A CASE STUDY ON THE POSSIBILITY OF SOIL MELIORATIVE STATE ASSESSMENT BY REMOTE SENSING DATA

Summary. The paper presents the results of the pilot study on the relationship between spatial bare-soil normalized difference vegetation index and content of salts in the dark-chestnut soil of the South of Ukraine. The study is based on the results of soil analyses, conducted in 2016 within the framework of sweet corn cultivation technology investigation at the Agricultural Cooperative Farm of Bilozersky district, Kherson oblast. The results of laboratory analyses were connected to bare-soil values of the spatial vegetation index, obtained at OneSoil platform. The relationship was estimated through rank correlation and polynomial regression analysis. As a result, very strong relationship was established between bare-soil vegetation index and toxic salts content in the soil, while moderate inter-connection was found out between the index and total salts content in the soil. The second-grade polynomial models, developed in the intercourse of regression analysis, proved to have very good fitting quality and accuracy of the salts content prediction, with the mean absolute percentage error 2.93-4.27%. Thus, bare-soil normalized difference vegetation index is suitable and prospective for meliorative surveys of the dark-chestnut soils of the South of Ukraine.

Introduction.Most lands of the South of Ukraine, located in Kherson, Zaporizhzhia, Mykolaiv regions, were damaged to some extent in the intercourse of the war, thus, the development of activities and steps for soil restoration is required.To provide scientifically sound recommendations on soil quality restoration it is necessary to conduct soil surveys first.However, large-scale soil surveys require much capital investments, as well as great labor and time expenditures, and qualified staff for the surveys conduction.
Although in-field investigation of soil quality is the first-choice method to obtain reliable data on the conditions of soil, remote sensing is another great option for conduction of large-scale soil surveys with less expenditures of time and labor [1].
It has been proved by the previous studies that remote sensing data, for example, spatial normalized difference vegetation index, are prospective for determination of humus content, and some other fertility properties, as well as some physical and chemical qualities of soils [2][3][4].Current study is desired to investigate whether remote sensing data could be applied for the estimation of meliorative properties of soils, especially, the content of total and toxic salts in the arable (0-30 cm) layer.

СЕКЦІЯ XI. АГРАРНІ НАУКИ ТА ПРОДОВОЛЬСТВО
Materials and methods.The study is based on the results of soil analyses, conducted in 2016 under sweet corn crops, cultivated at drip-irrigated experimental field at the Agricultural Cooperative Farm "Radianska Zemlia" of Bilozersky district, Kherson oblast.The analyses were carried out in the analytic laboratory of the Institute of Climate-Smart Agriculture of NAAS.The samples of the dark-chestnut soil for the estimation of meliorative properties were collected before sowing and after harvesting of the crop, when the field was tilled and remained free from vegetating plants.The points, where the soil samples were taken, were further tied to the remote sensing data on normalized difference vegetation index (NDVI) by geotagging.The values of NDVI were derived using OneSoil platform from the cloud-free images, downloaded for the closest date to the date of the soil samples collection.Total and toxic salts content were determined in the soil samples and connected to corresponding values of the spatial NDVI.The pairs of the experimental data were statistically processed by the means of polynomial regression analysis to find out the relationship between the spatial bare-soil vegetation index and true values of salts content in the soil arable layer [5].Besides, rank correlation analysis was also performed to establish the strength and direction of the inter-connection [6].Second-grade polynomial models for the prediction of total and toxic salts content in the arable layer of the soil were developed and evaluated through the calculation of mean absolute percentage error (MAPE) by the latest recommendations for models' classification by this indicator [7].All the calculations and work on visualization were performed using BioStat v.7 and Microsoft Excel 365 software.
Results.The results of soil surveys on the meliorative state of the dark-chestnut soil together with the data on spatial NDVI is presented in Table 1.These data were used as the inputs for polynomial regression and rank correlation analysis.Rank correlation analysis revealed presence of strong relationship between the toxic salts content and bare-soil NDVI, and moderate relationship between the toxic salts content and bare-soil NDVI (Table 2).The statistics, alongside with the accuracy evaluation by the values of MAPE, for the regression analysis and second-grade polynomial models for the prediction of salts content in the arable layer of the darkchestnut soil is provided in Table 3.It is evident that there is a possibility for accurate estimation of the meliorative state of the dark-chestnut soil using the spatial vegetation index.Visual approximation is provided in the Figures 1 and 2. Discussion.This study is not the first to try spatial NDVI for the evaluation of soil salinity patterns.For example, there is a study, devoted to similar subject, conducted for the conditions of Saudi Arabia [8].Some researchers applied improved NDVI-SI (salinization index) and SDI (salinization detection index) to estimate meliorative conditions of soils [9].Similar study was conducted in 2019 for Syrdarya province of Uzbekistan, where scientists used spatial NDVI within the framework of geoinformation system, combined with in-field determination of the content of total dissoluble salts (TDS) in the soils, to derive the relationship between TDS and NDVI and make soil mapping [10].Apart from NDVI, scientists use other, more specific spatial indices, e. g., salinity index (SI) and salinity detecting model (SDM) in remote monitoring of soils salinity [11].
Current study provides some additional insights on the issue of remote soil salinity monitoring, especially, for the soils of the South of Ukraine, as this subject has not been studied deeply in the country until now.Although small initial dataset, it opens new prospects for further development of this direction, providing valuable initial information for conduction of deeper investigations.
Conclusions.Polynomial regression analysis, accompanied by rank correlation analysis, revealed that there is strong relationship between the content of salts in the arable layer of the dark-chestnut soil and bare-soil NDVI values.The model provides good fitting quality and high predictive accuracy, especially, in case of toxic salts.Further investigations in this direction will be conducted to deepen the preliminary results and acquire scientific knowledge on the studied problem.

Table 1
Toxic and total salts content in the arable layer of the dark-chestnut soil with

Table 2
Rank correlation for the relationship between the salts content and bare-soil NDVI