Application of a Low-Cost Electronic Nose to Differentiate Between Soils Polluted by Standard and Biodegradable Hydraulic Oils

Abstract

Detection of soil pollution by petroleum products is necessary to remedy threats to economic and human health. Pollution by hydraulic oil often occurs through leaks from forestry machinery such as harvesters. Electronic noses equipped with gas sensor arrays are promising tools for applications of pollution detection and monitoring. A self-made, low-cost electronic nose was used for differentiation between clean and polluted samples, with two types of oils and three levels of pollution severity. An electronic nose uses the TGS series of gas sensors, manufactured by Figaro Inc. Sensor responses to changes in environmental conditions from clean air to measured odor, as well as responses to changes in sensor operation temperature, were used for analysis. Statistically significant response results allowed for the detection of pollution by biodegradable oil, while standard mineral oil was difficult to detect. It was demonstrated that the TGS 2602 gas sensor is most suitable for the studied application. LDA analysis demonstrated multidimensional data patterns allowing differentiation between sample categories and pollution severity levels.

1. Introduction

Soil quality assessment and pollution monitoring are essential for sustainable land use and environmental protection. Among the various pollutants, petroleum hydrocarbons and pesticides represent particularly persistent and hazardous threats to both ecosystems and human health. Conventional chemical analysis methods—such as gas chromatography–mass spectrometry (GC-MS)—offer high precision and sensitivity but are often costly, time-consuming, and laboratory-bound.

In recent years, electronic noses (e-noses) [1] have emerged as a promising alternative for rapid and cost-effective detection of soil contaminants. These multisensor systems mimic the human olfactory mechanism and can detect volatile organic compounds (VOCs) associated with specific chemical or biological processes in soil. Numerous studies have demonstrated their utility for identifying and distinguishing between hydrocarbon- and pesticide-based contaminants [2,3], as well as for characterizing soil health parameters such as humic substances [4] or moisture levels [5].

Electronic nose technologies vary widely in design, ranging from custom sensor arrays (e.g., 25-sensor arrays in [2]; novel configurations in [6]) to low-cost, open-source platforms such as SENose [7]. Studies have also explored the use of e-noses for differentiating oil types in diverse soil matrices [8], assessing the potential reuse of contaminated land [9], and developing hierarchical or UAV-integrated monitoring systems [10].

In the context of soil contamination originating from forestry machinery, such as harvesters, the capability of e-noses for rapid and precise detection becomes particularly relevant. Harvesters utilize various types of hydraulic oils, lubricants, as well as fuels. Studies by Kong et al. [2] have demonstrated the rapid and simultaneous detection of petroleum hydrocarbons and organic pesticides in soil using a custom e-nose. The authors described methods for soil sampling and controlled contaminant addition, which allowed for effective differentiation between these two types of substances.

The ability of electronic noses to differentiate oil types and their concentrations within various soil matrices has been confirmed in studies where an artificial olfactory system successfully distinguished between soils contaminated with new engine oil and those with used oil across six different soil types [8]. Furthermore, the potential for evaluating hydrocarbon soil pollution (gasoline and diesel) using an e-nose across 10 different soil types, considering signal changes over time, has been shown [11]. Zaytsev et al. focused on coding the smell patterns of crude oil, paving the way for the specific identification of various contamination sources [12]. Importantly, e-noses not only detect the presence of pollutants but also have the capability to specifically recognize certain types of pesticides, as demonstrated in recent research [13].

In the context of forest environment monitoring, where access to laboratories can be limited, the application of low-cost and portable solutions becomes crucial. An example is a platform that costs under 50 USD and can be used for monitoring soil gas emissions, including diesel-contaminated soil [7]. While other custom sensor systems also exist [6], along with approaches for soil organic matter determination using artificial olfaction [14], the emphasis is on practical solutions tailored to specific field needs.

This study aims to investigate the application of a low-cost electronic nose to the differentiation between soil polluted by a standard and biodegradable hydraulic oil, commonly used in forestry machinery such as harvesters. We will focus on the development and validation of an e-nose system, considering various concentrations of soil contamination under the most common forest conditions in Poland, to provide an effective tool for pollution monitoring and management in forestry.

