Relationships between Soil Electrical Conductivity and Sentinel-2-Derived NDVI with pH and Content of Selected Nutrients

: Site-speciﬁc crop management demands maps which present the content of the main macronutrients. Such maps are prepared based on optimized soil sampling within management zones, which should be quite homogenous according to nutrient content, especially the content of potassium and phosphorus. Delineation of management zones is very often conducted using soil apparent electrical conductivity (EC) or other variables related to soil condition, including satellite-derived vegetation indices. In this study conducted in North-Western Poland, relationships between soil electrical conductivity and the satellite-derived normalized difference vegetation index (NDVI) of various crops (wheat, barley, and rapeseed) with soil pH and content of P, K, and Mg were evaluated. Strong relationships were observed between NDVI of cereals with potassium content in soil. Correlation coefﬁcients for wheat ranged from 0.37 to 0.60 for average potassium content for three years and from 0.05 to 0.63 for barley. Stronger relationships were observed for the years 2018 and 2019 when NDVI was based on Sentinel-2 data, while weaker for year 2017 when Landsat 8 NDVI was used. Relationships between EC and macronutrients content were similar to those observed with NDVI. Satellite-derived NDVI of cereals can be used as a variable for the delineation of within-ﬁeld management zones. The same relationships were much weaker and not consistent for winter rapeseed. and in were negative (for NDVI for years 2018 and 2018) or positive but weak (for NDVI for year 2017). These correlations for all winter wheat ﬁelds between NDVI and pH for years 2017–2019 − 0.45 while between NDVI The correlation coefﬁcient between was equal to − 0.29 and between EC and phosphorus equal to − 0.68.


Introduction
Evaluation of available nutrients content such as potassium, magnesium, and phosphorus as well soil pH is very important in crop production because it allows for the optimization of soil fertilization and liming. Soil sampling and chemical analysis demand expenses which are higher if such soil sampling is conducted frequently and at high spatial density. Accurate evaluation of soil physico-chemical properties is very important in site-specific crop management, where variable-rate fertilization is applied. Delineation of management zones allows obtaining more accurate soil maps because soil properties within such zones are more homogenous. One composite soil sample represents each management zone. Areas of such management zones should be homogenous according to the most important agronomical soil properties such as soil texture, content of soil organic carbon, pH, and content of the most important nutrients such as potassium, phosphorus, and magnesium. Delineation of management zones can be based on proximal sensing (e.g., evaluation of electrical conductivity-EC) or remote-sensing data (e.g., satellite-derived spectral indices). One of the most common spectral indices used for delineation of management zones is NDVI (Normalized Difference Vegetation Index) from satellite sensors of high or medium spatial resolutions such as, e.g., Sentinel-2 [1,2], Landsat 8 [3,4], or PlanetScope [5].

Measurements of Soil Properties
EC scanning (soil scanning) has been surveyed in the summer of 2015 to allow the division into zones before soil sampling in subsequent years, following crop harvest, in the fields included in this study in conditions with relatively low soil moisture to better evaluate the differences between various soil textures and the relative relationships between EC and physico-chemical soil properties [37]. Fields were cultivated with disc harrow, and soil humidity was moderate. Geonics EM-38 (MK-1, first generation) was used in the vertical (ECL) mode, providing up to about 1 m soil penetration which was presented in mS/m (millisiemens per metre). Because the most important effect on the EC results has topsoil, the result of the measurement is mainly connected with the physicochemical properties of the arable layer of soil [38]. The unit was calibrated before each new field according to the manufacturer's procedures (Q and P zeroing). The scanner was installed on dielectric (polyethylene) sledge pulled by a pick-up truck in 15-20 m passages and with the speed of 15-20 km/h. EC values were recorded with 1 Hz frequency with geographical coordinates based on a DGPS receiver in a field computer with FarmWorks Mobile software.
After collecting data (ESRI SHP shapefile format), errors caused by abnormal measurements (values below zero) were filtered and cleaned in GIS software (FarmWorks Office) [39]. Proofed points were interpolated using the inverse distance weighting (IDW) method using squared-weighted interpolation where the power value was equal to 2 to create contour maps. Management zones were manually delineated based on EC values to obtain similar EC values within each zone.
One composite soil sample (consisting of 10-12 cores/subsamples) was collected from every management zone from a depth of 5-25 cm by automatic soil sampler Wintex 1000 [40] to provide the necessary quality of probes. Standard, "Z" shape of transects (zig-zag survey lines) was applied. Navigation inside zones was provided by a PC Tablet with a GPS receiver and FarmWorks Mobile software.

