When satellite yield prediction tools started showing up at field days a few years back, I was skeptical. Another piece of ag tech promising to revolutionise farming while probably just complicating things.

But when a Horsham grain grower I know decided to trial one of these systems, I followed along. Two seasons later, here’s what actually happened.

The Setup

The farm runs about 2,500 hectares of wheat, barley, and canola across several properties west of Horsham. Good country, but variable—some patches are consistently strong, others struggle regardless of what you do.

The satellite system promised to predict yield variability at paddock level, helping with harvest logistics and marketing decisions. No upfront hardware cost, just an annual subscription based on area covered.

Setting it up meant defining paddock boundaries in the platform and providing historical yield data from the header monitor. The vendor processed satellite imagery automatically and delivered predictions through a web dashboard.

Season One: The Learning Curve

First season was mixed. The predictions weren’t bad for the main paddocks—within about 15% of actual yields on most fields. But some outliers were way off, particularly on country that had been recently developed or where irrigation had changed the game.

The platform struggled with anything that didn’t have at least three years of consistent cropping history. It basically needed to learn each paddock before it could predict accurately.

The grower found the within-paddock variability maps more useful than the total yield predictions. Seeing where crops were stressed two months before harvest helped prioritise areas for the header crew.

Season Two: Better, But Not Perfect

Second season showed improvement. Predictions on established paddocks came within 8-10% of actual, which is genuinely useful for forward contracting decisions.

The grower started using the data for logistics planning—which silos to prepare, how much road freight to book, when to start harvest on which blocks. Having a reasonable yield estimate weeks before harvest helped smooth out the usual chaos.

Still not great on new country or paddocks with unusual conditions. The algorithm doesn’t understand why a particular patch might perform differently—it just sees historical patterns.

What It Actually Cost

The subscription ran about $8 per hectare annually. For 2,500 hectares, that’s $20,000 a year.

Was it worth it? The grower reckons the logistics improvements and better marketing timing saved probably $25,000-30,000 across the two seasons. Not a massive margin, but positive.

The real value might come from building multi-year datasets that improve decision-making over time. Still too early to judge that properly.

The Honest Assessment

Satellite yield prediction isn’t magic. It works best on consistent country with good historical data. It’s less useful on variable or newly developed land.

The technology is improving year over year. What was a novelty five years ago is becoming more practical. But it’s not a replacement for ground-truthing, walking crops, and knowing your country.

For larger operations with tight margins on logistics and marketing, probably worth a look. For smaller farms or highly variable country, the return mightn’t justify the cost yet.

The grower is continuing the subscription. Not because it’s transformational, but because it’s one more data point in a business that needs every edge it can get.