Nonhomogeneous Gaussian regression

Last updated

Non-homogeneous Gaussian regression (NGR) [1] [2] is a type of statistical regression analysis used in the atmospheric sciences as a way to convert ensemble forecasts into probabilistic forecasts. [3] Relative to simple linear regression, NGR uses the ensemble spread as an additional predictor, which is used to improve the prediction of uncertainty and allows the predicted uncertainty to vary from case to case. The prediction of uncertainty in NGR is derived from both past forecast errors statistics and the ensemble spread. NGR was originally developed for site-specific medium range temperature forecasting, [1] but has since also been applied to site-specific medium-range wind forecasting [4] and to seasonal forecasts, [5] and has been adapted for precipitation forecasting. [6] The introduction of NGR was the first demonstration that probabilistic forecasts that take account of the varying ensemble spread could achieve better skill scores than forecasts based on standard model output statistics approaches applied to the ensemble mean.

Contents

Intuition

Weather forecasts generated by computer simulations of the atmosphere and ocean typically consist of an ensemble of individual forecasts. Ensembles are used as a way to attempt to capture and quantify the uncertainties in the weather forecasting process, such as uncertainty in the initial conditions and uncertainty in the parameterisations in the model. For point forecasts of normally distributed variables, one can summarize an ensemble forecast with the mean and the standard deviation of the ensemble. The ensemble mean is often a better forecast than any of the individual forecasts, and the ensemble standard deviation may give an indication of the uncertainty in the forecast.

However, direct output from computer simulations of the atmosphere needs calibration before it can be meaningfully compared with observations of weather variables. This calibration process is often known as model output statistics (MOS). The simplest form of such calibration is to correct biases, using a bias correction calculated from past forecast errors. Bias correction can be applied to both individual ensemble members and the ensemble mean. A more complex form of calibration is to use past forecasts and past observations to train a simple linear regression model that maps the ensemble mean onto the observations. In such a model the uncertainty in the prediction is derived purely from the statistical properties of the past forecast errors. However, ensemble forecasts are constructed with the hope that the ensemble spread may contain additional information about the uncertainty, above and beyond the information that can be derived from analysing past performance of the forecast. In particular since the ensemble spread is typically different for each successive forecast, it has been suggested that the ensemble spread may give a basis for predicting different levels of uncertainty in different forecasts, which is difficult to do from past performance-based estimates of uncertainty. Whether the ensemble spread actually contains information about forecast uncertainty, and how much information it contains, depends on many factors such as the forecast system, the forecast variable, the resolution and the lead time of the forecast.

NGR is a way to include information from the ensemble spread in the calibration of a forecast, by predicting future uncertainty as a weighted combination of the uncertainty estimated using past forecast errors, as in MOS, and the uncertainty estimated using the ensemble spread. The weights on the two sources of uncertainty information are calibrated using past forecasts and past observations in an attempt to derive optimal weighting.

Overview

Consider a series of past weather observations over a period of days (or other time interval):

and a corresponding series of past ensemble forecasts, characterized by the sample mean and standard deviation of the ensemble:

.

Also consider a new ensemble forecast from the same system with ensemble mean and ensemble standard deviation , intended as a forecast for an unknown future weather observation .

A straightforward way to calibrate the new ensemble forecast output parameters and produce a calibrated forecast for is to use a simple linear regression model based on the ensemble mean , trained using the past weather observations and past forecasts:

This model has the effect of bias correcting the ensemble mean and adjusting the level of variability of the forecast. It can be applied to the new ensemble forecast to generate a point forecast for using

or to obtain a probabilistic forecast for the distribution of possible values for based on the normal distribution with mean and variance :

The use of regression to calibrate weather forecasts in this way is an example of model output statistics.

However, this simple linear regression model does not use the ensemble standard deviation , and hence misses any information that the ensemble standard deviation may contain about the forecast uncertainty. The NGR model was introduced as a way to potentially improve the prediction of uncertainty in the forecast of by including information extracted from the ensemble standard deviation. It achieves this by generalising the simple linear regression model to either:

[1]

or

[1] [2]

this can then be used to calibrate the new ensemble forecast parameters using either

or

respectively. The prediction uncertainty is now given by two terms: the term is constant in time, while the term varies as the ensemble spread varies.

Parameter estimation

In the scientific literature the four parameters of NGR have been estimated either by maximum likelihood [1] or by maximum continuous ranked probability score (CRPS). [2] The pros and cons of these two approaches have also been discussed. [7]

History

NGR was originally developed in the private sector by scientists at Risk Management Solutions Ltd for the purpose of using information in the ensemble spread for the valuation of weather derivatives. [1]

Terminology

NGR was originally referred to as ‘spread regression’ rather than NGR. [1] Subsequent authors, however, introduced first the alternative names Ensemble Model Output Statistics (EMOS) [2] and then NGR. [8] The original name ‘spread regression’ has now fallen from use, EMOS is used to refer generally to any method used for the calibration of ensembles, and NGR is typically used to refer to the method described in this article. [4] [7]

Related Research Articles

<span class="mw-page-title-main">Least squares</span> Approximation method in statistics

The method of least squares is a parameter estimation method in regression analysis based on minimizing the sum of the squares of the residuals made in the results of each individual equation.

