Members’ Publication

Fire Hazard has become a regular phenomenon in the densely populated cities of Bangladesh due to haphazard urban development, high building density, unplanned city growth etc. The cities of Bangladesh have experience around terrifying 16000 fire incidents and 1590 death in the last ten year. The city of Rajshahi, one of the oldest municipalities of the country, is also prone to fire hazard and many factors such as- unplanned building construction, existence of many old buildings, slum and squatter settlement, presence of potential industrial fire source, narrow road width, pond filling etc. This research aims to assess the fire hazard risk of residential buildings of Rajshahi City Corporation and mapping the spatial variation of fire risk to identify proper planning interventions. This study used Entropy- Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) integrated method for fire hazard risk assessment based on 9 most relevant factors including proximity to road, distance to water body, distance from fire station, distance from fuel station, distance to hospital, space between adjacent structure, building storey, distance to high voltage electric pole, distance to gas pipeline. From the fire hazard risk score calculated from above mentioned method and factors, ward wise fire risk mapping of Rajshahi City Corporation is done using ArcGIS software. The result shows that only 2.72% of residential buildings of Rajshahi city have low level of fire risk whereas 54.64% of residential building falls in high level of fire risk. About 42.64 % of residential building is marked as moderate fire risky in Rajshahi city. This result expected to be helpful in deciding disaster risk sensitive land use planning interventions of Rajshahi City.

Land cover classification has great contribution in many stages of city planning, urban management and understanding human-environment interaction. It is widely used to analyse spatial pattern of land cover change of an area and further used in simulating land cover change and its impact on environment. The aim of this paper is to utilize machine learning algorithms such as support vector machine (SVM), k- nearest neighbor, deep neural network, decision tree in terms of land cover classification from multispectral satellite data. For this study, Landsat 8 OLI/TIRS sensor data was used for land cover classification. To evaluate the performance of these algorithms in land cover classification, Cohen kappa and overall accuracy score are used using 155 random known points from google earth. The result shows that DNN scores 88% cohen kappa and 92% overall accuracy whereas SVM gains 84% cohen kappa ,89% overall accuracy; decision tree gains 80% cohen kappa, 86% overall accuracy score and k-nearest neighbor gains 84% cohen kappa, 89% overall accuracy. This result clearly indicates that DNN has better ability in classification of low-resolution satellite data than other machine learning algorithms which has the scope to use in different stages of urban planning and management. (PDF) Land Cover Classification from Multispectral Remote Sensing Image Using K-Nearest Neighbor, Deep Neural Network, Decision Tree and SVM Algorithms: A machine learning-based comparison approach.