The electronic nose device used in the research, equipped with TGS (Taguchi Gas Sensors), was constructed as a universal device. The applied gas sensors have demonstrated their capability to detect volatile organic compounds in broad applications, not only in environmental monitoring or industrial safety, but also in areas like food quality control. TGS sensors, are utilized for assessing meat freshness [15] and characteristics [16], detecting contamination [17], differentiation of origin [18], or monitoring ripening and degradation processes [19]. These examples underscore the versatility and potential of TGS sensors in various scientific and industrial fields, demonstrating their ability to provide rapid and reliable data on the composition of volatile substances.

2. Materials and Methods

2.1. Sample Preparation

Soil for the study was collected from a single location (51.030381° N, 18.935184° E) within a pine stand, growing on a fresh mixed coniferous forest site. The soil was sampled using a spade to a maximum depth of 10 cm. The soil primarily consisted of sand. Before the sample preparation, the collected forest soils were stored in plastic bags, which remained open for most of the time. They were sealed only during transport, aiming to minimize alterations to the natural conditions within the soil before the measurements began. Subsequently, the soil was manually cleared of any organic contaminants, such as roots or needles. After initial cleaning, the soil was left to air-dry completely. Once dried, in the laboratory, the soil was again manually cleaned to remove any remaining impurities.

The next step involved precise weighing of soil portions using a laboratory balance. Exactly 285 g of the prepared soil was transferred into each jar. Subsequently, appropriate amounts of pure, biodegradable hydraulic oil (bio-oil) or conventional hydraulic oil, commonly used in forestry machinery such as harvesters, were added to the samples using a syringe. After the oils were added, the contents of each jar were thoroughly mixed to ensure the homogeneous distribution of the contaminant.

Shell Tellus S2 VX46 [20] (Shell plc, London, UK) was used as the conventional hydraulic oil. This is a high-performance hydraulic fluid formulated with Group II base oils. This classification indicates that the base oil is derived from highly refined crude oil, subjected to intensive hydrocracking processes. The result is a high purity, low sulfur content with a high level of saturated hydrocarbons. It is characterized by excellent oxidative and thermal stability, good anti-wear properties, and the ability to operate across a wide temperature range.

The second product used was Komatsu HE Gen II Natura [21] (Kamatsu Ltd., Tokyo, Japan), a bio-hydraulic oil. This is an environmentally friendly hydraulic fluid formulated with fully synthetic esters. It is enhanced with ashless (zinc-free) additives. A key feature of its composition is a very high content of renewable raw materials, with an average bio-based content exceeding 80% (ASTM D 6866). It provides excellent wear protection, a high viscosity index (VI), and shear stability, ensuring efficient hydraulic system performance across a wide range of temperatures and loads. It is readily biodegradable (at least 60% within 28 days) and exhibits minimal toxicity to aquatic organisms.

Throughout the entire experiment, the sample jars remained uncapped, ensuring free air access to each sample. This approach aimed to replicate natural environmental conditions in the laboratory as closely as possible. It allowed for processes such as evaporation, oxidation, and other natural transformations that would occur in the soil in an outdoor environment after the introduction of an oil substance. In our opinion, this also allowed the inclusion of more variability between samples used by consecutive days of measurements.

Following these preparations, a total of 21 samples were obtained, comprising 7 variants, each with 3 repetitions:

A

Control (clean soil);

B

Soil with bio-oil 0.1 mL;

C

Soil with bio-oil 0.5 mL;

D

Soil with bio-oil 1.0 mL;

E

Soil with standard oil 0.1 mL;

F

Soil with standard oil 0.5 mL;

G

Soil with standard oil 1.0 mL.

The minimum oil content was determined based on current Polish law, specifically the Regulation of the Minister of the Environment of 1 September 2016, on the method of conducting soil surface contamination assessment [22]. According to this regulation, the permissible concentrations of the sum of C12–C35 hydrocarbons (oil fraction) for category III land (forests) are 300 mg/kg of dry soil matter. Thus, samples containing 0.1 mL of oil or bio-oil represented the maximum permissible contamination level in 285 g of prepared soil. In subsequent samples, the contamination content increased fivefold (0.5 mL) and tenfold (1.0 mL), respectively.

