Precise and reliable yield forecast tools could play a foundational role in supporting policy formulation and decision-making processes in agriculture. For farmers, reliable yield forecasts might represent a source for considering changes in management practices during the reproductive phase of the field crops.
For this study, a final yield forecast tool was developed by utilizing satellite imagery of current corn growing conditions across the state. The main steps to forecast yields were:
1) Data collection (2009-2017 period)
a) USDA-National Agricultural Statistics Service (NASS) county-level data was collected to determine yield distributions at the district level.
b) MODIS (820-ft x 820-ft pixel size) satellite information with an image every 8 days until the beginning of August
c) CropLand Data Layer or spatial distribution of the different crops throughout the state
2) Building and Validation of Yield Forecasting Models (YFM)
a) For building the models, NDVI (Normalized Difference Vegetation Index) was utilized, collected from corn producing regions within a county.
b) Yields at the county-level were estimated before harvest from satellite information and compared against the final yields reported by USDA-NASS at the county-level.
3) Building and Validation Layer for Corn 2018
Information from the CropLand Data Layer is available for corn in Kansas from 2006 until 2017 – but current corn locations across the state are not available for the 2018 season until next year. Therefore, a complex statistical technique was employed (random forest prediction) to predict corn geo-locations across the state for the current season.
Figure 1. Geolocation for corn fields predicted by the Random Forest classification model for 2018.
4) Validation of previous years
Forecasted yield at the state-level from previous years (2009-2016) was validated by comparing those yields with the August yield estimation and the final yield from USDA-NASS. The yield forecasted via the satellite model was quite precise in predicting the yields from past-seasons.
5) 2018 Kansas Corn Yield Forecast
As a last step on the yield estimation, the yield and crop map models previously built were utilized to forecast the 2018 corn yield for Kansas. Satellite information from planting until the present was aggregated to provide a more reliable yield prediction.
Based on the satellite yield model developed by our team, the state-level yield prediction will be 130 bushels per acre (termed in Figure 2 as “Sat. October”), which is close to both the August and September yield estimations of 129 and 131 bushels per acre released by USDA-NASS.
Figure 2. Forecasting corn yields derived from satellite data for the state of Kansas.
As a last step, we are currently working on a providing yield estimates at county-level for our next yield forecast, so stay tuned for further details on this coming release.
Ignacio A. Ciampitti, Crop Production and Cropping Systems Specialist
Rai Schwalbert, KSUCROPS Production, Dr. Ciampitti’s Lab
Luciana Nieto, KSUCROPS Production, Dr. Ciampitti’s Lab