Multi-Temporal Remote Sensing for Forest Conservation and Management: A Case Study of the Gran Chaco in Central Argentina

Abstract

Anthropogenic alteration of tropical and subtropical forests is a major driver of biodiversity loss; notably, the Chaco Forest, which is the largest dry forest in the Americas, is among the most impacted regions. Sustainable forest management, a key objective of the UN’s 15th Sustainable Development Goal (SDG), underscores the need for advanced monitoring tools. This study integrates Sentinel-2 remote sensing (RS) spectral indices with field data to analyze forests under varying management regimes and levels of alteration in a representative area of the Chaco region (Chancaní Provincial Reserve and surrounding areas of the West Arid Chaco). Forest structure types and conservation levels were linked to monthly spectral index behavior using linear mixed models. Spectral indices such as the BI (Brightness Index), NDWI_Gao (Normalized Difference Water Index), and MCARI_Sent (Modified Chlorophyll Absorption in Reflectance Index) effectively differentiated forest stands by conservation status and structural alteration. This combined RS and field data approach proved highly effective for detecting and characterizing forests with diverse conservation and sustainability conditions. The methodology demonstrates significant potential as a reliable RS-based tool for monitoring forest health and supporting progress toward SDG targets, particularly in regions like the Chaco Forest, which face extensive anthropogenic pressures.

1. Introduction

Forest ecosystems have played a major role in human history, and forest management has accompanied population growth and development worldwide for thousands of years [1]. The extent of anthropogenic alteration and degradation of natural forests in many tropical and subtropical ecosystems is one of the main causes of biodiversity loss worldwide [2,3]. Forest degradation is a process that diminishes the biodiversity and functioning of these ecosystems [4]. Rather than reducing forest area, degradation involves the deterioration of the structure and composition of woody plant biomass, leading to a decline in the forest’s overall quality [4]. Thus, degradation directly results in a long-term loss of the forest’s capacity for carbon storage [5] and its ability to provide goods and services [6].

Identifying effective strategies for forest management is a crucial challenge of the Anthropocene Epoch. In response to this challenge, the United Nations included forest management and governance strategies in the 2030 Sustainable Development Agenda (SDG) [7]. The 15th SDG states, “Protect, restore and promote sustainable use of terrestrial ecosystems, sustainably manage forests, combat desertification, and halt and reverse land degradation and halt biodiversity loss”. Achieving this goal requires significant efforts to develop effective monitoring tools that illustrate sustainable management practices and support policymakers [1].

In this context, remote sensing (RS) could provide valuable support as an effective source of data and methods for forest monitoring, offering a cost-effective option for frequent observation of large areas [8]. RS provides multi-temporal data at various spatial scales in a cost-effective, spatially contiguous, and timely manner, supporting standardized assessments of forest cover and fragmentation [1,9]. However, while RS is particularly effective for producing forest cover maps with explicit information on forest distribution, using RS to quantify forest degradation is more challenging. In such cases, the area remains forested but undergoes structural, compositional, and functional alterations [10]. Recent advancements in satellite imaging, algorithms, and computing now enable high-resolution, large-scale mapping of forest disturbances [8]. While most studies focus on boreal, temperate, and tropical wet forests, the mapping of forest alterations in subtropical semi-arid regions has only recently gained attention [11,12]. To accurately assess forest characteristics in bio-climatic regions with strong seasonality, RS techniques can be combined with approaches that capture spatial and temporal vegetation patterns [13]. For instance, analyzing ecosystem phenological properties, which are driven by climatic seasonality [14], is an effective method for mapping intense seasonal biomass variations, such as those that characterize subtropical forests [15,16].

The implementation of RS for mapping forest characteristics worldwide, particularly in subtropical and developing countries where natural forests face significant pressure, greatly benefits from the availability of free satellite imagery with good spatial resolution and high revisit frequency, such as that provided by the SENTINEL satellite constellation [8,16]. These studies primarily rely on remote sensing data, as further research is needed to integrate remote sensing with conservation and alteration metrics obtained through field measurements [17,18].

The Gran Chaco, one of the largest seasonally dry subtropical forests in the world, spans Argentina, Paraguay, Brazil, and Bolivia [19,20]. It is also one of the most threatened ecosystems in Latin America [3]. Forest fragmentation and degradation, driven by unsustainable practices such as logging, firewood collection, charcoal production, and livestock grazing [9,21,22,23,24], have significantly impacted the Chaco Forest. Numerous studies have documented how the overuse of Chaco forests leads to biodiversity loss [25,26,27,28,29,30], changes in surface and groundwater hydrological cycles [31,32,33,34,35,36], and soil loss, which contributes to the advancement of land desertification [33,35]. Furthermore, forest degradation leads to increased carbon dioxide emissions into the atmosphere [36,37,38,39].