Measurements of Soil Properties
EC scanning (soil scanning) has been surveyed in the summer of 2015 to allow the division into zones before soil sampling in subsequent years, following crop harvest, in the fields included in this study in conditions with relatively low soil moisture to better evaluate the differences between various soil textures and the relative relationships between EC and physico-chemical soil properties [37]. Fields were cultivated with disc harrow, and soil humidity was moderate. Geonics EM-38 (MK-1, first generation) was used in the vertical (EC L ) mode, providing up to about 1 m soil penetration which was presented in mS/m (millisiemens per metre). Because the most important effect on the EC results has topsoil, the result of the measurement is mainly connected with the physico-chemical properties of the arable layer of soil [38]. The unit was calibrated before each new field according to the manufacturer's procedures (Q and P zeroing). The scanner was installed on dielectric (polyethylene) sledge pulled by a pick-up truck in 15-20 m passages and with the speed of 15-20 km/h. EC values were recorded with 1 Hz frequency with geographical coordinates based on a DGPS receiver in a field computer with FarmWorks Mobile software.
After collecting data (ESRI SHP shapefile format), errors caused by abnormal measurements (values below zero) were filtered and cleaned in GIS software (FarmWorks Office) [39]. Proofed points were interpolated using the inverse distance weighting (IDW) method using squared-weighted interpolation where the power value was equal to 2 to create contour maps. Management zones were manually delineated based on EC values to obtain similar EC values within each zone.
One composite soil sample (consisting of 10-12 cores/subsamples) was collected from every management zone from a depth of 5-25 cm by automatic soil sampler Wintex 1000 [40] to provide the necessary quality of probes. Standard, "Z" shape of transects (zig-zag survey lines) was applied. Navigation inside zones was provided by a PC Tablet with a GPS receiver and FarmWorks Mobile software.
Sampling was repeated every season from 2017 to 2019 after the crop harvest, according to the time of harvesting particular crops, starting from the middle of July (winter barley), end of July (winter oil rapeseed), and beginning of August (winter wheat).
Four-hundred thirty-eight samples (each sample consisted of soil mixed from 10-12 cores) were tested for P 2 O 5 , K 2 O, Mg (available forms), and soil pH. Chemical analysis was performed according to standard procedures, i.e., pH was measured using the potentiometric method in potassium chloride solution (KCl) [41], available phosphorus and potassium were measured using the Egner-Riehm method [42], and the content was presented in mg of P 2 O 5 and K 2 O per 100 g of soil. Content of available magnesium was measured using Schachtschabel methodology [43] and presented in mg Mg per 100 g of soil.

Crop Management
In the studied farmm three crops were planted, i.e., winter wheat, winter barley, and winter rapeseed. Figure 2 presents the fields with these crops in years 2017-2019. Fertilization applied was adjusted to nutrient requirements assuming expected grain yields of winter wheat and winter barley of 8 tons per hectare (t/ha) and 4.5 t/ha of winter rapeseed. Doses of phosphorus, potassium, and magnesium mineral fertilizers are presented in Table 1. For fields where pH was low and liming was required, calcium minerals containing about 34% of calcium carbonate (CaCO 3 ) were applied at the rate of 1-1.5 t/ha.

Statistical Analysis
Descriptive statistics such as means, standard deviations (SD), and coefficients of variations (CV) were calculated for the variables in the study. Relationships between pairs of variables were calculated using Pearson's correlation coefficients and for selected pairs of

Satellite Data
Mean values of normalized difference vegetation index (NDVI) were calculated using data acquired by Landsat 8 (for year 2017) and Sentinel-2 (for years 2018 and 2019) satellites for areas of management zones using zonal statistics in QGIS software. For the analyses, satellite imagery (C2 Level 2 product for Landsat-8 and Level-2A product for Sentinel-2) from the following dates: 19 May 2017, 3 June 2018, and 5 June 2019 was used. Such dates were selected because all three crops (winter wheat, winter barley, and winter rapeseed) were, in the region of study, during intensive growth stages (winter cereals are after anthesis or during milk maturity while winter rapeseed is in the end of flowering or during the pod-filling stage). NDVI was calculated based on red (central wavelength 655 nm for Landsat 8 and 665 nm for Sentinel-2) and near-infrared (central wavelength 865 nm for Landsat 8 and 833 nm for Sentinel-2) bands with spatial resolution 30 m for Landsat 8 and 10 m for Sentinel-2 [44]. Pixels located at the borders of the management zones were excluded from the analyses. Because NDVI for various crops cannot be compared, for the analyses for all crops together, the standardized value of NDVI was used. Standardization was performed for each crop separately, i.e., from each value mean NDVI for each crop was subtracted and then divided by the standard deviation.

