Python-based Convolutional Neural Network Analysis of Satellite Imageries to Develop a Landscape Ecology Triangle

Abstract

This research aims to understand the applicability of Python-based Convolutional Neural Networks (CNNs) in combination with satellite imageries to classify the landscape types and create the Landscape Ecology Triangle for Bangladesh. Prevailing as a field that involves the manual interpretation and empirical observation of landscapes, landscape ecology is becoming infused with deep learning, thus creating new approaches to the analysis of spatial data. Unlike previous research that has concentrated on the characterization of landscape, this research classifies seven types of landscape: City, District Town, Upazila Town, Paurasabha, Urban Fringe, Rural Fringe, and Rural. This study also used secondary data from the World Bank databases and other reliable sources, in addition to satellite imageries, to classify and describe the landscape types. Data about the proportions of the landscape elements especially the built-up areas were collected and compiled from a literature review that included the relevant research articles. These data were used to train a specific CNN model that was intended to identify landscapes according to their ecological and structural properties. The derived Landscape Ecology Triangle offers a conceptual model that not only categorizes different types of landscapes but also quantifies the proportion of various landscape components within each type, thereby offering a better understanding of the landscape processes. When applied to the classification of landscape types, the approach can give an understanding of the best places for living based on ecological conditions and the suitability of the environment, thus promoting further ecological studies and better practices of land use in Bangladesh.