LIQUID SOURCE NITROGEN AS A MORE SUSTAINABLE TECHNOLOGY FOR OAT FERTILIZATION WITH COMPUTATIONAL SIMULATION RESOURCE

Purpose: The objective of the study is to validate the technology of liquid nitrogen source by spraying of foliar absorption in comparison to the standard source urea of root absorption in the oat crop. Adapt the fuzzy logic model and training an artificial neural network to simulate oat productivity, under conditions of nitrogen use with the combined action of rainfall and thermal sum accumulated during the crop cycle. Method/design/approach: The study was conducted in Augusto Pestana, RS, Brazil, in a randomized block design with four replications in a 2x4 factorial, for 2 nitrogen sources (liquid and solid) with 4 doses (0, 30, 60 and 120 kg ha-1), respectively. Solid (urea) and liquid (N-Top®) nitrogen sources were applied at the phenological stage of expanded oat fourth leaf. Results and conclusion: The liquid source nitrogen technology presents results similar to the use of the standard urea source, confirming the possibility of using the nutrient by foliar absorption. The simulation of grain yield by fuzzy logic with the input variables nitrogen dose, thermal sum and precipitation is not adequate for the formulated rule base. The input variables used in the artificial neural network proved to be appropriate in the simulation of oat productivity, with


INTRODUCTION
The agricultural and food importance and the health benefits promoted by white oats (Avena sativa L.) in the world context have reflected the need to increase production with cost reduction and greater sustainability together with new management technologies (Silva Júnior et al., 2021;Allwood et al., 2021). As it is a grass, adequate supply of nitrogen reflects an increase in productivity (Trautmann et al., 2017;Mantai et al., 2021). However, the most common source, urea, with high solubility, presents high losses by leaching and volatilization, generating increased production costs and environmental pollution Yang et al., 2020). A scenario aggravated by conditions of greater rainfall and higher air temperature during fertilization, reducing the use of nitrogen (Marolli et al., 2018;Trautmann et al., 2022).
Studies show that the dissipation of nitrogen into the environment has caused contamination of groundwater and surface water, in addition to causing atmospheric pollution (Husk et al., 2017;Anas et al., 2020). It is noteworthy that approximately 60% of the nitrogen provided in the form of fertilizers in non-leguminous crops is lost to the environment and is no longer absorbed (Cassman, Dobermann, & Walters 2002;Rodrigues et al., 2021). Therefore, increasing the productivity of crops with less impact on the environment is a major current challenge for global food production (Abbade, 2020;Sarkar et al., 2020). In this perspective, an alternative to the conventional use of solid urea source would be the technology of nitrogen application by spraying for absorption via foliar, management that seeks to ensure greater efficiency of use with the use of small amounts per hectare, facilitating the absorption with reduction of losses Mortate et al., 2018).
Nitrogen is a mobile element because it is easily absorbed and translocated within a plant tissue (Das & Avasthe, 2018). Allied to this, the supply of nitrogen by spraying can cause leaf cooling, a condition that would lead to the opening of the stomata and, consequently, facilitate the absorption of the nutrient, regardless of the soil moisture condition (Alshaal & El-Ramady 2017;Wang et al., 2022). On the other hand, the cultivation of oats is a complex system that involves controlled and uncontrolled variables, a condition that demands studies to help the farmer to identify more stable and efficient doses in different environmental conditions and to validate the technology of application of nitrogen by foliar absorption in oats (Mamann et al., 2018;Alessi et al., 2021). Thus, the use of computational modeling can serve as a basis for improving the understanding of environmental conditions and management of agricultural crops based on the simulation, optimization and validation of technologies (Marolli et al., 2018;Mamann et al., 2020). Among the modeling techniques, fuzzy models and artificial neurals networks stand out (Malaman & Amorim 2017;Dornelles et al., 2018).
Through fuzzy logic techniques, it is possible to describe complex systems of non-linear behavior, produced from rules, which must be elaborated by specialists, providing their experience for the elaboration of a system (Mamann et al., 2018;Scremin et al., 2020a). While the use of artificial neural networks represents a tool based on the functioning of the biological nervous system, formed by neurons that can be distributed in several interconnected layers, which through training store knowledge and generalize the information seized, being able to solve complex problems and model the behavior of the variables involved (Azarpour et al., 2015;Scremin et al., 2020 b). In agriculture, neural networks can be used to develop prediction models in complex systems and estimate desired parameters, enhancing process optimization and decision making (Huang et al., 2010;Silva et al., 2014).
