Integration of In Situ and Sentinel-2 Data to Assess Soil Quality in Forest Monitoring: The Case Study of the Vesuvius Fires

Abstract

The climatic conditions in southern Italy favor the occurrence and spread of forest fires, with severe long-lasting consequences on the local flora and fauna. On the one hand, biological and chemical in situ measurements are typically used to accurately investigate the evolution of the land affected by fires, with limited spatial coverage. On the other hand, Remote Sensing (RS) is a mature technology to complement the in situ campaigns on large regions with adequate revisit time. In this paper, we evaluate the capability of Sentinel-2 data to spatially and temporally extend post-fire in situ analysis on a fire-affected area. In particular, we estimate the soil quality index from Sentinel-2 data and achieve a remarkable coefficient of determination ($R^2=0.79$) and low relative error ($e_r=0.06$), highlighting the robustness of the proposed approach. Furthermore, the soil water content and the total iron (Fe) concentrations emerged as pertinent indicators detectable through Near-Infrared and Short-Wave Infrared Sentinel-2 bands. The obtained results prompted an investigation into the post-fire evolution of soil properties, thanks to RS data, in a large area covered by diverse vegetation types. The obtained results encourage a deeper synergic use of in situ and remotely observed data, enabling a comprehensive understanding of soil quality dynamics in fire-affected regions.

1. Introduction

Forests, as natural environments teeming with diverse flora and fauna, shape local ecosystems and influence both the local and global climate. The global trend of forest loss, driven by natural forces and human activities, brings consequences regarding critical environmental and development issues, including food security, biological diversity conservation, and climate change [1,2]. Forest fires are one of the main causes of forest loss. They arise from different factors, including climate conditions and weather patterns, vegetation states, and human activities. Approximately half of the global forest loss can be attributed to a combination of natural and anthropogenic drivers that are directly associated with fires for agricultural practices and deforestation [3,4]. The impact of forest fires goes beyond the loss of trees, affecting vegetation and soil in direct and indirect ways. High temperatures, combustion of plant material, and the reactions of vaporization and condensation modify fundamental soil characteristics, such as pH, nutrient cycles, organic matter amount, and water availability. These changes have evident consequences on soil organisms and activities [5]. The impoverishment of the overall ecosystem also increases hydro-geological instability, particularly in mountainous areas with steep slopes, threatening human settlements situated downstream and in proximity to burned areas. Over the last few decades, the growing frequency of fires has emphasized the limits of existing preventive measures and the need to find alternative approaches to reinforce existing strategies and provide relevant authorities with new instruments. This scenario calls for continuous multi-scale fire disturbance monitoring spanning the pre-fire, fire, and post-fire phases. The first phase includes monitoring of the vegetation state and the weather conditions preceding a fire outbreak. The second phase provides comprehensive insights into the fire dynamics, the fire intensity, the weather, and other relevant environmental parameters during the combustion process. The third phase includes diverse analyses designed to classify burned areas, comparing data collected before and after the fire event. The focus of this work is on post-fire monitoring, aiming to investigate the restoration capabilities of sites affected by severe fires. The traditional methods to better understand the post-fire soil recovery was in situ measurements. Post-fire monitoring is crucial to comprehend the connection between fire severity and subsequent response and recovery of the environment. For example, by tracking soil nutrients and decomposition rates, we gain insights into carbon sequestration and global nutrient cycles [6]. The post-fire alterations of soil pH modify the availability of nutrients and pollutants [7]. The high temperatures generated during a fire reduce soil affinity with water, increasing runoff and risk of erosion [8]. The fire changes on soil abiotic properties alter the abundance of soil organisms and control the recovery rate of microbial populations [9]. In addition, fires alter soil microorganism activities and exert an indirect effect on the nutrients and soil organic matter cycles [9]. Post-fire monitoring is typically carried out through in situ measurements, yielding highly precise but localized in space and time results. This method is constrained by limited spatial and temporal coverage due to the expensive and time-consuming nature of data collection and laboratory analysis. In particular, the high costs of the sampling campaigns result in infrequent data collection for small areas, leading to entire decades without measurements. These limits can be mitigated by the use of Earth Observation (EO) instruments, which play a crucial role in monitoring land use and land cover, detecting changes, managing risks (such as floods, subsidence, and landslides), and monitoring forest health, loss, and recovery [1,2,10,11,12]. In [13], efforts to integrate local and remote measurements have been summarized, highlighting the necessity to extend in time and space the information acquired with a campaign of in situ measurements to cover larger regions captured by satellite data. Moreover, the added value provided by in situ data lies in its ability to corroborate the information obtained through satellite acquisitions. Indeed, satellite remote sensing provides synoptic views of large areas (including those inaccessible and/or dangerous to humans [14,15]) that can be used for (i) fire risk analysis (pre-fire phase) [16], (ii) active fire detection (fire phase) [17], (iii) burned area mapping (post-fire phase) [18], and (iv) post-fire vegetation recovery [19,20]. In the literature, the time series of Normalized Difference Water Index (NDWI) and Enhanced Vegetation Index (EVI) [21] derived from MODIS data have been used to relate fire ignition events, the status of vegetation, and climate data [22]. The joint use of Short Wavelength Infra-Red (SWIR) and Near Infra-Red (NIR) spectral bands have provided encouraging results to detect active fires [15,23,24,25]. In particular, the SWIR frequencies are beneficial because they are responsive in the presence of active fires and transparent for smoke clouds during the active fires [14,26]. Burned area mapping is accomplished in the literature through change detection analysis [18,23,25], a method that compares images of the same scene on the ground acquired at different times. For the investigation of the vegetation recovery, Normalized Difference Vegetation Index (NDVI), and other vegetation multispectral indices, have been used to observe a long time series of satellites data, such as Landsat-8, Sentinel-2, or Sentinel-1 [20,27,28]. In this paper, we propose a new methodology for integrating in situ measurements and EO data. The study focuses on the Vesuvius volcano in Campania (southern Italy), which has been affected by severe fires in recent years, and in particular by a massive event in July 2017. First, we estimate the relationship between information retrieved from selected satellite imagery (Sentinel-2 data) and in situ measurements. This is done by investigating the emerging correlation between the spectral bands of Sentinel-2 and selected chemical and biological parameters acquired in the field. Through this process, we identify the most suitable multispectral indices to the investigation of the soil quality index evolution in the years following the fire. The main added value of this research lies in the opportunity to extend the spatial and temporal dimensions of biological and chemical parameters acquired on site, offering an augmented comprehension of the evolution of post-fire response of Mediterranean forests.

The overall structure of the paper is as follows. In Section 2, the investigated area, remote sensing data, and soil biological and chemical analyses are described. The proposed methodology of integration is detailed in Section 3. In Section 4, we show the main results with visual and numerical assessment using four Sentinel-2 images acquired on four different dates between 2018 and 2019 (see

Table 1. Dates related to the on-ground measurements and Sentinel-2 acquisition data.

Sampling Times	Soil Campaigns	Sentinel-2 Acquisitions
Summer	12 June 2018	2 June 2018
Fall	15 October 2018	25 October 2018
Winter	8 February 2019	7 February 2019
Spring	3 May 2019	18 April 2019

). Finally, we present the conclusion and the future perspectives of this work.

2. Materials

2.1. Study Area

The Vesuvius National Park, established in 1995, is located 12 km southeast of Naples (Campania, Italy) and covers an area of 8482 ha (Figure 1a,b). This area is characterized by Mediterranean climatic conditions, as reported in [29]. The choice of this area is motivated by the huge variety of flora and fauna [30] and by the landscape damage caused by fires. In recent decades, fires have affected 1.5 million hectares of Italian forests, with more than 15,000 fires. In terms of burned area, the most affected Italian regions are Sicily and Sardinia, while in terms of fires, Calabria and Campania are the most affected [31].

