Prospects for Satellite Spectral Monitoring for Automation of Processes for Assessing Agricultural Soil Use

Abstract

Means of technical vision as innovative solutions have found wide use for automation of technological processes in agriculture in general and in crop production in particular. Their introduction becomes especially important when introducing the market of agricultural land in Ukraine when it is quite possible that it is misused by tenants or owners. The use of satellite monitoring may be an effective solution, as high and ultra-high resolution (resolution) satellite images have become available to farmers in recent years. The aim of the work was to assess the prospects of satellite spectral monitoring to automate the processes of assessing agricultural soil use. The research was conducted on the production fields NUBiP of Ukraine. During 20162021, the fields were occupied by different crops - winter and spring. Mostly cereals were grown, some fields were occupied by sunflowers, corn for grain and silage, perennial grasses. Archival data on multispectral images from a specialized Landsat 8 satellite were used for the research. It is established that satellite spectral monitoring turned out to be suitable for automation of processes of technological soil erosion monitoring. Using a series of satellite images, it was possible to identify a field for which agricultural practices in crop production were carried out at a higher level and, accordingly, the soil has a higher fertility. To ensure one-year image accuracy, it is necessary to use images with a resolution suitable for precision atmospheric correction on terrestrial objects with stable and known spectral indices.

Keywords

• 1 Soil quality
• satellite monitoring
• automation

1. Actuality

Means of technical vision as innovative solutions have found wide use for automation of technological processes in agriculture in general and in crop production in particular. Their introduction becomes especially important when introducing the market of agricultural land in Ukraine when it is quite possible that it is misused by tenants or owners. The review article Hongkun Tian at al [ 1 ] (2020) on the prospects of technical vision shows that the specificity of agricultural production is the diversity and instability of the forms of the studied objects, when in addition to their geometry should be taken into account spectral indicators. Certain technological operations with the use of technical vision devices have been successfully completed with the automation of agricultural production. What is an example of the identification of apples in the crown of trees, presented in the work of I. Smirnov at al [ 2 ] (2021), strawberries on the ridges, described in the work of D. Khort at al [ 3 ] (2020), tomatoes, considered in work I Korobiichuk et al [ 4 ] (2017). At the same time, the introduction of automation in agricultural practices is uneven, as evidenced in the analytical work Kirtan Jha at al [ 5 ] (2019), devoted to the prospects and needs of agricultural automation, which shows the need to strengthen developments to determine the state and especially soil fertility. The problem of soil fertility reproduction is extremely relevant not only on the scale of individual farms, but at the state level for European countries, which was covered in the article by Hakkı Emrah Erdogan at al [ 6 ] (2021). Ground-based devices such as the Dutch company SoilCares (https://www.soilcaresfoundation.com) and the experimental device described in Sérgio H.G. SILVA at al [ 7 ] (2021), intended for soil analysis, however, do not provide scalability of studies, as they require direct contact with the test sample. The use of satellite monitoring may be an effective solution, as high and ultra-high resolution (resolution) satellite images have become available to farmers in recent years.

The aim of the work was to assess the prospects of satellite spectral monitoring to automate the processes of assessing agricultural soil use. 1.1.

The state of the issue

The issue of spectral monitoring using satellite platforms is especially relevant for tropical regions, which in the context of growing global food shortages in the future may become additional agricultural land. The potential for successful use of satellite monitoring for soil science (Pedological assessment) was predicted in the work of José A.M. Demattê at al [ 8 ] (2014). Practical implementation was shown in the works of Wanderson de S. Mendes at al [ 9 ] (2019) and Raúl R. Poppiel at al [ 10 ] (2019). In staging articles, which showed both the presence of many methodological problems and promising ways to analyze not a single image, but the dynamics of changes in the characteristics of the images over time. One of the problems was the low resolution of existing Landsat-5 satellite images, but in recent years several new satellite platforms have been launched into orbit, such as Landsat-8, with higher image resolution, and, accordingly, new opportunities are emerging for researchers. . In the work of Nélida Elizabet Quiñonez Silvero at al [ 11 ] (2021) on the prediction of soil properties in Brazil, it was possible to determine the type of parent rock and to some extent the organic content. The authors were forced to work in conditions of significant shortage of open soil in the trails and changes in the humidity of the upper layer, so to assess the soil used the concept of "soil line", which led to the possibility of significant error even in numerous measurements. For similar climatic conditions in India, Kishan Singh Rawat at al [ 12 ] (2019) developed a modified water balance model (MWCM) based on spectral data from the LANDSAT-8 satellite. As in the previous work, the developed solution is based on the concept of the ground line, for the initial data uses the NDVI index. In the European part, soil monitoring of fields not occupied by vegetation can be carried out most often in spring and autumn. Moreover, the soil is mostly in the air-dry state, which contributes to the objectivity of its direct spectral evaluation. The open ground is characteristic of certain plantations, in particular perennials with keeping the rows unoccupied. Such objects, namely vineyards, were considered in the work of A. Brook at al [ 13 ] (2020), where the prospects of water erosion of the upper layer were successfully assessed. The authors compared the intensity of the color components compared data from satellites and UAVs, for the implementation of atmospheric correction as reflector panels used gravel roads. Considering the national specifics of Ukraine, the prospects of such standards in the production fields are currently insufficient. An alternative, as shown in the work of V. Lysenko at al [ 14 ] (2018), can be sections of dirt roads, the identification of which can be carried out according to the method described in the work of S.A.Shvorov at al [ 15 ] (2018).

