Using regional normalized difference vegetation index for the large-scale yield prediction of potato, vegetables, fruits, and berries, cultivated in Kherson region of Ukraine

Abstract

Remote sensing is a universal tool for various agricultural applications, e.g., environmental monitoring, crop mapping, vegetation cover monitoring, drought events monitoring, disease and insects spreading, floods and desertification scales, etc. (Teke et al., 2013). Today it is one of the irreplaceable constituents of global food security provision, as well as ecological safety (Karthikeyan et al., 2020). It is also widely implemented for irrigation and fertilization management, prediction of yields in plant science, using different spatial vegetation indices as predictors. Yield prediction is an important part of current agricultural science because it is crucial for rational agricultural planning, adjustment of agrotechnology’s, plant products supply forecasting and provision of food security. Crop yield prediction should be an additional instrument in the field of agrarian economy, supporting strategic planning of plant production and food import-export policy (Nyéki & Neményi, 2022). To satisfy the task mentioned above, methods for regional large-scale yield predictions for major crops should be developed. There are different ways for large-scale yield predictions, based on various empirical approaches and simulations, e.g., using photosynthetically active radiation amounts (PAR), leaf area indices (LAI), solar induced chlorophyll fluorescence parameters (SICF), light, water, and nutrients use efficiency, etc. (Karthikeyan et al., 2020). Simulation models are usually realised in the form of specialised software applications; the most popular of them are DSSAT (Jones et al., 2003), WOFOST (Van Diepen et al., 1989), APSIM (Holzworth et al., 2014), and CERES (Timsina & Humphreys, 2006). But simulation models and software are not limited to this list. However, remote sensing-based predictions are now gaining popularity and demand, as they are comparatively simple and reliable enough to provide relevant information on possible scenarios of crop production in a certain agricultural location. Most scientists use vegetation indices to develop remote sensing-based crop yield predictions. The normalized difference vegetation index (NDVI) and Enhanced Vegetation Index (EVI) have the highest popularity in scientific community (Baret & Guyot, 1991; Liu, 1995). Other remote sensing-based approaches utilise normalised difference water index (NDWI), two-band EVI (EVI2), Green-Red Vegetation Index (GRVI), and vegetation condition index (VCI) as predictors (Gao, 1996; Jiang et al., 2008; Motohka et al., 2010). Besides, current science has huge mathematical and statistical apparatus to build up predictive models with the highest accuracy, relevance, and fitting quality. The choice of certain mathematical approach in the yield prediction depends mainly on the number of inputs used in the model, their distribution pattern, and the aims of prediction. Sometimes, more sophisticated and mathematically strong methods are not applicable for predictions because of small sample size, or its great inequality, or unnormal distribution of data, etc. Artificial neural networks, for example, notwithstanding their great performance and accuracy, are inappropriate in many scientific and practical purposes because it is impossible to derive the way to solve the prediction task (so called “black box nature”), as in regression modelling, making the latter relevant even considering its relative “out-of-date” status (Karthikeyan et al., 2020). The main purpose of this study was to establish the relationship between the values of the regional Normalised Difference Vegetation Index (NDVI), calculated for the croplands of the Kherson region, and the average annual yields of potatoes, vegetables, fruits and berries, cultivated there. Mathematical models were also developed to predict the yields of the crops mentioned above on a regional scale.