Title : Yield prediction of maize crop grown under irrigated and rainfed conditions using remote sensing
Abstract:
Maize is considered as a staple food supporting over 200 million people in developing countries. Close to 60% of the cropping area in Southern Africa is planted with maize of which half of it comes from South Africa making the country a prominent supplier of maize in the region. Of the three main provinces where maize is grown in South Africa, the Free State Province accounts for the largest portion (44.3%) of the total 15.8 metric tons produced in the country in 2020/2021 growing season followed by Mpumalanga (22.4%) and North West (17%) provinces. Maize is predominantly cultivated as a rainfed crop in the semi-arid parts of South Africa, where climatic variability has a direct effect on maize production arising from rainfall and temperature variation which results in seasonal shift of crop growing period. Maize irrigation has also been on the rise with 241 000 ha recorded in 2015/2016
growing season. It is projected that the global warming will likely reach 1.8 ?C under low greenhouse gas emission scenario according to the IPCC. Some studies reported a significant increase of temperature by 0.13 °C per decade in the period from 1960 and 2003 in South Africa. The unabated global warming is therefore, increasingly becoming of a great concern for agricultural production with significant yield reduction already reported in the world. For every 1?C increase in global mean temperature, maize production is indicated to decrease by 7.4% and recurring drought incidences and climatic variabilities are often the main factors responsible for the reduction. It is, therefore, important to effectively predict crop yield before harvest to facilitate decision on whether or not to import maize to ensure the national food security as well as to bench mark market prices.
The Agricultural Research Council-Natural Resources and Engineering (ARC-NRE) together with the research partners, the Institute for Geodesy and Cartography (IGIK) in Poland investigated the forecast of crop yield using the multi-spectral MODIS satellite data, which consisted of 8-Day L3 Global 250-meter (m) surface reflectance (MOD09Q1) for NDVI estimation as well as the combined Fraction of Photosynthetically Active Radiation (FPAR), and Leaf Area Index (LAI) product (MCD15A3H) with 500-meter spatial resolution (pixel size). Maize yield data for white and yellow maize cultivars was obtained from 5766 maize fields cultivated under rain-fed and irrigation system, in the Free State, Mpumalanga and North West provinces during the 2018/2019 and 2019/2020 growing seasons. The data were statistically analyzed and tested for significant differences among all three provinces and
within the two maize varieties. A Random Forest (RF) model was used to predict maize yield using the MODIS satellite data. The results showed that there is a strong correlation between the actual field yield data and the maximum NDVI, LAI and fPAR during the mid-season (November – December) of the maize growing period with a strong to moderate correlation coefficient of 0.8, 0.7 and 0.6, respectively.
The Random Forest algorithm was able to predict the yield from the multispectral MODIS satellite data with a high accuracy and R2 value between the predicted and actual yield of 0.85.
Acknowledgement: This work is based on the research supported wholly by the National Research Foundation of South Africa (Grant Numbers: 118679).