Specifically, we analyzed the effects of the wildfires, which occurred from late June to mid-July 2017, on the considered forests [32,33]. The dates used to carry out the analysis and the study conducted, both for the on-ground data and for the satellite data, are reported in Table 1. In summer 2017, wildfire occurred at different intensities in the different park areas. In particular, 11% of the total forested area burned at high fire severity, 13% burned at moderate–low fire severity, 33% burned at low fire severity, and 12% remained unburned [33]. Fire caused the consumption of more than 50% (approximately 3000 ha) of the existing plant cover [33] and left burnt and unburnt areas inside each plant cover typology. Before fire, holm oaks (Quercus ilex L.), pines (Pinus nigra L.), herbs and shrubs (such as Myrtus communis L., Laurus nobilis L., Viburnum tinus L., Cistus sp., Ginesta sp.), and black locust (Robinia pseudoacacia L.), an invasive species, dominated the different areas of the Vesuvius National Park [34].

2.2. Remote Sensing Data

The European Forest Fire Information System (EFFIS) [35] and the Advanced Fire Information System (AFIS) [36], at the moment, only make use of data acquired by MODIS [14] and VIIRS satellite sensors [15], with good revisit time requirements (1 day for VIIRS and 1–2 for MODIS) and a very limited spatial resolution on the ground (hundreds of meters). In order to analyze in greater detail the area under consideration of limited size, compared to the VIIRS and MODIS satellites, we focused our attention on the Sentinel-2 data with a worse temporal resolution (5 days), but a decidedly higher spatial resolution (10–20 m for the bands generally used in fire monitoring). In

Table 2. The 13 Sentinel-2 bands.

Spatial Resolution [m]	Bands (Bands Number)	Wavelength Range [nm]
10	Blue (2), Green (3),	490–842
	Red (4), and NIR (8,8A)
20	Vegetation Red Edge (5,6,7),	705–2190
	and SWIR (11,12)
60	Coastal Aerosol (1), Water Vapour (9),	443–1375
	and SWIR (10)

, we report the spectral range information about the S-2 bands. The choice of Sentinel-2 (S-2) as Multispectral/Optical source is dictated by the fact that S-2 is widely employed in applicative scenarios thanks to its capability to provide information about vegetation, risk management (floods, forest fires, and so on), water monitoring, and much more [17,37,38,39]. In addition, a plethora of indices, such as the Normalized Difference Vegetation Index (NDVI) [40,41,42], the Normalized Difference Water Index (NDWI) [43,44,45], and others have been developed to provide quantitative estimates of selected surface covers [17]. In particular, Sentinel-2 images were used to establish fire risk using a plethora of spectral indices [46,47] and to monitor the burned area and the post-fire regeneration [48,49,50]. To enhance the applicability of the proposed technique, we made use of the Google Earth Engine (GEE), a freely available cloud computing portal designed for rendering the data collection of satellite imagery over four decades around the entire world [10,51]. GEE offers valuable tools for applications on a large scale in hydrology, food security, and other areas. Free access to algorithms and satellites facilitate the development of new applications in interdisciplinary projects. Furthermore, parallelization significantly accelerates the processing.

In 2018 and 2019, soil samples were collected from various spatial locations (twenty, as mentioned previously) in the Vesuvius volcano on four specific dates (see Table 1). Then, we selected the Sentinel-2 acquisition date as close as possible to the sampling campaign within the study area. The biological and chemical analyses for twenty stations were investigated, and we took into account the capability of Sentinel-2 data in estimating these parameters. For the specific investigated study areas, we used Sentinel-2 (S-2) standard Level-2A data, which represent the Bottom-Of-Atmosphere (BOA) in cartographic geometry (UTM/WGS84 projection) that is preferred to be used in multi-temporal analysis, such as that conducted in [52].

2.3. Soil Sampling

The soils of Vesuvius National Park show a silty-clay texture and are classified as Lepti-Vitric Andosols [53]. In the present study, the sampling points all fell in the area with high or very high fire intensity, in which vegetation was completely destroyed. The surface soils (0–10 cm depth) were sampled, in burnt areas, at 20 sites equally distributed among 4 different vegetation covers (5 sites for each vegetation): Herbaceous and Shrubs (H), Black Locust (BL), Pines (P), and Holm Oak (HO). After litter removal at each site, 8 soil cores were collected and mixed to obtain a homogeneous sample. The soil samples were placed in sterile flasks and transported on ice to the laboratory. The soil campaigns were performed, within two consecutive days, in June 2018, October 2018, February 2019, and May 2019, after seven days without rainfall to minimize the climatic variability. Two years after the fire, an overall evaluation of the vegetation covers highlighted that the species composition of areas covered by holm oaks and herbaceous vegetation remain similar to pre-fire conditions, whereas those covered by pines and shrubs showed higher presence of black locust specimens as compared to the pre-fire conditions, as reported in [54]. Details regarding vegetation has been added in the Material and Method Section.

2.4. Soil Chemical Analyses

In the laboratory, the soil samples were sieved (>2 mm) and analyzed for pH, water, and organic matter contents, as well as for the total C and N contents. Soil pH was measured in a soil:distilled water (1:2.5 = v:v) suspension by electrometric method; Water Content (WC) was determined gravimetrically by drying fresh soil at 105 °C until reaching constant weight [7]; and total carbon, nitrogen, and $C_{org}$ concentrations were determined by Elemental Analyzer (Thermo Finnigan, CNS Analyzer) on dried and pulverized samples. Organic Matter (OM) content was calculated by multiplying by 1.724 the $C_{org}$ [55] measured on dried and pulverized samples, previously saturated with HCl (10%, v:v). Total Al, Ca, Cu, Fe, K, Mg, Mn, Na, and Pb concentrations were measured in oven-dried (75 °C) and ground soil samples. Samples were digested by hydrofluoric acid (50%) and nitric acid (65%) at a ratio of 1:2 (v:v) in a microwave oven (Milestone-Digestion/Drying Module mls 1200). The available Ca, Cu, Fe, K, Mg, Na, Ni, and Pb fractions were extracted according to [56]. Briefly, 50 mL of diethylenetriamine pentacetic acid (DTPA), CaCl₂, and triethanolamine (TEA) solution at pH 7.3 ± 0.05 were added to 25 g of oven-dried (75 °C) soil samples. The soil suspensions were shaken for 2 h and filtered with a Whatman 42 filter. Total concentrations and available fractions of elements were measured by Inductively Coupled Plasma Mass Spectrometry (ICP-MS, Aurora M90—Bruker Daltonics, Bremen, Germany). The accuracy of element measurements was checked by concurrent analysis of standard reference material (BCR CRM 142R—Commission of the European Communities, 1994). The overall element recovery ranged from 80 to 120% for all the investigated soil samples. All the above described were performed on three technical replicates.

