The use of remote sensing to derive maize sowing dates for large-scale crop yield simulations

Ehsan Eyshi Rezaei,Stefan Siebert,Olena Dubovyk,Gohar Ghazaryan,Javier González,Natalie Cornish

doi:10.1007/s00484-020-02050-4

Abstract

One of the major sources of uncertainty in large-scale crop modeling is the lack of information capturing the spatiotemporal variability of crop sowing dates. Remote sensing can contribute to reducing such uncertainties by providing essential spatial and temporal information to crop models and improving the accuracy of yield predictions. However, little is known about the impacts of the differences in crop sowing dates estimated by using remote sensing (RS) and other established methods, the uncertainties introduced by the thresholds used in these methods, and the sensitivity of simulated crop yields to these uncertainties in crop sowing dates. In the present study, we performed a systematic sensitivity analysis using various scenarios. The LINTUL-5 crop model implemented in the SIMPLACE modeling platform was applied during the period 2001–2016 to simulate maize yields across four provinces in South Africa using previously defined scenarios of sowing dates. As expected, the selected methodology and the selected threshold considerably influenced the estimated sowing dates (up to 51 days) and resulted in differences in the long-term mean maize yield reaching up to 1.7 t ha−1 (48% of the mean yield) at the province level. Using RS-derived sowing date estimations resulted in a better representation of the yield variability in space and time since the use of RS information not only relies on precipitation but also captures the impacts of socioeconomic factors on the sowing decision, particularly for smallholder farmers. The model was not able to reproduce the observed yield anomalies in Free State (Pearson correlation coefficient: 0.16 to 0.23) and Mpumalanga (Pearson correlation coefficient: 0.11 to 0.18) in South Africa when using fixed and precipitation rule-based sowing date estimations. Further research with high-resolution climate and soil data and ground-based observations is required to better understand the sources of the uncertainties in RS information and to test whether the results presented herein can be generalized among crop models with different levels of complexity and across distinct field crops.

Highlights

According to household surveys, more than 13 million people have limited access to food in South Africa (BvenuraManagement decisions, such as the selection of sowing dates, can considerably affect a summer crop’s yield in semiarid regions (Aguirrezábal et al 2009)
Low precipitation thresholds resulted in relatively similar sowing dates for all studied provinces at day 320 (Fig. 3a), while the estimated sowing date differed among the studied provinces when using high precipitation thresholds
The earliest remote sensing (RS)-derived sowing dates were obtained for Mpumalanga (290–326) (Fig. 3b), and the RS-based sowing dates for North West and Free State were 11 days earlier than the sowing dates estimated for the other provinces (Fig. 3b)

Summary

Introduction

More than 13 million people have limited access to food in South Africa Management decisions, such as the selection of sowing dates, can considerably affect a summer crop’s yield in semiarid regions (Aguirrezábal et al 2009). Shifting the maize sowing date in the dry regions of South Africa by only 15 days could increase crop yield by 10% (Abraha and Savage 2006). The effects of management practices, such as the effect of the optimized sowing date on crop yield, are generally assessed by conducting field experiments (Chen et al 2011). Testing the suitability of management practices such as optimized sowing dates and cultivar selection is a challenging issue at larger scales (Therond et al 2011)

Objectives

Methods

Results

Conclusion