Anjie Digital Technology and Graphcore use IPU for weather forecasting, precise irrigation, disaster prevention and mitigation

Anjie Digital and Graphcore work together to use IPU for weather forecasting, precision irrigation and disaster prevention and mitigation

November 23, 2021, Beijing, China - Today, Anjie Digital, a wholly-owned subsidiary of Anjie Zhongke, and Graphcore (to be announced) officially announced a strategic partnership. The two parties will build on the current ET0 scientific computing cooperation project for meteorology, irrigation and disaster prevention, and extend it to jointly create IPU-based integrated solutions for different industries and scenarios. It will be applied to agriculture, industry, meteorology, rural revitalization, disaster prevention and mitigation and other fields, to solve the challenges and pain points faced by these fields in digital transformation and AI transformation, and release the industry's data value on the IPU system.

The initial cooperation between Anjie Digital and Graphcore was a scientific computing project based on the ET0 model on the IPU. The technical teams of both parties used meteorological data to predict the weather on the ET0 model. Compared with the computing platform CPU used by Anjie before, Anjie's technical team achieved a 60.5-fold performance improvement on the IPU, which made large-scale accurate weather forecasting possible. In the next stage, in addition to expanding the scale and time range of meteorological forecasts, with the use of richer data sets, both parties will use ET0 based on evaporation and transpiration for forest fire prevention, precision agricultural irrigation, natural disaster prediction and insurance disaster reduction applications.

Achieving fast and accurate weather forecasting on IPU

The 26th Climate Change Conference held recently once again sounded the alarm for the climate crisis for mankind. Frequent extreme weather events around the world, such as forest fires, droughts, floods, hurricanes, extreme cold and heat, have become the primary issues that all countries need to solve. In August this year, the China Meteorological Administration released the "China Climate Change Blue Book: The Risk of Extreme Weather and Climate Events Has Further Intensified". The blue book shows that with the increase and intensification of extreme weather and climate events such as high temperature and heavy precipitation, my country's climate risk level tends to rise. From 1961 to 2020, my country's extreme heavy precipitation events have shown an increasing trend, and extreme high temperature events have increased significantly since the mid-1990s; from 1991 to 2020, my country's average climate risk index (6.8) increased by 58% compared with the average value (4.3) from 1961 to 1990. This year's rainstorm disaster in Henan is a typical example of extreme weather events.

Advanced AI technology and scientific computing are being explored to deal with such extreme weather events. Among many meteorological elements, Anjie Digital Technology has selected ET0 (Reference evapotranspiration) based on evaporation to make accurate weather forecasts. ET0 interpolates and calculates weather elements including temperature, humidity, air pressure and wind speed, which is of great value to meteorological data mining, evaluation of numerical weather forecasting level and climate change and other related research.

Anjie Zhongke CEO Huang Siyuan

Huang Siyuan, CEO of Anjie Zhongke, said: "In the ET0 project, our biggest challenge is that the basic data we get is sparse and inaccurate, and sometimes the intermediate parameters are missing. In this case, we need a lot of data cleaning and data governance work, which requires a lot of computing power. The main body of ET0 calculation is the Kriging interpolation algorithm. We have to interpolate one square kilometer through more than 4,000 Chinese meteorological stations. Our country is approximately 10 million square kilometers, so we need 10 million points. To interpolate 10 million points from 4,000 points is either inaccurate or the calculation workload is very large. We also assisted with topographic data to obtain more accurate interpolation. Therefore, the calculation workload is even greater. In the end, we chose the new computing power represented by Graphcore IPU to help improve our computing speed. On IPU, we have currently achieved a 60.5-fold increase in the speed of a single algorithm."

"Climate change means that our past knowledge and laws no longer work, so all our calculations must be oriented towards short-term, real-time calculations. This places high demands on the difficulty and intensity of the calculations. Graphcore's IPU technology gives us the confidence to cope with more climate change in the future and achieve shorter-term predictions and responses. This technology has indeed become the foundation of Anjie Digital Technology. This is why we hope to establish a long-term strategic partnership with Graphcore and apply IPU-based solutions to other fields to benefit the public." said Huang Siyuan.

The above figure shows the results of using ET0 on the IPU for the meteorological data throughout 2015. The change in color from blue to red represents the change in ET0 value from low to high. The color difference in different areas directly reflects the difference in the statistical values ​​(highest/lowest/average) of various climate elements (air temperature, humidity, wind speed, atmospheric pressure) on that day. The historical/forecasted ET0 value of each point can be used for the identification and statistics of extreme weather in that year and the prediction of crop irrigation volume in the future.

Precision agriculture irrigation using ET0 on IPU

In addition to meteorology, evapotranspiration is also widely used in the field of irrigation. To achieve precision irrigation, the only technical difficulty that needs to be overcome is to provide real-time and future forecasts of evapotranspiration covering the whole country. Anjie Digital has cooperated with the National Meteorological Information Center to obtain historical and real-time data from meteorological stations across the country, and calculate the evapotranspiration measured in square kilometers across the country through interpolation.

"This is a huge amount of calculation. The 60.5-fold performance improvement on the Graphcore IPU means that we can provide a real-time technology for all 7 to 15-day ET0 calculations with an accuracy of one square kilometer. This improvement is actually huge for us, allowing us to provide real-time control for all irrigation controllers in the future. This has a far-reaching impact on China's entire irrigation water consumption." said Huang Siyuan.

Applying IPU-powered innovation to natural disaster prevention

Changes in evapotranspiration can predict potential natural disasters. For example, if soil and vegetation evapotranspiration decreases, the risk of forest fires will increase. If a region has abnormal evapotranspiration in the near future, we should be alert to extreme rainfall or drought.

Landslides and mudslides can occur in some soil environments, but not in other areas. Generally speaking, disasters are less likely to occur in areas with good plant coverage. In addition to the obvious reason that plant roots restrict soil movement, the latest calculations of Anjie Digital show that in fact, plant evaporation and transpiration bring a lot of water from the soil to the air, while also reducing the risk of slope landslides. With the excellent performance of IPU, Anjie Digital will be able to conduct faster and more innovative research.

Lu Tao, President of Graphcore Greater China and Global Chief Revenue Officer

Lu Tao, President of Graphcore Greater China and Global Chief Revenue Officer, said: "Anget Digital is a highly innovative Chinese company that applies data technology and data science to different industries and social fields. The cooperation between Graphcore and Anget Digital will enable us to take root in China and serve China more deeply, and apply IPU technology to agriculture, meteorology, natural disaster prediction and social services, and help the digital construction of infrastructure and the transformation of industrial AI."

"We will continue to actively work with customers and ecological partners including Anjie Digital, and strive to apply IPU technology to AI to benefit the people, AI to benefit farmers, and AI to develop businesses, to benefit society and the people, and to help my country's industrial digital transformation." Lu Tao said.