Statistical Analysis
Descriptive statistics such as means, standard deviations (SD), and coefficients of variations (CV) were calculated for the variables in the study. Relationships between pairs of variables were calculated using Pearson's correlation coefficients and for selected pairs of variables linear regression was applied. Statistical analyses were preformed using Statistica 13.3 program [45]. Significance level for all the analyses was set at 0.05 probability level.

Characteristics of Chemical Soil Properties
Soil reaction as well content of available forms of phosphorus, potassium, and magnesium in most of the studied area was favorable for crops because the content of available nutrients (P, K, and Mg) in the soil was sufficient for most of the area ( Table 2). In most of the studied fields, soil reaction was optimal or near to optimal for crops, i.e., soil reaction was from moderately acidic to neutral. Such a soil reaction was characterized by 91.6% of all management zones included in the study. For only six management zones (1.4%), a strong acidic soil reaction was observed and for 31 management zones (7.1%), the soil reaction was slightly alkaline. Content of phosphorus was, for almost the entire studied area, from medium to very high according to the recommendations for Poland [46]. Only 19 management zones (4.3%) were characterized by low phosphorus content in soil. Potassium content was from medium to very high for 90% of the management zones and only for 10% (44 of 438) was low and very low. In the case of magnesium, only 15 (3.4%) management zones showed low content and for the rest of the area the content of magnesium was from medium to very high according to the recommendations for Poland [47]. The studied fields were characterized by a near-to-optimal value of pH (means about 6.2-6.4) and relatively low variability of pH (CV about 8%) ( Table 3). Mean content of available phosphorus (in mg P 2 O 5 per 100 g of soil) for all three years of the study (2017-2019) was equal to 17.2 mg, which means a high content according to the recommendations for agricultural crops [46,47]. Mean content of available potassium (in mg K 2 O per 100 g of soil) was equal to 18.14 mg, which means medium content according to the recommendations for agricultural crops. In the case of magnesium, average content was equal to 10.45 mg/100 g of soil, which means very high content for agricultural crops. Variability of the content of P, K, and Mg was similar and much higher in comparison with the variability of pH. Coefficients of variations ranged from 25.3% to 28.5% for these three nutrients (P, K, and Mg). Table 3. Means, standard deviations (SD), ranges (min-max), and coefficients of variation (CV) for the studied variables.

Relationship between pH and Nutrients
Relationships between pH and each nutrient in subsequent years were positive and strong (Table 4), which means that these soil-chemical properties were quite stable during the period of the study. the study and averages for three years (significant correlations at 0.05 significance level are in bold).

Relationships between NDVI and EC with pH and Content of Nutrients
Because various crops (wheat, barley, and rapeseed) were cultivated in each year, standardized NDVI (standard score) was used for calculation of correlation coefficients between NDVI for all crops together with pH and content of nutrients in soil. Standardization was performed separately for each crop based on mean NDVI and standard deviation. The results presented in Table 5 Table 6. Correlation coefficients between EC and standardized NDVI with soil pH and nutrients content for winter wheat (significant correlations at 0.05 significance level are in bold; color background scale for correlations the same as in Table 4). Significant relationships were observed between all of the pairs of chemical soil properties. pH of soil was correlated positively with content of phosphorus and negatively correlated with content of potassium and magnesium. Moreover, content of phosphorus was negatively correlated with content of potassium and magnesium. Potassium and magnesium were positively correlated. The relationships were similar for each year separately as well for average values for all years (2017-2019).

