Skip to content

R&D

Current Research

UAV-based geospatial modeling to detect species in Texas and assess the effects of urban development

M.S. research, Department of Natural Resources Management, Texas Tech University (Advisor: Dr. A. H. M. Ali Reza)

Graduate Research Assistant
Spring 2027 – present

Optimized geohazard monitoring: A DInSAR, InSAR, and explainable machine learning approach for land subsidence quantification in the deltaic metropolis of Bangladesh

Undergraduate thesis, Department of Urban and Regional Planning, KUET (Supervisor: Mafrid Haydar)

B.URP Thesis
2025

Research Themes

Land subsidence and InSAR

Measuring ground deformation in the deltaic city of Khulna with DInSAR and InSAR time series, and explaining its drivers and spatial susceptibility with interpretable machine learning. This was the focus of my undergraduate thesis.

  • Optimized geohazard monitoring: A DInSAR, InSAR, and explainable machine learning approach for land subsidence quantification in the deltaic metropolis of Bangladesh (Undergraduate thesis)
  • Coupling InSAR and explainable machine learning to assess land subsidence in a deltaic metropolitan of Bangladesh (submitted)
  • Optimized quantification of land subsidence using DInSAR and InSAR techniques in deltaic metropolitan of Bangladesh (submitted)

Climate-related hazard susceptibility

Data-driven susceptibility mapping of forest fire, coastal erosion, floods, drought and thunderstorms in Bangladesh, including projections under future climate scenarios, using remote sensing predictors and ensemble machine learning.

Urban growth and land transformation

Modeling where cities expand and how land use changes, by combining night-time lights, thermal and spectral imagery with machine learning and geostatistics.

Water, agriculture and ecosystems

Assessing natural spring-based water potential, crop production and soil erosion impacts on ecosystem services, linked to my interest in remote sensing of groundwater resources and groundwater modeling.

Transportation, environment and public health

Field-based studies of traffic congestion, intersection performance and road-side pollution, and of public health, carried out through coursework and with the Center for Environmental Research.

  • Origin-destination-based traffic congestion analysis in Bangladesh (submitted)
  • Optimizing intersection efficiency through strategic signalization at two substantial intersections in Bangladesh (in preparation)
  • Spatial distribution of pollutants and their relationships with environmental and economic factors across road sections in Bangladesh using machine learning (in preparation)
  • Advancing diabetes prediction: Explainable artificial intelligence (XAI) for enhanced clinical insights (submitted)

Research Projects with Code

Forest Fire Susceptibility Modeling in the Chittagong Hill Tracts

Machine learning and deep learning models that predict forest fire susceptibility in the Chittagong Hill Tracts from environmental and human factors, to help prioritize risk zones for forest management.

Problem: Forest fires in the Chittagong Hill Tracts degrade forest cover and livelihoods, but fire-prone areas are not mapped systematically, which makes prevention hard to target.

  • Susceptibility modeling with CatBoost, Random Forest, SVM, XGBoost, GBDT and CART
  • Neural network models (ANN, DNN, CNN)
  • ROC/AUC evaluation and factor importance analysis

The repository holds one script per model. Each script trains a classifier on a table of conditioning factors, reports accuracy and variable importance, and plots the ROC curve. The ANN, DNN and CNN scripts also export importance scores, so the deep learning models can be compared with the tree-based ones.

Python · scikit-learn · pandas · matplotlib

GitHub repository · Related paper in Ecological Indicators

Open-source code
Paper published 2024

Climate-Scenario Forest Fire Risk Projection (2040–2100)

Ensemble tree-based models that assess current forest fire susceptibility in the Chittagong Hill Tracts and project future fire risk under RCP climate scenarios for 2040–2100.

Problem: Fire risk depends on climate, so maps built on present-day conditions alone may understate where risk will grow. Planning needs projections of how susceptible zones shift under future climate.

  • Random Forest, LightGBM, XGBoost, AdaBoost, CatBoost, Bagging, Gradient Boosting and Decision Tree classifiers
  • 19 conditioning factors, including aridity index, climatic water deficit, LAI, NDVI, NDMI, NDWI, NDBI, LST, evapotranspiration, rainfall, humidity, temperature, wind speed and terrain
  • Projection of susceptibility under RCP scenarios for 2040–2100
  • ROC/AUC evaluation and feature importance

Each model is a separate script with the same 70/30 train–test design, so accuracy, AUC and factor importance can be compared directly across algorithms before the selected models are applied to future climate layers.

Python · scikit-learn · LightGBM · XGBoost · CatBoost · pandas · matplotlib

GitHub repository · Related paper in Ecological Indicators

Open-source code
Paper published 2025

Data-Driven Urban Expansion Susceptibility Mapping

A Random Forest and geostatistical approach to identify areas at high risk of urban sprawl in the major urban units of Bangladesh.