2.2. Electronic Nose Device

The electronic nose device used in the experiment belongs to the category of low-cost equipment [23,24]. It consists of a set (Table 1) of the TGS series of gas sensor arrays manufactured by Figaro Engineering Inc. (Osaka, Japan) [25]. The technical details of the electronic nose construction were described in previous papers [26,27].

Table 1. List of sensor models used in electronic nose and their target gases and odors [25].

Sensor Model	Target Gas Detection
TGS 2600	Has a high sensitivity to low concentrations of gaseous air contaminants, such as hydrogen and carbon monoxide, which exist in cigarette smoke. The sensor can detect hydrogen at a level of several ppm.
TGS 2602	Has high sensitivity to low concentrations of odorous gases, such as ammonia and H2S, generated from waste materials in office and home environments. The sensor also has a high sensitivity to low concentrations of VOCs, such as toluene, emitted from wood finishing and construction products.
TGS 2603	Has high sensitivity to low concentrations of odorous gases, such as amine-series and sulfurous odors, generated from waste materials or spoiled foods such as fish.
TGS 2610	Uses filter material in its housing, eliminating the influence of interference gases such as alcohol, resulting in a highly selective response to LP gas.
TGS 2611	Uses filter material in its housing, which eliminates the influence of interfering gases such as alcohol, resulting in a highly selective response to methane gas.
TGS 2612	Has high sensitivity to methane, propane, and butane, making it ideal for LNG and LPG monitoring. Due to its low sensitivity to alcohol vapors (a typical interference gas in the residential environment), the sensor is ideal for consumer market gas alarms.
TGS 2620	Has high sensitivity to organic solvents and other volatile vapors, making it suitable for organic vapor detectors/alarms.

Two types of sensor response were analyzed in the presented research: (i) Change in sensor resistance caused by the presence of the measured gas. In this case, the sensor resistance, when the sensor is exposed to odor conditions, is related to the sensor resistance in clean air conditions. (ii) Change in sensor resistance caused by a change in sensor heater voltage, when the sensor is exposed to the studied gas for a sufficiently long time to reach a stable response level. A detailed description of the electronic nose operation can be found in the references [24,26].

2.3. Measurements by Electronic Nose

Measurements using the electronic nose (e-nose) were conducted over five consecutive days. The first series of measurements took place on the same day that the hydraulic oil or bio-oil was added to the soil samples. Throughout the entire experiment, the sample jars remained uncapped, ensuring free air access to each sample.

The measurement of each individual sample lasted approximately 12 min and consisted of a series of 1100 individual readings. The first 50 measurements were recorded under clean air conditions to establish a baseline. The subsequent 650 measurements were performed with the sensor chamber shutter open, capturing the response during the adsorption phase. The final 400 measurements were taken with the chamber closed again; this constituted the cleaning phase and preparation for measuring the next sample. The first 700 individual readings were used for analysis. Figure 1 illustrates how the sample measurement was performed using the electronic nose. The order in which the sample measurements were performed was randomized, using random numbers generated by Microsoft Excel. Before commencing the actual sample measurements, two test measurements were also conducted. A total of 105 actual measurements were performed.

Figure 1. Measurement setup of the used electronic nose applied to the contaminated soil sample.

2.4. Electronic Nose Data Analysis

Differentiation between the sample categories by electronic nose measurements requires the extraction of main features from the sensors’ response curves. We follow the method described in the previous paper [24]. Several features are extracted from the sensor response curves, such as the final level of the response, area under the response curve (which is equivalent to the average response), slopes of the response at the beginning of the change of sensor conditions, and at the end of observation.

The selection of the main features from the sensor response curves allows a significant reduction of the dimensionality of data; however, it still leaves multidimensional data, which may be difficult to analyze and find patterns. A common approach for further dimensionality reduction is the application of linear discriminant analysis (LDA) to choose a data projection to a lower-dimensional space that emphasizes differences between the studied treatments. LDA facilitates a comparison and interpretation of the similarities in the patterns in the data. For such a dimension reduction task as inputting data of all the above-mentioned features extracted from the sensors, response curves were used.

