Quenching Belgrade's thirst

The LizzaPAKP Graphical User Interface showing a regional model overview. Highlighted points denote the places where the water level is measured. Image courtesy Milos Ivanovic.

Belgrade may be growing, but the city’s groundwater source is not. Almost 900 million people worldwide do not have access to clean water, according to the UN. Managing Belgrade’s water resources to avoid augmenting that statistic is a complex task, one that researchers in Serbia are tackling with the help of grid and high-performance computing-enabled models of the local water system.

The Belgrade groundwater source, also known as an aquifer, consists of 139 wells located on the Sava River, one of the longest rivers in Europe. The common problem aquifers such as this have are in the aging of its wells, which has a detrimental effect on the water supply.

To help with this problem, researchers at the Jaroslav Cerni Institute for the Development of Water Resources, in Belgrade, led by developer Velibor Isailovic from the University of Kragujevac in Serbia,  developed a 3D numerical application called LizzaPAKP. The application helps urban planners identify the capacity of the Belgrade groundwater source, study the causes of water-well aging, predict water flow, and propose an efficient plan for well regeneration.

LizzaPAKP is made up of two parts. The first part, Lizza, is a Graphical User Interface that enables an end user to build a 3D model. Lizza models can simulate a variety of key factors ranging from terrain contours to rock properties, such as porosity (a measure of how porous a material is).

The second part, PAKP, is a numerical solver, which uses Darcy’s law of flow. This is a geological law that describes the rate at which a fluid flows through a permeable medium. Darcy's law states that this rate is both directly proportional to the drop in vertical elevation between two places in the medium and indirectly proportional to the distance between them. In the case of Belgrade’s water supply, Darcy’s law can be used to predict how water flows between the system’s many wells.

To properly simulate Belgrade’s water system, the researchers and planners needed to run a large number of independent compute jobs – something ideal for grid computing. In 2008, the LizzaPAKP application, originally used on single desktop computers, was ported to the grid.

“One of our recent examples was running a large number of independent finite element simulations to find leakage paths at the bottom of the dam. And to understand the dam’s energy production under various and changing circumstances,” said Milos Ivanovic, a researcher from the University of Kragujevac.

Not all of their simulations are suited to grid computing, however.

“We use high-performance computing clusters in cases where we have to deal with extra-large finite element models, which require large memory and processing power, as well as closely-coupled hardware resources,” Ivanovic said.

The LizzaPAKP GUI showing a detailed model of the potential isolines (contour lines) for a radial-type-well with 4 laterals. Image courtesy Milos Ivanovic.

Users of LizzaPAKP at the Jaroslav Cerni Institute calculated the amount of time it would take to exhaust a specific well according to the flow rate from other wells. This enabled engineers to install new canal laterals at specific locations to sustain these ageing wells. They are also planning future investments in the placement of new canal laterals to maximize the amount of water output.

Ivanovic and his colleagues published a research paper on their work last year in the Journal of Hydrology and are currently working on more papers about how grid and evolutionary algorithms can be applied to hydrology problems.

Ivanovic and his colleagues aren’t done with LizzaPAKP yet, however. There is ongoing work to keep the application user friendly, for example.

“Our current users are involved in the development process, submitting bug reports and requesting new features,” Ivanovic said. “The entire application, especially its graphical user interface, is designed the way they wanted it to be.”

The current challenge that Ivanovic is working on is managing the large amount of post-processing and visualization data that LizzaPAKP transfers to a user’s desktop after each grid job is run.

One solution would be the use of novel web applications to take some of this data load off a user’s computer and onto web servers instead. But there are other solutions to explore.

“We are experimenting with elastic grid job schedulers like Work Binder, developed during the SEEGRID-2 project,” Ivanovic explained. “[The schedulers] turned out to be a big improvement regarding ease of use and the rate of grid resource utilization.”