Integrating Remote Sensing and Ground Data to Assess the Effects of Subsoiling on Drought Stress in Maize and Sunflower Grown on Haplic Chernozem

Abstract

In drought-prone regions without irrigation systems, effective agrotechnologies such as subsoiling are crucial for enhancing soil infiltration and water retention. However, the effects of subsoiling can vary depending on crop type and environmental conditions. Despite previous research, there is limited understanding of the contrasting responses of C3 (sunflower) and C4 (maize) crops to subsoiling under drought stress. This study addresses this knowledge gap by assessing the effectiveness of subsoiling as a drought mitigation practice on Haplic Chernozem in Northern Bulgaria, integrating ground-based and remote sensing data. Soil physical parameters, leaf area index (LAI), canopy temperature, crop water stress index (CWSI), soil moisture, and yield were evaluated under both conventional tillage and subsoiling for the two crops. A variety of optical and radar descriptive remote sensing products derived from Sentinel-1 and Sentinel-2 satellite data were calculated for different crop types. Consequently, the use of machine learning, utilizing all the processed remote sensing products, enabled the reasonable prediction of LAI, achieving a coefficient of determination (R²) after a cross-validation greater than 0.42 and demonstrating good agreement with in situ observations. Results revealed differing responses: subsoiling had a positive effect on sunflower, improving LAI, water status, and slightly increasing yield, while it had no positive effect on maize. These findings highlight the importance of crop-specific responses in evaluating subsoiling practices and demonstrate the added value of integrating unmanned aerial systems (UAS) and satellite-based remote sensing data into agricultural drought monitoring.

1. Introduction

Drought is a widespread natural hazard that increasingly affects larger territories of the world and poses a serious threat to agricultural productivity and food security [1,2,3]. In the context of climate change and the increased frequency of extreme weather events, the adoption of effective soil management strategies has become critical. Farmers in drought-prone areas are more likely to adopt drought mitigation techniques that improve soil water storage and crop resilience to stress. In regions with a moderate continental climate, such as Northern Bulgaria, summer droughts are a major driver of variability in summer crop yields [3]. In areas lacking irrigation systems, soil management practices such as subsoiling have been proposed to enhance soil water retention and root development, thereby reducing runoff and evaporation losses.

Subsoiling, also referred to as deep ripping or deep tillage, is a non-inversion tillage practice that mechanically loosens compacted subsoil layers (plough pans) without turning over the topsoil [4]. Typically performed at depths ranging from 30 cm to over 50 cm, subsoiling is designed to improve root penetration and water infiltration, particularly in soils affected by compaction. In our study, subsoiling was applied at 50 cm depth. The effectiveness of subsoiling, however, is highly variable and depends on several factors, including soil texture, the presence of restrictive layers, tillage depth, soil moisture content at the time of tillage, and the frequency and duration of dry spells [4]. While it can increase the availability of plant-available water in the root zone, it may also lead to potential conflicts over water resources, especially under future climate conditions with greater variability in precipitation [5].

Evaluating drought mitigation strategies requires not only an understanding of soil physics and crop physiology but also effective monitoring tools that operate at multiple scales. The integration of multi-physical and climatic indicators with social or management-related variables has proven more effective for drought early warning and information systems [1,2]. Assessing the effect of subsoiling thus involves analyzing changes in soil compaction, soil moisture recharge, and plant response, both in time and space. However, this evaluation is complicated by field heterogeneity, which is best captured through remote sensing techniques.

One of the most widely used indicators for crop water stress is canopy temperature, measured via infrared thermometry since the 1960s [6]. Modern applications increasingly rely on unmanned aerial systems (UAS) that produce precise geospatial models of land surface temperature (LST), offering high-resolution data for assessing crop health and stress over large fields [7]. These UAS-based thermal measurements, when combined with meteorological data and vegetation indices, enable the precise calculation of indicators like the crop water stress index (CWSI) and their relation to soil water status [8,9]. The advantages of UAS include high spatial and temporal resolution, flexible deployment, and integration with RGB and multispectral sensors for simultaneous monitoring of structural and physiological crop traits.

Despite the recognized potential of subsoiling and remote sensing tools, there is a lack of integrated studies evaluating their combined use in assessing drought mitigation effects across different crop types. In particular, the comparative response of C3 (e.g., sunflower) and C4 (e.g., maize) crops to subsoiling under drought conditions remains poorly understood.

Significant advancements have been made in recent years in soil moisture and leaf area index retrieval using remote sensing data, particularly from synthetic aperture radar (SAR) and optical data sources. Specifically, machine learning approaches have been developed to enhance retrieval accuracy. The integration of Sentinel-1 (SAR) and Sentinel-2 (optical) data has shown promising results, particularly under the canopy of agricultural crops, leveraging backscattering models and machine learning algorithms [10,11]. Recent remote sensing studies have highlighted the dynamic nature of soil moisture retrieval research, emphasizing the need for the continued innovation and integration of diverse remote sensing technologies [12,13,14].

The main objective of this study is to evaluate the effectiveness of subsoiling at 50 cm depth on Haplic Chernozem as a drought mitigation practice for maize and sunflower cultivation. We integrate ground-based measurements of soil and crop parameters with UAS-derived LST data and satellite-based LAI models to assess differences in crop water stress under conventional tillage and subsoiling. The novelty of this study lies in its combined utilization of high-resolution thermal imagery, in situ soil moisture and LAI data, and crop-specific analysis to understand the differential responses of C3 and C4 crops to deep tillage under drought-prone conditions in Northern Bulgaria.