2.5. Soil Biological Analyses

Biological analyses were performed on soil samples stored at 4 °C within three days of the soil sampling. The Microbial Biomass (MB) was evaluated by SIR, the substrate-induced respiration method [57], as CO₂ evolution from soil after the addition of 2 mL of a D-glucose solution (75 mM) and incubation in sealed vials (30 mL) for 4 h at 25 °C in the dark. CO₂ efflux from each sample was measured using an infrared gas analyzer (LI-COR LI6262, Lincoln, NE, USA). The Fungal Biomass (FB) was evaluated through the membrane filter technique [58], after staining with aniline blue, determining hypha length with an optical microscope (Optika, B-252) by the intersection method [59]. Basal respiration (Resp) was estimated as CO₂ evolution from the samples according to Froment [60]. The results obtained by the biological analyses were used to calculate two indices: the metabolic quotient ($qC{O}_2$), i.e., the degree of activity of the microbial biomass, and the Coefficient of Endogenous Mineralization (CEM), i.e., the rate of organic carbon mineralization. The $qC{O}_2$ was calculated as the ratio between microbial respiration (C-CO₂) and Microbial Biomass (MB), whereas the CEM was calculated as the ratio between microbial respiration (C-CO₂) and $C_{org}$ [61]. Hydrolase Activity (HA) was determined by adding a solution of potassium phosphate (pH 7.6) and fluorescein diacetate (FDA) to 1 g of fresh soil. The reaction mixture was incubated at 30 °C for 20 min. At the end of incubation, the fluorescein was extracted with acetone and centrifuged at 5000 rpm for 5 min. The absorbance of the supernatant was measured at 490 nm and the results were expressed as mmol of FDA produced for 1 g of dry soil in 1 min [62]. Dehydrogenase Activity (DHA) was determined by adding 2,3,5-triphenyltetrazolium chloride (TTC) dissolved in 0.1 M Tris-HCl buffer (pH 7.5) to 1 g of fresh soil. The reaction mixture was incubated at 30 °C for 24 h in the dark. At the end, the triphenylformazan (TFF) was extracted with acetone, and the extract was centrifuged at 3500 rpm for 15 min. The absorbance of the supernatant was measured at 546 nm, and the results were expressed as mmol of TFF produced for 1 g of dry soil in 1 min [53]. $\beta$-glucosidase activity ($\beta$-glu) was determined by adding a solution of Modified Universal Buffer (MUB) and p-nitrophenyl b-D-glucopiranoside (PNP) to 1 g of soil. The mixture was then incubated at 37 °C for 1 h, after which the enzymatic reaction was stopped by cooling on ice for 15 min. Then, a solution of CaCl₂ and tris-hydroxymethilaminomethane-sodium hydroxide (THAM-NaOH) was added. The absorbance of the supernatant was measured at 420 nm, and the results were expressed as mmol of PNP produced for 1 g of dry soil in 1 min [63,64]. The biological analyses were performed on three technical replicates.

2.6. Phytotoxicity Assays

The soil phytotoxicity was evaluated, on three technical replicates, according to [65,66] using a monocotyledon (Sorghum saccharatum L.) and a dicotyledon (Lepidium sativum L.). The phytotoxicity tests were carried out on fresh and sieved (2 mm) soil samples. Ten seeds for each species were placed in Petri dishes (Ø 90 mm) containing 10 g of equivalent fresh soil, subsequently saturated with water. OECD standard soil [67] was used as a negative control. After 72 h of incubation in the dark (at 25 °C), the number of germinated seeds and the total root length were measured. The results were expressed as the effect percentage of Germination Index (GI), compared to a standard soil [67].

3. Proposed Method

The workflow of the proposed method is illustrated in Figure 2. To ensure its general applicability, we divided the data into two distinct sets: one for training and validating the model and one for testing the model developed through the training phase. The training and validation process is represented in part (I) of Figure 2. The analysis was performed on two datasets, each independently pre-processed. Both datasets, the 10 m and 20 m S-2 bands and the chemical and biological field and soil campaign, share a common starting day, denoted as $t_0$. Following this, each dataset undergoes processing, after which a correlation analysis is performed between the biological/chemical parameters and the remote sensing S-2 indices, as shown in the workflow. A linear regression approach is then employed to estimate the proposed $SQI$ (Soil Quality Index) using S-2 indices. The considered LS estimation is supported by the strong linear relationships between the remote sensing S-2 indices and the biological/chemical parameters. The testing phase, represented in part (II) of Figure 2, evaluates the $SQI$ estimated using the S-2 data by means of the trained LS model across the entire National Park of Vesuvius. This evaluation is conducted over a five-year time series (2016–2020), allowing us to assess not only the ground-based data points but also the broader area of interest. Furthermore, this time-series analysis begins a few months before the 2017 fire, acquiring insight into the forest’s condition prior to the event.

3.1. Soil Quality Index (SQI) Calculations

An integrated Soil Quality Index ($SQ{I}_{TOT}$) was calculated considering all the chemical, biological, and ecotoxicological parameters ranked by linear scoring technique according to [54,66,68,69,70]. The scores, ranging from 0 to 1, were assigned by applying “more is better” or “less is better” functions that reflect the low and high quality of the soil. The “more is better” function was applied to soil variables recognized for their key role in soil fertility and importance in soil organic matter dynamics (organic matter, water and total C and N contents, MB, FB, Resp, and CEM) [71,72]. The “less is better” function was applied to the soil variable representing a stress for soil quality (total concentrations and available fractions of metals, $qC{O}_2$) [73,74,75]. The maximum pH score was attributed to value of 7 [70], whereas the percentage effects for the phytotoxicological assays were 100%. These scores were assigned by considering the “more is better” or the “less is a better” function depending on whether the parameter value is below or above the threshold value or the optimal range. For each site, the $SQI$ was calculated, summing the parameter scores and dividing for the number of parameters, as reported by [71]:

$$SQI={\displaystyle \frac{1}{N}}\sum _{i=1}^N{S}_i$$

(1)

where $SQI$ is soil quality index, $S_i$ is the score assigned to each parameter, and N is the number of the investigated parameters.

3.2. Selection of the Minimum Dataset ($SQ{I}_{MDS}$)

In order to select the soil parameters that were significantly influenced ($p<0.05$) by either vegetation cover or sampling time or their interactions, a two-way Analysis of Variance (ANOVA) was performed [76]. The soil parameters significantly influenced by at least one of the factors were used in the standardized Principal Component Analysis (PCA), which provides as outputs the eigenvalues and weighted loadings [77]. The PCs with eigenvalues >1 were considered for identifying the MDS, as those with eigenvalues <1 were impacted less [78]. Parameters with absolute values within 10% of the highest weighted loading were selected for the MDS within each PC. When, for each PC, more than one parameter showed the values to be selected for the MDS, two approaches were performed for the selection. The first approach retained in the MDS, for each PC, the parameters that were not correlated, eliminating the redundancy. The second approach retained in the MDS, for each PC, the parameters having the highest correlation sum [79,80]. The Shapiro–Wilk test assessed the normality of the distribution of the datasets. The ANOVA or Kruskal–Wallis test, according to the data distribution, was performed to compare the differences in Soil Quality Index and S(MDS) at different vegetation cover and sampling times.

3.3. Correlation Coefficient Between Soil Quality Indices and Sentinel-2 Indices

The correlation coefficient ${\rho }_{XY}$ is a statistical way to quantify interdependence (linear relationship) between variable quantities. The correlation is defined as follows:

$${\rho }_{XY}={\displaystyle \frac{E[(X-{\mu }_X)·(Y-{\mu }_Y)]}{{\sigma }_X·{\sigma }_Y}}$$

(2)

In our study, we proceeded to calculate the correlation between the $SQI$ parameter (Y) and all potential S-2 indices (X). The S-2 Normalized Difference Indices (NDIs) are defined as follows:

$$NDI(i,j)={\displaystyle \frac{B_i-B_j}{B_i+B_j}} with i\ne j$$

(3)

where $B_i$ and $B_j$ are the reflectances of all bands at 10 m and 20 m resolutions. To the best of our knowledge, the S-2 indices that are already known in the literature are reported in

Table 3. Normalized Difference Indices (NDI) for Sentinel-2 data.