Statistical significance of differences between the treatment groups has been evaluated using the one-way ANOVA method and Tukey’s pairwise post-hoc tests. The statistical tests were used for demonstration of significant difference between the treatment groups using the selected (the most often used) features extracted from the sensor responses.

Electronic nose data allow the application of machine learning models, in which multidimensional data are used to find patterns for differentiating between the categories of interest. Classification models using the random forest technique [28] were applied to differentiate between treatment groups. For estimation of the model’s performance, the so-called out-of-bag score (OOB) was used. The OOB score applies a similar concept to the commonly used cross-validation method. During the random forests training procedure, the observations not used for training can be used for independent testing of the model’s performance. It has been shown that OOB estimation converges with leave-one-out cross-validation [29]. As input for classification modeling, all features extracted from the sensor response curves were used.

Data processing and analysis were performed using Python 3.10 with the application of SciPy [30], Statmodels [31], and Scikit-learn [32] modules.

3. Results

3.1. Identification of Pollution Type

Measurements by a gas sensor array, constituting an electronic nose, can be used for the identification of soils polluted by oil and differentiation between the two types of pollutants.

The main feature extracted from the first type of sensor response is the magnitude of the sensor conductivity at the stable state, when the sensor is exposed to the measured gas. However, it should also be recalled that the magnitude relative to the conductivity in the clean air conditions is used for such analysis. In addition, an important remark is that the magnitudes of sensor responses cannot be compared between various sensor types, as the magnitude of the response depends on the sensor’s internal construction, the sensing materials, etc. The same sensor can compare responses to various types of odors.

In Figure 2, a comparison of sensor responses to the presence of studied sample categories is presented for each electronic nose sensor. One should keep in mind that sample categories consist of various samples with different levels of pollution; thus, the spread of the data distribution can be naturally expected.

Figure 2. Sensor response to the presence of sample odor, versus three considered sample categories. The sensor response is expressed in arbitrary units without numbers, as the magnitude of the sensor response has no meaningful interpretation and should not be compared between various sensors. The sensor type is indicated on the y-axis. The box spans from the first to the third quartile. Horizontal line in the box represents sample median, × sign represents mean value, the whiskers show 1.5 inter-quartile distance from the mean, and the circles outside whiskers represent outlier observations.

In Figure 2, one can notice the difference between the three studied categories of samples. This is especially visible for the sensors TGS 2600, TGS 2602, TGS 2611, and TGS 2620, for which the data representing the distribution of biodegradable oil is located differently than the data representing the Control samples. Also, such visual analysis suggests that data collected by the TGS 2602 sensor allows differentiation between soil samples polluted by biodegradable oil and standard oil.

Table 2 presents the results of statistical tests confirming the above-mentioned observation. For the three mentioned sensors, there is a statistically significant difference (at a p-value threshold of 0.05) between Control and Bio samples, as well as between the Bio and Standard samples—the later only for the case of the TGS 2602 sensor.

Table 2. (A) Results of one-way ANOVA test for comparison between three considered sample categories using sensor response to the presence of measured odor (Figure 2). (B) Results of post-hoc Tukey’s test for comparison between pairs of categories.

