ABSTRACT Cuba is immersed in the use of wind energy. However, for its development it has required various efforts in different fields, including the improvement of tools that make the wind predictable and, in turn, wind generation, such is the case of very short-term forecasts. For this reason, this paper compares the wind forecast of the Weather Research and Forecasting model (WRF) at 3 km spatial resolution a Long Short-Term Memory (LSTM) model type. The comparison and evaluation of the forecasts of the models is carried out with data from the Gibara I and II wind farms and the Los Cocos wind survey mast, located in Holguín, Cuba, with wind speed measurements every 10 minutes at a height of 50 m. The LSTM were built by first training the observations and then combining the observations with the WRF model forecast. The results of the comparison were carried out for three study cases and indicate that both LSTM models present better results than the WRF model, although the differences do not exceed 1 m/s. However, for the case studies, the WRF model behaves well reproducing the daytime cycle, but with a MAE greater than 4 m/s.

RESUMEN Cuba se encuentra inmersa en el empleo de la energía eólica. Sin embargo, para su desarrollo ha requerido diversos esfuerzos en diferentes campos, incluyendo el perfeccionamiento de herramientas que hagan predecible el viento y a su vez la generación eólica, tal es el caso de los pronósticos a muy corto plazo. Por tal motivo, en el presente trabajo se compara el pronóstico del viento producido por Weather Research and Forecasting Model (WRF) a 3 km de resolución espacial y un modelo de inteligencia artificial del tipo Long Short Term Memory (LSTM). La comparación y evaluación de los pronósticos de los modelos se realiza con datos de los parques eólicos Gibara I y II y la torre de prospección eólica Los Cocos, ubicadas en la provincia de Holguín, Cuba. Allí se realizan mediciones de la velocidad del viento cada 10 minutos a una altura de 50 m. El LSTM se construyó entrenando primero las observaciones y luego combinándolas con el pronóstico del modelo WRF. Los resultados de la comparación se realizaron para tres casos de estudio e indican que ambos modelos LSTM presentan mejores resultados que el modelo WRF, aunque las diferencias no superan 1 m/s. Sin embargo, para los casos de estudio, el modelo WRF se comporta bien reproduciendo el ciclo diurno, pero con un MAE superior a 4 m/s.

ABSTRACT The Short-range prediction system, better known as SisPI, offers very short-term data on wind, temperature, precipitation, solar radiation and other meteorological variables. The first version of the model (SisPI1.0) underwent changes in its internal configuration: modification in the solar radiation parameters, incorporation of the “shallow convection” parameterization and extension of the forecast period up to 48 hours. The changes made room for its second version (SisPI2.0), experimentally operational since September 2020, at the Institute of Meteorology (INSMET). The SisPI2.0 had to be evaluated in order to know the effects of the change made, especially, you needed to know its ability to quantitatively forecast precipitation. It will also, be necessary to know the quality of its forecast with respect to SisPI1.0, and the ability to simulate the weather 48 hours in advance. The study performs the spatial verification of the SisPI2.0 forecast based on the GPM observations, and the selection of four study cases of the rainy season of 2021. The MODE was used as an evaluation method. SisPI2.0 precipitation forecast turned out to be more effective in object identification than SisPI1.0, but quantitatively they make similar errors, in addition SisPI2.0 suggests to have good ability to forecast precipitation with a period of up to 48 hours in advance.

RESUMEN El Sistema de Pronóstico Inmediato, más conocido como SisPI, ofrece datos de viento, temperatura, precipitación, radiación solar y otras variables meteorológicas a muy corto plazo. La primera versión del modelo (SisPI1.0) sufrió cambios en su configuración interna: modificación en las parametrizaciones de radiación solar, incorporación de la parametrización de “shallow convection” y ampliación del plazo de pronóstico hasta 48 horas. Los cambios dieron lugar a su segunda versión (SisPI2.0), operativa experimentalmente desde septiembre del 2020, en el Instituto de Meteorología (INSMET). El SisPI2.0 debía ser evaluado para conocer los efectos de los cambios realizados, en especial, para conocer su habilidad para pronosticar cuantitativamente la precipitación. Era necesario investigar, además, la calidad de su pronóstico respecto al SisPI1.0, y la habilidad para simular el estado del tiempo con 48 horas de antelación. El estudio realiza la verificación espacial del pronóstico del SisPI2.0 a partir de las observaciones del GPM, y la selección de cuatro casos de estudio del período lluvioso de 2021. Se empleó el MODE como método de evaluación. El pronóstico de la precipitación de SisPI2.0 resultó ser más efectivo en la identificación de objetos que el de SisPI1.0, pero cuantitativamente cometen similares errores, además el SisPI2.0 sugiere tener buena habilidad para pronosticar la precipitación con un plazo de hasta 48 horas de antelación.