In addition to traditional factors of soil erosion, such as wind and water, in intensive agriculture there is also technological erosion associated with changes in soil properties, primarily a decrease in organic matter content. According to the data covered in the work of Yawen Li at al [ 16 ] (2021), the issues of technological erosion are insufficiently studied. Thus, the authors found that for the garden erosion was higher than for industrial fields, which contradicts the results presented by Zhongwu Li at al [ 17 ] (2017).

The above analysis of the literature allows us to draw the following conclusions:  satellite platforms can be used to assess the condition of soils, but ready-made solutions, especially regarding the nature of erosion, have not been identified;  since the values of the intensity values of the color components are informative, for atmospheric correction it is possible to use as reflective panels of roads with artificial surface, as well as rolled soil;  to determine the condition of the soil, it is advisable to consider the dynamics of changes in spectral indicators over time; erosion of both traditional (wind, water) and technological nature is possible in the fields, which must be considered when organizing research.

2. Organization of the experiment

The research was carried out on the production fields of NUBiP (https://nubip.edu.ua/en) of Ukraine "Velykosnitynske training and research farm. OV Muzychenko ”(Kyiv region; coordinates Lat: 50.09080, Lng: 30.02997). During 2016-2021, the fields were occupied by different crops - winter and spring. Mostly cereal grains were grown, some fields were occupied by sunflower, corn for grain and silage, perennial grasses (Table 1). The soil of the territory is podzolic chernozem.

To maintain soil fertility in some fields after harvesting the main crop sown green manure (leies). In field 3, organic fertilizers (manure from cattle) were applied.

2.1. Initial data of spectral satellite monitoring

At present, there is free access to archival data of the results of spectral imaging from the Landsat 4-5 and 8 satellites (provided by NASA / USGS). The highest resolution is in the spectral systems of Landsat 8 and is 30 m (15 for the panchromatic band). The frequency of images is 16 days. According to the data presented in Hengbiao Zheng et al (2020) [ 18 ] regarding the identification of plantings, it was determined that the minimum possible size of the object for visual identification in the optical range is 13 × 13 pixels, respectively. Therefore, the use of local roads for atmospheric correction of dirt roads is not realized due to their small width. Therefore, the results of atmospheric correction directly from the image provider were used, namely, channels blue B2 (0.450–0.515 µm), green B3 (0.525–0.600 µm), red B4 (0.630–0.680 µm) and near infrared B5 (0.845–0.885 µm). . For ease of perception of information, consumers used not monochrome, but color images with an additive model of RGB color formation (channels 4, 3, 2) for the visible range of the spectrum and in false color composite (channels 5,4,3). The use of the infrared range is due to the need to assess the density and condition of plantations, as the analysis should be carried out for the top layer of soil. For research in the expert mode, images were selected where there are no clouds in the experimental fields and, accordingly, the shadow from them (Fig. 1).

To take into account the possibility of the impact of water erosion of the soil at the choice of experimental sites, they were checked for the presence of stable puddles due to the terrain. An archive of high-resolution satellite images 0.5 m / pixel obtained (Fig. 2) from the archive data of the Google Earth Pro service (ver: 7.3.3.7786) was used for verification.

Mathematical data processing

Processing of satellite images was performed using MathCad. Spectral monitoring data were saved in Jpeg format. For the data processing algorithm, two options were considered both directly for finding the average value and with the approximation of experimental data.