2. Materials and Methods

2.1. Study Area

2.1.1. Location

This study was carried out in two agricultural fields located south of the village of Kalipetrovo, Silistra region, in the northeast part of the Danube Plain of Bulgaria (Figure 1). The topography is flat, with an altitude of 129 m. The region belongs to the European moderate-continental climate zone, but it differs from rest part of the Bulgarian Danube plain with respect to precipitation. The region is one of the driest in the country, with a mean annual precipitation sum of 500 mm, the frequent occurrence of prolonged (18–20 day) drought periods, and lower spring and summer precipitation [15]. The mean air temperature of the coldest month (January) is −1.7 °C, and the mean air temperature of the hottest month (June) is 23.0 °C [15].

2.1.2. Meteorological Characteristics and Drought Indicators

The data on the meteorological parameters (precipitation, air temperature and humidity, solar radiation, wind speed) was collected in 2024 with the automated meteorological station Davis Vantage Pro2. The reference evapotranspiration (ETo) was calculated via the Penman–Monteith method as described in the FAO56 methodology [16]. The meteorological conditions of the experimental year are compared with the long-term characteristics obtained over the long-term period (1951–2024 for air temperature and precipitation and 1982–2024 for the reference evapotranspiration) (Figure 2). February, April, June, and August 2024 were extremely hot (Figure 2b), and June and July were extremely dry.

The other indicators for drought used in this study are the standard precipitation index (SPI), calculated for 3- and 12-month timescales to reflect agricultural and hydrological droughts [17,18], and the Palmer drought severity index (PDSI) [19,20]. The SPI is recommended by the World Meteorological Organization to be used by all national meteorological and hydrological services [21]. The SPI measures precipitation anomalies after transforming the fitted “gamma” probability distribution of long-term series into a normal distribution such that the mean SPI value is zero for location and period in question, and the standard deviation is 1. As part of its water balance calculation, the PDSI incorporates data for monthly air temperature and precipitation, as well as for soil water holding capacity. The PDSI has a timescale of approximately nine months and thus mostly reflects the long-lasting consequences of drought occurrence.

2.1.3. Soil Profile Characteristics

According to the 1:10,000 soil map [22], the studied area was homogenous and occupied by Haplic Chernozems, formed on loess material. A deep (1.5 m) soil profile was revealed in the field S_CT (44.06964 N, 27.2589 E, 129 m a.s.l.) in March 2024. The basic soil properties of this profile are presented in

Table 1. Basic soil properties.

Horizon	Depth, cm	Sampling Depth, cm	Munsell Color Moist	SOC, %	Total Carbonates, %	pH in H2O	Sand (2–0.063 mm), %	Silt (0.063–0.002 mm), %	Clay (<0.002 mm), %	Soil Texture Class
Ap	0–35	0–5	10YR 3/2	1.1	0	6.5	1.6	64.1	34.3	SiCL
		15–20	10YR 3/2	1.1	0	6.6	6.4	61.4	32.1	SiCL
		30–35	10YR 3/2	1.2	0	6.4	3.4	63.9	32.7	SiCL
A	35–58	45–50	10YR 4/3	0.7	0	6.7	4.1	62.0	33.9	SiCL
B1	58–82	68–73	10YR 5/3	0.2	0.4	7.1	5.6	63.0	31.4	SiCL
B2	82–110	93–98	10YR 5/4	0.2	traces	7.4	3.4	67.5	29.1	SiCL
Ck	110–150	125–130	10YR 6/3	0.2	16.9	8.0	8.5	66.0	25.5	SiL

. The soil has a homogeneous silty clay loam (SiCL) texture to a depth of 1 m and a coarser texture below this depth, where significant carbonates content was determined (Table 1). The dominant fraction is the silt (61–68%). The clay content, which is an important factor in terms of soil structure stability, decreases from 34% to 29% at a 1 m depth. The content of SOC in the A horizon is low (0.7–1.2%). The soil reaction is slightly acidic to neutral (pH 6.4–7.0) to a depth of 82 cm and slightly alkaline below this depth due to the presence of carbonates.

2.2. Tillage Practices

This study was conducted on two adjacent agricultural fields: one cultivated with sunflower and the other with maize. The sunflower field (designated as “S”), encompassing an area of 45.0 hectares, and the maize field (designated as “M”), encompassing an area of 44.2 hectares, were each subdivided into two approximately equal parts to facilitate the comparison of two different tillage systems. These systems were labeled as “RT” for deep subsoiling (ripping) and “CT” for conventional (traditional) tillage practices (Figure 1b).

Deep subsoiling operations were performed on the subplots S_RT and M_RT on 15 November 2023. The soil was tilled to a depth of 50 cm using a Case IH Magnum 340 tractor equipped with a Lemken Karat 9 subsoiler. In contrast, traditional tillage practices were applied on subplots S_CT and M_CT. This included plowing to a depth of 35 cm, conducted on the same date (15 November 2023), using the same tractor fitted with a Lemken Diamant Plow. Subsequent secondary tillage operations included disking to a depth of 20–25 cm on 25 and 26 February 2024, and cultivation to a depth of 12 cm, performed on 3 April 2024 in the sunflower field and on 7 April 2024 in the maize field. These operations were carried out using a Lemken KORUND cultivator attached to the Case IH Magnum 340 tractor.

The sunflower grown during the experimental period was hybrid Futura, selected by Syngenta (Basel, Switzerland). Sowing took place on 6 April 2024. For maize sowing was undertaken using the hybrid PR 9367, selected by Pioneer (Johnston, IA, USA), with sowing carried out on 11 April 2024. Sowing for both crops was performed with a Väderstad Rapid 400S seed drill (manufactured by Väderstad, Sweden), coupled with a Deutz Agrotron 260 tractor (manufactured by Deutz Fahr, SDF Group, Lauingen, Germany). The seeding rate for both crops was standardized at 62,000 seeds per hectare.