Resumo Os modelos de previsão de tempo baseiam-se em métodos computacionais aplicados para prever os fenômenos meteorológicos a partir da integração numérica de equações que descrevem o movimento dos elementos químicos que compõem a atmosfera. Entre as variáveis previstas por esses modelos a precipitação é uma das mais importantes, e é a mais difícil de se obter bons resultados no posicionamento, intensidade e extensão dos eventos. Uma das fontes de incerteza na qualidade dessas previsões é o próprio método de avaliação empregado, pois os métodos convencionais não avaliam adequadamente os aspectos espaciais. O presente estudo tem como objetivo principal propor e validar uma metodologia de avaliação diagnóstica orientada a objeto, empregando o algoritmo Method for Object-Based Diagnostic Evaluation (MODE), dentro do Sistema Comunitário de Avaliação de Modelos Numéricos de Previsão de Tempo e Clima (SCANTEC). Experimentos utilizando casos idealizados são apresentados, os quais permitem validar a implementação do método e demonstrar as suas vantagens com relação às métricas convencionais de avaliação. São também apresentados os resultados desse método na comparação das previsões reais de precipitação gerados por modelos utilizados rotineiramente no CPTEC, como prova de conceito, exemplificando o uso ao explorar as métricas disponíveis. Os resultados são discutidos e demostram o grande potencial da metodologia orientada a objeto para o aprimoramento e evolução dos produtos de previsão numérica de tempo.

Abstract Numerical Weather Prediction Models (NWPM) are based on computational methods applied to predict meteorological phenomena from the numerical integration of equations that describe the movement of the atmospheric components. Among the variables predicted by NWPM, precipitation is one of the most important, and it is also the most difficult to obtain good results in the positioning, intensity, and extension of the events. One of the sources of uncertainty in the quality of these forecasts is the evaluation method because conventional methods do not evaluate the spatial aspects appropriately. The objective of this study is to propose and validate a methodology of object-based diagnostic evaluation using Object-Based Diagnostic Evaluation (MODE) algorithm implemented in the Community System for the Evaluation of Numerical Models of Weather and Climate Prediction (SCANTEC). Experiments using idealized cases are presented, which allow validate the implementation of the method and show its advantages over conventional evaluation metrics. The results of this method are also presented in the comparison of the real precipitation forecasts generated by routinely used models in CPTEC as proof of concept, exemplifying its use to explore the available metrics. The results demonstrate the great potential of the object-based methodology for the evolution of the NWPM.

Based on the use of the numerical model of the Weather Research Forecast (WRF), to predict the wind speed in the place where a wind farm is located, a methodology is presented to make energy forecasts with 30 hours in advance, taking into account The power curve of the wind turbines given by the manufacturer and the estimated according to the actual generation of them. The nests made to perform the run with the WRF model showed that the best forecasted wind speed values were obtained for the 1 km resolution domain and that the values given by the statisticians mean value, absolute mean error and Pearson correlation , indicated that the energy forecasts obtained are acceptable and that the WRF model demonstrated its ability to generate wind forecasts with high resolution and good reliability for our country.

Basado en el empleo del modelo numérico del tiempo Weather Research Forecast (WRF), para pronosticar la rapidez del viento en el lugar donde se encuentra ubicado un parque eólico, se muestra una metodología para realizar pronósticos energéticos con 30 horas de antelación, teniendo en cuenta la curva de potencia de los aerogeneradores dada por el fabricante y la estimada de acuerdo a la generación real de los mismos. Los anidamientos realizados para efectuar la corrida con el modelo WRF, mostró que los mejores valores de rapidez del viento pronosticados se obtuvieron para el dominio de 1 km de resolución y que los valores dados por los estadígrafos valor medio, error medio absoluto y correlación de Pearson, indicaron que los pronósticos energéticos obtenidos, son aceptables y que el modelo WRF demostró su habilidad para generar pronósticos de viento con alta resolución y buena confiabilidad para nuestro país.

São apresentados os resultados de uma investigação sobre o processo de convecção de Rayleigh - Bénard partindo do cálculo dos números adimensionais de Rayleigh (Ra) e de Richardson (Ri), com o objetivo de encontrar critérios que sejam representativos do disparo da convecção nas condições tropicais e insulares de Cuba. Os resultados são comparados com os obtidos empregando-se a técnica de ondeleta, e as observações de nebulosidade obtidas das imagens de satélites meteorológicos. Utilizaram-se as medições do mês de Junho do ano 2007 na torre de gradiente em Camagüey, e foram obtidos resultados como o número Ra crítico, (Ra = 1.32 x 10³). A velocidade vertical positiva foi empregada como condição necessária para o disparo da convecção. Encontrou-se que o Ra é um parâmetro indicador do tipo de convecção que deve formar-se, enquanto que o Ri é um indicador de sua intensidade.

Results from an investigation on the physical process of Rayleigh-Bernard convection, based on the calculus of the dimensional Rayleigh (Ra) and Richardson (Ri) numbers, with the objective of finding criteria representative of the onset of convection in Cuba's tropical and insular conditions are shown. The obtained results are compared to those based on the Wavelets Technique and cloudiness derived from meteorological satellite images. Observations on June 2007 at Camagüey gradient Tower were used, and some results were derived, such as the critical (Ra = 1.32 x 10³). Positive value of the vertical velocity was set as the necessary condition for the onset of convection. It was found that Ra is an indicator of the convection type that should form, whereas Ri is an indicator of its intensity.