From flood-prone area risk assessment to pollution dispersion analysis.

Anticipate, predict and trigger alerts with lead-in times

Models are simplified representations of real-world processes, and they have proved to be very effective tools for planning and decision-making. There are two model types. Mechanistic models implement the physical equations of real-world processes to determine the relationships between inputs and outputs. Data-driven models use statistical techniques to determine those same relationships. Hybrids are a combination of both types of models.

Which type of model is most effective depends on its usage. With well-known physical processes, mechanistic models generally excel. ADASA's experts, from hydraulic engineers to chemical engineers, effect mechanistic approaches via software packages that first build the model, then iteratively calibrate and test it until the model behaviour is close to the mimicked system.

Data-driven is apt when system process knowledge scarcity combines with high-quality data availability. Our data scientists implement data-driven models that make use of libraries of statistical methods. These analyse, clean, and process the data, then construct several models and validate them, finally choosing the most accurate one.

Hydrological model

ADASA has a long history of model implementation. Examples include:

  • Flood Early Warning Systems (FEWS) use rainfall predictions as inputs alongside catchment rainfall runoff and river hydraulics models for flooding forecasting and to predict river flows.
  • Flood analysis of multiple scenario planning purposes such as return period, land use, and more.
  • Water pollution dispersion analysis.
  • Decision Support Systems for irrigation schemes use a combination of river, open-channel and pressurised pipes network models to recommend optimum operation of reservoir valves, canal regulator structures and pumping stations.
  • Water distribution network models calibrated with telemetry data for operational purposes.
  • Models to automatically detect and correct the flow time series.
  • Water demand forecasting systems based on historical data. Impact assessment of variations in water demand patterns.

Our specialists develop complex, hydrological and hydraulic models that continuously simulate all the processes of a basin’s hydrological cycle with great detail and variability, even in large basins. We use distributed efficient models based on physical processes such as Tetis, Topkapi, SHE or Topmodel as well as simpler aggregate or semi-distributed models such as HEC-HMS or NAM, and even conceptual and empirical models.

When we build operational tools like Flood Early Warning Systems or Decision Support Systems, the modelling software packages run automatically through online forecast and warning platforms such as Delft-FEWS. Gridded datasets are typical inputs for these systems, along with numerical weather predictions and radar outputs, and spatially distributed data like rain gauges.

The next phase sees all datasets validated. Data with disparate spatial and temporal scales are transformed (e.g. spatial interpolation to derive areal precipitation), we run the catchment runoff that predicts the hydrographs, which allows the hydraulic model to run and predict flood areas.

Our data scientists have experience using Python with various libraries, including NumPy, SciPy, Pandas, matplotlib, sci-kit learn and TensorFlow. They use these to build data-driven models through supervised learning techniques — e.g., multiple linear regression, logistic regression, random forests, and deep neural networks — for prediction or classification use cases. They also utilise unsupervised learning techniques — such as k-means and multivariate Gaussian distribution — for clustering use cases such as anomaly detection.

ADASA has extensive experience in the development of hydraulic models for pressurised water distribution networks (EPANET and Bentley WaterGEMS) and also for dam operation, as well as 1D/2D transient flow models. These simulate the evolution of a flooding event in a river: models include HEC-RAS, MIKE, IBER, SOBEK, ISIS, Floodway or Dambreak.


  • Improved decision-making thanks to high confidence levels in prediction outcomes.
  • Better planning via analysis of multiple ‘What if?’ scenario outcomes.
  • More efficient maintenance procedures.
  • Cost savings.

Related Solutions

See our work


Contact our experts