Band Pair (i,j)	Full Index Name	Acronym	Reference
(2,5)	Bare Soil Index	BSI	[81]
(2,6)	Normalized Difference Built-Up Index	NDBI	[82]
(11,8)	Built-Up Index	BUI	[82]
(2,8)	Green Normalized Difference Vegetation Index	GNDVI	[83]
(3,8)	Green–Red Vegetation Index	GRVI	[84]
(4,5)	Normalized Difference Chlorophyll Index	NDCI	[85]
(4,8)	Normalized Difference Vegetation Index	NDVI	[86]
(5,11)	Normalized Difference Water Index	NDWI	[45]
(5,12)	Modified Normalized Difference Water Index	MNDWI	[87]
(6,7)	Normalized Difference Snow Index 1	$NDS{I}_1$	[88]
(11,12)	Normalized Difference Snow Index 2	$NDS{I}_2$	[88]
(7,11)	Normalized Difference Moisture Index	NDMI	[89]
(8,12)	Normalized Burn Ratio	NBR	[90]
(8A,5)	Chlorophyll Index—Red Edge	CIre	[91]
(8,11)	Land Surface Water Index	LSWI	[92]
(8,4)	Two-band Enhanced Vegetation Index	EVI2	[93]
(8,4)	Soil Adjusted Vegetation Index	SAVI	[94]

. The 20 m spatial resolution bands are reported to a higher resolution of 10 m using a simple bilinear upsampling approach. Then, this ${\rho }_{XY}$ coefficient is used to find the strength of the linear relationship between two variables, as shown in our analysis in

Table 4. Selected correlations between biological/chemical parameters and S-2 data.

NDI(i,j)	${\mathit{SQI}}_{\mathit{MDS}}$	${\mathit{SQI}}_{\mathit{TOT}}$	pH	WC	${\mathit{Fe}}_{\mathit{tot}}$	${\mathit{qCO}}_2$
i = 2; j = 5	−0.4168	0.2055	0.1693	−0.1246	−0.1884	0.1192
i = 2; j = 6	−0.5545	0.2748	0.023	−0.3161	−0.251	0.2166
i = 2; j = 7	−0.5548	0.2474	0.0592	−0.372	−0.2371	0.2393
i = 2; j = 8	−0.5082	0.219	0.1276	−0.3016	−0.1995	0.2296
i = 2; j = 8A	−0.5568	0.2531	0.0575	−0.328	−0.2158	0.171
i = 3; j = 5	−0.4247	0.2222	0.2987	−0.1751	−0.0381	0.0001
i = 3; j = 6	−0.6053	0.3183	0.0547	−0.4058	−0.2115	0.1827
i = 3; j = 7	−0.5999	0.2822	0.0931	−0.4621	−0.2051	0.2183
i = 3; j = 8	−0.5486	0.2457	0.1737	−0.3895	−0.1537	0.2049
i = 3; j = 8A	−0.6042	0.2914	0.0958	−0.4136	−0.1706	0.1298
i = 4; j = 5	−0.5463	0.2439	0.0198	−0.2808	−0.2024	0.0514
i = 4; j = 6	−0.5498	0.2688	−0.0767	−0.3682	−0.2357	0.1617
i = 4; j = 7	−0.5544	0.2476	−0.0343	−0.419	−0.2277	0.1925
i = 4; j = 8	−0.5124	0.2177	0.0347	−0.3571	−0.1885	0.1841
i = 4; j = 8A	−0.5619	0.2531	−0.042	−0.3826	−0.207	0.1255
i = 5; j = 6	−0.4112	0.2419	−0.1339	−0.3499	−0.2147	0.2007
i = 5; j = 7	−0.4387	0.2241	−0.0717	−0.4264	−0.2128	0.243
i = 5; j = 8A	−0.4491	0.2347	−0.0866	−0.3839	−0.1804	0.1558
i = 5; j = 12	0.2411	0.2052	0.1732	0.3017	0.4545	−0.0444
i = 6; j = 7	−0.2927	0.0512	0.181	−0.4716	−0.1051	0.2639
i = 6; j = 12	0.3869	−0.0015	0.1865	0.3993	0.4232	−0.1362
i = 7; j = 8A	0.1723	−0.0618	−0.0102	0.3752	0.2475	−0.4609
i = 7; j = 11	0.3996	−0.0189	0.188	0.4421	0.3955	−0.1807
i = 7; j = 12	0.4155	−0.0109	0.1486	0.4533	0.4193	−0.1701
i = 8; j = 12	0.3986	0.0149	0.0919	0.4234	0.3901	−0.1663
i = 8A; j = 11	0.4109	−0.0088	0.2075	0.4138	0.3868	−0.1064
i = 8A; j = 12	0.4228	−0.0005	0.1585	0.4343	0.4161	−0.1185
i = 11; j = 12	0.4318	0.0225	0.0652	0.4527	0.4723	−0.1361

. Our preliminary findings indicate a good correlation between the chemical/biological parameters and the S-2 bands, with correlation values ranging from 0.5 to 0.8. This suggests that certain bands provide more informative data than others for the specific scenario. As a result, we were only able to find some significant correlations between the two sets of data for a few S-2 indices, which are typically used to monitor forest fires, as detailed in the Introduction Section.

3.4. Relationship Between Biological/Chemical Parameters and Remote Sensing Data

The main purpose was to explore the potential of a set of S-2 indices using the 10 m and 20 m bands and extrapolate helpful biological and chemical information for large-scale forest monitoring and prevention. In order to study the correlation between biological/chemical parameters and the S-2 data, a set of 20 sites (described in Figure 3 as colored irregular polygons) are selected around the 20 spots where ground acquisitions were taken.

The red polygons represent regions around burned spots; the blue polygons represent regions around unburned spots. In addition, we selected, for comparison purposes, two regions (the green polygons in Figure 3) representing wide unburned regions significantly distant from the burned area. These polygons have been drawn by combining the GPS positions acquired during the on-field measurements and a very high-resolution (cm) image available on GEE. The polygons’ total area is equal to approximately 2 km². Once the polygons are drawn, we consider the correlation coefficient between the average values of all indices based on 10 m and 20 m S-2 bands in these irregular polygons and biological/chemical parameters to assess the nature and strength of the relationships. The values range from 0 to 1, and the linear relationship is characterized by a value close to 1. The correlation values are reported in

Table A1. Correlations between S-2 indices and $A{l}_{TOT}$, $M{n}_{TOT}$, $F{e}_{TOT}$, $M{g}_{TOT}$, $C{a}_{TOT}$, $N{a}_{TOT}$, $K_{TOT}$, $C{u}_{TOT}$, and $P_{TOT}$—Part I.

