Systems, methods and computer program products of the present invention compare the historical forecast information, and more particularly, the predicted weather components, generated by the plurality of weather models to observed weather data to determine the historical performance of the weather models.
解答例
According to one aspect of the invention, the historical performance is the accuracy with which a weather model predicts a particular weather component.
Thereafter, each model is weighted based upon the historical performance of that model in predicting a weather component at a particular geographic location or a range of geographical locations.
The weighted weather models are then combined to generate a multi-model superensemble.
The multi-model superensemble utilizes the historical performance of every weather model in forecasting weather components to generate a weather forecast for a future period of time.
More specifically, according to the present invention, a multi-model superensemble is developed using a plurality of forecasts from a variety of weather and climate models.
Along with observed (or benchmark) analysis fields, these forecasts are used to derive statistics on the past behavior of the models.
These statistics, combined with model forecasts, enables the construction of a superensemble forecast.
More specifically, given a set of past model forecasts, the present invention uses a multiple regression technique to regress the model forecasts against observed (analysis) fields.
Least-squares minimization of the difference between the model and the analysis field is used to determine the weights of each model component at any geographic location and vertical level.
According to one embodiment of the present invention, there is disclosed a method for generating an accurate weather forecast model.
The method includes the steps of collecting historical forecast information from a plurality of weather models, wherein the historical forecast information includes at least one predicted weather component, and wherein the historical forecast information corresponds to a past period of time.
The method also includes accumulating observed weather data, wherein the observed weather data corresponds to a plurality of known weather values, wherein at least one known weather value of the plurality of known weather values corresponds to the at least one predicted weather component, and wherein the observed weather data corresponds to the past period of time.
The method further includes comparing the historical forecast information to the observed weather data to determine the historical performance of each weather model of the plurality of weather models, and generating a multi-model superensemble of the weather models, wherein the multi-model superensemble is based upon the historical performance of each weather model of the plurality of weather models.
According to one aspect of the invention, comparing the historical forecast information to the observed weather data to determine the historical performance of each weather model includes comparing the at least one known weather value to at least one predicted weather component.
According to another aspect of the invention, comparing the at least one known weather value to the at least one predicted weather component includes calculating at least one weight factor for the at least one predicted weather component.
Furthermore, comparing the at least one known weather value to the at least one predicted weather component can include calculating at least one weight factor for the at least one predicted weather component by least squares minimization.
According to yet another aspect of the present invention, generating a multi-model superensemble of the weather models includes generating a multi-model superensemble based upon a combination of weather models weighted by their respective historical performances.
Additionally, generating a multi-model superensemble of the weather models can include generating a multi-model superensemble based upon a summation of the at least one weight factor for the at least one predicted weather component of each of the plurality of weather models.
The method can further include collecting future forecast information from the plurality of weather models corresponding to a future period of time, and wherein generating a multi-model superensemble includes generating a multi-model superensemble based upon the historical performance of each weather model of the plurality of weather models and the future forecast information.
Moreover, generating a multi-model superensemble can include weighting the future forecast information from the plurality of weather models based upon the historical performance of each weather model of the plurality of weather models.
According to another embodiment of the present invention, there is disclosed a method for generating accurate weather forecasts.
The method includes collecting historical forecast information from a plurality of weather models, wherein the historical forecast information includes at least one predicted weather component, and wherein the historical forecast information corresponds to a past period of time.
The method also includes accumulating observed weather data corresponding to a plurality of known weather values, wherein at least one known weather value of the plurality of known weather values corresponds to the at least one predicted weather component, and wherein the observed weather data corresponds to the period of time.
The method further includes comparing the historical forecast information to the observed weather data to determine the historical performance of each weather model of the plurality of weather models, and calculating at least one weight for each weather model, based upon the historical performance of each weather model in forecasting the at least one predicted weather component.
Finally, the method includes combining the weights for each weather model with future forecast information from the plurality of weather models, wherein the future forecast information corresponds to a future period of time, to generate a multi-model superensemble forecast.
According to one aspect of the invention, generating a multi-model superensemble forecast includes combining the weather models, wherein each model is weighted based on its respective weight.
According to another aspect of the invention, comparing the historical forecast information to the observed weather data to determine the historical performance of each weather model includes comparing the at least one known weather value to the at least one predicted weather component.
According to yet another aspect of the present invention, comparing the at least one known weather value to the at least one predicted weather component includes calculating at least one weight factor for the at least one predicted weather component.
Additionally, in the method of the present invention, comparing the at least one known weather value to the at least one predicted weather component can include calculating at least one weight factor for the at least one predicted weather component by least squares minimization.