(A)
Sensor			F-Statistics		p-Value
TGS 2600			3.68		0.0290
TGS 2602			5.61		0.0050
TGS 2603			0.88		0.4171
TGS 2610			3.12		0.0488
TGS 2611			3.86		0.0245
TGS 2612			0.20		0.8227
TGS 2620			4.55		0.0131
(B)
TGS 2600
group1	group2	meandiff	p-adj	lower	upper	reject
Control	Bio	0.0095	0.0317	0.0007	0.0183	True
Control	Standard	0.0047	0.4113	−0.0041	0.0134	False
Bio	Standard	−0.0048	0.1836	−0.0112	0.0016	False
TGS 2602
group1	group2	meandiff	p-adj	lower	upper	reject
Control	Bio	0.0207	0.0212	0.0026	0.0388	True
Control	Standard	0.0049	0.7899	−0.013	0.0229	False
Bio	Standard	−0.0157	0.0154	−0.0289	−0.0025	True
TGS 2603
group1	group2	meandiff	p-adj	lower	upper	reject
Control	Bio	−0.0115	0.4839	−0.0351	0.0122	False
Control	Standard	−0.0039	0.9177	−0.0274	0.0196	False
Bio	Standard	0.0076	0.5519	−0.0097	0.0248	False
TGS 2610
group1	group2	meandiff	p-adj	lower	upper	reject
Control	Bio	0.0114	0.0568	−0.0003	0.0231	False
Control	Standard	0.0053	0.5237	−0.0063	0.0169	False
Bio	Standard	−0.0061	0.2072	−0.0146	0.0024	False
TGS 2611
group1	group2	meandiff	p-adj	lower	upper	reject
Control	Bio	0.0133	0.027	0.0012	0.0253	True
Control	Standard	0.0066	0.3956	−0.0054	0.0185	False
Bio	Standard	−0.0067	0.168	−0.0155	0.0021	False
TGS 2612
group1	group2	meandiff	p-adj	lower	upper	reject
Control	Bio	0.0001	0.9996	−0.0109	0.0112	False
Control	Standard	−0.0019	0.9131	−0.0128	0.0091	False
Bio	Standard	−0.002	0.8261	−0.01	0.0061	False
TGS 2620
group1	group2	meandiff	p-adj	lower	upper	reject
Control	Bio	0.0103	0.0158	0.0016	0.0191	True
Control	Standard	0.0049	0.3715	−0.0038	0.0136	False
Bio	Standard	−0.0054	0.1099	−0.0118	0.0009	False

In Figure 3, we present a similar type of results, comparing the sensor’s response to the change in sensor heater voltage, versus the studied sample categories. The results are quantitatively similar to the previous ones, indicating that there is a possibility to distinguish between the soil samples polluted by biodegradable oil and the two other types of samples. Comparison between other pairs of samples does not indicate differences. The possibility of differentiating between the samples can be achieved only using data collected by the TGS 2602 sensor. This visually observed pattern in the data could be confirmed as a statistically significant difference using ANOVA and Tukey’s test; the result is presented in Table 3.

Figure 3. Sensor response to the change in sensor heater voltage, versus three considered sample categories. The sensor response is expressed in arbitrary units without numbers, as the magnitude of the sensor response has no meaningful interpretation and should not be compared between various sensors. The sensor type is indicated on the y-axis. The box spans from the first to the third quartile. Horizontal line in the box represents sample median, × sign represents mean value, the whiskers show 1.5 inter-quartile distance from the mean, and the circles outside whiskers represent outlier observations.

Table 3. (A) Results of one-way ANOVA test for comparison between three considered sample categories using sensor response to the change in sensor heater voltage in the presence of measured odor (Figure 3). (B) Results of post-hoc Tukey’s test for comparison between pairs of categories, only the results for the TGS 2602 sensor are presented.

(A)
Sensor			F-Statistics		p-Value
TGS 2600			0.53		0.5895
TGS 2602			5.70		0.0046
TGS 2630			0.50		0.6088
TGS 2610			0.03		0.9742
TGS 2611			0.03		0.9692
TGS 2612			0.21		0.8133
TGS 2620			0.50		0.6092
(B)
TGS 2602
group1	group2	meandiff	p-adj	lower	upper	reject
Control	Bio	0.0096	0.0288	0.0008	0.0184	True
Control	Standard	0.0016	0.8993	−0.0071	0.0104	False
Bio	Standard	−0.008	0.0106	−0.0144	−0.0016	True

Figure 2 and Figure 3 and Table 2 and Table 3 present the results of the comparison between the main treatment groups for the case of independently considering responses of individual sensors. Also, we present one selected feature (final response magnitude), as such, in our opinion, it is the most representative and commonly used in other research.

To evaluate the possibility of differentiation between the studied categories of samples, we applied a Random Forest method. The received performance of the trained model reached an accuracy of 44.3% for differentiation between the three categories of samples. In the case when we were interested only in the differentiation between the unpolluted and polluted samples, the obtained accuracy was 82.5%.

3.2. Differentiation Between Levels of Pollution Severity

The prepared samples of polluted soil with various amounts of oil content were added. The more detailed visualization presented in Figure 4 and Figure 5 allows the observation of patterns in data demonstrating the dependence of the sensor’s response on the amount of added oil, which corresponds to the severity of soil pollution.