Relationships between NDVI and EC with pH and Content of Nutrients
Because various crops (wheat, barley, and rapeseed) were cultivated in each year, standardized NDVI (standard score) was used for calculation of correlation coefficients between NDVI for all crops together with pH and content of nutrients in soil. Standardization was performed separately for each crop based on mean NDVI and standard deviation. The results presented in Table 5  Positive correlations observed between EC with average (for 2017-2019) macronutrients content were with magnesium (r = 0.518) and potassium (r = 0.286) and negative between EC with phosphorus and pH (r = −0.358 and r = 0.057). The direction of the relationships was similar for both NDVI and EC, which confirmed that these two variables are positively correlated (the strongest positive correlation between EC and NDVI was observed in 2019, r = 0.287).
Correlation coefficients were calculated not only for total area of the study (all fields together), but also separately for each crop, for all fields with the same crop species (Tables 6-8).
The results proved moderate or strong positive correlations between NDVI (depending on the year) and EC with content of potassium and magnesium for winter wheat ( Table 6). Most of the correlation coefficients with NDVI ranged from 0.36 to 0.63 for fields with winter wheat while with EC ranged from 0.63 to 0.70. Table 8. Correlation coefficients between EC and standardized NDVI with soil pH and nutrients content for winter rapeseed (significant correlations at 0.05 significance level are in bold; color background scale for correlations the same as in Table 4). Most of the correlations observed between NDVI and EC with pH and phosphorus content in soil were negative (for NDVI for years 2018 and 2018) or positive but very weak (for NDVI for year 2017). These correlations for all winter wheat fields between NDVI and pH for years 2017-2019 ranged from −0.45 to 0.28, while between NDVI and phosphorus ranged −0.75 to 0.23. The correlation coefficient between EC with average pH for years 2017-2019 was equal to −0.29 and between EC and phosphorus equal to −0.68.

Standardized
Correlation coefficients between NDVI and EC with content of potassium and magnesium for barley for all fields together (Table 7) proved similar relationships to those observed for winter wheat. Positive correlations were observed between NDVI and EC with potassium and magnesium in 2018 and 2019; negative or very weak correlations were observed between NDVI and EC with pH and phosphorus content in soil. The correlations for all barley fields and average content of K and Mg for years 2017-2019 were stronger between potassium with NDVI (r = 0.63 for year 2019 and 0.23 for 2018 and very weak in 2017, r = 0.05) than between magnesium with NDVI (r = 0.40 in 2019, 0.11 in 2018 and negative in 2017, r = −0.35). Correlations between EC with K and Mg were positive, respectively r = 0.18 and 0.40. Correlations between NDVI and EC with pH and P were rather weak. For all barley fields and average content P in soil for years 2017-2019, the strongest correlation was with NDVI in 2019 (r = −0.33), and for years 2017 and 2018 they were weaker (−0.18 and −0.16 respectively). The correlations for content of nutrients for individual years were quite consistent with these, which were observed for average content of nutrients, i.e., correlations between NDVI and EC with K and Mg were positive and significant, while between NDVI and EC with pH and P were negative or very weak. Stronger correlations were observed for NDVI in comparison with correlations with EC, but only for the year 2019, while for the other two years they were weaker.
Correlation coefficients between NDVI and EC with content of potassium and magnesium for winter rapeseed for all fields together (Table 8) proved similar relationships (positive correlations) to these observed for winter wheat and barley; however, the correlations were slightly weaker. The correlations for all rapeseed fields and average content of K for years 2017-2019 were the strongest between potassium with NDVI for the year 2019 (r = 0.53) and in the same year the strongest between magnesium with NDVI (r = 0.32). For other years were weaker. Correlations between EC with K and Mg were positive but weaker (respectively r = 0. 23 and 0.17). Correlations between NDVI and EC with pH and P were weaker. For all rapeseed fields and average P for years 2017-2019, positive but weak correlations were observed between phosphorus with EC (r = 0.18). The correlations between nutrients and NDVI and EC for years were similar to those which were observed for average content of nutrients.
The relationships are presented in graphical form in Figures 3 and 4 with regression equations and coefficients of determination. Increase of potassium content by 1 unit (mg of K 2 O per 100 g of soil) was related to an increase of the standardized NDVI for all crops together by about 0.08 units. In case of magnesium, an increase of magnesium content by 1 unit (mg of Mg per 100 g of soil) was related to increase of the standardized NDVI by about 0.05 units. Increase of phosphorus content by 1 unit (mg of P 2 O 5 per 100 g of soil) was related to an decrease of standardized NDVI of winter wheat by about 0.05 units. A negative relationship between pH and NDVI based on linear regression proved that an increase of pH by one unit was related to the decrease of the standardized NDVI by about 0.25 units. Moreover, an increase of potassium by 1 unit (mg K 2 O per 100 g of soil) for fields with winter wheat was related to an increase of EC by about 0.24 units. For magnesium, an increase of content by 1 unit (mg of Mg per 100 g of soil) was related to an increase of EC by 0.68 units. An increase of phosphorus by one unit (mg of P 2 O 5 per 100 g of soil) was related to a decrease of EC by 0.3 units. The relationship between EC and pH was negative but very weak.