NDI(i,j)	${\mathit{Al}}_{\mathit{TOT}}$	${\mathit{Mn}}_{\mathit{TOT}}$	${\mathit{Fe}}_{\mathit{TOT}}$	${\mathit{Mg}}_{\mathit{TOT}}$	${\mathit{Ca}}_{\mathit{TOT}}$	${\mathit{Na}}_{\mathit{TOT}}$	${\mathit{K}}_{\mathit{TOT}}$	${\mathit{Cu}}_{\mathit{TOT}}$	${\mathit{P}}_{\mathit{TOT}}$
i,j = 2,3	−0.3459	0.0708	−0.2784	−0.2397	−0.2921	−0.2895	−0.4203	0.0968	0.0264
i,j = 2,4	−0.0555	−0.0057	−0.083	−0.0701	−0.1989	−0.2301	−0.0467	−0.0046	−0.0446
i,j = 2,5	−0.3699	−0.1742	−0.1884	−0.0621	−0.0682	−0.1941	−0.3193	−0.1507	−0.2253
i,j = 2,6	−0.4936	−0.1325	−0.251	−0.1438	−0.0035	−0.1312	−0.426	−0.1185	−0.1634
i,j = 2,7	−0.5018	−0.1627	−0.2371	−0.1048	0.0477	−0.1293	−0.4342	−0.1361	−0.1921
i,j = 2,8	−0.4488	−0.1699	−0.1995	−0.1238	0.0291	−0.1201	−0.3631	−0.1583	−0.1995
i,j = 2,8A	−0.4757	−0.1665	−0.2158	−0.0831	0.055	−0.1053	−0.403	−0.1527	−0.1975
i,j = 2,11	−0.0856	−0.3825	0.0865	0.1023	0.0854	0.0608	0.0216	−0.3861	−0.4214
i,j = 2,12	0.0867	−0.4197	0.2006	0.18	0.0994	0.1458	0.1777	−0.4323	−0.4559
i,j = 3,4	0.3277	−0.0809	0.2363	0.2053	0.1664	0.1372	0.4134	−0.1078	−0.0615
i,j = 3,5	−0.2348	−0.3061	−0.0381	0.1085	0.1476	−0.0104	−0.0952	−0.2952	−0.3425
i,j = 3,6	−0.4826	−0.1782	−0.2115	−0.0943	0.1092	−0.0641	−0.3677	−0.1726	−0.2006
i,j = 3,7	−0.4996	−0.2057	−0.2051	−0.0603	0.1527	−0.0746	−0.3912	−0.1833	−0.2281
i,j = 3,8	−0.4342	−0.2148	−0.1537	−0.0764	0.1356	−0.0622	−0.3053	−0.2112	−0.2382
i,j = 3,8A	−0.4596	−0.2157	−0.1706	−0.0238	0.1743	−0.0338	−0.3398	−0.2103	−0.2387
i,j = 3,11	0.0343	−0.4743	0.2056	0.2083	0.2117	0.1765	0.1926	−0.4899	−0.5047
i,j = 3,12	0.2114	−0.4892	0.3116	0.2761	0.2032	0.2487	0.3371	−0.5124	−0.5161
i,j = 4,5	−0.4368	−0.2042	−0.2024	−0.0581	0.0077	−0.1134	−0.3809	−0.1751	−0.2483
i,j = 4,6	−0.4745	−0.1186	−0.2357	−0.1379	0.0349	−0.0827	−0.414	−0.1049	−0.1406
i,j = 4,7	−0.4897	−0.149	−0.2277	−0.1049	0.0797	−0.087	−0.4267	−0.1228	−0.1707
i,j = 4,8	−0.4382	−0.1576	−0.1885	−0.1182	0.0653	−0.0792	−0.3595	−0.1463	−0.1803
i,j = 4,8A	−0.466	−0.1546	−0.207	−0.0842	0.0907	−0.0595	−0.3978	−0.1409	−0.1773
i,j = 4,11	−0.0943	−0.4515	0.1149	0.127	0.1499	0.1295	0.0279	−0.4559	−0.488
i,j = 4,12	0.1189	−0.5009	0.2567	0.2264	0.1611	0.2279	0.2205	−0.5166	−0.5354

,

Table A2. Correlations between S-2 indices and $A{l}_{TOT}$, $M{n}_{TOT}$, $F{e}_{TOT}$, $M{g}_{TOT}$, $C{a}_{TOT}$, $N{a}_{TOT}$, $K_{TOT}$, $C{u}_{TOT}$, and $P_{TOT}$—Part II.

NDI(i,j)	${\mathit{Al}}_{\mathit{TOT}}$	${\mathit{Mn}}_{\mathit{TOT}}$	${\mathit{Fe}}_{\mathit{TOT}}$	${\mathit{Mg}}_{\mathit{TOT}}$	${\mathit{Ca}}_{\mathit{TOT}}$	${\mathit{Na}}_{\mathit{TOT}}$	${\mathit{K}}_{\mathit{TOT}}$	${\mathit{Cu}}_{\mathit{TOT}}$	${\mathit{P}}_{\mathit{TOT}}$
i,j = 5,6	−0.3942	−0.0051	−0.2147	−0.1717	0.0216	−0.0676	−0.3533	−0.0066	−0.0095
i,j = 5,7	−0.4338	−0.0578	−0.2128	−0.1286	0.0797	−0.0806	−0.387	−0.0395	−0.0651
i,j = 5,8	−0.3485	−0.0728	−0.1482	−0.1352	0.0636	−0.0655	−0.2831	−0.0761	−0.0824
i,j = 5,8A	−0.3976	−0.0598	−0.1804	−0.0987	0.1026	−0.0352	−0.3426	−0.0614	−0.0666
i,j = 5,11	0.2394	−0.5341	0.3569	0.2403	0.2252	0.2827	0.3726	−0.5684	−0.5559
i,j = 5,12	0.423	−0.5272	0.4545	0.3251	0.2071	0.3463	0.5123	−0.5661	−0.5454
i,j = 6,7	−0.3551	−0.1935	−0.1051	0.079	0.2277	−0.1179	−0.3159	−0.1154	−0.2086
i,j = 6,8	−0.0709	−0.1608	0.0703	0.0119	0.1174	−0.0323	0.0089	−0.1677	−0.1767
i,j = 6,8A	−0.0106	−0.2792	0.1872	0.3992	0.4235	0.2265	0.0659	−0.2826	−0.2917
i,j = 6,11	0.4583	−0.3486	0.3961	0.2913	0.1362	0.2362	0.5132	−0.3702	−0.3584
i,j = 6,12	0.5023	−0.3549	0.4232	0.3154	0.1284	0.271	0.537	−0.3804	−0.3643
i,j = 7,8	0.1379	−0.0485	0.1344	−0.033	−0.0134	0.0289	0.1954	−0.1017	−0.0558
i,j = 7,8A	0.3725	0.006	0.2475	0.2027	0.0613	0.2345	0.3854	−0.0801	0.0133
i,j = 7,11	0.5056	−0.29	0.3955	0.2586	0.0817	0.2422	0.5485	−0.326	−0.2961
i,j = 7,12	0.5331	−0.3079	0.4193	0.2884	0.0862	0.2715	0.5583	−0.3439	−0.3148
i,j = 8,8A	0.0667	0.0518	0.0009	0.1441	0.0463	0.1022	0.0161	0.0576	0.063
i,j = 8,11	0.4596	−0.2752	0.3506	0.2723	0.0873	0.2362	0.4826	−0.2929	−0.279
i,j = 8,12	0.503	−0.2993	0.3901	0.2996	0.091	0.2684	0.5156	−0.322	−0.3042
i,j = 8A,11	0.4841	−0.3206	0.3868	0.2453	0.0761	0.2188	0.5276	−0.3426	−0.3291
i,j = 8A,12	0.5211	−0.332	0.4161	0.2842	0.0833	0.2592	0.5466	−0.358	−0.3404
i,j = 11,12	0.5846	−0.3574	0.4723	0.3577	0.112	0.3424	0.5798	−0.3923	−0.3651

,

Table A3. Correlations between S-2 indices and WC, pH, C, N, and OM—Part I.

