The changes in the weather play a significant role in people's planning. It has attracted the attention of several study communities due to the fact that it has an impact on human life all over the world. But weather forecasting is a challenging task because it is dependent on a variety of factors such as wind speed, wind direction, global warming, etc. Deep learning-based solutions have seen a lot of success in the geospatial domain over the last few years. In the past few years, a variety of deep learning-based weather forecasting models have been proposed. The forecasting techniques used traditionally are highly parametric and so are complex. In this chapter, deep learning techniques which are used for weather forecasting, such as Multilayer Perceptron, Jordan Recurrent Neural Network, Elman Recurrent Neural Network, etc., are discussed in detail. This chapter presents a comparative analysis of various deep learning-based weather forecasting models that are currently available.
TopIntroduction
Artificial Neural Networks (ANN) have grown in popularity in recent years, and are now being used in a variety of tasks such as pattern recognition, time-series analysis, classification, and regression. Weather prediction is considered as a procedure regarding science and technology that is predicting weather conditions for a specific time and location in advance. Accurate weather forecasts can aid individuals in their short-, medium-, and long-term planning efforts, as well as in emergency situations. As a result, before making a selection on where to go, it is vital to be aware of current weather patterns. Therefore, it increases the requirement of handiness of tools used for weather prediction accurately. This requirement is more noticeable if forecasting is considered for short-term forecasting also known as nowcasting. Prediction is on the basis of the sliding window algorithm. The outcome of every month is calculated to check accuracy.
The different atmosphere attributes like temperature, wind and humidity which affects different aspects of human livelihood. In the last few decades, a wide variety of weather prediction models have been deployed (Yadav,2013). The majority of these models are based variants of fuzzy logic and artificial neural networks. This chapter discusses some of the existing weather forecasting models along with their pros and cons. Makhamisa et al. (Makhamisa, 2020) discusses three neural network models, namely Elman recurrent neural network (ENN), Jordan recurrent neural network (JNN) and Multi-layer perceptron (MLP) for rainfall forecasting which is also discussed in (Goodfellow,2016, Nielsen,2015 and Lewis,2016). Using these methods, it is possible to compare the model's performance over time which helps us to determine the best model for forecasting the weather (Yadav,2013). ENN, JNN and MLP models are used to describe the rise of rainfall and solar irradiations. Therefore, work here can be summed up as prediction of weather using deep learning techniques. But the question that needs to be addressed is regarding the accuracy of forecasting techniques.
Figure 1. Weather forecasting working
The weather forecasting process is illustrated in Figure 1. First, data is gathered and submitted for observation, then a perfect model is selected and submitted to weather consulting, and last, findings are announced, which provide short- and long-term predictions, respectively.
The chapter here is organized as follows. Section 2 discusses some existing deep learning-based methods used for the weather forecasting. Section 3 includes various regression models used for the weather forecasting. Section 4 present the comparative analysis of different weather forecasting models and Section 5 highlights the conclusion and future work.