← Back to Publications List

Data-Driven Forecasting of Rainfall and Temperature Using Multi-Output LSTM Networks

Students & Supervisors

Student Authors
Raif Tanjim
Bachelor of Science in Electrical & Electronic Engineering, FE
Shahed Islam Swapnw
Bachelor of Science in Electrical & Electronic Engineering, FE
Tanvir Ahmed
Bachelor of Science in Electrical & Electronic Engineering, FE
Md. Nazmun Nur
Bachelor of Science in Electrical & Electronic Engineering, FE
Supervisors
Abu Shufian
Lecturer, Faculty, FE

Abstract

Accurate long-term forecasting of rainfall and temperature is important for climate sensitive countries such as Bangladesh, where agriculture, water management, and disaster preparedness depend on monsoon behavior. Many existing studies focus on single variables, short horizons, or station specific data and often do not exploit long historical records or advanced feature engineering. This paper presents a multi output Long Short-Term Memory (LSTM) framework that jointly forecasts monthly rainfall and temperature using a 120 year nationally aggregated climate dataset. The model incorporates lagged inputs, rolling averages, anomaly features, and cyclical month encodings so that both short term fluctuations and slow seasonal cycles are represented. A sliding window formulation is used with twelve-month input sequences and single step multi variable outputs. Performance is assessed on a held-out test period and compared with classical time series baselines using RMSE and R². Results show high accuracy and stable residuals for temperature, along with reasonable skill in reproducing seasonal rainfall patterns, although the highest monsoon peaks remain underestimated. The framework provides a reproducible pipeline for data preprocessing, feature construction, and model training, and it demonstrates the value of deep learning for supporting climate risk assessment and planning in monsoon dominated regions.

Keywords

Bangladesh climate prediction deep learning LSTM rainfall forecasting temperature forecasting

Publication Details

  • DOI: N/A
  • Type of Publication:
  • Conference Name: 4th IEEE International Conference on Robotics, Automation, Artificial-Intelligence and Internet-of-Things 2025
  • Date of Conference: 27/11/2025 - 27/11/2025
  • Venue: Military Institute of Science and Technology (MIST), Dhaka, Bangladesh
  • Organizer: IEEE Bangladesh Section (IEEE BDS) and IEEE Robotics and Automation Society (RAS) Bangladesh Chapter.
View on Publisher