#insights

Artificial intelligence (AI) is being used more and more in water management to improve the efficiency, sustainability and profitability of water supply and sanitation services, as well as for land management when extreme weather phenomena occur, such as floods and droughts.

The availability of large amounts of historical data of different kinds – and the ongoing capture of new data – provides an unbeatable starting point for the application of AI and ML techniques to improve land and water management and associated infrastructures.

Algorithms based on artificial neural networks (ANN) have proven to be effective in solving complex problems for the optimization of water resources planning and management. Some of the benefits of using algorithms include their ability to adapt to non-linear problems, their flexibility to handle multiple objectives, and their ability to handle uncertainty and variability in the system.

These types of neural networks are data-driven models trained based on the input-output relationships of a model based on physics or field measurements. This type of emulator has the advantage that it can reproduce the complex nonlinear behavior between the input and the desired output. This is one of the reasons why emulators are already commonly applied in the field of hydrology. In addition, genetic algorithms can provide a set of optimal solutions instead of a single solution, which can help better understand the trade-offs between different objectives in decision making.

Below, we discuss some of the areas where artificial intelligence is currently being applied, with the aim of improving the efficiency, sustainability and profitability of water supply and sanitation services.

Flood management

Floods are one of the most damaging and frequent natural hazards and are expected to affect more people and infrastructure more severely in the future due to climate change, land use change and population growth. Some of the fields in which the currently described techniques are used are:

• Historical study of floods: In the case of having satellite images of past flooding events, the use of neural networks is very effective for the analysis and processing of images and the delimitation of flooded areas.
• Flood prediction due to atmospheric phenomena: The use of adequate evacuation schemes has the potential to significantly reduce the consequences of a flood event, that is, reduce damage and the number of losses of lives and livestock in areas at risk. Sophisticated two-dimensional (2D) hydraulic models are typically used to simulate floodplain delineation as a decision-making tool. Although these models have been shown to be accurate in predicting flood wave propagation and extents in areas with complex dynamic interactions, these models cannot be used for short-term prediction due to high computational demands and long simulation times. For this reason, emulators (e.g. artificial neural networks) have gained a lot of attention in recent years. In small hydrographic basins, the problem in flood prediction is greatly accentuated due to the short period to transmit accurate warnings in time and the small spatial scale for atmospheric prediction.
• Prediction of failures in flood protection infrastructures: The development of neural networks to predict the locations of failures of containment structures based on real or predicted hydrographs in a very short period (i.e. 1 second) is very important, helping to predict possible flooded areas allowing timely evacuation of people in areas at risk.
• Flood risk management of existing railway infrastructure: The existing railway infrastructure networks must manage the growing challenge of flooding due to the lack of local hydraulic capacity in the system, the degradation of drainage assets, the change in land use or coverage of the contributing basin or the increase in hydraulic load due to climate change. These flooding events cause serious problems for rail transportation, resulting in extensive delays in train services, signaling equipment failure, ballast carryover, and structural destabilization of the track. The development of machine learning (ML) algorithms to analyze historical data of road flooding incidents to identify possible links between rainfall parameters, the condition of drainage assets and road flooding allows for optimal management of assets, improving safety and performance, through more objective proactive interventions.

Water supply: prediction of availability and demand

Likewise, these techniques are useful today when it comes to guarantee water supply, helping to predict water availability and demands.

Water management is a complex issue, especially considering climate change forecasts. Semi-arid regions are characterized by long periods of low precipitation, increasing temperatures, decreasing water resources. Efficient management requires a proper understanding of water availability and is key to drought mitigation. Physics-based models are often used for simulations and predictions of available water, but these methods have a major limitation because they cannot be applied in cases of missing data. However, the use of machine learning techniques allows a positive approach to be obtained with fewer variables than physics-based models. The automated machine learning model for water resources prediction consists of three parts:

• The acquisition, completion, processing, and cleaning of data.
• Performing an intelligent search for the best algorithms and hyperparameters for a specific objective (e.g., water table prediction), based on different autoregressive hydrological behaviors.
• Using the results to train and predict the required horizons (e.g.: 3, 6, 9 or 12 months).

The model is pre-trained using the most up-to-date data (up to the month before the period to be predicted). Using this pre-trained model, a multi-step (up to 12 months) recursive prediction is performed using the input time series data, using the monthly moving average of the past years for explanatory variables other than precipitation. In the case of precipitation, different scenarios are built – low, regular, and high levels of rain. This module is executed every time the input data is updated – ideally, every month.

Predicting drinking water demands for supply network management: A reliable short-term forecast model is essential to properly manage a water distribution system. In recent years, neural networks have gained special attention, especially with the rise of the deep learning approach. Neural network algorithms can evaluate the influence of using different sets of inputs, analyzing many combinations of past observations, temporal/calendar variables and also meteorological variables. The results show that the neural network-based deep learning model provides an effective solution for consumption forecasting.

The proposed methodologies consider simulating all the different combinations between sets of historical observations, meteorological variables (temperature, precipitation, etc.) and temporal variables (month, day, hour, working day/holiday, etc.). In this type of methodologies, historical observation plays a crucial role to train a model efficiently.

Optimizing water treatment processes

The large number of variables that influence the effectiveness of wastewater treatment makes it difficult to predict the outcome of a process that has not yet been tested. However, machine learning (ML) models have been successfully applied in some studies related to wastewater treatment plants to predict effluent concentration from influent loads and operating conditions.

Monitored ML is the most common approach in both the drinking water treatment and wastewater treatment sectors. The main objective of these models is to predict a target variable (output) as a function of a series of variables (inputs), feeding the models with data series that include values of both types of variables. This technology allows us to perceive interactions between different variables that may not be visible to the naked eye. This could help in decision making when designing wastewater treatments, allowing for faster optimization of processes.

AI offers great opportunities to improve the efficiency, safety, and sustainability of Sener’s projects, in areas such as water, aerospace and defence, mobility or energy. But AI also involves great challenges and responsibilities, which is why it is necessary a commitment to use it in an ethical, transparent, and respectful manner.