Both crops were cultivated within a three-year crop rotation system: maize was sown following wheat; sunflower was sown following maize; and wheat followed sunflower. This rotational scheme is designed to optimize soil fertility, reduce pest and disease pressures, and enhance overall crop productivity under regional agroecological conditions.

2.3. Ground-Based Data Collection

The ground-based data set from each field includes basic soil properties, compaction indicators, soil hydraulic properties, soil moisture data, leaf area index (LAI), and crop yields. The available archive climate records of the National Institute of Meteorology and Hydrology and the daily meteorological data for 2024 obtained by an automated meteorological station were used to describe the drought occurrence.

2.3.1. Soil Sampling and Applied Analytical Methods for Soil Properties

The soil physical properties under all variants were studied at depths of 0–5, 15–20, and 45–50 cm in March and June 2024. Soil texture fractions were determined by the sieving and pipette methods set out in [23]. The concentration of soil organic carbon (SOC, %) was determined by the modified Tjurin method [24,25]. The acidity of soil was measured in a soil–water suspension of 1:2.5 with a pH meter. The soil compaction and soil hydraulic properties were determined on soil cores in 4 replicates. The field capacity (FC) (soil water retention at pF 2.0) was determined by a suction plate method using shot filters (G5) connected to a hanging column, as described by [26]. The wilting point (soil water retained at pF 4.2) was determined using fine earth (<2 mm) samples using pressure membrane apparatus (Soilmoisture equipment Corp., Goleta, CA, USA). The hygroscopic water content at pF 5.6 at the water adsorption parts of the pF curves was determined using the vapor pressure method with controlled relative humidity (75%) in desiccators containing a saturated solution of NaCl. The obtained soil water retention curves are presented in Figure S1.

The soil compaction was assessed by several indicators: soil bulk density (Db); total porosity (Pt); soil aeration capacity (AC), i.e., volume of air filled pores at water retained at pF2.0; and packing density (PD). These indicators, along with their threshold values for optimal and critical values, are described in detail in [27].

The packing density (PD, g cm⁻³) is widely used as an indicator of the status and vulnerability to compaction of soils with different textures [28,29]. The compaction classes are as follows: high, PD > 1.75 g cm⁻³; medium, 1.40 g cm⁻³ < PD < 1.75 g cm⁻³; and low, PD < 1.40 g cm⁻³ [28]. It is calculated as described in [28]:

PD = Db + 0.009 × clay

(1)

where Db (g cm⁻³) is the soil bulk density, and clay is the content of soil particles (<0.002 mm).

The determination of the total porosity (Pt), relative field capacity (RFC), aeration capacity (AC), and plant-available water capacity (PAWC) is described in [27,30]. In brief, their estimations are as follows:

Pt = (1 − Db/Ds) × 100

(2)

RFC = θ_pF2.0/Pt

(3)

AC = Pt − θ_pF2.0

(4)

PAWC = θ_pF2.0 − θ_pF4.2

(5)

where Pt is the total porosity (%vol.), Db and Ds denote the measured soil bulk and particle densities (g cm⁻³), and θ_pF2.0 and θ_pF4.2 denote the volumetric water contents (%) at field capacity (pF2.0) and the wilting point (pF4.2).

The saturated hydraulic conductivity (Ksat) was determined in laboratory conditions using 200 cm³ metal rings via the falling head method [31].

2.3.2. Soil Moisture Data

The volumetric water content (VWC) of soil was recorded by TEROS 12 (Meter Group, München, Germany) sensors installed at 15–20 and 45–50 cm soil depths. Additionally, at the time and locations of the LAI measurements (Figure 1b), the VWC in the top soil was measured by a FieldScout TDR350 soil moisture meter (Spectrum Technologies Inc., Aurora, IL, USA).

2.3.3. Crop Characteristics

The Leaf Area Index (LAI) was recorded monthly during the growing season for both maize and sunflower. Measurements were taken at 64 locations in the maize field and 59 location in the sunflower field using an AccuPAR LP-80 Ceptometer (Meter Group, München, Germany). The device is a portable, linear PAR sensor that uses 80 individual sensors spaced 1 cm apart to measure light interception levels within plant canopies. It calculates LAI by comparing the photosynthetically active radiation (PAR) above and below the canopy. At each measurement point, 5 to 10 PAR readings were taken above and below the canopy to calculate an average LAI value.

The plant height and yield were measured using a wooden frame of the known area, which was randomly placed within the crop. All plants within the frame were cut and analyzed. This sampling was conducted in four replicates. Grain yield was standardized to a moisture content of 10% for maize and 4.6% for sunflower.

2.4. Remote Sensing Data

Two sources of remote sensing data were used—a thermal image created by an unmanned aerial system (UAS), obtained on 2 July 2024, and a satellite data for LAI, acquired on 5 July 2024.

2.4.1. Thermal Survey with UAS

For the purposes of this study, we conducted targeted mapping of the land surface temperature (LST) of the studied area. We used the EbeeX platform (Figure 3a) from AgEagle, (Wichita, KS, USA) equipped with an integrated thermal and photogrammetric sensor DueT (Figure 3b), which enables the generation of a digital thermal geospatial model with high spatial accuracy and resolution. This is an innovative device and one of the lightest integrated thermal radiometric and photogrammetric sensors available, featuring synchronization technology for effective thermal mapping of small-to-medium-sized areas. The thermal sensor itself is a FLIR Tau 2, featuring a fixed focal length of 40 mm (equivalent to 35 mm), an image frequency of 30 Hz, automatic FFC calibration, a spectral range of 7.5–13.5 μm, and a sensitivity of 50 mK. It is technologically integrated with a calibrated S.O.D.A photogrammetric camera, which, in turn, synchronously generates georeferenced RGB images with a resolution of up to 20 Mpx and with a focal length of 28 mm (35 mm equivalent), which complement, and are systematically integrated with, thermal infrared images with a resolution of 640 × 512 px with a focal length of 40 mm (35 mm equivalent). The camera is controlled by the unmanned aerial system, with an adaptive automated mechanism that adjusts and configures the sensor parameters according to the flight conditions. The two sensors work in tandem and are synchronized, with direct image georeferencing, which is greatly facilitated by the RTK/PPK functionality of the flight platform.

