A strong and stable agriculture sector is important at a national level as a source of food, and beneficial as a source of economic gain through exports. Early prediction of harvest yields at national and regional scales is therefore important as an influencer of commodity trading pricies, a means of ensuring adequate food supplies, managing logistics, and as a component of balance of trade. At a smaller scale, accurate estimates of likely yields allow farmers to better manage budgets, plan seed retention versus sales, assist with insurance cover judgements, and crop management decisions through the latter part of the growing season. Crop yield prediction is possible using remote sensing from satellite imagery, and accurate predictions of wheat yield are shown below to be possible months before harvest.
Crop yield at the end of a growing season is the result of a complex set of factors. Weather (rain, sunshine, and other factors), soil type, retained soil moisture, nutrient availability, applications made, sowing date, and crop variety are just some of the many factors affecting final production figures. Hence, accurately predicting yield is difficult and is less reliable, generally, the further from harvest the prediction is made.
Traditionally, weather and plant-phenology based models were used to predict yields, or simple "growing degree days" estimates served as rough indications that could be compared to eye-based judgements in field. However, there has always been a desire to have more accurate estimates and to have them as early as possible in the growing season.
That requirement from the agriculture community, from our agribusiness customers, individual farmers and corporate customers, drove the development of our wheat yield prediction service. With the backing of the European Space Agency (ESA) and validation from Elders plc over several years covering very different seasons, an on-demand service has been created that lets total area yields to be predicted (e.g. paddock, farm) as well as maps on a 10m x 10m scale. Although total yield is generally all that is sought, the ability to predict at a pixel level what the yield will be across a paddock provides the opportunity for late-season intervention and decision assistance.
Our approach is to fuse weather data with remotely-sensed multispectral imagery (from Sentinel-2) and process from the start of the growing season until the date the prediction is required. Prediction can be performed any time after the crop starts showing, and accuracy increases when closer to harvest. The earliest we offer the service in a growing season is 3 months after planting, and updated estimates are encouraged as the season progresses to allow for evolving conditions, weather and any applications made.
In the example output from our service shown below, a paddock in South Australia illustrates the expected yield at harvest for the 2020 season (left), predicted in late August 2020, and the actual yield map from the harvester in early December 2020 (right).
Although the satellite imagery is lower resolution than the map generated by the header at harvest, sufficient detail is available to aid decision making and to map poorer and better performing areas of the paddock.
By using this service clients:
For more details about our products and services, please get in touch.