The first algorithm involved processing in two stages: the first determined the average intensity of the color component in the area, and the second to remove random objects removed pixels in which the intensity of the color components differed from the average by more than 10 units. If the area of the error plots exceeded 10%, a second algorithm based on the approximation of experimental data was used. For approximation, we used the Gaussian distribution according to the method described in N. Pasichnyk at al [ 19 ] (2021). The second approach makes it possible to estimate the presence of several objects at the same time in the experimental area, although it requires a multiple of larger computing power (Fig. 3).

Statistical processing showed that the data are described by the Gaussian distribution, the coefficient of determination was ≥0.95. Comparing the results obtained by the first algorithm, the data difference did not exceed 5%. Despite the presence of plantings in field 3, the value of the standard deviation w is virtually identical, respectively, when processing spectral data with a resolution of 30 m / pixel to assess the condition of the soil, the second algorithm will not have fundamental advantages.

Statistical data processing was performed using a specialized software product OriginPro Sp4 (Origin Lab Corporation). 1800 1600 1400 ls1200 e ixp1000 f o re 800 b um600 N 400 200 0 0

R2, xc=18, w=2.8 R3, xc=20, w=2.3 G2, xc=16, w=2.2 G3, xc=28, w=1.3 B2, xc=12, w=2.5 B3, xc=21, w=1.9 10

20 Color component intensity 30 40

The results and discussion

Selection of suitable data

Remote detection and assessment of the degree of erosion of a technological nature can be done for soil that is in an air-dry state, because the color of dry chernozem corresponds to gray gradations, and moist soil is close to black, which is difficult to interpret. The available satellite image processing programs estimate cloudiness and temperature, not humidity. According to the authors, some of the pictures, namely from 11/21/2017 and 12/20/202020, were taken when the soil was in a wet state, as evidenced by the low values of the intensity of the color components (in Table 1, these items are highlighted in gray). At higher image resolutions, it will be possible to reliably assess the moisture content of the topsoil The results of spectral data processing are shown in table 2. by assessing the color of dirt roads. For further calculations, data were used in which the value of any color component was more than 10 units for the 8-bit color model.

2.5. Evaluation of the

atmospheric correction data quality of of spectral

Based on the data given in Table 2, in the period 27.08 - 08.11.2018 there were favorable weather conditions for satellite monitoring, so we managed to take a series of images for the fields, the results of which are shown in Figure 4.

As can be seen from the above data, for the visible range there is a trend to reduce the intensity of the color components, which could be explained by the gradual moistening of the soil and, accordingly, its darkening. However, in the case of soil moisture, for the infrared channel, according to the results of A. J. Richardson at al [ 20 ] (1977), there should be an increase, but a declining trend. According to the authors, the explanation for this is the imperfect atmospheric correction, which must be carried out using artificial ground or natural reflector panels. Because this is not always easy to implement, especially for low-resolution images such as Landsat 8, it may be appropriate to focus on a series of images over several years.

2.6. The results of statistical data processing

Different types and subtypes of soils can have different values of color intensity of color components, so setting a limit value that corresponds to soil without plants is a debatable issue. In addition, different crops during the growing season have different indicators of the intensity of the color component in both the optical and infrared ranges. Therefore, to determine the soil parameters, data was filtered based on the value in the infrared channel. The results are shown in Figure 5. iR R G

B 80 90 100 110 120

iR filtration Figure 5: The dependence of the mean value of the intensity of the red color component, calculated for the condition of the mean value for pixels if the value of the iR pixel ≤iR filtration

First-order Exponential Decay equations were used to approximate the experimental data. For the green and blue components of color, dependences of a similar nature were obtained.

The analysis of the obtained data showed that for the third field the color is significantly darker, which is obviously a consequence of more organic matter in the soil. 2.7.

Direction of further research

In addition to the Landsat v5-8 agricultural satellites, there are alternative solutions, such as Sentinel-2 with higher image resolution, for which it is easier to choose acceptable optical templates.

Establishing the state of moisture of the upper soil layer will be of fundamental importance for the automated determination of the state of soil erosion based on the results of spectral monitoring. It is necessary to develop a mathematical algorithm that can assess the suitability of the data.

Agricultural satellites are shooting in automatic mode, not taking into account the state of clouds. The systems provide an assessment of the state of clouds throughout the photograph, but there is a high probability that for the experimental area the state of clouds may not correspond to the average value. According to the authors, research on the introduction of machine learning to assess the suitability of images for cloud parameters in the experimental areas is promising. 3. Conclusions

1. Satellite spectral monitoring proved to be suitable for automation of processes of technological soil erosion monitoring.