Problem: Rapid, unplanned urban growth in Bangladesh consumes agricultural land and outpaces infrastructure. Planners need to know where expansion is most likely to happen next.

  • Random Forest susceptibility model with ROC/AUC evaluation
  • 20 factors, including DEM, slope, curvature, roughness, LST, NDVI, NDWI, LAI, land use/land cover, rainfall, population, and distance to rivers, roads, settlements, industry and growth centres
  • Geostatistical analysis of the predicted surfaces

The script trains the Random Forest model on GIS-derived factors, reports accuracy and the relative importance of each driver, and plots the ROC curve.

Python · scikit-learn · pandas · matplotlib · GIS

GitHub repository · Related paper in Journal of Urban Management

Open-source code
Paper published 2025

Academic and Field Projects

Traffic Congestion Modelling Based on Origin and Destination in Khulna

A field study of traffic congestion along the Phultola–Rupsha Ghat corridor of Khulna City, using a congestion index built from intersection counts, traffic volume, spot speed and traffic density.

  • Traffic volume, spot speed and density surveys by road segment
  • Congestion index (CI) for origin, sub-origin and destination segments
  • Weekday and weekend comparison

Key result: The highest congestion index was 5.79 on weekdays and 4.78 on weekends, both around City Medical and Moylapota Mor. A manuscript from this study has been submitted to Transportation Journal (Wiley).

Field survey

B.URP project
KUET

Site Suitability for Rainwater Harvesting in Drought-Prone Rangpur

A geospatial land suitability analysis to locate sites for rainwater harvesting reservoirs across 58 subdistricts in the drought-prone northwest region of Rangpur, Bangladesh.

  • Land suitability analysis with eight factors (DEM, land use/land cover, slope, precipitation, NDVI, NDWI, land surface temperature, distance to water)
  • Weighted overlay of remote sensing and GIS layers

Key result: Only 6 of 58 subdistricts were found suitable for rainwater harvesting.

GIS · Remote sensing

B.URP project
KUET

Regional Economic Composition of Rangpur Division

A combined spatial and non-spatial analysis of industrial structure, employment change and inequality across the districts of Rangpur Division.

  • Location quotient and shift-share analysis
  • National average index, Herfindahl-Hirschman index and entropy index
  • Gini coefficient and Lorenz curve (2011 vs 2022)
  • PCA and AHP to rank development factors

Key result: Saw mills, rice mills and tailoring are basic industries in Rangpur district. Wage inequality increased between 2011 and 2022, and population was the most important development factor.

Spatial analysis · Statistics

B.URP project
KUET

Smart Village Development — Betaga vs Sharafpur, Khulna

A comparative survey of smart village practices in a reference area (Betaga Union) and a study area (Sharafpur Union), Khulna, and their effect on rural quality of life.

  • Indicator selection from global and local literature (education, energy, health, infrastructure, agriculture)
  • Questionnaire surveys in both unions
  • Correlation, Mann-Whitney U and chi-square tests

SPSS · Field survey

B.URP project
KUET

Child-Friendly Neighbourhood Environments in Khulna

A framework for measuring child-friendliness at the neighbourhood level in Khulna through indicators of provision, protection and participation.

  • Indicator framework adapted from international practice to the local context
  • Participatory workshops and community mapping
  • Quantitative household surveys

Participatory mapping · Field survey

B.URP project
KUET

Single-Use Plastic and Consumer Behaviour in Fultala Market, Khulna

A survey of 50 business owners on plastic consumption and awareness of alternatives at Fultala Market, Khulna.

  • Structured survey of 50 market businesses
  • Correlation analysis of plastic distribution, cost and availability of alternatives

Key result: Mean daily plastic use was 0.5 kg per business, highest among vegetable sellers. Plastic distribution and cost showed a moderate inverse correlation.

Field survey · Statistics

B.URP project
KUET

Research Interests

Methods and Tools

Remote sensing: InSAR / DInSAR, Sentinel-1 SAR backscatter, Spectral indices (NDVI, NDWI, NDBI, NDMI, LST), Night-time lights

Machine learning: Random Forest, XGBoost, LightGBM, CatBoost, AdaBoost, SVM, ANN / DNN / CNN, Explainable ML

GIS & platforms: ArcGIS, Google Earth Engine

Programming & analysis: Python (pandas, scikit-learn, matplotlib), R, SPSS, Correlation, Mann-Whitney U and chi-square tests, Location quotient, shift-share, AHP, PCA

Design: AutoCAD, SketchUp, Adobe Illustrator

Fieldwork: Questionnaire surveys, Participatory surveys, Focus group discussions, Traffic volume & spot-speed surveys, UAV data collection

Collaborators