The combination of simulation techniques and use in solving problems related to agriculture has been highlighted and, therefore, the use of fuzzy logic and artificial neural networks are presented as an efficient tool in the process of simulating the productivity of oats, as it allows involving the non-linear effects of environmental conditions on agroecosystems. Therefore, it can simulate, compare and validate the technology of liquid nitrogen applied via foliar in real cultivation conditions. The objective of the study is to validate the technology of liquid nitrogen source by spraying of foliar absorption in comparison to the standard source urea of root absorption in the oat crop. Adapt the fuzzy logic model and training an artificial neural network to simulate oat productivity, under conditions of nitrogen use with the combined action of rainfall and thermal sum accumulated during the crop cycle.

MATERIALS AND METHODS
The study was carried out in the years . The soil in the experimental area is classified as a Typical Distroferric Red Latosol, with a deep, well-drained profile and a dark red color. The climate in the region, according to the Köppen classification, is Cfa (humid subtropical), with well-distributed rainfall during the year with volumes close to 1600 mm per year, with higher volumes of precipitation in winter. Ten days before sowing, soil analysis was carried out showing the following chemical characteristics (pH= 6.2; P=33.9 mg dm -3 ; K= 200 mg dm -3 ; MO= 3.0 %; Al= 0 cmolc dm -3 ; Ca = 6.5 cmolc dm -3 e Mg=2.5 cmolc dm -3 ).
Sowing was carried out with a seeder-fertilizer in two cropping systems, high and low residual N release, corn/oat and soybean/oat system, respectively, between the first and third week of June in different years, using the white oat cultivar URS Guará, with a population density of 400 viable seeds m -2 . The experimental design was randomized blocks with four replications in a 2x4 factorial model, for 2 nitrogen sources (liquid and solid) with 4 nutrient rates (0, 30, 60 and 120 kg ha -1 ), respectively. The experimental unit consisted of a plot of five rows, 5 meters long and spaced 0.20 meters apart, totaling 5 m².
At sowing, 45 and 30 kg ha -1 of P2O5 and K2O were applied based on the levels of P and K in the soil for expected grain yield of 3 t ha -1 , respectively, and 10 kg ha -1 of N (except in the standard experimental unit). During the execution of the study, applications of tebuconazole fungicide FOLICUR® CE were applied at a dosage of 0.75 L ha -1 . Weed control was carried out by applying the metsulfuron-methyl herbicide ALLY® at a dose of 2.4 g ha -1 and additional weeding whenever necessary.
The nitrogen source for root absorption in the soil was urea (45% N) applied in cover and for foliar absorption was the commercial product N-Top® (28% N) in liquid form with a density of 1.3 g ml -1 . The liquid source nitrogen used comes from a substance with a high concentration of organic compounds, mainly humic and fulvid acids. The urea was applied by spray and the N-Top® was sprayed with a volume of water of 200 L ha -1 . In each of the sources, the different doses of nitrogen indicated in the study were dimensioned, converted to the area of the experimental unit of 5 m 2 . Nitrogen application by the leaves was carried out with a knapsack sprayer at a constant pressure of 30 lb in -2 , using compressed CO 2 , with "cone" jet nozzles, with the spraying time sized to apply the different doses of the nutrient. The application of the treatments in the sources of absorption by root and leaf was in the phenological stage V4, considering the oat plant with four expanded leaves.
To estimate the grain yield (GY, kg ha -1 ), the three central rows of each plot were cut at the harvest maturity stage with grain moisture around 22%. Afterwards, the plants were threshed on a stationary threshing machine and the grains were taken to the laboratory for moisture correction to 13%, and yield conversion to kg ha -1 . Meteorological data were obtained by the Automatic Total Station installed 500 meters from the experiment. The productivity values together with the air temperature and rainfall information were used to classify the agricultural years in favorable and unfavorable to the cultivation of oats. The maximum and minimum temperature information was also used to calculate the thermal sum. The thermal sum (TS) was obtained from plant emergence using the model: = maximum temperature; = minimum temperature; n = number of days from emergence to harvest; = base temperature. The base temperature of the oats used in the study was 4°C (Pedro Júnior et al., 2004).