2. Using a series of satellite images, it was possible to identify a field for which agricultural practices in crop production were carried out at the highest level and, accordingly, the soil has a higher fertility.

3. To ensure the accuracy of one-year images, it is necessary to use images with a resolution suitable for precision atmospheric correction on terrestrial objects with stable and known spectral indices.

4. References

References

[1] Hongkun Tian ( 2020 ) Computer vision technology in agricultural automation - A review / Hongkun Tian , Tianhai Wang, Yadong Liu, Xi Qiao, Yanzhou Li // Information Processing in Agriculture, Vol. 7 ( 1 ), рр. 1-19 https://doi.org/10.1016/j.inpa. 2019 . 09 .006.

[2] I. Smirnov ( 2021 ) Neural network for identifying apple fruits on the crown of a tree / I. Smirnov , A. Kutyrev , N.Kiktev // International scientific forum on computer and energy Sciences (WFCES 2021 ), E3S Web Conf . Vol. 270 , 01021 , https://doi.org/10.1051/e3sconf/2021270010 21.

[3] Khort D. ( 2020 ) Computer vision system for recognizing the coordinates location and ripeness of strawberries (Conference Paper) / D.Khort, A. Kutyrev , I. Smirnov , V. Osypenko , N.Kiktev // Communications in Computer and Information Science DSMP 2020 , Vol. 1158 , рр. 334 - 343 , https://doi.org/10.1007/978-3- 030 -61656- 4_ 22 .

[4] I. Korobiichuk ( 2017 ) Energy-Efficient Electrotechnical Complex of Greenhouses with Regard to Quality of Vegetable Production . In: Szewczyk R., Kaliczyńska M . (eds) / Korobiichuk I., Lysenko V. , Reshetiuk V. , Lendiel T. , Kamiński M. // Recent Advances in Systems, Control and Information Technology . SCIT 2016. Advances in Intelligent Systems and Computing , vol. 543 , рр. 243 - 251 . Springer, Cham. https://doi.org/10.1007/978-3- 319 - 48923-0_ 30 .

[5] Kirtan Jha ( 2019 ) A comprehensive review on automation in agriculture using artificial intelligence / Kirtan Jha , Aalap Doshi, Poojan Patel, Manan Shah // Artificial Intelligence in Agriculture, Vol. 2 , рр . 1- 12 , https://doi.org/10.1016/j.aiia. 2019 . 05 .004.

[6] Hakkı Emrah Erdogan ( 2021 ) Soil conservation and sustainable development goals(SDGs) achievement in Europe and central Asia: Which role for the European soil partnership? / Hakkı Emrah Erdogan , Elena Havlicek, Carmelo Dazzi, Luca Montanarella, Marc Van Liedekerke, Borut Vrščaj , Pavel Krasilnikov, Gulchekhra Khasankhanova, Ronald Vargas // International Soil and Water Conservation Research, Vol. 9 ( 3 ), рр. 360 - 369 , https://doi.org/10.1016/j.iswcr. 2021 . 02 .003.

[7] Sérgio H.G. Silva ( 2021 ) Proximal sensorenhanced soil mapping in complex soillandscape areas of Brazil / Sérgio H .G. Silva, David C. Weindorf , Wilson M. Faria, Leandro C. Pinto, Michele D. Menezes , Luiz R.G. Guilherme, Nilton Curi // Pedosphere, Vol. 31 ( 4 ), pp. 615 - 626 , https://doi.org/10.1016/S1002- 0160 ( 21 ) 60007 - 3 .

[8] José A. M. Demattê ( 2014 ) Spectral pedology: A new perspective on evaluation of soils along pedogenetic alterations / José A . M. Demattê , Fabrício da Silva Terra // Pedosphere, Vol. 31 ( 4 ), рр. 615 - 626 , https://doi.org/10.1016/S1002- 0160 ( 21 ) 60007 - 3 .

[9] Wanderson de S. Mendes ( 2019 ) Is it possible to map subsurface soil attributes by satellite spectral transfer models? / Wanderson de S. Mendes, Luiz G. Medeiros Neto, José A.M. Demattê , Bruna C. Gallo, Rodnei Rizzo, José L. Safanelli, Caio T. Fongaro // Geoderma, Vol. 343 , рр. 269 - 279 , https://doi.org/10.1016/j.geoderma. 2019 . 01 . 025.