Figure 4. Sensor response to the presence of sample odor versus seven considered sample categories. The sensor response is expressed in arbitrary units without numbers, as the magnitude of the sensor response has no meaningful interpretation and should not be compared between various sensors. The sensor type is indicated on the y-axis. The box spans from the first to the third quartile. The horizontal line in the box represents the sample median, × sign represents the mean value, the whiskers show 1.5 inter-quartile distance from the mean, and the circles outside whiskers represent outlier observations.

Figure 5. Sensor response to the change in sensor heater voltage versus seven considered sample categories. The sensor response is expressed in arbitrary units without numbers, as the magnitude of the sensor response has no meaningful interpretation and should not be compared between various sensors. The sensor type is indicated on the y-axis. The box spans from the first to the third quartile. The horizontal line in the box represents sample median, × sign represents mean value, the whiskers show 1.5 inter-quartile distance from the mean, and the circles outside whiskers represent outlier observations.

In Figure 4, we can observe an additional pattern in the data when compared to the results presented in Figure 2. The response of the TGS 2600, TGS 2602, TGS 2610, TGS 2611, and TGS 2620 sensors for various studied categories of soil pollution severity exhibits a clear trend of increased sensor response level with an increase in polluting oil concentration. Such a trend is observed only for samples polluted with Bio-oil, which is consistent with the previously discussed results.

A similar pattern in data can be noticed in Figure 5. An increase in sensor response with an increase in the severity of soil pollution can be observed only for samples polluted with biodegradable oil for data collected with the TGS 2602 sensor.

Linear Discrimination Analysis of the Data

In Figure 6, we present the results of Linear Discrimination Analysis of the collected data. Three principal components were selected to visualize patterns in the data, demonstrating possibilities of differentiation between the studied sample categories. The method allows for finding a representation of data that maximizes linear separation between the treatments and also gives information on the proportion of variability captured by the components.

Figure 6. Distribution of observations obtained by electronic nose measurements as an LDA projection of features extracted from sensor response curves. Two projection axes are shown in each subfigure: (a) C1 vs. C2, (b) C1 vs. C3, and (c) C2 vs. C3. The variability explained is indicated in the axis labels. Confidence ellipses for two standard deviations are shown.

The analysis revealed that captured variability by the first three most important LDA-based components is 41%, 21%, and 13%, which explains 75% of the variability in the data overall.

The first pattern that we can observe in Figure 6 is a clear distinction between the control and samples polluted by biodegradable oil. Such a distinction is along all three components, which can be noticed by examining each of the subfigures; however, the first component seems to be dominant. Differentiation of biodegradable oil from control along the secnd component is less pronounced, which can be noticed in subfigure (b) and even more in subfigure (c), in which only the most polluted samples can be differentiated.

Differentiation between the control and standard hydraulic oil is more difficult, as we can observe that the dots (and ellipses) representing these two categories of data are placed close together. However, Figure 6b demonstrates that such a separation is possible using the first and second LDA components.

Differentiation between samples polluted with two types of hydraulic oil is also possible, which can be noticed, especially in Figure 6a.

It could also be noticed that the collected data allowed differentiation between the samples with various levels of pollution by biodegradable oil, which is especially possible by using the first and second LDA components (Figure 6a).

Differentiation between the various levels of pollution by standard oil was found to be much more difficult. The distributions of measurement data points overlap, but still, some distinction can be noticed if we use the first and third LDA components (Figure 6b).

4. Discussion

4.1. Soil Contamination by Lubricating Oils in Forest Environments—Sources, Impacts, and Mitigation Strategies

Soil pollution by oils—both mineral-based (derived from petroleum) and biodegradable (e.g., plant-based esters used as environmentally friendly alternatives)—poses a significant environmental concern in forest areas subject to either industrial use or recreational visitation. This issue arises during both commercial forestry operations and the increasing pressure of nature-based tourism.

4.2. Forest Machinery as a Source of Pollution

Timber harvesting, biomass extraction, and silvicultural maintenance involve the use of heavy machinery such as harvesters, forwarders, skidders, and trucks. Even well-maintained fleets are susceptible to unintentional leaks of hydraulic fluids, fuels, and lubricants. Mineral oils, particularly those containing long-chain hydrocarbons, are toxic to soil environments; they disrupt microbial activity, suppress enzymatic function, reduce nutrient cycling, and may bioaccumulate through soil–plant–organism pathways [33,34].