Based on the data obtained, analysis of variance of main effects and interaction between source and doses of liquid and solid nitrogen on grain yield was carried out, regardless of the growing season in the cropping systems. Descriptive statistics of rainfall, thermal sum and grain yield of oats per agricultural year and joint analysis were also performed. The mean test by Scott e Knott was performed to compare the effects of the succession system (soybean/oat and corn/oat) on grain yield. In fuzzy modeling and artificial neural networks, the amplitude of variation of meteorological and productivity variables was taken into account considering the joint analysis of the three agricultural years. In the simulation by fuzzy logic and artificial neural network, the input variables were used: nitrogen doses, sum of rainfall and thermal sum during the cycle, with grain yield being the output variable.
The system based on fuzzy rules was implemented by the Fuzzy Logic Toolbox of the MATLAB® software, using the Mamdani inference method, with the use of the connective "and (^)", for evaluation of the rules, by the triangular membership function and defuzzification by the method of the smallest value of the maximum association function of the aggregate. For the simulation, mean values of the meteorological elements of the three years of study were used for each nitrogen dose and succession system. The fuzzification process occurred in 4 successive modules. In module 1 (fuzzification), the information of the input variables was mathematically modeled using fuzzy sets. With the help of an agronomist with experience in oat cultivation, the classes and class intervals were determined for each input and output variable of the model, as well as the rule base that includes the fuzzy uncertainty logic. For the nitrogen doses, the interval domain of 0.120 was considered; for rainfall, the range domain was from 400 to 700 e; for the thermal sum, the range domain was from 1000 to 2000. The doses were classified as low (L), medium (M), high (H) and very high (VH). The variables of rainfall and thermal sum were classified as low (L), medium (M) and high (H). The output variable (grain yield) in the soybean/oat system was classified as very low (VL), low (L), medium (M) and high (H) and in the corn/oat system, the classification very high (VH) was added. Due to the number of input and class variables, 36 linguistic rules were generated in the different succession systems. In module 2 (rule base), the variables were adjusted in their linguistic classifications, where each rule base satisfied the following structure: where Ai e Bi are the fuzzy sets. The expression A is in Ai mean that μ A i (a)ϵ [0,1].
Both the Ai and Bi are the Cartesian product of fuzzy sets, that is, In this case, each fuzzy sets A ij e B ik represented a linguistic term for the j-th input variable and k-th output variable, and expression A is in A i which means: In module 3 (inference), the logical connectives used to establish the fuzzy relation for modeling the rule base were defined. The relationship between linguistic variables was characterized by the operator (MIN) of the fuzzy system. In each rule, a fuzzy relation R i with degree of pertinence for each pair (a, b) was considered: The relation between each rule is characterized by the operator (MAX), of the fuzzy relation R that represents the model determined by a rule base obtained by the MAX union of each individual rule, so that for each pair (a,b) is obtained: where ^ represents the MIN operator. By Mamdani's method the membership function of B is given by: If the input is a unitary classical set, then μ A (a) = 1 e μ Ai (a) ≤ 1. So, the above expression results in: Therefore, the fuzzy set B represents the action for each input A.
In module 4 (defuzzification), the state of the fuzzy output variable provides the numeric value. One of the main defuzzification methods is the center of mass for continuous variables, given by the expression: and of discrete variables by the expression: The fuzzy controller is described as a function f: R n → R m , since given an input value, there is only one corresponding output value.
In the simulation via artificial neural networks, the Toolbox Neural Network by MATLAB® software was used. The multi-layer Perceptron architecture was chosen, with the input layer having three neurons, the hidden layer having five neurons and the output layer having one neuron. Following the interpretation of Soares et al. (2015) who studied the prediction of corn yield by artificial neural network, the architecture chosen was the one that presents a relationship between the number of training samples and the number of hidden connections greater than two, as indicated by Masters (1993) of smaller error mean relative validation. The "NE-NCE-NS" notation was used to represent the network architecture, where NE = number of input variables, NCE = number of neurons in the hidden layer and NS = number of neurons in the output layer. The tan-sigmoid function (Tansig) was used to activate the neurons and the network training was performed by the backpropagation algorithm with the LevembergMarquard method.
The data for the simulation were normalized, because the activation function defined for each neuron in the output layer provides values in this range. The data are normalized using the equation: being: pn = the normalized value; pmin = the smallest observed value of the data; pmax = the largest observed value of the data.
Data were randomly divided: 70% for training, 15% for testing and 15% for validation. Two simulation models were generated, one model for each succession system. The verification of the efficiency of the models took place through the observation of the real data obtained by bioexperimentation and those simulated by fuzzy logic and artificial neural network, as well as by the absolute error obtained by these modeling techniques.