[10] Raúl R. Poppiel ( 2019 ) Pedology and soil class mapping from proximal and remote sensed data / Raúl R . Poppiel, Marilusa P.C. Lacerda , José A.M. Demattê , Manuel P. Oliveira, Bruna C. Gallo, José L. Safanelli // Geoderma, Vol. 348 , рр. 189 - 206 , https://doi.org/10.1016/j.geoderma. 2019 . 04 . 028.

[11] Nélida Elizabet Quiñonez Silvero ( 2021 ) Soil variability and quantification based on Sentinel-2 and Landsat-8 bare soil images: A comparison / Nélida Elizabet Quiñonez Silvero, José Alexandre Melo Demattê, Merilyn Taynara Accorsi Amorim, Natasha Valadares dos Santos, Rodnei Rizzo , José Lucas Safanelli, Raul Roberto Poppiel, Wanderson de Sousa Mendes, Benito Roberto Bonfatti // Remote Sensing of Environment, Vol. 252 , 112117 , https://doi.org/10.1016/j.rse. 2020 . 112117 .

[12] Kishan Singh Rawat ( 2019 ) Synergetic methodology for estimation of soil moisture over agricultural area using Landsat-8 and Sentinel-1 satellite data / Kishan Singh Rawat, Sudhir Kumar Singh , Raj Kumar Pal // Remote Sensing Applications: Society and Environment, Vol. 15 , 100250 , https://doi.org/10.1016/j.rsase. 2019 . 100250 .

[13] Brook , A. ( 2020 ) A smart multiple spatial and temporal resolution system to support precision agriculture from satellite images: Proof of concept on Aglianico vineyard / A . Brook , V. De Micco, G. Battipaglia , A. Erbaggio , G. Ludeno, I. Catapano , A . Bonfante // Remote Sensing of Environment, Vol. 240 , 111679 , https://doi.org/10.1016/j.rse. 2020 . 111679 .

[14] Lysenko , V. ( 2018 ) Information Support Of The Remote Nitrogen Monitoring System In Agricultural Crops / V. Lysenko , O. Opryshko , D. Komarchuk , N. Pasichnyk , N. Zaets , A .Dudnyk // International Journal of Computing Vol 17 ( 1) рр . 47 - 54 .

[15] Shvorov , S. A. ( 2018 ) UAV Navigation and Management System Based on the Spectral Portrait of Terrain / Sergey A Shvorov, Dmytro S Komarchuk, Natalia A Pasichnyk, Oleksiy A Opryshko, Yurii A Gunchenko , Svitlana D Kuznichenko // 2018 IEEE 5th International Conference on Methods and Systems of Navigation and Motion Control , MSNMC 2018 , рр. 68 - 71 , https://doi.org/10.1109/MSNMC. 2018 . 8576 304.

[16] Yawen Li ( 2021 ) Interactive effects of land use and soil erosion on soil organic carbon in the dry-hot valley region of southern China / Yawen Li, Xingwu Duan , Ya Li , Yuxiang Li , Lanlan Zhang // CATENA, Vol. 201 , 105187 , https://doi.org/10.1016/j.catena. 2021 . 10518 7.

[17] Zhongwu Li ( 2017 ) Response of soil organic carbon and nitrogen stocks to soil erosion and land use types in the Loess hilly-gully region of China / Soil and Tillage Research, Vol. 166 , рр . 1- 9 , https://doi.org/10.1016/j.still. 2016 . 10 .004.

[18] Hengbiao Zheng ( 2020 ) Early season detection of rice plants using RGB, NIR-GB and multispectral images from unmanned aerial vehicle (UAV) / Hengbiao Zheng , Xiang Zhou, Jiaoyang He, Xia Yao, Tao Cheng, Yan Zhu, Weixing Cao, Yongchao Tian Computers and Electronics in Agriculture , Vol. 169 , 105223 , https://doi.org/10.1016/j.compag. 2020 . 1052 23.

[19] Pasichnyk , N. ( 2021 ) Technologies for Environmental Monitoring of the City / N. Pasichnyk , D. Komarchuk , O. Opryshko , S. Shvorov , V. Reshetiuk , O.Bahatska // 2021 IEEE 16th International Conference on the Experience of Designing and Application of CAD Systems (CADSM) , pp. 40 - 43 , https://doi.org/10.1109/CADSM52681. 2021 . 9385213 .

[20] Richardson , A. J. ( 1977 ) Distinguishing vegetation from soil background information / Richardson, A .J., Wiegand , C.L. // Photogrammetric Engineering and Remote Sensing Vol. 43 ( 2 ), рр. 1541 - 1552 .