Deep learning-based effective fine-grained weather forecasting model

• DOI: 10.1007/s10044-020-00898-1
• Publication Date: 2020-06-22T19:03:51Z
• Authors: Pradeep Hewage; Marcello Trovati; Ella Pereira; Ardhendu Behera
• Keywords: Ensemble forecasting; Probabilistic Forecasting

Abstract It is well-known that numerical weather prediction (NWP) models require considerable computer power to solve complex mathematical equations obtain a forecast based on current conditions. In this article, we propose novel lightweight data-driven forecasting model by exploring temporal modelling approaches of long short-term memory (LSTM) and convolutional networks (TCN) compare its performance with the existing classical machine learning approaches, statistical dynamic ensemble method, as well well-established research (WRF) NWP model. More specifically Standard Regression (SR), Support Vector (SVR), Random Forest (RF) are implemented Autoregressive Integrated Moving Average (ARIMA), Auto (VAR), Error Correction Model (VECM) approaches. Furthermore, Arbitrage Forecasting Expert (AFE) method in article. Weather information captured time-series data thus, explore state-of-art LSTM TCN models, which specialised form neural network for prediction. The proposed deep consists number layers use surface parameters over given period time forecasting. assessed two different regressions, namely multi-input multi-output single-output. Our experiment shows produces better results compared WRF model, demonstrating potential efficient accurate up 12 h.