As a result of applying the system, which utilizes a specialized photogrammetric processing protocol based on the structure–from–motion method, a precise orthorectified and georeferenced model of the Earth’s surface temperature was created, with a spatial resolution of 25 cm/pixel and a positional accuracy of within 5 cm (Figure 4).

A separation of vegetated from non-vegetated pixels is performed by employing supervised classification in ArcGIS-Pro based on the support vector machine (SVM) machine learning algorithm [32]. Samples are collected based on high-resolution orthophotography, with a pixel size of 40 cm, imaged from UAS on the same date as the thermal imagery acquisition.

The canopy temperature data were used to calculate the crop water stress index (CWSI) and assess the homogeneity of the crop cover.

2.4.2. LAI Retrieval Modeling

Sentinel-1 and Sentinel-2 data were acquired on 5 July 2024.

Data was processed via the web-based satellite platform Sentinel-Hub^® by PlanetLabs (Sinergise Solutions, Ljubliana, Slovenia) [33]. The reason is that Sentinel-Hub by PlanetLabs (Sinergise Solutions, Ljubliana, Slovenia)^® is an easy-to-use, web-based platform for processing large amounts of satellite datasets, with a primary focus on data from ESA’s Copernicus Sentinel missions, as well as a coding repository, which facilitates the calculation of the products from satellite imagery [34]. Two collections were used for this purpose—Sentinel-2 (S2) L2A and Sentinel-1 (S1) GRD, representing optical and synthetic aperture radar (SAR) measurements around, or at the date of, the in situ measurements. Calculated products from optical S2 L2A data comprise calibrated reflectance for bands B01, B02, B03, B04, B05, B06, B07, B08, B8A, B11, and B12, as well as the following set of spectral indices: LAI, MSI, NDMI, NDWI, and SAVI, with a spatial resolution of 10 m. Regarding the SAR data from S1, assessing both the descending and the ascending orbits provides two different high- and low-incidence angles for testing. The Sentinel-1 satellite passes over the test sites at approximately 16:10 local time at the ascending node and around 04:05 at the descending node, which is in the early morning, when the relative humidity is higher near the ground. The S1 SAR data processing via Sentinel-Hub by PlanetLabs (Sinergise Solutions, Ljubliana, Slovenia)^® comprises normalization with terrain flattening, speckle incoherent multiplicative filtering (refined Lee) with a sliding window of 5 × 5 pixels, and calculation of the gamma nought (γ0) backscatter coefficient, aiming for a square pixel size of 10 m [35]. The following polarimetric features from both VV and VH polarizations were calculated: γ0_VV and γ0_VH, as well as the polarization ratio: γ0_VH/γ0_VV. The geocoded products were delivered as GeoTIFFs, 32-bit floating-point. The data was inspected in GIS.

Leaf area index (LAI) measurements and remote sensing data from Sentinel-1 and Sentinel-2 are used as inputs to develop a nonparametric model. To estimate LAI, we employed a quantile regression forests (QRF) model [36,37]. QRF is an extension of random forest regression [38] and provides conditional quantiles for estimating epistemic uncertainty. We applied the band analysis tool (BAT) [39] to determine the most responsive spectral bands for estimating crop height and to identify the minimum number of bands required to maintain acceptable accuracy. BAT used a backward band reduction approach, beginning with the complete set of available bands. In each iteration, the least important band was eliminated and the model was recalibrated. Accuracy was assessed at every step for each band subset, allowing us to pinpoint the most effective bands.

The QRF modeling was conducted using version 3.32 of the ARTMO toolbox [40] (https://artmotoolbox.com/, accessed on 6 May 2025). The model was set up using ARTMO’s default and recommended settings to ensure robust regression performance. In addition, QRF derives conditional quantiles from the ensemble of decision tree outputs, enabling the estimation of epistemic uncertainty [37] and capturing the variability in the model’s predictions. This capability allows QRF to provide not only predictions but also insights into the confidence of those predictions.

We trained three models: Model 1 (M1), which utilized only sunflower samples; Model 2 (M2), which employed only maize data; and Model 3 (M3), which utilized a combined dataset containing both sunflower and maize samples.

To evaluate the accuracy and generalizability of our predictive model, we used 3-fold cross-validation for M1 and M2 and 5-fold cross-validation for M3. This approach was tailored to the size and composition of each dataset, ensuring a reliable evaluation of model performance. Additionally, an independent test set, separate from the training data, was set aside to evaluate the final model. The dataset was randomly divided, with 70% used for training and 30% for testing, while ensuring that the proportion of conventional tillage and subsoiling samples in each set reflected their original distribution in the full dataset. This approach ensures a robust evaluation of model performance and its ability to generalize to new data. Model performance was measured using five metrics: coefficient of determination (R²); root mean square error (RMSE); normalized RMSE (nRMSE), relative RMSE (rRMSE); and mean absolute error (MAE). The formulas for these metrics, as described by Richter et al. (2012) [41], align with those used in the ARTMO toolbox. The optimal model for LAI estimation was then used to generate a high-resolution LAI map.

2.5. Crop Water Stress Index

The obtained data for the canopy temperature on 2 July 2024 are used to calculate the difference (ΔT) between the canopy temperature of the crop (Tc) and the ambient air (Tair = 31 °C)

ΔT = Tc − Tair

(6)

The threshold value for the occurrence of water stress depends on the vapor pressure deficit, the crop type, the crop development stage, and the soil hydraulic properties. This crop water stress indicator can be used to compare the tillage effect on both crops at the same meteorological and crop conditions.

