is placed on the boundary the other half in the continuous part.

Probability is a value independent of the ensemble. It is adjusted based on the fraction of past observations falling on the discrete point.

Probability is fixed to a value.

Discrete probability using logistic regression. Probability is of the form: logit(P) = a + b * m + c * f, where a, b, c are constants, m is the ensemble mean, and f is the fraction of ensemble members lying on the discrete point.

Reduces an ensemble to a deterministic forecast.

Computes the mean of the distribution by sampling several quantiles

Connects each pair of coordinates by a straight line. Interpolation outside the domain of the points gives the value of the nearest end point. In this case the slope is 0. When multiple points have the same x-value but different y-values, the average of the smallest and largest y-value is used when interpolating at this x-value.

Fits a polynomial of specified order using least-squared error. Under development.

Connects each pair of coordinates by a step function. The interpolation point immediately below is used. The slope is 0 at all points, and a fixed large value at the interpolation points. Interpolation outside the domain of the points gives the value of the nearest end point. When multiple points have the same x-value but different y-values, the average of the smallest and largest y-values is used when interpolating at this x-value.

Create a neighbourhood pf locations that are at a similar elevation

Creates a neighbourhood using the nearest location. Curved earth distances are calculated.

Returns the deviation of the enemble mean from the climatological value. Positive value means the ens mean is larger than normal.

Returns the median of the ensemble

Returns a specific moment of the ensemble

Returns the fraction of members equal to some preset threshold

Schemes representing a verification metric for a single forecast/observation pair By default the metric will not be computed if the obs is invalid. Overwrite to change this. By default the metric will not be computed if the fcst is invalid. Overwrite to change this.

Observation minus forecast (deterministic)

Computes the CDF at a specified threshold

Returns number from 2x2 contingency table formed by a set threshold.

Probability integral transform value corresponding to the observation. In other words, the CDF of the forecast at the verifying value.

Gives the absolute value (i.e. turns -5 into +5)