NDI(i,j)	WC	pH	C	N	OM
i,j = 2,3	−0.0122	−0.0683	−0.2253	0.1592	−0.3458
i,j = 2,4	0.2359	0.3398	−0.1077	0.4277	−0.349
i,j = 2,5	−0.1246	0.1693	−0.4565	0.1665	−0.3708
i,j = 2,6	−0.3161	0.023	−0.356	0.0783	−0.2815
i,j = 2,7	−0.372	0.0592	−0.3155	0.1217	−0.252
i,j = 2,8	−0.3016	0.1276	−0.3317	0.1946	−0.3175
i,j = 2,8A	−0.328	0.0575	−0.3383	0.1084	−0.2808
i,j = 2,11	0.0106	0.2193	−0.4314	0.2316	−0.4246
i,j = 2,12	0.1323	0.2076	−0.403	0.2169	−0.3919
i,j = 3,4	0.1866	0.3179	0.1685	0.1456	0.1188
i,j = 3,5	−0.1751	0.2987	−0.492	0.084	−0.2502
i,j = 3,6	−0.4058	0.0547	−0.3989	0.0006	−0.2367
i,j = 3,7	−0.4621	0.0931	−0.3492	0.0638	−0.21
i,j = 3,8	−0.3895	0.1737	−0.3801	0.1452	−0.2975
i,j = 3,8A	−0.4136	0.0958	−0.3773	0.0437	−0.2381
i,j = 3,11	0.0039	0.2804	−0.4592	0.1892	−0.3807
i,j = 3,12	0.1438	0.2462	−0.3929	0.1756	−0.3282
i,j = 4,5	−0.2808	0.0198	−0.5261	−0.0308	−0.2832
i,j = 4,6	−0.3682	−0.0767	−0.3416	−0.0452	−0.1964
i,j = 4,7	−0.419	−0.0343	−0.3085	0.0095	−0.1791
i,j = 4,8	−0.3571	0.0347	−0.3293	0.0795	−0.2451
i,j = 4,8A	−0.3826	−0.042	−0.3306	−0.0104	−0.1994
i,j = 4,11	−0.0591	0.1446	−0.4984	0.1385	−0.4071
i.j = 4,12	0.102	0.1472	−0.4594	0.1412	−0.3813

,

Table A4. Correlations between S-2 indices and WC, pH, C, N, and OM—Part II.

NDI(i,j)	WC	pH	C	N	OM
i,j = 5,6	−0.3499	−0.1339	−0.1754	−0.0701	−0.1367
i,j = 5,7	−0.4264	−0.0717	−0.1609	0.0059	−0.1299
i,j = 5,8	−0.3357	0.0251	−0.1983	0.1005	−0.2258
i,j = 5,8A	−0.3839	−0.0866	−0.1781	−0.0212	−0.1525
i,j = 5,11	0.1501	0.2056	−0.3495	0.2388	−0.4035
i,j = 5,12	0.3017	0.1732	−0.2662	0.2005	−0.3057
i,j = 6,7	−0.4716	0.181	−0.0308	0.2705	−0.0578
i,j = 6,8	−0.1454	0.3387	−0.1598	0.3918	−0.3134
i,j = 6,8A	−0.1794	0.2542	−0.0284	0.262	−0.1025
i,j = 6,11	0.3653	0.2396	−0.081	0.215	−0.1384
i,j = 6,12	0.3993	0.1865	−0.069	0.1752	−0.1175
i,j = 7,8	0.1277	0.2363	−0.1447	0.2358	−0.2827
i,j = 7,8A	0.3752	−0.0102	0.0121	−0.1011	−0.0112
i,j = 7,11	0.4421	0.188	−0.0691	0.1468	−0.1166
i,j = 7,12	0.4533	0.1486	−0.0591	0.1253	−0.1031
i,j = 8,8A	0.0801	−0.2421	0.1525	−0.2904	0.2767
i,j = 8,11	0.3982	0.1065	−0.0186	0.0641	−0.0176
i,j = 8,12	0.4234	0.0919	−0.0253	0.0662	−0.0359
i,j = 8A,11	0.4138	0.2075	−0.0774	0.1818	−0.1284
i,j = 8A,12	0.4343	0.1585	−0.0651	0.1481	−0.1112
i,j = 11,12	0.4527	0.0652	−0.0544	0.0775	−0.0703

,

Table A5. Correlations between S-2 indices and $SQ{I}_{MDS}$, $SQ{I}_{TOT}$, Resp, $qC{O}_2$, $V_{TOT}$, $C{r}_{TOT}$, and $N{i}_{TOT}$—Part I.

NDI(i,j)	${\mathit{SQI}}_{\mathit{MDS}}$	${\mathit{SQI}}_{\mathit{TOT}}$	Resp	${\mathit{qCO}}_2$	${\mathit{V}}_{\mathit{TOT}}$	${\mathit{Cr}}_{\mathit{TOT}}$	${\mathit{Ni}}_{\mathit{TOT}}$
i,j = 2,3	−0.2018	0.0941	0.1595	0.2	0.0838	0.2667	−0.1243
i,j = 2,4	0.0556	0.0228	0.1599	0.1951	−0.019	0.0406	−0.3033
i,j = 2,5	−0.4168	0.2055	0.0223	0.1192	−0.1579	0.1733	0.1637
i,j = 2,6	−0.5545	0.2748	0.1786	0.2166	−0.1082	0.1125	0.3069
i,j = 2,7	−0.5548	0.2474	0.2295	0.2393	−0.1366	0.0295	0.3523
i,j = 2,8	−0.5082	0.219	0.2172	0.2296	−0.1507	−0.0081	0.3299
i,j = 2,8A	−0.5568	0.2531	0.1354	0.171	−0.1425	0.0764	0.3589
i,j = 2,11	−0.2015	0.2697	−0.016	0.0586	−0.3787	0.0833	0.285
i,j = 2,12	−0.0524	0.2458	−0.0747	0.0169	−0.4236	0.1138	0.227
i,j = 3,4	0.2531	−0.0844	−0.0533	−0.071	−0.1044	−0.255	−0.0625
i,j = 3,5	−0.4247	0.2222	−0.1045	0.0001	−0.2942	0.0181	0.3195
i,j = 3,6	−0.6053	0.3183	0.1515	0.1827	−0.1531	0.0366	0.4084
i,j = 3,7	−0.5999	0.2822	0.2211	0.2183	−0.1785	−0.0505	0.4473
i,j = 3,8	−0.5486	0.2457	0.2006	0.2049	−0.1955	−0.0921	0.4341
i,j = 3,8A	−0.6042	0.2914	0.1007	0.1298	−0.1913	−0.0034	0.4615
i,j = 3,11	−0.1612	0.2906	−0.0762	−0.003	−0.4754	−0.0084	0.3615
i,j = 3,12	0.0063	0.2441	−0.1283	−0.0397	−0.4981	0.0363	0.2773
i,j = 4,5	−0.5463	0.2439	−0.0486	0.0514	−0.1769	0.2005	0.3217
i,j = 4,6	−0.5498	0.2688	0.1351	0.1617	−0.0913	0.1112	0.3547
i,j = 4,7	−0.5544	0.2476	0.1925	0.1925	−0.1201	0.033	0.393
i,j = 4,8	−0.5124	0.2177	0.1799	0.1841	−0.1352	−0.0024	0.3761
i,j = 4,8A	−0.5619	0.2531	0.101	0.1255	−0.1274	0.0779	0.4084
i,j = 4,11	−0.2613	0.3167	−0.0523	0.0249	−0.443	0.0957	0.4015
i,j = 4,12	−0.074	0.2843	−0.1209	−0.0211	−0.5025	0.1315	0.3247

and

Table A6. Correlations between S-2 indices and $SQ{I}_{MDS}$, $SQ{I}_{TOT}$, Resp, $qC{O}_2$, $V_{TOT}$, $C{r}_{TOT}$, and $N{i}_{TOT}$—Part II.