The crop water stress index (CWSI) is calculated based on the canopy temperatures (Tc) data, extracted from the map, as described in Section 2.4:

$$\mathrm{C}\mathrm{W}\mathrm{S}\mathrm{I}={\displaystyle \frac{\mathrm{T}\mathrm{c}-\mathrm{T}\mathrm{p}\mathrm{o}\mathrm{t}}{\mathrm{T}\mathrm{p}\mathrm{o}\mathrm{t}-\mathrm{T}\mathrm{d}\mathrm{r}\mathrm{y}}}$$

(7)

where Tpot and Tdry correspond to the canopy temperatures at the maximum rate of transpiration and no transpiration at all, respectively. Jackson et al. [42] proposed a theoretical justification for the water stress index based on the ratio of the actual (ETc) to the maximum (ETmax) evapotranspiration of the crop:

CWSI = 1 − ETc/ETmax

(8)

We test the estimations of Tpot and Tdry as 0.5% and 99.5% percentiles of the canopy temperature histograms, as proposed in [7,43,44]. The assumption is that at the time of measurement, at least 0.5% of the plants were not transpiring at all, and at least 0.5% of the plants were transpiring at a potential (maximal) level. The occasional water spills of the underground pipe created well-watered conditions in some parcels in the sunflower field, which was used for Tpot estimation (Figure 4). In our case, a better approach to determining Tdry is to fix it to a 5 °C difference with the ambient air temperature [45]. CWSI is also related to other indicators of water stress, such as leaf water potential, transpiration rate, CO₂ exchange, stomatal conductance, and the relative depletion of plant-available soil water.

2.6. Statistical Analyses

The descriptive statistical analyses and one-way ANOVA analyses were performed using STATGRAPHICS Centurion 18 software and Excel.

3. Results

The effectiveness of the applied tillage depends on various factors, including soil properties, the occurrence of drought conditions, and crop responses.

3.1. Effects of Subsoiling on Soil Physical Properties

The homogeneity of the studied fields is an important condition for comparing the effects of the applied agrotechnologies on the soil properties. Soil texture is a basic property that influences the other soil physical properties. The texture fractions do not statistically differ between the top 0–5 and 15–20 cm layers and the studied variants (Table S1). The silt fraction does not change significantly with depth, and it is 63.8 ± 1.8% (Cv = 1.4%) at a depth of 0–20 cm and 63.1 ± 0.8% (Cv = 1.3%) at a depth of 45–50 cm (Table S1). The average clay content value in the 0–20 cm layer is 33.2 ± 0.7% (Cv = 2.1%) and increases by 1% in the 45–50 cm layer to 34.3 ± 0.4% (Cv = 1.3%) on account of the sand fraction (Table S1).

The data for soil bulk density (Db), total porosity (Pt), field capacity (FC), relative field capacity (RFC), wilting point (WP), plant-available water capacity (PAWC), and aeration capacity (AC), measured in June 2024, are presented in

Table 2. Soil parameters under conventional tillage (CT) and subsoiling (RT) of sunflower (S) and maize (M) fields, June 2024. Db—soil bulk density; Pt—total porosity; FC—field capacity; RFC—relative field capacity; WP—wilting point, PAWC—plant-available water capacity; AC—aeration capacity. p—significance level of F ratio of ANOVA analyses of the variants (different letters (a, b) show significant (p < 0.05) differences between all variants).

Depth, cm	Variant	Db, g cm−3	Pt, %vol.	FC, %wt	RFC	WP, %wt	PAWC, %vol.	AC, %vol.
0–5	S_CT	1.20 a	55.9 a	26.2 a	0.56 a	11.3 a	17.9 a	24.5 b
	S_RT	1.25 a	53.7 a	27.3 a	0.64 ab	11.2 a	20.1 a	19.5 ab
	M_CT	1.23 a	54.6 a	26.5 a	0.60 a	11.6 b	18.3 a	22.0 ab
	M_RT	1.31 a	51.8 a	29.6 b	0.76 b	11.3 a	24.0 b	13.1 a
	F	1.01	0.99	8.38	3.26	8.13	7.06	2.51
	p	0.44	0.44	0.008	0.08	0.008	0.01	0.13
15–20	S_CT	1.23 a	54.2 a	26.7 a	0.61 a	11.8 a	18.3 a	21.4 a
	S_RT	1.50 b	44.3 b	24.4 b	0.82 b	12.0 a	18.6 a	7.7 b
	M_CT	1.48 b	45.1 b	24.7 b	0.81 b	12.4 b	18.1 a	8.7 b
	M_RT	1.51 b	43.9 b	24.2 b	0.83 b	12.1 ab	18.3 a	8.4 b
	F	12.4	12.1	15.4	9.95	7.9	0.4	10.3
	p	0.002	0.002	0.001	0.004	0.01	0.75	0.004
45–50	S_CT	1.56 a	42.8 a	23.1 a	0.84 a	12.3 a	16.8 a	6.8 a
	S_RT	1.49 b	45.2 b	23.8 a	0.78 b	12.7 ab	16.6 ab	9.7 b
	M_CT	1.47 b	46.0 b	23.5 a	0.75 b	12.9 b	15.7 b	11.4 b
	M_RT	1.49 b	45.2 b	23.7 a	0.78 b	12.8 ab	16.3 ab	9.9 b
	F	4.59	5.23	1.1	9.6	2.2	2.6	8.3
	p	0.004	0.03	0.39	0.005	0.16	0.12	0.008

. The compaction indicators (Db, Pt, RFC, and AC) show that subsoil compaction under S_RT, M_CT, and M_RT occurred even at 15–20 cm, while under S_CT, it was detected in the deeper layer (45–50 cm) (Table 2).