RESULTS AND DISCUSSIONS
In figure 1, of the meteorological conditions during the oat cultivation cycle, the moment of nitrogen application in 2016 was in milder temperature conditions with adequate rainfall distribution throughout the cultivation cycle. These conditions facilitate the use of nitrogen while providing greater expression of grain yield (Table 1), year classified as favorable to cultivation (FY). The year 2017 (Figure 1) was characterized by a period of restricted rainfall at the beginning of the development cycle and strong variations in air temperature, with values close to 30 o C and below zero, together with the formation of frost in the same week. From the middle of the cycle onwards, very high temperatures and rainfall concentration close to the grain harvest (Figure 1). The observed restrictions hinder the adequate use of nitrogen at the same time that they restrict important biological processes, reducing grain productivity, a year classified as unfavorable (UY) for cultivation (Table 1). In 2018 (Figure 1), the moment of Nfertilizer application occurred with reduced soil moisture and higher temperatures, which may have contributed to the loss of nutrient by volatilization, reducing the efficiency of use by oats. Frost formation was observed at the beginning of elongation (60 days after emergence) with two large intervals without rain in the vegetative phase. From the flowering phase, with adequate rainfall distribution until the end of the cycle. The productivity obtained together with the meteorological information contributed so that 2018 is also classified as unfavorable (UY) for cultivation (Table 1).  The uncertainty of meteorological conditions during the cultivation of agricultural species represents a potential risk to the guarantee of productivity with economic return to the farmer (Lu et al., 2019;Harkness et al., 2020). It is noteworthy that for greater efficiency in the use of nitrogen by oats through the use of the standard source (urea), the presence of moisture in the soil is necessary, together with a milder air temperature (Arenhardt et al., 2017;Kraisig et al., 2017;Kraisig et al., 2021). Rains without large volumes and well distributed throughout the cycle characterize a favorable environment for the expression of higher grain productivity (Souza, Gerstemberger, & Araujo, 2013;Reginatto et al., 2021). Allied to this, temperatures up to 25°C are considered suitable for the cultivation of oats (Castro, Costa, & Ferrari, 2012;Mantai et al., 2017). These conditions impact on the greater efficiency of nutrient absorption by plants due to the greater mobility of the nutrient in the soil, as well as on the plant metabolism, after being absorbed by the roots (Andrews, Raven, & Lea 2013;Mantai et al., 2021). However, unfavorable agricultural years characterized by poorly distributed and large amounts of precipitation, together with high temperatures, can cause nutrient losses through leaching and volatilization (Scremin et al., 2020a;Trautmann et al., 2022). These losses that negatively impact productivity, together with the high variation in food prices, intensify the challenge of ending hunger in the world and achieving food security (Griggs et al., 2013;Ouro Salim, Guarnieri, & Leitão, 2021). Apart from causing a greater environmental impact related to water contamination and air pollution due to the increase in greenhouse gases, contributing to climate change and destruction of the ozone layer (Kanter et al., 2020;Vries, 2021).
In Table 2, from the summary of the analysis of variance by the joint analysis of the agricultural years, the nitrogen sources did not change the grain yield, with a statistical difference obtained only for the effect of the doses, regardless of the succession system. The absence of interaction source versus doses of nitrogen shows that the behavior of the productivity in function of the doses does not depend on the source used in the fertilization. Thus, the analyzes proceeded to interpret the behavior of productivity by nitrogen doses, regardless of the source of the nutrient. In Table 3, of the descriptive statistics, the volumes of rainfall in the different agricultural years were similar, although the form of distribution during the months of the cycle was different (Table 1). The thermal sum values indicated greater variation between the years of cultivation, with a higher result for the year 2017, which showed the lowest productivity (Table 1). Although the year 2018 has a lower thermal sum than the year 2016 (favorable to cultivation), the irregular distribution of rainfall hampered adequate development. The results presented so far indicate that the effect of higher air temperature and precipitation irregularity ( Figure 1; Table 1) were decisive on the expression of grain yield. Also in Table 3, the contribution of the different succession systems to oat productivity is evident, in which residual nitrogen through biological fixation and high straw mineralization due to the reduced C/N ratio of soybeans (soy/oat system) collaborate greater nutrient availability to plants, promoting greater grain yield. The results shown in Table 3 are in line with those obtained by Mantai et al. (2021) and Pansera et al. (2022) who dimensioned the great contribution of the succession system with the soybean crop on the productivity and industrial quality of oat grains. Wendling et al. (2007) comment on the benefits of available nitrogen for biological fixation and rapid decomposition of straw on the soil, promoting greater vegetative growth in wheat. Thus, cultural precedent exerts great influence on nutrient dynamics affecting the expression of production components in cereals.