NDI(i,j)	${\mathit{SQI}}_{\mathit{MDS}}$	${\mathit{SQI}}_{\mathit{TOT}}$	Resp	${\mathit{qCO}}_2$	${\mathit{V}}_{\mathit{TOT}}$	${\mathit{Cr}}_{\mathit{TOT}}$	${\mathit{Ni}}_{\mathit{TOT}}$
i,j = 5,6	−0.4112	0.2419	0.2252	0.2007	0.0158	0.0466	0.2599
i,j = 5,7	−0.4387	0.2241	0.2976	0.2430	−0.0336	−0.0475	0.3253
i,j = 5,8	−0.3723	0.172	0.2606	0.2179	−0.058	−0.0906	0.3019
i,j = 5,8A	−0.4491	0.2347	0.1798	0.1558	−0.0382	0.0070	0.3541
i,j = 5,11	0.0749	0.2815	−0.0160	0.018	−0.5454	−0.0146	0.338
i,j = 5,12	0.2411	0.2052	−0.1080	−0.0444	−0.5463	0.0441	0.2139
i,j = 6,7	−0.2927	0.0512	0.3994	0.2639	−0.1701	−0.3459	0.3893
i,j = 6,8	−0.0883	−0.0665	0.1827	0.1256	−0.1660	−0.3275	0.2262
i,j = 6,8A	−0.1952	−0.0100	−0.2552	−0.2484	−0.2758	−0.2055	0.444
i,j = 6,11	0.3608	−0.0076	−0.1745	−0.1353	−0.3721	−0.0537	0.0184
i,j = 6,12	0.3869	−0.0015	−0.1913	−0.1362	−0.3794	0.0024	0.0051
i,j = 7,8	0.0822	−0.0979	−0.0510	−0.0288	−0.0676	−0.1282	−0.0049
i,j = 7,8A	0.1723	−0.0618	−0.6109	−0.4609	−0.0166	0.2223	−0.0607
i,j = 7,11	0.3996	−0.0189	−0.2456	−0.1807	−0.3169	0.0215	−0.0553
i,j = 7,12	0.4155	−0.0109	−0.2429	−0.1701	−0.3349	0.0566	−0.0506
i,j = 8,8A	0.013	0.064	−0.2846	−0.2246	0.0584	0.2506	−0.0312
i,j = 8,11	0.3719	0.0156	−0.2279	−0.17	−0.2956	0.0662	−0.0538
i,j = 8,12	0.3986	0.0149	−0.2341	−0.1663	−0.3216	0.0869	−0.0483
i,j = 8A,11	0.4109	−0.0088	−0.1469	−0.1064	−0.3458	−0.0209	−0.0521
i,j = 8A,12	0.4228	−0.0005	−0.1753	−0.1185	−0.3578	0.0288	−0.0470
i,j = 11,12	0.4318	0.0225	−0.2265	−0.1361	−0.3834	0.1240	−0.0229

.

3.5. Least Square (LS) Approach on Remote Sensing Data

In this paper, we adopt the LS approach to calculate a $SQI$ from the more relevant S-2 indices. The strong correlation between these two datasets is utilized in this method. Our research is focused on the S-2 band combinations that are more closely correlated with the physical parameters. We particularly restricted our selection to S-2 indices with a correlation to the $SQI$ of at least 0.5. The squared residuals arising from discrepancies in the observed $SQI$ value and the estimated $SQI$ value, based on this model, tend to be zero since the LS approach seeks to establish a straight hyper-plane that minimizes the squares of the errors produced by the results of the related equations. We have obtained experimental evidence thanks to this methodology. As a result, we gathered all potential NDIs from the Sentinel-2 data. We then calculated the average S-2 NDI values over all polygons that were taken into consideration, and we associated this set of data as the input of our workflow, with the $SQI$ values for the associated polygons serving as the output. The LS technique receives all of these average values of the chosen indices as input, and the resulting output is the observed $SQI$ parameter. We balanced all NDIs under consideration using the LS approach to obtain the smallest error between the estimated and observed $SQI$. The 20 boxes (5 sites for each vegetation covers: Herbaceous and Shrubs (H), Black Locust (BL), Pines (P) and Holm Oak (HO)) were divided into two datasets that were randomly selected. The training set consists of sixteen boxes (4 for each vegetation), from which the LS-based model parameters are derived. The correctness of the model is then determined using a distinct set of stations known as the validation or testing set. This data partition guarantees that the model is trained and tested on several sets of data (the remaining four: 1 for each vegetation), providing a fair assessment of the model’s ability to generalize to new, previously unknown data.

3.6. Metrics

To assess the strength of the relationships between the $SQI$ estimated by remote sensing acquisitions and the observed $SQI$, we considered the coefficient of determination ($R^2$) and relative error ($e_r$) as statistical measures. The $R^2$ and the $e_r$ are defined as follows:

$$R^2=1-{\displaystyle \frac{{\sum }_{i=1}^M{(y_i-{\widehat{y}}_i)}^2}{{\sum }_{i=1}^M{(y_i-\underline{y})}^2}}$$

(4)

$$e_r={\displaystyle \frac{{\sum }_{i=1}^M{(y_i-{\widehat{y}}_i)}^2}{{\sum }_{i=1}^M{y}_i^2}}$$

(5)

The $R^2$ values range from 0 to 1, with values close to 1 indicating a good model fit. Instead, when the $e_r$ value is close to 0, the estimated parameter fits the data coherently.

4. Results

4.1. Soil Quality Index ($SQ{I}_{TOT}$) and Soil Quality Index Minimum Dataset ($SQ{I}_{MDS}$)

In this research, a soil quality index (called $SQ{I}_{TOT}$) was derived from a comprehensive set of abiotic and biotic parameters: pH, water content, total C and N, organic matter, macro- and micronutrient concentrations, microbial activity, enzymatic assays, and phytotoxicity tests [7,29]. As shown in Figure 4, $SQ{I}_{TOT}$ ranged from 0.40 to 0.50 and did not reveal any statistically significant differences among vegetation cover types (herbaceous, black locust, pines, holm oak) or across the four sampling times (summer, fall, winter, spring). Figure 4a further illustrates the relative stability of $SQ{I}_{TOT}$ values. While this index offers a broad overview of soil status, its correlation with Sentinel-2 multispectral data proved limited. A more refined approach was then adopted through the use of an $MDS$ strategy, guided by standardized principal component analysis to isolate key parameters (pH, soil water content, total iron concentrations, and qCO₂). This reduced soil quality index, $SQ{I}_{MDS}$, ranged between 0.40 and 0.62, as displayed in Figure 4b. Despite again showing no statistically significant variation among vegetation covers or sampling periods, $SQ{I}_{MDS}$ exhibited stronger sensitivity to fire-driven changes. Correlation analyses with Sentinel-2 multispectral bands underscored the importance of Near-Infrared (NIR) and Shortwave Infrared (SWIR) wavelengths in detecting shifts in soil moisture and iron, both of which can be significantly altered by high-temperature events.

4.2. Numerical Assessment and Visual Inspection

The numerical assessment is conducted using the described metrics. First, we report in Table 4 the correlations between the parameters measured in situ and the indices from remote sensing. From the correlation analysis, it is clearly possible to see that the $SQ{I}_{MDS}$ is the one that manages to obtain the highest correlations with the S-2 multispectral data. Furthermore, the indices involving the red-edge and SWIR bands are the most informative for the current problem. From this, we proceed with the analysis of the estimate of the in situ parameters obtained from the satellite data.

As reported in Table 4, and further supported by the results in

Table 5. Coefficient of determination and relative error. The bold values indicate that the $SQ{I}_{MDS}$ index performs better in terms of $R^2$ and $e_r$ compared to $SQ{I}_{TOT}$.