The profile distributions of PD, AC, and RFC indicate well-expressed compaction from 30 to 50 cm soil depth (Figure 5a–c). The top 0–5 cm soil layer is the most loosened, especially in March under CT before sunflower sowing. The equilibrium state of soil bulk density in Bulgaria, as well as in other European countries, occurs in June [46,47]. Although the increased Db, the top layer remains well aerated (AC > 10%), and in most cases, it can be considered water limited (RFC < 0.6) [27]. The lowest values of PD and RFC, as well as the highest values of AC, were observed under S_CT in spring, when the soil is prepared for sowing.

The PAWC is assessed as “good” (between 15 and 20%) along the soil profile (Figure 5d) and optimal (24% vol.) in the top layer under subsoiling of maize (Table 2).

It has to be mentioned that although the PD, AC, and RFC indicate the highest compactness at a depth of 45–50 cm under S_CT, sampled in March, the saturated hydraulic conductivity remained moderately high (5.13 ± 5.2 cm h⁻¹) at this depth (Figure 5b). This confirms the suggestion of Raynolds et al. [27] that several indicators should be used to describe the compactness of the soil.

3.2. Meteorological Drought

The onset of a drought (PDSI < −1.0) in 2022 occurred in July, when the PDSI reached −1.36 (Figure 6). In the winter and summer of 2023, the drought in most of the months was classified as moderate (−2.0 ≤ PDSI ≤ −2.99), reaching −3.03 in October. The 2024 experimental year was dry until September. According to PDSI and SPI12, the drought was severe in July and August. While PDSI and SPI12 reflect the long-lasting drought effects, SPI3 accounts for the shorter fluctuations. According to SPI3, the drought was extreme in July 2024 (SPI3 = −1.71) at the time of the mapping of the land surface temperature (LST) with the UAS.

3.3. Effects of Subsoiling on Soil Moisture

The registered volumetric water content (VWC) with TEROS12 sensors confirmed the occurrence of soil drought (Figure 7). During the last ten days of June and in August, the VWC was near or even below the wilting point. The lower VWC in sunflower than in maize indicates the higher water consumption by this crop, especially under conventional tillage.

The volumetric water content (VWC) of the top soil (0–10 cm), measured by FieldScout, was, on average, higher by 1.8%vol. in the subsoiling variant than in the conventional-tillage sunflower field on 2 July (

Table 3. Volumetric water content (VWC) in the top 0–10 cm soil layer, measured on 2 July 2024 by the FieldScout TDR350. n—number of observation locations; FC—field capacity; WP—wilting point; p—significance level of the F ratio in the ANOVA analyses of the variants (different letters (a, b) show significant (p < 0.05) differences between all variants); Cv—coefficient of variation.

Parameter	Sunflower		Maize		F (p)
	CT	RT	CT	RT
VWC
n	30	27	17	47
mean, %vol.	14.5 a	16.3 ab	18.5 b	14.4 a	3.7 (0.014)
std, %vol.	3.7	4.8	3.7	5.7
Cv, %	25.7	29.8	19.9	39.5
min, %vol.	8.2	7.1	12.5	3.5
max, %vol.	22.3	26.4	25.5	29.2
FC, %vol.	31.4	34.3	32.6	38.7
WP, %vol.	13.5	14.1	14.3	14.8

). The surface VWC of conventional-tillage (M_CT) maize fields was significantly higher (by 4%vol.) than that of the subsoiling (M_RT) variant. This can be explained by the optimal water holding soil properties of the M_CT variant—i.e., the higher field capacity, the soil bulk density, the optimal RFC, and not-so-high AC (Table 2). Despite the high variability (Cv = 19.9–39.5%) of the VWC due to the prolonged duration of the field survey, all maximum values were below the FC of the top soil.

3.4. Crop Response

The development of sunflower under subsoiling (S_RT) progressed more slowly compared to the traditionally cultivated variant (Figure 8, Table S2). The measured leaf area index (Figure 8) and plant height (

Table 4. Parameters of sunflower (S) and maize (M) under conventional tillage (CT) and subsoiling (RT), different letters (a, b) show significant (p < 0.05) differences between all variants.

Crop	Variant	Seeding Density (Plants ha−1)	Plant Height (cm)		Yield, kg ha−1
			30 May 2024	2 August 2024
Sunflower	S_CT	56,750 ± 3304	14.5 ± 1.9	166.1 a ± 1.5	2000 a ± 8 (2000 *)
	S_RT	58,250 ± 2754	12.3 ± 2.1	166.4 a ± 6.7	1950 a ± 31 (2250 *)
Maize	M_CT	58,250 ± 3774	16.5 ± 1.7	239.0 a ± 5.6	1800 a ± 14
	M_RT	56,500 ± 3109	16.5 ± 1.0	215.8 b ± 18.0	1675 a ± 5

* Yield estimated after harvesting the whole area.

) were lower in May and June for the S_RT variant. However, by July, a higher LAI was recorded compared to the conventionally tilled variant. The statistically significant differences in LAI between the S_CT and S_RT variants across the sunflower stages (Table S2) can be explained by the different rates of development of the taproot system and the rate of water consumption in the deep soil layers. In the case of subsoiling, the higher compaction at 15–20 cm and the preferential water infiltration through the subsoiler slits in the early stages preserve the water stored in the deep layers for the flowering period. In the case of maize, a statistically significant difference was observed only in July, when the LAI in the M_CT was greater than in the M_RT. Maize has a fibrous root system, characterized by numerous shallow roots, which cannot take advantage of the subsoiling until the reproductive stage [48].

