The effective functioning of smart drainage systems for urban flooding alleviation, resource-efficient management of water resources, and infrastructure strengthening requires precise rainfall-runoff predictions. Smart systems enable climate adaptation and are sensitive to the needs of the environment. Urban processes that are hydrological in nature are characterized by their complexity, non-linearity, data deficiency, and ever-changing timelines. In order to overcome these issues, this paper presents a transformative approach by utilizing Bidirectional Transformers owing to the potential of deep learning with attention mechanisms in temporal sequence tasks. Through historical rainfall/runoff data, the model builds complex relationships with dependencies by determining them contextually - it propels through both forward and backward driving through sequences, which provide better contextual understanding po hydrological patterns. The proposed model underwent training and evaluation processes that incorporated datasets collected from drain IoT devices bordering the city and weather monitoring stations. The transformer's data was validated against benchmark metrics of RMSE, NSE, and MAE in reference to the LSTM and GRU networks in a traditional setting. It is displayed in the results that with regard to shifting weather, the Bidirectional Transformers endured the most significant gains by outperforming expectational models in both short and long-term runoff forecasts. Moreover, the model’s attention weights allow for explainability context since it reveals the temporal data that most affect the future runoff levels. This model improves the existing framework of rainfall-runoff modeling systems as well as sets a foundation for incorporating intelligent forecasting tools into smart real-time drainage systems. This advancement allows city authorities and planners to manage flood risks more actively, optimize the use of drainage systems, and practice effective urban water management. The research emphasizes the importance of the application of Transformer-based models in monitoring environmental land systems, which would create new opportunities of development in the area of hydrological AI and smart technologies for cities.
Barhoumi et al. (Fri,) studied this question.