Short-Term Temperature and Rainfall Prediction at Local and Global Spatial Scale

Uncertainty in weather dynamics makes it important to build accurate weather prediction systems because it can save lives by better preparing people for an upcoming incidence. The aim of present research study is to present a comprehensive review of recent scientific works for short term temperature and rainfall prediction on both local and global spatial scale. The literature shows that some meteorological factors like Atmospheric pressure, precipitation, dew point temperature, solar radiation, vapor pressure, cloud cover, snowfall, humidity, wind velocity and wind direction are potential measures to predict future temperature and rainfall. We focused on recent applications of machine learning and deep learning models like Deep Echo State Network, recurrent neural network, convolutional recurrent neural network, and graph convolutional network, Autoencoders, Multi layer Perceptron and Long short-term memory. Applications of multimodal learning, reservoir computing and multitask learning have shown noticeable enhancement in the prediction accuracy of other state of art the models. Fine capability of CNN to extract suitable patterns from numeric weather data is also reported. The time interval of data recording also affects the prediction accuracy greatly. More frequently recorded input data worked better than less frequently recorded data. The use of electromagnetic sensors instead of satellite and radar setups is reliable as well as cost effective for collecting data for prediction. Evaluation indices related to hit rate of rainfall and no rainfall, caching rate, overlooking rate and Swing-and-miss rate can be considered as statistical measures along with other statistical metrics in case of rainfall prediction.