Predicting Wheat Yield at the Field Scale by Combining High-Resolution Sentinel-2 Satellite Imagery and Crop Modelling

Yan Zhao,Graeme L Hammer,Miao Zhang,Bingfang Wu,Andries B Potgieter

doi:10.3390/rs12061024

Abstract

Accurate prediction of crop yield at the field scale is critical to addressing crop production challenges and reducing the impacts of climate variability and change. Recently released Sentinel-2 (S2) satellite data with a return cycle of five days and a high resolution at 13 spectral bands allows close observation of crop phenology and crop physiological attributes at field scale during crop growth. Here, we test the potential for indices derived from S2 data to estimate dryland wheat yields at the field scale and the potential for enhanced predictability by incorporating a modelled crop water stress index (SI). Observations from 103 study fields over the 2016 and 2017 cropping seasons across Northeastern Australia were used. Vegetation indices derived from S2 showed moderately high accuracy in yield prediction and explained over 70% of the yield variability. Specifically, the red edge chlorophyll index (CI; chlorophyll) (R2 = 0.76, RMSE = 0.88 t/ha) and the optimized soil-adjusted vegetation index (OSAVI; structural) (R2 = 0.74, RMSE = 0.91 t/ha) showed the best correlation with field yields. Furthermore, combining the crop model-derived SI with both structural and chlorophyll indices significantly enhanced predictability. The best model with combined OSAVI, CI and SI generated a much higher correlation, with R2 = 0.91 and RMSE = 0.54 t/ha. When validating the models on an independent set of fields, this model also showed high correlation (R2 = 0.93, RMSE = 0.64 t/ha). This study demonstrates the potential of combining S2-derived indices and crop model-derived indices to construct an enhanced yield prediction model suitable for fields in diversified climate conditions.

Highlights

Wheat (Triticum aestivum L.) is the 5th largest staple crop consumed by people worldwide [1]
These values varied between the two years and reflected the crop stress experienced by the crop at flowering and grain filling due to water limitation
77% of the rainfall occurred during the non-cropping period, while extremely low rainfall occurred during the booting to flowering period in 2017 (Figure 2)

Summary

Introduction

Wheat (Triticum aestivum L.) is the 5th largest staple crop consumed by people worldwide [1]. Significant volatility exists in wheat production due to large fluctuations in yield and area planted across the broad cropping region of Australia, often associated with the high climatic variability arising from El Niño or La Niña events [4]. Within this context, crop prediction has proved to be a valuable tool in assisting decision makers at a range of scales to improve yield outcomes and enhance profitability [5,6]. Crop productivity influences the entire supply chain so that accurate and timely monitoring of crops and early prediction of yield are crucial for crop management, marketing, insurance and financial decisions. Information on possible crop failures and potential yield reductions could aid policy and decision making aimed at moderating consequences of serious production shortages

Objectives

Results

Discussion

Conclusion