Crowdsourced supercomputer helps TU Delft researchers predict local rainfall in Africa

Nieuws - 04 november 2019

Volunteers of IBM's World Community Grid are helping TU Delft map sub-Saharan rainfall with unprecedented detail for an entire rainy season. This could yield a new forecasting methodology that benefits small farms. The start of this one-year crowdsourcing endeavour is especially exciting for TU Delft Global fellow Camille le Coz, who has prepared the numerical weather model that will be tested with the help of the Grid’s crowdsourced supercomputing power.

In Africa, agriculture relies heavily on localised rainfall, which is difficult to predict. Satellite data, information from ground stations and weather models can be used for predicting local rainfall by making detailed calculations of the atmosphere. In collaboration with the Trans-African Hydro-Meteorological Observatory (TAHMO), researchers at TU Delft will start making these calculations this week by using crowdsourced computing power through the IBM World Community Grid.

Erratic

“African rain storms are erratic and vary strongly from place to place, while many Africans rely on rain for their livelihood. If we want to come to grasps with rain in Africa, we have to be able to make very detailed calculations of the atmosphere” says Prof. Nick van de Giesen from TU Delft. For detailed and reliable weather predictions, large and varied data are required. Satellite observation provides weather information, but rainfall needs to be measured on the ground as well. The satellite may well spot clouds but it will still be unclear if it is raining, and if it is, how much rain is falling.

Van de Giesen: “That’s why ten years ago, we started the Trans-African Hydro-Meteorological Observatory (TAHMO). To obtain rainfall data from the ground, we put up low-cost technology weather stations and started accumulating ground data. We now have weather stations in Ghana, Kenya and Uganda and fieldwork projects are under way.”

Grid size

Present weather models that are run for Africa have a grid-size of tens of kilometers, implying that a complete storm fits in one calculation cell. This is like having one pixel for a complete face, one cannot make out what happens inside. Together with the poor data situation on the ground, this makes that rainfall predictions in Africa for tomorrow are as (un)reliable as ten day rainfall predictions in Europe. Ideally, one would like to run a weather model for the whole continent at a scale of one kilometer in order to resolve the individual rainstorms.

Super computing power

"Such calculations would not be possible with standard computers, but IBM’s World Community Grid makes this possible. By dividing the continent into many small areas, each area sufficiently small to be calculated by a single computer, rainfall throughout Africa can be determined with the necessary detail” says PhD candidate Camille Le Coz, who works on the project from the TU Delft side.

IBM’s World Community Grid volunteers assist in research projects by having their device perform research calculations when it’s idle. Within the context of this research, this means that, for the first time, weather over the complete African continent will be calculated at a fine enough scale to simulate individual rainstorms. Researchers at IBM and TU Delft have spent substantial effort to prepare a numerical weather model for the World Community Grid. The project will run for one year after which the researchers will know how much better weather predictions become when sufficient computational power is available.

/* */