← Back to Publications List

Long Term Hydroclimatic Trends and Logistic Regression Based Flood Month Classification in Bangladesh

Students & Supervisors

Student Authors
Md. Ibtihazzaman
Bachelor of Science in Computer Science & Engineering, FST
Sadman Sakib
Bachelor of Science in Computer Science & Engineering, FST
Supervisors
Mahfuza Khatun
Associate Professor, Faculty, FST

Abstract

Monthly hydroclimatic observations from a total of 33 meteorological stations throughout Bangladesh over the period 1948-2013 are analyzed to quantify long run changes in key variables and to assess how routine measurements relate to flood timing. Each station month is tagged as flood or non-flood, enabling analysis of both associated drivers and predictability from standard observations. Station wise annual means of rainfall and relative humidity are evaluated using Kendall’s τ and the Theil Sen slope, with false discovery rate adjustment across sites. Findings point to broadly increasing trends in both variables over the study window. Flood months are further identified through a logistic-regression model that uses monthly rainfall, temperature, humidity, wind speed, cloud cover, bright sunshine hours and basic geographic descriptors (latitude, longitude, altitude) complemented by a seasonal indicator for month of year. To avoid look ahead bias, performance is measured with a time ordered train test split, and spatial generalization is probed with a leave one station out design. On the temporal holdout, discrimination is very strong (ROC-AUC ≈ 0.99) with excellent precision recall behaviour (average precision ≈ 1.00). Under leave one station out evaluation, performance remains high (ROC-AUC ≈ 0.94; average precision ≈ 0.70) indicating good transferability alongside expected local variation. Annual maximum rainfall is also modeled with a generalized extreme value distribution to characterize rare, high intensity events and to quantify uncertainty in 10 and 50 year return levels using bootstrap intervals. Taken together, the trend assessment, flood month classification and extreme value analysis provide a transparent, reproducible view of flood hazard in Bangladesh and offer an interpretable foundation for strengthening early warning systems and long-term risk planning.

Keywords

Flood prediction Hydroclimatic trends Logistic regression Extreme value analysis Bangladesh.

Publication Details

  • Type of Publication:
  • Conference Name: International Conference on Applied Statistics and Data Science (ICASDS) 2025
  • Date of Conference: 28/12/2025 - 28/12/2025
  • Venue: University of Dhaka, Dhaka 1000, Bangladesh
  • Organizer: Institute of Statistical Research and Training (ISRT), University of Dhaka
View on Publisher