In statistics, the Gauss–Markov theorem states that the ordinary least squares (OLS) estimator has the lowest sampling variance within the class of linear unbiased estimators, if the errors in the linear regression model are uncorrelated, have equal variances and expectation value of zero. The errors do not need to be normal, nor do they need to be independent and identically distributed. The requirement that the estimator be unbiased cannot be dropped, since biased estimators exist with lower variance. See, for example, the James–Stein estimator, ridge regression, or simply any degenerate estimator.

<span class="mw-page-title-main">Beta distribution</span> Probability distribution

In probability theory and statistics, the beta distribution is a family of continuous probability distributions defined on the interval [0, 1] or in terms of two positive parameters, denoted by alpha (α) and beta (β), that appear as exponents of the variable and its complement to 1, respectively, and control the shape of the distribution.

In statistical inference, specifically predictive inference, a prediction interval is an estimate of an interval in which a future observation will fall, with a certain probability, given what has already been observed. Prediction intervals are often used in regression analysis.

In econometrics, the autoregressive conditional heteroskedasticity (ARCH) model is a statistical model for time series data that describes the variance of the current error term or innovation as a function of the actual sizes of the previous time periods' error terms; often the variance is related to the squares of the previous innovations. The ARCH model is appropriate when the error variance in a time series follows an autoregressive (AR) model; if an autoregressive moving average (ARMA) model is assumed for the error variance, the model is a generalized autoregressive conditional heteroskedasticity (GARCH) model.

<span class="mw-page-title-main">Regression analysis</span> Set of statistical processes for estimating the relationships among variables

In statistical modeling, regression analysis is a set of statistical processes for estimating the relationships between a dependent variable and one or more error-free independent variables. The most common form of regression analysis is linear regression, in which one finds the line that most closely fits the data according to a specific mathematical criterion. For example, the method of ordinary least squares computes the unique line that minimizes the sum of squared differences between the true data and that line. For specific mathematical reasons, this allows the researcher to estimate the conditional expectation of the dependent variable when the independent variables take on a given set of values. Less common forms of regression use slightly different procedures to estimate alternative location parameters or estimate the conditional expectation across a broader collection of non-linear models.

Ridge regression is a method of estimating the coefficients of multiple-regression models in scenarios where the independent variables are highly correlated. It has been used in many fields including econometrics, chemistry, and engineering. Also known as Tikhonov regularization, named for Andrey Tikhonov, it is a method of regularization of ill-posed problems. It is particularly useful to mitigate the problem of multicollinearity in linear regression, which commonly occurs in models with large numbers of parameters. In general, the method provides improved efficiency in parameter estimation problems in exchange for a tolerable amount of bias.

<span class="mw-page-title-main">Ordinary least squares</span> Method for estimating the unknown parameters in a linear regression model

In statistics, ordinary least squares (OLS) is a type of linear least squares method for choosing the unknown parameters in a linear regression model by the principle of least squares: minimizing the sum of the squares of the differences between the observed dependent variable in the input dataset and the output of the (linear) function of the independent variable. Some sources consider OLS to be linear regression.

Weighted least squares (WLS), also known as weighted linear regression, is a generalization of ordinary least squares and linear regression in which knowledge of the unequal variance of observations (heteroscedasticity) is incorporated into the regression. WLS is also a specialization of generalized least squares, when all the off-diagonal entries of the covariance matrix of the errors, are null.

In statistics, the residual sum of squares (RSS), also known as the sum of squared residuals (SSR) or the sum of squared estimate of errors (SSE), is the sum of the squares of residuals. It is a measure of the discrepancy between the data and an estimation model, such as a linear regression. A small RSS indicates a tight fit of the model to the data. It is used as an optimality criterion in parameter selection and model selection.

<span class="mw-page-title-main">Simple linear regression</span> Linear regression model with a single explanatory variable

In statistics, simple linear regression (SLR) is a linear regression model with a single explanatory variable. That is, it concerns two-dimensional sample points with one independent variable and one dependent variable and finds a linear function that, as accurately as possible, predicts the dependent variable values as a function of the independent variable. The adjective simple refers to the fact that the outcome variable is related to a single predictor.

In statistics, stochastic volatility models are those in which the variance of a stochastic process is itself randomly distributed. They are used in the field of mathematical finance to evaluate derivative securities, such as options. The name derives from the models' treatment of the underlying security's volatility as a random process, governed by state variables such as the price level of the underlying security, the tendency of volatility to revert to some long-run mean value, and the variance of the volatility process itself, among others.