At the time of the thermal imaging (2 July), both crops were in the flowering stage: sunflower was at BBCH growth stage 63–65 (flowering), while maize was at BBCH stage 61–65 (flowering, anthesis).

In terms of plant height, statistically significant differences were observed in August for maize but not for sunflower. The differences in the plant density and yield between the two tillage variants for both crops were not statistically significant. Overall, weaker crop development was observed under deep ripping, but this was generally not significant, except for LAI.

The yields of sunflower slightly decreased in the S_RT variant compared to the S_CT variant by 2.5%, which was not confirmed after harvesting the entire area. Additionally, yields decreased in the M_RT variant compared to the M_CT variant by 6.9%. The lower crop density of maize by −3% under the M_RT variant should also be considered.

3.5. LAI Retrieval

We evaluated three distinct models, each based on a different set of independent variables. The dataset of the M1 model consists only of sunflower samples, and the best-performing M1 model utilizes two bands (S2_B8A, S1_VV), as shown in

Table 5. Performance metrics and results from the top-performing models.

Dataset	MAE	RMSE	RRMSE	NRMSE	R2
M1 Cross-validation	0.71	0.85	38.01	21.81	0.29
M1 Training	0.51	0.64	28.69	16.46	0.63
M1 Testing	0.50	0.67	28.96	20.91	0.50
M1 all sunflower samples	0.51	0.65	28.78	16.71	0.59
M2 Cross-validation	0.35	0.43	24.95	16.87	0.24
M2 Training	0.25	0.32	18.35	12.41	0.62
M2 Testing	0.37	0.45	24.16	22.81	0.33
M2 all maize samples	0.28	0.36	20.44	14.11	0.49
M3 Cross-validation	0.45	0.64	32.82	16.39	0.40
M3 Training	0.29	0.40	20.84	10.40	0.79
M3 Testing	0.51	0.64	29.79	19.04	0.33
M3 all sunflower and maize samples	0.36	0.49	24.31	12.53	0.64

. The dataset of the M2 model consists only of maize samples, and the best-performing M2 model uses two bands (S2_NDWI, S2_SAVI), as shown in Table 5. The dataset of the M3 model comprises a combined dataset containing both sunflower and maize samples. The best-performing M3 model utilizes nine bands (S2_B4, S2_B6, S2_B8, S2_B8A, S2_B11, S2_MSI, S2_NDMI, S2_SAVI, S1_VV), as shown in Table 5.

Based on the NRMSE values from the testing dataset, the M3 model demonstrated the best overall performance. However, to evaluate the model’s performance specifically on sunflower and maize samples, we calculated the performance metrics separately, as presented in

Table 6. Performance metrics and results from the best-performing M3 model applied separately to sunflower and maize samples.

Dataset	MAE	RMSE	RRMSE	NRMSE	R2
M3—all sunflower samples	0.47	0.60	26.48	15.38	0.65
M3—all maize samples	0.25	0.36	20.29	14.00	0.49

.

The predicted LAI from the ML model for the four test sites is shown in Figure 9.

The statistical analyses of the obtained modeled LAI are presented as histograms (Figure 10) and basic statistical parameters (

Table 7. Statistical parameters of the modeled leaf area index (LAI).

Parameter	Sunflower		Maize
	Conventional	Subsoiling	Conventional	Subsoiling
mean, m2 m−2	1.94	2.38	1.89	1.61
stdev, m2 m−2	0.47	0.56	0.22	0.24
Cv, %	24.1	23.6	11.9	15.1
Min, m2 m−2	1.08	1.11	1.17	1.09
Max, m2 m−2	3.26	3.34	2.35	2.32

).

The modeled LAI was higher in S_RT than in S_CT, and higher in M_CT than in M_RT.

3.6. Effect of Subsoiling on Canopy Temperature

The canopy temperature obtained from the UAS platform on 2 July 2024 (Figure 4) were processed in order to separate the bare soil from the vegetation. The separation of the vegetation pixels from the soil by employing the ML SVM algorithm demonstrated higher accuracy for the Site-2 than for Site-1. The reason for this is the quality of the aerial photography at Site-2, where the spectral characteristics of the vegetation interfere with those of the soil, leading to a misleading clustering of vegetation pixels. Despite this, the classification results showed a very good separation of soil from vegetation pixels (Figure 11).

The classification sets of vegetation pixels were statistically processed to derive a distribution pattern. The resulting histograms are Gaussian in nature and clearly demonstrate differences between crop types and tillage practices (Figure 12).

3.7. Crop Water Stress Indicators

The comparison of the crop water stress is conducted with respect to the mode of the canopy temperature, derived from histograms (Figure 12,

Table 8. Parameters and water stress indices (CWSI) based on the canopy temperature Tc histograms.

Parameter	Sunflower		Maize
	Conventional (S_CT)	Subsoiling (S_RT)	Conventional (M_CT)	Subsoiling (M_RT)
0.5% percentile, Tc °C	27.9	27.8	30.7	32.0
99.5% percentile, Tc °C	37.2	34.9	36.5	37.7
Mode of Tc, °C	33.5	31.7	33.0	34.8
ΔT (Mode), °C	2.5	0.7	2.0	3.8
Tpot, °C	27.9	27.8	30.7	30.7
Tdry, °C	36.0	36.0	36.0	36.0
CWSI (Mode)	0.69	0.47	0.44	0.76

). As can be seen, the water stress for both crops varies regarding the applied tillage. A drought mitigation effect is observed under S_RT and M_CT. The applied histogram approach for the determination of Tpot and Tdry, in the case of uncontrolled moisture conditions, has to be applied with caution as it can overestimate or underestimate the water stress depending on the temperature distribution. For this reason, Tdry has been fixed to 36 °C, which coincides with the temperature difference, ΔT = 5 °C (Tair = 31 °C), at the measurement time. As mentioned above, ΔT = 5 °C is often used as a threshold estimate for stopping transpiration.

