Publicação
Reconstitution of weather time series with an analog ensemble model
| Resumo: | The numeric weather prediction (NWP) that is currently used is based on global circulation models (GCM), which may be used for weather forecasting within horizons of 15 days, commonly. Yet, GCM lacks the spatial resolution required for engineering applications such as wind energy. Additionally, weather forecasts and hindcasts are often affected by phase errors. This study presents the use of a post-processing technic applied to the forecasting of weather time series. The technic is based on identifying analog ensembles from another time series of observations and using these to refine the forecast. To evaluate the skill of the method it was applied to ten weather stations. The focus of the study is to create data for reanalysis in places that lack weather measurements. To be able to evaluate the skill of the method, data from one station was used to forecast six variables at another station. This study used five years of training data to predict two years of forecast. As the analysis required a significant computational power, the studies were divided into two major approaches. The first approach had only one variable in the training period. The results were good for the variables that are easier to predict but had poor results in predicting variables with high level of abrupt changes. The second approach used multiple variables for the training period. The results were found to be significantly better. Although quantitatively there is error in the forecast characterized by a mean absolute error of 0.49 m/s for the wind speed, qualitatively the forecast was able to follow the behavior of the observed curve. It was found that the method can be very sensitive to the initial calibration, which may hinder the results. |
|---|---|
| Autores principais: | Santos, Maycon Meier dos |
| Assunto: | Analog ensemble Weather forecast Time series Post-processing method |
| Ano: | 2019 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Instituto Politécnico de Bragança |
| Idioma: | inglês |
| Origem: | Biblioteca Digital do IPB |
| Resumo: | The numeric weather prediction (NWP) that is currently used is based on global circulation models (GCM), which may be used for weather forecasting within horizons of 15 days, commonly. Yet, GCM lacks the spatial resolution required for engineering applications such as wind energy. Additionally, weather forecasts and hindcasts are often affected by phase errors. This study presents the use of a post-processing technic applied to the forecasting of weather time series. The technic is based on identifying analog ensembles from another time series of observations and using these to refine the forecast. To evaluate the skill of the method it was applied to ten weather stations. The focus of the study is to create data for reanalysis in places that lack weather measurements. To be able to evaluate the skill of the method, data from one station was used to forecast six variables at another station. This study used five years of training data to predict two years of forecast. As the analysis required a significant computational power, the studies were divided into two major approaches. The first approach had only one variable in the training period. The results were good for the variables that are easier to predict but had poor results in predicting variables with high level of abrupt changes. The second approach used multiple variables for the training period. The results were found to be significantly better. Although quantitatively there is error in the forecast characterized by a mean absolute error of 0.49 m/s for the wind speed, qualitatively the forecast was able to follow the behavior of the observed curve. It was found that the method can be very sensitive to the initial calibration, which may hinder the results. |
|---|