Weather warnings are important to protect life and property. Weather condition is state of atmosphere at given time in terms of weather variables like temperature, pressure, wind direction etc. Modern weather forecasting involves a combination of computer models, observation, and knowledge of trends and patterns. Weather forecasting is the application of science and technology to predict the state of the atmosphere for a given location, and are made by collecting quantitative data about the current state of the atmosphere and using scientific understanding of atmospheric processes to project how the atmosphere will evolve.
Recently, there has been growing interest in the possibility of using neural networks for both weather forecasting and the generation of climate datasets. The building blocks of neural network model architecture provide general insight into the relative importance of model properties for post-processing ensemble forecasts. Specifically, the results indicate that encoding local information is very important for providing skillful probabilistic temperature forecasts.
In paper Neural networks for post-processing ensemble weather forecasts, authors propose a flexible alternative based on neural networks that can incorporate nonlinear relationships between arbitrary predictor variables and forecast distribution parameters that are automatically learned in a data-driven way rather than requiring pre-specified link functions key components to this improvement are the use of auxiliary predictor variables and station-specific information with the help of embeddings. They trained neural network can be used to gain insight into the importance of meteorological variables thereby challenging the notion of neural networks as uninterpretable black boxes. This approach can easily be extended to other statistical post-processing and forecasting problems. Authors also showed that a trained machine learning model can be used to gain meteorological insight. It allowed to identify the variables most important for correcting systematic temperature forecast errors of the ensemble. In this context, neural networks are somewhat interpretable and give us more information than we originally asked for. While a direct interpretation of the individual parameters of the model is intractable, this challenges the common notion of neural networks as pure black boxes.
In work Weather Forecasting Model using Artificial Neural Network , authors a model that can reduce this processing cost by working on raw data. Since the have 10 inputs, a 5 hidden-layer network with 10 or 16 neurons/ layer and a tan-sigmoid transfer function for hidden layers seemed to do generalize much better over 750 and 1460 samples as compared to a single hidden-layer network with the same number of neurons. In this paper the have already discussed the method to analyze and handle overfitting while aiming for accuracy in prediction. Finally, the prediction that they made for the maximum temperature can be extended to other weather factors like humidity, wind speed etc. using the same model and precautions discussed. Further measures to optimize the performance of such a weather forecasting model can be based on various macro and micro-environmental factors. This study can be best used to develop supportive statistical plots and concentrate on the trend of weather over a long period of time in a particular area.
The paper An Effective Weather Forecasting Using Neural Network gives comparison between gradient descent and LM algorithm. Levenberg-Marquardt training normally used for small and medium size networks. The gradient descent algorithm is generally very slow because it requires small learning rates for stable learning. The momentum variation is usually faster than simple gradient descent, because it allows higher learning rates while maintaining stability, but it is still too slow for many practical applications. These two methods are normally used only when incremental training is desired. Authors propose a new technique of weather forecasting by using Feed-forward ANN. The data is taken from Rice Research center (Kaul) Haryana. This is the fastest method among other weather forecasting methods. As there are many BP algorithm but among them Levenberg BP has better learning rate.
In work Weather and climate forecasting with neural networks: using general circulation models (GCMs) with different complexity as a study ground authors have tested the use of neural networks for forecasting the “weather” in a range of simple climate models with different complexity.
They use a bottom–up approach for assessing whether it should, in principle, be possible to do this. For this they have used a deep convolutional encoder–decoder architecture that Scher developed for a very simple general circulation model without seasonal cycle. The network is trained on the model in order to forecast the model state 1d ahead. This process is then iterated to obtain forecasts at longer lead times. Authors specifically assess how the complexity of the climate model affects the neural network’s forecast skill and how dependent the skill is on the length of the provided training period. Additionally authors showed that using the neural networks to reproduce the climate of general circulation models including a seasonal cycle remains challenging – in contrast to earlier promising results on a model without seasonal cycle.