<p>Natural disasters disrupt both human habitats and vital infrastructures, leading to significant damage and sometimes permanent alterations to the environment. Such disasters can arise from diverse natural phenomena, such as earthquakes, cyclones, floods and wildfires. Despite various efforts by researchers to use Deep Learning (DL) for disaster identification and classification, the complexity and imbalance of image data present ongoing challenges. In response to these challenges, a new LightDD (Lightweight Deep Disaster Detection) model is proposed to streamlined deep Convolutional Neural Network (CNN) tailored for disaster detection. Notably, the proposed LightDD model is lightweight, comprising just 0.11 million parameters. The model is divided into two main segments: the first employs three convolutional modules for feature extraction, with filters sizes in descending order (128, 64, and 32), capped by average global pooling. This configuration allows the model to extract complex features using high-level filters and efficiently condense them in subsequent layers, preserving a rich feature representation while reducing the overall number of parameters. The second segment is dedicated to disaster classification and consists of a singular module of fully connected layers. The LightDD model was evaluated on a disaster dataset, achieving an accuracy rate of 91.44%.</p>