In the joint analysis of the three agricultural years, the mean rainfall and thermal sum were 533 mm and 1590 degrees day -1 accumulated, respectively. In this configuration, the work amplitude for oat productivity simulation involving fuzzy modeling and artificial neural networks was from 510 to 590 mm of precipitation and from 1330 to 1840 degrees day -1 accumulated. Therefore, an observed range of productivity ranging from 1876 to 4201 kg ha -1 in the soybean/oat system and from 879 to 4414 kg ha -1 in the corn/oat system (Table 3). From there, the rule base was built together with the expert, as shown in Table 4, for the development of simulations via fuzzy logic. In Table 5, in the year 2016 (favorable to cultivation) in a soybean/oat system, observing as an example the dose of 60 kg ha -1 of nitrogen, there is an observed grain yield of 3852 kg ha -1 . The simulation with fuzzy logic showed a value of 3240 kg ha -1 of grains, indicating an absolute error around 600 kg ha -1 . At this same nutrient dose, the artificial neural network showed a simulation result with 3742 kg ha -1 , generating a reduced absolute error around 100 kg ha -1 of grains. This trend of proximity of the observed and simulated values was also observed for the other nitrogen points in the different agricultural scenarios in the soybean/oat system, qualifying the artificial neural network in comparison to the fuzzy logic.
Also in Table 5, with reference to the year 2018 (unfavorable for cultivation) in a corn/oat system, the dose of 60 kg ha -1 showed an observed productivity of 2433 kg ha -1 . The simulation by fuzzy logic, grain yield was 2580 kg ha -1 , an absolute error of 147 kg ha -1 . On the other hand, the simulation by artificial neural network showed a value of 2415 kg ha -1 , an absolute error of only 18 kg ha -1 , very close to the real point obtained. In the other doses of nitrogen in the corn/oat system, similar results are obtained, a condition that confirms the greater efficiency of the artificial neural network in the simulation of productivity compared to fuzzy logic.
The efficiency of using simulation techniques by neural network in agriculture is confirmed in studies developed by Dornelles et al. (2018) in the optimization of sowing density and simulation of oat grain yield, Scremin et al. (2020) in the simulation of oat grain productivity throughout the development cycle and Ferreira et al. (2011), who used artificial neural networks as a price forecasting strategy in the context of agribusiness. Soares et al. (2015) concluded that multilayer artificial neural networks are efficient and can be used as a tool to estimate corn grain yield. These same authors state that the use of this modeling tool can be considered an alternative for estimating the productivity of agricultural crops, thus making it an important tool in the sector.
Although the simulation technique by fuzzy logic results in absolute errors greater than those obtained by the artificial neural network, it is a method that allows the predictability of the productivity of oat grains in the use of nitrogen, configuring the need for continuity of research in the area of prediction of productive systems. Fuzzy logic has the advantage of not requiring experimental data to carry out simulations, as the specialist can create, based on his experience, the basis of rules and intervals for simulations (Assis Silva and Lima 2009;Silva et al., 2014). However, care is needed regarding the formulation of the rule base to ensure higher quality results obtained with the lowest absolute error. Simulation results by fuzzy logic are found in studies developed by Silva et al. (2014) who used fuzzy logic to estimate wheat grain yield as a function of nitrogen fertilization. Scremin et al. (2020), in the simulation of oat grain productivity by using the hydrogel biopolymer together with nitrogen fertilization. Mamann et al. (2018) state that the fuzzy model allows estimating biological and wheat grain productivity values under the conditions of use of the hydrogel as a function of nitrogen doses, without significant difference between the means obtained experimentally and those obtained through fuzzy logic.

FINAL CONSIDERATIONS
The liquid source nitrogen technology presents results similar to the use of the standard urea source, confirming the possibility of using the nutrient via foliar absorption.
The simulation of grain yield by fuzzy logic with the input variables nitrogen dose, thermal sum and precipitation is not adequate for the formulated rule base.
The input variables used in the artificial neural network proved to be appropriate in the simulation of oat productivity. The high training power of the network allows you to generate consistent simulation results.
Although the technical efficiency of liquid source nitrogen is proven, the high cost of this technology does not guarantee economic efficiency for recommendation on a commercial scale.
New research related to the topic can collaborate to define more sustainable management of nitrogen use in agricultural crops and make it possible to understand the mechanisms of nutrient absorption by leaves in the development of productivity, in line with the Sustainable Development Goals of the UN 2030 Agenda.