In statistics, the variance inflation factor (VIF) is the ratio (quotient) of the variance of a parameter estimate when fitting a full model that includes other parameters to the variance of the parameter estimate if the model is fit with only the parameter on its own. The VIF provides an index that measures how much the variance of an estimated regression coefficient is increased because of collinearity.

The term generalized logistic distribution is used as the name for several different families of probability distributions. For example, Johnson et al. list four forms, which are listed below.

<span class="mw-page-title-main">Normal-inverse-gamma distribution</span>

In probability theory and statistics, the normal-inverse-gamma distribution is a four-parameter family of multivariate continuous probability distributions. It is the conjugate prior of a normal distribution with unknown mean and variance.

Demand forecasting, also known as demand planning and sales forecasting (DP&SF), involves the prediction of the quantity of goods and services that will be demanded by consumers or business customers at a future point in time. More specifically, the methods of demand forecasting entail using predictive analytics to estimate customer demand in consideration of key economic conditions. This is an important tool in optimizing business profitability through efficient supply chain management. Demand forecasting methods are divided into two major categories, qualitative and quantitative methods:

Experimental uncertainty analysis is a technique that analyses a derived quantity, based on the uncertainties in the experimentally measured quantities that are used in some form of mathematical relationship ("model") to calculate that derived quantity. The model used to convert the measurements into the derived quantity is usually based on fundamental principles of a science or engineering discipline.

<span class="mw-page-title-main">Errors-in-variables model</span> Regression models accounting for possible errors in independent variables

In statistics, an errors-in-variables model or a measurement error model is a regression model that accounts for measurement errors in the independent variables. In contrast, standard regression models assume that those regressors have been measured exactly, or observed without error; as such, those models account only for errors in the dependent variables, or responses.

The generalized functional linear model (GFLM) is an extension of the generalized linear model (GLM) that allows one to regress univariate responses of various types on functional predictors, which are mostly random trajectories generated by a square-integrable stochastic processes. Similarly to GLM, a link function relates the expected value of the response variable to a linear predictor, which in case of GFLM is obtained by forming the scalar product of the random predictor function with a smooth parameter function . Functional Linear Regression, Functional Poisson Regression and Functional Binomial Regression, with the important Functional Logistic Regression included, are special cases of GFLM. Applications of GFLM include classification and discrimination of stochastic processes and functional data.

In statistics, linear regression is a model that estimates the linear relationship between a scalar response and one or more explanatory variables. A model with exactly one explanatory variable is a simple linear regression; a model with two or more explanatory variables is a multiple linear regression. This term is distinct from multivariate linear regression, which predicts multiple correlated dependent variables rather than a single dependent variable.

References

  1. 1 2 3 4 5 6 7 Jewson, S.; Brix, A.; Ziehmann, C. (2004). "A new parametric model for the assessment and calibration of medium‐range ensemble temperature forecasts". Atmospheric Science Letters. 5 (5): 96–102. arXiv: physics/0308057 . doi: 10.1002/asl.69 .
  2. 1 2 3 4 Gneiting, T.; Raftery, A.; Westveld, A.; Goldman, T. (2005). "Calibrated Probabilistic Forecasting Using Ensemble Model Output Statistics and Minimum CRPS Estimation". Monthly Weather Review. 133 (5): 1098. doi: 10.1175/MWR2904.1 .
  3. "Calibrating and Combining Ensemble Predictions" (PDF). NOAA Earth System Research Laboratory. U.S.A.: National Oceanic and Atmospheric Administration.
  4. 1 2 Thorarinsdottir, T.; Johnson, M. (2012). "Probabilistic Wind Gust Forecasting Using Nonhomogeneous Gaussian Regression". Monthly Weather Review. 140 (3): 889–897. doi: 10.1175/MWR-D-11-00075.1 .
  5. Lalic, B.; Firany Sremac, A.; Dekic, L.; Eitzinger, J. (2017). "Seasonal forecasting of green water components and crop yields of winter wheat in Serbia and Austria". The Journal of Agricultural Science. 156 (5): 645–657. doi: 10.1017/S0021859617000788 . PMC   6199547 . PMID   30369628.
  6. Scheuerer, M. (2013). "Probabilistic quantitative precipitation forecasting using Ensemble Model Output Statistics". Quarterly Journal of the Royal Meteorological Society. 140 (680): 1086–1096. arXiv: 1302.0893 . doi:10.1002/qj.2183. S2CID   88512854.
  7. 1 2 Gebetsberger, M.; Messner, J.; Mayr, G.; Zeileis, A. (2018). "Estimation Methods for Nonhomogeneous Regression Models: Minimum Continuous Ranked Probability Score versus Maximum Likelihood". Monthly Weather Review. 146 (12): 4323–4338. doi: 10.1175/MWR-D-17-0364.1 .
  8. Wilks, D. S. (2006-08-22). "Comparison of ensemble-MOS methods in the Lorenz '96 setting". Meteorological Applications. 13 (3): 243. doi: 10.1017/s1350482706002192 . ISSN   1350-4827.