4. Discussion

The effectiveness of subsoiling as a drought mitigation measure is strongly influenced by the depth of tillage, soil texture, timing, and crop type. In this study, subsoiling was applied at a depth of 50 cm, which is considered effective for loosening compacted subsoil layers and increasing water infiltration and root access to deeper water pools [5]. However, no clear loosening effect was detected from the soil physical measurements. This may be attributed to the fact that soil sampling was performed between the subsoiler slits, potentially missing zones where compaction was alleviated.

Subsoil compaction at 15–20 cm was observed in all treatments except for sunflower under conventional tillage. The absence of compaction in this variant may be due to favorable soil structure resulting from recent plowing, more favorable soil moisture at the time of tillage, and the residual root system of the previous crop (maize), which may have improved porosity. The lower packing density (PD) and higher aeration capacity (AC) in this treatment support this interpretation.

Subsoiling effects on plant growth varied according to crop type and growth stage. For sunflower, a C3 crop, subsoiling initially delayed early growth but later promoted better development, as seen in the higher leaf area index (LAI) and lower canopy temperatures during flowering. In contrast, maize (a C4 crop) performed better under conventional tillage, particularly in terms of water status and yield. This suggests that subsoiling had a positive effect on sunflower but a negative effect on maize under the specific dry conditions of 2024.

These findings are in line with the broader literature, which shows contrasting drought responses between C3 and C4 crops. Although C4 plants like maize generally exhibit higher water use efficiency (WUE) [49], they are often more sensitive to combined heat and drought stress, especially during the reproductive stages [1,2]. Our results confirm this; sunflower responded better under subsoiling, while maize suffered reduced yield despite deeper tillage. Studies have shown that sunflower mainly absorbs water from the upper and middle soil layers (0–50 cm), whereas maize increasingly relies on the deeper zone (50–70 cm) during tasseling and grain filling [48]. Since our soil moisture monitoring extended only to 50 cm, the lack of information from the 50–70 cm layer represents a limitation of the current study, as it may have missed critical stress signals for maize.

Furthermore, while we cited several studies comparing C3 and C4 crop responses under drought conditions [43,44,45,46,47,48,49], we now emphasize a direct comparison between our findings and previous results. For example, Bellasio et al. [45] demonstrated that C4 crops like maize are more sensitive to rapid dehydration than C3 crops like sunflower. Similarly, He et al. [46] showed that root water uptake in sunflower is more stable in the topsoil, while maize depends on deeper layers. This reinforces our observation that sunflower outperformed maize under deep tillage despite the dry conditions, likely due to better root access within the sampled profile.

Overall, the indicators used in our study—soil compaction parameters (Db, PD, AC), volumetric water content (VWC), LAI, canopy temperature, and yield—provided a robust assessment of subsoiling impact. These metrics revealed that subsoiling at 50 cm offered limited physical improvements in the soil profile but did lead to improved crop water status in sunflower. The differing effects between maize and sunflower underline the importance of crop-specific evaluation when implementing subsoiling as a drought mitigation practice.

This study highlights important considerations regarding the evaluation of subsoiling as a drought mitigation strategy. Although subsoiling at 50 cm depth showed beneficial effects on sunflower (a C3 crop) under dry conditions, its effectiveness for maize (a C4 crop) appears to be limited, potentially due to insufficient water availability in deeper soil layers during critical growth stages. One limitation of the current study is the lack of soil moisture and compaction data below 50 cm, which restricts our ability to assess the full extent of the subsoiling impact, particularly in the 50–70 cm layer, which was identified in previous research as crucial for maize water uptake. Moreover, soil sampling was conducted between the subsoiler slits, which may have caused an underestimation of the actual loosening effect. Future research should therefore incorporate deeper soil monitoring to better capture subsoil water dynamics, especially for C4 crops, and apply targeted sampling directly along the rip lines to more accurately assess soil structural changes. In addition, the timing of the subsoiling relative to crop phenology should be considered, as its effectiveness may vary depending on the crop’s developmental stage and seasonal water demand.

5. Conclusions

This study integrated ground-based and remote sensing data to assess the impact of subsoiling at 50 cm depth on the soil physical properties, crop water status, and yield performance of maize (C4) and sunflower (C3) cultivated on Haplic Chernozem in a drought-prone region of Northern Bulgaria. By combining measurements of soil physical properties, soil moisture, LAI, canopy temperature, and crop yield with UAS-derived land surface temperature (LST) and satellite-based LAI models, we evaluated the effectiveness of subsoiling as a drought mitigation strategy.

The findings revealed that subsoiling did not significantly alter soil physical properties across the field, likely due to the sampling position relative to the rip lines. However, the responses of the crops differed markedly. Subsoiling had a positive effect on sunflower, improving leaf area development, reducing canopy temperature, and slightly increasing water content in the topsoil. In contrast, maize under subsoiling showed higher canopy temperature and reduced yield, suggesting increased water stress, possibly due to insufficient moisture availability in the deeper soil layers not monitored in this study.

This study contributes new insights into the species-specific response of C3 and C4 crops to deep tillage under drought conditions, highlighting that subsoiling can be effective for sunflower but may not benefit maize unless subsoil moisture is available during critical growth stages. The use of high-resolution UAS and satellite data also demonstrates the potential of remote sensing technologies for evaluating the spatial variability of drought impact and supporting precision agriculture.

From a practical standpoint, the results suggest that subsoiling should be carefully targeted, considering soil properties, crop type, rooting depth, and drought risk, to ensure its effectiveness. Future research should include deeper soil monitoring (50–70 cm), especially for C4 crops, and improve the sampling design to capture physical changes along subsoiling paths.