Efficiency of Variable Rate Application of Nitrogen Fertilizer Using Artificial Intelligence Model

Introduction. Winter wheat is a key grain crop, the productivity of which is largely determined by optimal nitrogen nutrition. Optimization of nitrogen nutrition for winter wheat in modern conditions of a large agroholding requires variable rate fertilizer application, however, agronomists face the problem of selecting doses and the lack of comprehensive, validated methods for a large number of fields. Existing approaches do not take into account the full range of factors that complicates decision-making process. The development of validation methods and artificial intelligence models for variable rate application of nitrogen fertilizers is critically important for increasing productivity and efficiency of winter wheat cultivation. The problem is that in the neural network, there are not validation methods and a model for variable rate application of nitrogen fertilizers to optimize the processes of cultivation of winter wheat and increase the agricultural land productivity. Aim of the Study. The study is aimed at developing and introducing a technology for variable rate application of nitrogen fertilizers to optimize the processes of winter wheat cultivation. Materials and Methods. On the fields of JSC Rassvet in the Ust-Labinsk district of the Krasnodar Territory, there was performed comparative field experiment for the variable rate application of nitrogen fertilizers (two top dressings). Two fields with 83 and 68 hectares of winter wheat were selected for the study. Each field was divided into 3 variants: the Russian scheme is when more fertilizers were applied to the zone of low productivity, the European scheme is when less fertilizers were applied to the zone of low productivity, and the economic scheme is when a single dose of fertilizers was applied. An LSTM model was developed based on the average vegetation indexes (EVI, NDWI, REP, SR) from Sentinel-2 satellite images obtained during nine months. To increase the training sample, there was made an attempt to synthesize the data at the pixel level; the final productivity forecasts were smoothed and converted into job cards for variable rate application of fertilizers with norms calculated based on historical correlation. Laboratory studies of plant selection were conducted at the Kuban State Agrarian University (Department of Operation and Technical Service and at the experimental station of the Kuban experimental farm). The object of the study was the technological process of variable rate application of fertilizers based on the artificial intelligence model and its impact on vegetation, yield and quality indicators of winter wheat. Results. The actual winter wheat yields (combine and biological) under the Russian scheme on the first field was 90.9...101.5 c/ha; European scheme – 89.2...96.4 c/ha; economic scheme – 89.9...90.9 c/ha. On the second field respectively 87.4...99.6 c/ha; 92.4...98.5 c/ha; 87.8...93.6 c/ha. The average yield on the first field is higher by 6,31 % on the Russian scheme and by 2,56% on the European scheme in comparison with the economic scheme; on the second field is higher by 5,25% on the European scheme and by 3,08 % on the Russian scheme in comparison with the economic scheme. Almost all the studied variants can be attributed to the 3rd statutory grade. The increase in grain-unit is directly proportional to the decrease in protein and gluten content in both fields of AI variants. Protein content for all AI variants (except for the variant of European scheme on the second field) is 0.23–1.5% less. Gluten content is 0.53–3.3% less in all AI variants (except for the European variant of the second field). The grain-unit of the AI variants is 0.33–1.6% more. Gluten content is 0.53–3.3% less for all AI variants (except for the European variant on the second field).. Discussion and Conclusion. Economic analysis of the farm on yield of grain when combine harvesting showed that when using European technology, production costs decreased by 400 thousand rubles compared to the control scheme, and fertilizer costs decreased by 2 567 rubles per one hectare. Revenue from one hectare increased by 6 401 rubles and attributable profit by 9 546 rubles. Gross profit also increased by 150 thousand rubles, and profitability increased by 5.3% compared to the control scheme. The results of the proposed validation methods and created model of variable rate application of nitrogen fertilizers were used in the work of neural network. Under real agricultural conditions, there was evaluated the efficiency of the neural network in terms of yield by creating a scale of performance in relation to existing methods of fertilizer application (increase in yield from 2.56 to 6.31%). The presented results of field experiments demonstrate the high practical significance of the proposed technology for variable rate application of nitrogen fertilizers, which requires cheching the results of farm tests on a larger number of fields.