Farmer's can't control the weather, and yet their livelihoods are so dependent on it.

One solution is to improve the accuracy with which we can predict the weather. Microsoft's AI for Earth program and Azure cloud support efforts to apply artificial intelligence (AI) to this challenge.

Managing Microclimates

For farmers, a regional or even local weather report is not precise enough. They need to understand how the weather will impact their farm -- their immediate microclimate. AI for Earth grantees The Yield, SunCulture, and International Crops Research Institute for the Semi-Arid Tropics (ICRISAT) are working solutions for this challenge. 

Australian technology company The Yield uses sensors to measure 12 key factors, including soil moisture, leaf wetness, light, wind, and rain. With AI technology from Microsoft, The Yield conducts predictive modeling to create a seven-day weather forecast for each farm's microclimate. Its mobile app helps farmers use the forecast to determine how, when, and where best to plant, irrigate, protect, feed, and harvest their crops.

Agricultural efforts benefit from deep insights, which AI can generate when comparing current and projected conditions against historical models. Nairobi-based SunCulture's AgOptimized software collects data from weather stations, sensors, and even satellites, then uses Azure-based AI to make recommendations to help farmers plan and optimize planting, irrigation, fertilizing, and pest control.

In another example, ICRISAT worked with Microsoft to develop a sowing app, which optimizes harvests by advising farmers in southern India about the best times to sow. Like many other agricultural analytics solutions, the sowing app combines data sourced from the world's best weather observation systems and applies AI techniques to focus results down to the individual farmer's crops. Unlike many similar tools, however, this sowing app delivers insights to farmers in their native languages via SMS, so they can access the guidance with basic cell phones.

Weather Pattern Recognition

Considering the sheer volume of historical weather pattern and climate data, machine learning in the cloud is a smart way to derive value without incurring prohibitive costs.

Lester Mackey, a statistical machine learning researcher at Microsoft,  teamed up with Judah Cohen and other colleagues to use machine learning techniques to tease out repeating weather and climate patterns from historical data. Their preliminary models, which generated  "subseasonal" forecasts of temperature and precipitation, outperformed standard models used by U.S. government agencies -- even when looking out months in advance. As opposed to general weather models, subseasonal models take into account many more localized and global variables, including daily temperatures, wind speeds, gulf stream wind patterns, and sea ice activity.

Mackey observes that historical weather data, once heavily employed in forecasting, fell away in favor of techniques that analyze how the atmosphere and oceans evolve. Mackey says the forecasts have improved over time, but now make little use of historical data.

Mackey and his team combined current techniques with historical information, winning top honors in forecasting competitions meant to develop significant improvements. "Machine learning is essentially learning from experience," says Mackey. "Even though we can't observe the future, we have many examples of past temperature patterns and the historical features that preceded them. Ample relevant data and the computational resources to extract meaningful inferences are what make machine learning so well-suited to the problem."

Learning from the Most Experienced

The Central Institution for Meteorology and Geodynamics (ZAMG) is the world's oldest meteorological center, with more than 270 stations across Austria. Customers around the globe have depended upon ZAMG to provide up-to-the-minute predictions, 10-day forecasts, and support for data-based forecasting strategies. Leveraging Microsoft's Azure cloud, ZAMG is able to ingest and process 100,000 data records per minute. 

ZAMG is using artificial intelligence and neural networks to predict favorable wind speed and solar energy conditions.

According to Günther Tschabuschnig, chief information officer at ZAMG, "Every single weather station constantly and meticulously monitors weather conditions and transmits all developments to our headquarters in Vienna every second. This creates enormous data sets that need to be processed, analyzed, and finally archived. Even with our traditional organization, technological progress does not stop."

Observes Tschabuschnig: "Digital transformation is changing all of our lives, and we strive to make the best possible use of technology. Big data analytics, modulation or simulation accompany us every day, and thanks to the cloud, we can now better understand and process these topics."