Weather Forecasts Using Deep Learning Models for Urban Areas in Nigeria

The data includes information from ten years, from 2014 up to 2023. Weather data is recorded monthly from the meteorological station at the Forestry Research Institute in Nigeria. Important data recorded includes temperature in Celsius, humidity, wind speed in km/h, and relative humidity. This data is recorded in seven columns and 120 rows. It includes temperature, humidity, wind speed, and relative humidity. Three models are applied in solving this problem. These models are time-aware long short-term memory networks, gated recurrent units, and the Transformer. These models have been improved with an attention-based approach to interpretability, inspired by RETAIN. The GRU model can forecast data up to six months into the future. This data shows an inverse correlation with temperature and relative humidity. Relative humidity goes down to 72% ± 5, indicating pre-rainy conditions. This occurs while temperatures peak at 31°C ± 0.8 in Month 4. In Month 6, temperatures drop to 28.2°C ± 0.5, and relative humidity rises to 85% ± 3, indicating that rain is on the way. Wind speeds decrease to 9.8-10.5 km/h in Months 3 and 4, when temperatures are at their peak and relative humidity is at its lowest. The weather forecasting model has shown how GRU, Time Aware, and even the Transformer can be applied in solving weather problems in Nigeria.