	${\mathit{R}}^2$	${\mathit{e}}_{\mathit{r}}$
Ideal value	1.00	0.00
$SQ{I}_{TOT}$	0.15	0.08
$SQ{I}_{MDS}$	0.79	0.06

, we can observe that the regression model employed to predict $SQ{I}_{MDS}$ gives good results in terms of both metrics. In fact, the coefficient of determination ($R^2$) is higher than 0.75, which is the theoretical threshold to be considered as a good correlation. The relative error (er) is equal to 0.06, which means a good capability to recover the observed values. Instead, the estimation of $SQ{I}_{TOT}$ presents an acceptable relative error, $e_r$ (0.08), and a very low $R^2$ value (0.15). The metrics underlined the low correlation coefficients obtained by the $SQ{I}_{TOT}$. Instead, the $SQ{I}_{MDS}$ showed a good estimation capability, although the correlation coefficients are slightly better when we consider some S-2 indices that used red-edge and SWIR bands. Figure 5 maps the estimated $SQ{I}_{MDS}$ over the study area at three distinct times, capturing (i) the immediate post-fire conditions (August 2017), where pronounced drops in soil quality become evident in the burnt zones, shown in Figure 5b, (ii) the progressive recovery detected about one year later in Figure 5c, and (iii) the further improvement observed by August 2019 in Figure 5d, reflecting gradual vegetation regrowth and soil stabilization.

The conducted analysis allows us to determine the vegetation recovery in the Vesuvius National Park area.

4.3. Time-Series of the Estimated $SQI$

To place these variations in a longer context, Figure 6 and Figure 7 show monthly series of both $SQ{I}_{MDS}$ and $SQ{I}_{TOT}$ from 2016 to 2020. We analyzed this time series from 2016 to 2020, to gain an understanding of the condition of the burned forests prior to the event in 2017 and three years after the event. The first two years after were because we had in situ data, and the third year was to see if the satellite-estimated trend continued in the direction assessed by the in situ data. In Figure 6 and Figure 7, we can see the trend of the two analyzed indexes ($SQ{I}_{MDS}$ and $SQ{I}_{TOT}$) over the period of time considered and the area under investigation. We plotted one value per month, evaluated as an average among all cloudless acquisitions of the current month. The $SQI$ values estimated with the proposed method for the entire 2016–2020 period are reported on the y-axis. With this analysis, we have highlighted that the estimated $SQ{I}_{MDS}$ index gives essential information on the soil quality and seems to be correctly estimated by a proper combination of S-2 indices. As we can observe in Figure 6, the fire of summer 2017 characterized the multi-temporal trend of the $SQ{I}_{MDS}$. Specifically, in Figure 6, the $SQ{I}_{MDS}$ curve drops when the first wildfires occurred and coherently decreased following degradation of the soil quality and the vegetation status. Instead, observing Figure 7 with the $SQ{I}_{TOT}$, the fire events are not evident, and there is no coherent trend with the vegetation status, and the vegetation recovery occurred in the following years after the ignition events of 2017. Further confirmation of the obtained results is given by the behavior of $SQ{I}_{MDS}$ in the presence of unburned areas, and in the presence of areas close to burned ones. We observe a periodic behavior associated with seasonal features of the considered forests in unburned cases. Instead, we note a slower drop of the trend in areas close to burned ones due to the closeness with the burned areas. Furthermore, due to the extension of the ignition events, we can assume that fires probably affected some pixels of these boxes.

5. Discussion

The present study evaluated the ability of Sentinel-2 data to complement in situ soil measurements in a Mediterranean forest area severely impacted by wildfires. This work is framed within a broader context where forest ecosystems play a major role in climate regulation, carbon storage, and biodiversity conservation, yet increasingly face significant losses due to anthropogenic drivers and natural disturbances [95]. As reported by [3,4], approximately half of the global forest loss can be traced back to a combination of natural and human-related activities associated with fires, generating immediate and long-term repercussions on soil properties. The obtained results show that Sentinel-2 satellite images can be fruitfully used to complement in situ data, in particular, to estimate the $SQI$ in a relatively large area (km²), covered by different vegetation. These findings confirm the added value of integrating in situ soil data and Earth observation imagery for forest monitoring, in line with previous research on remote sensing applications for environmental assessment [13,14,15]. Observing the post-fire evolution of soils is crucial since fire-driven alterations in pH, water repellency, nutrient cycles, and microbial communities can persist for extended periods and ultimately affect vegetation dynamics and erosion processes [5,8].

In the considered area, the surface conditions of vegetation and soil were variable over time due to a series of fires that occurred in the summer of 2017. The $SQ{I}_{TOT}$ results suggest that, while a large number of chemical and biological parameters can describe soil conditions in detail, their combined index may not always mirror short-term or localized changes, and the correlation with spectral data may remain weak. By contrast, $SQ{I}_{MDS}$ proved more effective, indicating that pH, soil moisture, total iron, and qCO₂ play a central role in capturing the impacts of high temperatures and ash deposition. This aligns with the notion that soil moisture deficits and nutrient loss (or redistribution) are hallmarks of wildfire damage, and that subsequent vegetation regeneration hinges on the availability of water and essential elements [96].

In Figure 5b, the reduced $SQ{I}_{MDS}$ values in recently scorched regions highlight the immediate post-fire degradation. Soil water content often decreases substantially, while iron oxidation states and pH undergo marked alterations [97,98]. The rebound documented in Figure 5c,d reveals an encouraging trend, linked to gradual recolonization by herbaceous and woody species. The continuous monitoring shown in Figure 6 highlights how $SQ{I}_{MDS}$ captures these restoration processes better than the broader $SQ{I}_{TOT}$ (Figure 7), likely because its selected parameters respond more rapidly to disturbance and recovery.

The integration between remote sensing and field data showed that indices based on B8, B11, and B12 S-2 bands are significantly correlated with soil water content and total Fe concentrations. Soil water content is often altered during a fire, negatively impacting the water retention soil capability [5]. Consequently, water infiltration in soil is reduced, increasing the risk of runoff [8]. In addition, the significant correlations between the soil water content and the combination bands (B3, B8, B11) of the NDWI and MNDWI indices suggested that these two indices have good capability in water detection, and they can be considered useful tools in environmental monitoring after fire disturbance. The found correlation between soil total Fe concentration and remote sensing data confirms the power of remote sensing to detect post-fire changes in soil characteristics. In fact, fire occurrence altered the total and available content of nutrients.

Despite the promising correlation coefficients and the practical advantages of mapping soil conditions at large scales, further refinements remain possible. Matching the timing of ground surveys more precisely with Sentinel-2 overpasses could minimize disparities between laboratory measurements and satellite acquisitions. Adding high-sensitivity markers, such as advanced enzymatic indicators or molecular proxies of microbial diversity, might improve the early detection of stress conditions [54]. Verifying the applicability of the $SQ{I}_{MDS}$ approach in different soil types, climatic regimes, or vegetation assemblages would strengthen the generalizability of these findings. Nonetheless, the insights gained here corroborate that a well-structured integration of in situ measurements and multispectral indices constitutes a valuable strategy for tracking soil quality in fire-prone areas. This integrated system can thus support risk assessment, inform land management decisions, and guide restoration efforts aimed at preserving biodiversity and ecosystem functionality.

6. Conclusions

In this study, we proposed a methodology for combining data from several sources, particularly in situ measurements and images from remote sensing for soil monitoring purposes. We have shown the potential of a simple method for estimating the Soil Quality Index (SQI) from the Sentinel-2 indices. In particular, we were able to make a large-scale estimate of the $SQI$ over the National Park of Vesuvius, the area of interest. This capability was highlighted by displaying the time series’ five-year trend. In fact, the incendiary event is clearly visible in the soil quality index trend, and the time-series curve nicely illustrates the vegetation’s ability to recover. The numerical and visual outcomes confirm the validity of this strategy and anticipate developments in the future, including the enhancement of the suggested algorithm and the execution of specific in situ measurements synchronized with the passage of the Sentinel-2 satellite. While the study provided promising results and information on soil quality in other areas close to those acquired (spatial extension) as well as on other dates (temporal extension), further on-ground measurements are required to verify the method stability elsewhere. Moreover, it recommends examining the transferability method to various areas with other geomorphological characteristics, requiring additional measurement campaigns.