While biodegradable lubricants—such as methyl esters of vegetable oils—are often recommended as safer alternatives, they are not without ecological risks. Although they degrade more readily than mineral oils, at elevated concentrations, they can still lead to microbial imbalances and localized eutrophication, especially in sensitive ecosystems.

4.3. Tourism-Related Soil Pressure and Oil Leakage

The growing popularity of mushroom foraging, hiking, and recreational vehicle use leads to increased traffic on forest roads, particularly by passenger cars and off-road vehicles. These vehicles are frequently parked in unauthorized areas—on forest clearings, sandy soils, or vegetated margins—posing a risk of leakage from engines and transmission systems. According to Bieganowski et al. [11], even short-term exposure of light-textured soils to motor oil significantly alters volatile organic compound (VOC) emissions and suppresses biological activity.

Light sandy soils, which are common in pine-dominated forests, exhibit low sorption capacity and high permeability. These physical properties facilitate the vertical migration of oil-derived pollutants into deeper horizons, and potentially into groundwater reservoirs [35,36,37].

4.4. Environmental Consequences and Food Safety Concerns

Oil-contaminated soils can lead to serious ecological disruptions. Reduced microbial biodiversity limits organic matter turnover, impairs root development, and affects overall forest productivity. Furthermore, pollutants may accumulate in edible forest products, including mushrooms. Studies in Polish forests have shown elevated levels of polycyclic aromatic hydrocarbons (PAHs) in fungal fruiting bodies collected from oil-contaminated sites [38], raising public health concerns.

4.5. Preventive and Remediation Measures

To address these risks, several mitigation strategies are recommended:

• Mandatory use of biodegradable lubricants in forestry equipment operating on public or protected lands, consistent with EU Green Public Procurement (GPP) policies [39].
• Establishment of designated forest parking areas with impermeable surfaces and runoff retention systems to capture and contain potential leaks.
• Implementation of rapid soil monitoring tools, such as portable electronic noses or VOC detection systems based on GC-MS, to enable early diagnosis of contamination.
• Environmental education campaigns, including signage and outreach materials for recreational forest users, to raise awareness of the ecological impact of oil spills on soil and non-timber forest products.

5. Summary and Conclusions

A self-made, low-cost electronic nose applying TGS-type gas sensors manufactured by Figaro corporation was used in the experiment. Two types of sensor data were collected: (i) time characteristics of sensor response to change conditions from clean air to the presence of the measured gas, and (ii) dynamic sensor response to change of sensing material temperature, when the sensor was in the steady response state as immersed in the measured gas.

The device was applied to measurements of soil samples polluted by two kinds of hydraulic oil: standard and biodegradable. The control (unpolluted sample) and samples with three concentrations of pollutants were prepared in three replications. The measurements were performed during five consecutive days in randomized order for the samples.

It was found that the collected signals of sensor response differ between control and biodegradable oil, with a statistical significance of p < 0.05. In the case of sensor response change due to change in exposure from clean air to the measured samples’ odors, differentiation was possible for the four applied sensor types: TGS 2600, TGS 2602, TGS 2611, and TGS 2620. In the case of the analysis of sensor response to a change in sensing element temperature, only sensor TGS 2602 allowed the collection of signals with a statistically significant difference. Sensor TGS 2602 was found to be the most suitable to differentiate between the measured sample categories.

It was found that the measurement results of biodegradable oil exhibit clear trends with the increase in pollution severity (amount of oil added to soil samples). Such an effect was not observed for standard oil, for which it was more difficult to differentiate these samples from the control.

The multi-dimensional analysis of data with Linear Discriminant Analysis revealed clear patterns, distinguishing between the two types of pollutants from the control sample. Additionally, various levels of pollution severity could be differentiated. However, that was more difficult for the case of standard (mineral) oil.

The obtained results encourage further research on the possible application of TGS types of gas sensors for the detection of soil pollution by hydraulic oils. Low-cost electronic noses can be useful tools for agriculture or forestry, where such types of pollution can occur and detection and remediation measures are needed.