This thesis aims to implement a prototype system to screen flooding photos from social media. These photos, associated with their geographic locations, can provide free, timely, and reliable visual information about flood events to the decision makers. This system is designed for the application to the real social media images, including several key functions: tweets downloading, image downloading, flooding photo detection, and human verification via a WebGIS application. In this study, a training dataset of 5,000 flooding photos was built based on an iterative method; a convolutional neural network (CNN) was then trained and applied to detect flooding photos. Also, the CNN can be re-trained by a larger training dataset after adding the verified flooding photos to the training set. The flooding photo detection result shows that the trained CNN achieved a total accuracy of 93% in a balanced test set (the flooding and non-flooding class have the same number of samples) and precisions of 46% -- 63% in the imbalanced real-time tweets (the number of flooding samples are over 20 times larger than non-flooding), demonstrating the feasibility of the proposed pipeline. The system is flexible to change the classifier, so that detecting other disasters (e.g., tornado) is possible. [The dissertation citations contained here are published with the permission of ProQuest LLC. Further reproduction is prohibited without permission. Copies of dissertations may be obtained by Telephone (800) 1-800-521-0600. Web page: http://bibliotheek.ehb.be:2222/en-US/products/dissertations/individuals.shtml.]