Chapter 12 - Artificial intelligence/machine learning techniques in hydroclimatology: A demonstration of deep learning for future assessment of stream flow under climate change

Abstract

Recent availability of large hydrometeorological data, both observed and simulated, offers a new set of challenges as well as a vast opportunity in the research domain of climate and hydrological sciences. The artificial intelligence (AI)/machine learning (ML) has a potential to extract hidden associations between climatological/meteorological forcings and hydrological responses. Thus, AI/ML remains a good choice to handle big data in such hydroclimatological/hydrometeorological studies. Deep learning (DL) techniques, a subset of AI/ML, are recently being popularized due to its ability to capture the highly complex hidden association, and thus, may be an even better choice for more reliable modeling and forecasting. This chapter highlights the potential of AI, ML, and DL techniques in the aforementioned studies. We pick out artificial neural network, decision tree, random forest, K-nearest neighbor (KNN), support vector regression (SVR), convolutional neural network, long short-term memory (LSTM), etc., for discussion. Toward the end, we demonstrate an illustrative example of modeling the nonlinear relationship between the meteorological forcings and stream flow variations for a rainfed river basin. One of the popular DL techniques, namely LSTM neural network is used and its performance is compared against a basic multiple linear regression model and a popular ML approach, namely SVR. Furthermore, using the developed DL-based LSTM model, the future assessment of stream flow is carried out to quantify possible impact of climate change, designated through shared socioeconomic pathways. Simulations from six different general circulation models are used to demonstrate as to how the streamflow may vary in future. The assessment will be useful in planning and management of water resources in different sectors, such as irrigation, hydropower generation, drought management, etc. Overall, the potential of AI, ML, and DL in this domain is enormous and can be beneficially utilized in many studies for different study regions.