Discussion
The results obtained in this study proved positive significant relationships between potassium and magnesium content in soil with NDVI and EC. Stronger relationships in this study were observed for cereals (winter wheat and winter barley) than for rapeseed. Since NDVI is strongly correlated with grain yield, we can conclude that the content of potassium and magnesium have a strong positive effect on the grain yield of winter wheat and winter barley. In this study, stronger relationships were observed between content of macronutrients with NDVI in the year 2019 than with EC, but for the other two years (2017 and 2018) the relationships were weaker. This means that delineation of management zones for stratified soil sampling can be efficient based on EC, which is directly connected with soil properties as well as based on NDVI, which is indirectly (by crop status) related to soil properties. EC is variable and can be used for the delineation management zones for site-specific crop management as a sole attribute or together with other auxiliary variables [17][18][19][20][21]. Such an approach demands proximal soil sensing using special equipment as well performing mapping of the fields at an appropriate time (very often after the crop harvest). Instead of EC maps, it is possible to use other mapping techniques of the field, including satellite remote sensing. Satellite remote-sensing data are available for free from, e.g., Sentinel and Landsat satellites. It allows the use of such data for the delineation of management zones without any additional measurements [48]. However, we should notice that cereals like winter wheat or barley are better crops for NDVI-based delineation of management zones in comparison with rapeseed because of stronger relationships with the contents of nutrients in soil. Moreover, the relationships between NDVI and soil properties can vary in different years.
In this study, negative significant relationships were observed between pH and phosphorus content in soil with NDVI and EC. Especially strong relationships were observed for winter wheat. The results were opposite to the expected because they mean that the higher pH and phosphorus content, the lower NDVI, and as a result the lower the grain yield. However, we should notice that negative correlations between potassium and magnesium content with phosphorus content were observed. Moreover, high phosphorus content in soil was related with high pH. High pH can cause lower ability for uptake by plants of phosphorus because of phosphorus fixation by calcium (precipitation of Ca phosphate minerals) [49][50][51]. Negative correlations between content of phosphorus content in soil with yields of various crops, including cereals, were found in some previous studies [27,52]. Similar results, i.e., a negative relation between NDVI with content of phosphorus in soil and pH, were observed in the study of Serrano et al. [27]. The highest phosphorus content (more than 50 mg P per kg) and the highest pH (5.6) were observed for the within-field management zone where NDVI as well yield potential was the lowest in comparison with medium-and high-potential management zones (P in the range of 20-30 mg per kg and pH in the range of 5.4-5.5). Another study [16] which presented relationships between soil properties with NDVI and grain yield of cereals (wheat and barley) proved a positive correlation between content of potassium with NDVI and grain yield (positive values of regression coefficients in multivariate model), but the relationship with content of phosphorus was not consistent, relatively weak, but in some cases positive and in some cases negative. A study on the delineation of management zones for a cotton field in Eastern China [53] proved negative relations between content of potassium as well pH with NDVI and cotton yield. The lowest pH (7.56) and content of available potassium (96 mg per kg) was observed for the management zone with the highest productivity, while the highest values (pH = 7.90 and 14 mg K per kg) were observed for the zone with the lowest productivity. Other factors such as content of nitrogen and organic matter were positively correlated with grain yield and NDVI. The results prove that potassium and phosphorus are not crucial for obtaining high yields. We should notice that usage of NDVI have some limitations including lack of availability of satellite images because of cloud cover [54], atmospheric effects, sensor factors, the saturation phenomenon [55], or other problems such as the existence of weeds which affect the value of the spectral index [56]. These problems should be avoided, e.g., by using multi-temporal satellite data.

Conclusions
Relationships between NDVI of cereals with macronutrients content in soil and pH were similar to relationships between EC and soil chemical properties. Because of this, both satellite-derived NDVI and EC are variables of similar usefulness for the delineation of within field management zones. The same relationships were much weaker and not consistent for winter rapeseed and because of that NDVI for rapeseed should, rather, not be used as a variable for the delineation of management zones for precision agriculture. Other spectral vegetation indices should be tested in future research on management zones delineation. In the case of EC, measurements at various soil moisture should be tested to find the optimal soil conditions for EC measurement for management zones delineation. Soil chemical properties such as pH and content of P, K, and Mg are quite stable in recent years and because of that delineated management zones can be used for stratified soil sampling in a period of about 5 years when these properties are quite stable in time.

Data Availability Statement:
The data presented in this study are available on request from the corresponding author. The data are not publicly available due to ongoing unpublished research.

Conflicts of Interest:
The authors declare no conflict of interest.