Prediction of multi-year winter wheat yields at the field level with satellite and climatological data

Michael Marszalek, Marco Körner, Urs Schmidhalter

Research output: Contribution to journalArticlepeer-review

19 Scopus citations

Abstract

Long-term yield mapping is key to Precision Farming and needs to consider yield-relevant factors such as the water demand and nitrogen uptake. For the prediction of winter wheat yields, we incorporated climatological data and daily crop water requirements (CWR), Sentinel-2 time series, and the derived indices such as the normalised difference red edge (NDRE), red edge inflection point (REIP) and the normalized difference water index (NDWI). These factors were evaluated by using stepwise linear regression (LR) and random forest (RF). Robust field data sets from 2016 to 2018 with weighed yields from three regions in southern Germany in the range of 49 dt ha−1 to 102 dt ha−1 were used for calibration and validation. Indices reflecting the field variability were not better suited for yield prediction than climatological data alone. A new approach using all raw bands of Sentinel-2 in combination with evapotranspiration and precipitation, delivered a reduced number of features and explained 84% of the yield variance with a RMSE of 5.6 dt ha−1. Alternatively, the NDWI-based CWR merged all significant parameters into one feature and explained 79% of the variance with a RMSE of 6.46 dt ha−1. A sufficiently precise prediction of yields at the field level could contribute to an optimised agricultural management.

Original languageEnglish
Article number106777
JournalComputers and Electronics in Agriculture
Volume194
DOIs
StatePublished - Mar 2022

Keywords

  • Crop Water Requirement
  • Machine Learning
  • Nitrogen
  • Sentinel-2
  • Winter Wheat
  • Yield prediction

Fingerprint

Dive into the research topics of 'Prediction of multi-year winter wheat yields at the field level with satellite and climatological data'. Together they form a unique fingerprint.

Cite this