<img height="1" width="1" style="display:none" src="https://www.facebook.com/tr?id=145304570664993&amp;ev=PageView&amp;noscript=1">

Nov 23, 2021

Environmental AI: how the IPU is helping Agilor watch our water

Written By:

Jiang Zhu

Join the IPU conversation

Join our Graphcore community for free. Get help and share knowledge, find tutorials and tools that will help you grow.

Join on Slack

Understanding our weather and climate has never been more important, as the world seeks to slow temperature rises and manage the impact of human behaviour on our environment.

The sophistication of meteorological models and the rate at which research progresses is hugely dependent on available compute.

Our latest work, with Chinese digital transformation specialists Agilor, demonstrates how Graphcore AI technology is helping to accelerate this vital field.

Agilor researchers achieved a dramatic performance improvement, compared to their existing compute platform, when modelling Evapotranspiration – the rate at which water moves from surfaces, such as soil and plants, into the atmosphere.

The company is now exploring further applications for the IPU across meteorology, natural disaster prevention and mitigation, precision irrigation in agriculture, and rural revitalisation.

Kriging complexity

A location's Reference Evapotranspiration value (ET0) is the product of several meteorological measurements, including temperature, humidity, air pressure and wind speed.

Once the ET0 values are calculated, they can be plotted on a map. However, there is a limit to the number of real-world data points that can be collected.

For a fuller picture, ET0 values in-between measurements locations are interpolated, using a technique known as Kriging, which can be highly computationally demanding.

Agilor’s initial work with the IPU focused on accelerating the Kriging process.

The team used PyKrige, an open-source model that is typically run on CPUs, but is heavily performance constrained by its inability to use multi-core, multi-thread processing.

Their IPU solution overcame these limitations, delivering a 60x performance gain with significant scope for further refinement.

A detailed explanation of Agilor’s work can be found in our technical deep dive blog.

New technology foundation

Agilor now plans to build on its work modelling Evapotranspiration on the IPU, exploring its use in forest fire prevention, precision irrigation in agriculture and management of natural disasters.

Agilor CEO Danny Huang said that it has never been more important to have access to leading-edge AI compute: “Climate change means that our past knowledge and laws no longer work and all of our computation must be short-term and real-time. We will have to deal with more difficult and intensive computation.

“With Graphcore IPU technology, we now have the confidence to address more climate changes in the future through shorter-term predictions and responses.

“This technology has indeed become Agilor’s technology foundation."


Redefining HPC

Agilor’s breakthrough work is the latest successful application of the Graphcore IPU in an area that has previously been dominated by CPUs or GPUs.

This convergence of AI and High Performance Computing (HPC) has accelerated dramatically in the past couple of years, with Graphcore technology being used across meteorology, cosmology, particle physics, and fluid dynamics.

Most recently, Graphcore researchers demonstrated a 5X performance advantage using the IPU compared to the leading GPU, when running a Multi-Layer Perceptron model, developed by the European Centre for Medium Range Weather Forecasting (ECMWF).

Interested in accelerating your HPC and AI-based workloads with cutting-edge compute? Check out our solutions for scientific researchers or apply to our academic programme for the opportunity to use IPUs for your research challenges.