Estimating Rainfall Intensity Using an Image-Based Deep Learning Model

• Develop a learning model to estimate rainfall intensity based on sensor images. Proposed provides estimates with MAPE between 13.5% and 21.9%. High spatiotemporal urban data can be acquired an extreme low cost. The proposed represents new measurement approach in the area. greatly facilitates real-time flood management. Urban flooding is major issue worldwide, causing huge economic losses serious threats public safety. One promising way mitigate its impacts develop risk management system; however, building such system often challenging due lack of high data. While some approaches (i.e., ground stations or radar satellite techniques) are available measure and/or predict intensity, it difficult obtain accurate desirable resolution using these methods. This paper proposes image-based deep spatial temporal resolution. More specifically, convolutional neural network (CNN) called CNN (irCNN) developed images collected from existing dense sensors smart phones transportation cameras) their corresponding measured values. trained irCNN subsequently employed efficiently sensors’ Synthetic real respectively utilized explore irCNN’s accuracy theoretically practically simulating intensity. results show that mean absolute percentage error ranging 21.9%, which exceeds performance other state-of-the-art modeling techniques literature. importantly, main feature cost acquiring alternative for estimating facilitate development manner.