Forecasting winter wheat yields using MODIS NDVI data for the Central Free State region

Abstract

Consumption of wheat is widespread and increasing in South Africa. However, global wheat production is projected to decline. Wheat yield forecasting is therefore crucial for ensuring food security for the country. The objective of this study was to investigate whether the anthesis wheat growth stage is suitable for forecasting dryland wheat yields in the Central Free State region using satellite imagery and linear predictive modelling. A period of 10 years of Normalized Difference Vegetation Index data smoothed with a Savitzky–Golay filter and 10 years of wheat yield data were used for model calibration. Diagnostic plots and statistical procedures were used for model validation and assessment of model adequacy. The period 30 days before harvest during the anthesis stage was established to be the best period during which to use the linear regression model. The calibrated model had a coefficient of determination of 0.73, a p-value of 0.00161 and a root mean squared error of 0.41 tons/ha. Residual plots confirmed that a linear model had a good fit for the data. The quantile-quantile plot provided evidence that the residuals were normally distributed, which means that assumptions of linear regression were fulfilled and the model can be used as a forecasting tool. Model validation showed high levels of accuracy. The evidence indicates that use of Moderate Resolution Imaging Spectroradiometer data during the anthesis growth stage is a reliable, cost-effective and potentially time-saving alternative to ground-based surveys when forecasting dryland wheat yields in the Central Free State.

Wheat Yield Prediction with Machine Learning based on MODIS and Landsat NDVI Data at Field Scale

“…These periods include the flowering and grain filling periods within the phenological period of the crop. These results are consistent with studies reporting a high correlation between NDVI and yield during flowering and grain filling (Mkhabela et al, 2011, Mashaba et al, 2017, Panek and Gozdowski, 2020. The most suitable period for wheat yield estimation in the study area is flowering and grain filling periods, which corresponds to 4-6 weeks before harvest.…”

“…There are several studies in the literature for wheat yield estimation using linear modeling through MODIS satellite imagery. Mashaba et al (2017) found a high correlation (R 2 : 0.73) between yield and NDVI during flowering, while Lopresti et al (2015) North of Buenos Aires province, Argentina showed a good correlation (R 2 :0.52) between wheat yield and MODIS-NDVI 1 month before harvest during flowering period. A similar result was given in a national study in the USA in the relationship between NDVI data obtained from 250 m spatial and 16 day temporal resolution MODIS images and yield (R 2 : 0.51) for the period between the end of May and the beginning of June, when the vegetation reached its peak level (Jahnson, 2016).…”

“…These empirical models are preferred because they require less data and are easy to apply on a regional scale. In the relationships established between VI and yield, different types of variables were used, such as periodic value (Lopresti, 2015), average value (Boken and Shaykewich, 2002;Mkhabela et al, 2011), cumulative value (Ren et al, 2008;Mashaba, 2017;Mirasi et al, 2019;Panek and Gozdowski, 2021) and maximum value (Becker-Reshef et al, 2010). It has been determined that seasonally integrated VIs can be predicted more accurately than a single period and have a high correlation with yield around the maximum time VI peaks (Guo et al, 2021;Ji et al, 2022).…”

Quantifying war-induced crop losses in Ukraine in near real time to strengthen local and global food security

“…For Argentina, Lopresti et al (2015) find an increase of yield of 109 kg/ha with each additional point of peak NDVI as measured by MODIS data (R 2 of 0.53). For10 seasons in the 2000-2013 period from South Africa's Free State, Mashaba et al (2017) find that MODIS-based NDVI values from up to 30 days before anthesis have the highest predictive power (with an R 2 of 0.73) with a 0.1 point change in the NDVI predicted to increase wheat yield by 1.21 t/ha (from a mean of about 2.2 t/ha).…”

A Spatialization Method for Grain Yield Statistical Data: A Study on Winter Wheat of Shandong Province, China

“…Lots of studies have explored the empirical relationship between NDVI and crop yield (Mkhabela et al, 2005;Huang et al, 2014;Mashaba et al, 2017). However, different studies have used different NDVI variables, such as original NDVI, average or accumulated NDVI over the growth period, average or accumulated NDVI over key growing stages, etc.…”