Gro’s Temperature Generative AI Model: Building Resiliency to Data Disruption

As global temperatures reach historic extremes, land surface temperature (LST) becomes an especially critical input when predicting the impact of heat stress on crop planting, development, production, and yield. In July, to complicate matters further, NASA’s two satellites that house MODIS sensors – which detect and measure land surface temperature among many other datasets – are experiencing an outage due to a data lapse from its Adaptive Processing System where data from several NASA satellites is processed for users.

In 2022, after witnessing a series of other satellite data lapses, Gro’s research and development teams created a Gro LST measure to safeguard against outages like the one we’re experiencing now, ensuring Gro’s measurement of land surface temperature is not disrupted by a single-source outage.

“Gro LST” is a generative AI model that incorporates observations from multiple LST and air temperature sources, shares information across them, and delivers a consistent and complete representation of a synthetic LST time-series. The result is an aggregated LST metric that can approximate a valid LST observation every day for all target geographies regardless of data availability from underlying LST and air temperature sources. While datasets ingested from the NASA MODIS sensors remain an input to Gro LST, they are one of many, ensuring Gro LST can continue to provide the same highly accurate LST values to our models. 

Today, Gro LST is a key input to Gro’s yield forecast models, preserving model performance during data disruptions or outages from other temperature sources. 

While these types of outages are relatively short-lived, they can be unpredictable and data lags can last for several weeks. Geospatial data disruptions or delays are caused by a number of factors including faulty satellites, government shutdowns, lags in availability, and even cloud cover. 

Given the current MODIS outage, the contingencies built into Gro LST have proved to be beneficial and we anticipate that there will be no practical impact to the quality of Gro’s yield forecasts.

To learn more about Gro models and methodology, contact a member of our team here.