Previous studies in the Chaco region have examined the effects of anthropogenic and natural disturbances on plant community phenology using RS data [11,40,41]. Most of these studies have focused on the Normalized Difference Vegetation Index (NDVI), widely recognized for its ability to assess the fraction of photosynthetically active radiation absorbed by vegetation [42,43,44]. Recently, however, advancements in satellite data and processing power have enabled the exploration of additional spectral indices with enhanced spatial and temporal resolution and faster processing speeds [18,45]. These new tools provide opportunities for detailed phenological analyses and the development of novel indicators for sustainable management specifically tailored to the Chaco region.

This study aims to explore the potential of spectral indices derived from Sentinel-2 remote sensing data, combined with field data, as indicators of forest structural integrity and degradation resulting from sustainable or unsustainable management practices in the Arid Chaco forests. We hypothesize that the phenological patterns of RS ecological indices vary depending on forest characteristics and the degree of alteration. To test this hypothesis, we calculated a series of spectral indices useful for RS-based ecosystem descriptions, including those related to photosynthetic activity and biomass, canopy water content and moisture, water stress, soil characteristics, chlorophyll and carotenoid content, and fire-induced stress.

We then performed a Variance Inflation Factor (VIF) analysis to identify and retain only the variables with low multicollinearity, ensuring a focus on minimally correlated variables. Next, we assessed the relationship between this set of low-correlated spectral indices and field-collected data on forest structures across ecosystems with varying levels of structural alteration. Specifically, based on a set of minimally correlated monthly RS spectral variables and field data, we addressed the following questions:

(a)

Do forests with different management regimes, levels of degradation (structural alteration), and dominant species exhibit different spectral phenology (i.e., the study of seasonal changes in vegetation as observed through spectral indices)?

(b)

Which indices effectively differentiate forests with different levels of structural degradation?

Validating our hypothesis—that in highly seasonal subtropical forests, the phenological patterns of remote sensing (RS) ecological indices vary with ecosystem characteristics and the degree of alteration—could provide further evidence of RS as a powerful proxy for modeling forest attributes. The integration of a statistically adequate number of field samples with monthly Sentinel RS ecological indices would strengthen our ability to conduct accurate, large-scale assessments at a significantly lower cost than relying solely on traditional field inventories.

2. Materials and Methods

2.1. Study Area

The study area is located in the southern sector of the dry Chaco in central Argentina (Figure 1) and is representative of the Chaco Phytogeographical Province [20]. It includes a protected area (Chancaní Provincial Reserve) and surrounding forests with different management regimes and levels of degradation (Figure 1A). Although these forests provide essential ecosystem services to local communities and contribute to the regional economy [46,47], making them valuable for conservation and sustainable management, the region’s permissive and liberal legal regulations fail to prevent deforestation and degradation [48].

The study area comprises two main geomorphological landscapes: foothills and plains (Figure 1B). The foothills, located on the western slopes of the Sierras de Pocho, range from 370 to 500 m above sea level and have permeable, poorly developed soils. The plains, the most extensive landform, feature gentle slopes of less than 1% and sedimentary deposits with sandy or clayey soils of low fertility, which are prone to erosion [49,50]. The region has a subtropical xeric climate [51], characterized by a dry season, high evapotranspiration [33], and a mean annual rainfall of 600 mm, mostly occurring between October and April. The average annual temperature is 18 °C [52].

The vegetation of the study area includes xerophilous subtropical forests [19] and shrublands [51,53]. In the foothills, forests are dominated by Aspidosperma quebracho-blanco Schltdl., accompanied by isolated individuals of Neltuma flexuosa (DC.) C. E. Hughes and G. P. Lewis, Sarcomphalus mistol (Griseb.) Hauenschild, Neltuma torquata (Lag.) DC., Neltuma chilensis (Molina) C. E. Hughes and G. P. Lewis, and Celtis iguanaea (Jacq.) Sarg. The tree layer can reach heights of up to 15 m and is dominated by A. quebracho-blanco, while the shrubby layer, reaching heights of up to 4 m, is primarily composed of Larrea divaricata Cav., Mimozyganthus carinatus (Griseb.) Burkart, Senegalia gilliesii (Steud.) Seigler and Ebinger, and Celtis pallida Torr.

In the western plains, forests are co-dominated by A. quebracho-blanco and N. flexuosa, with shrub cover exceeding 60%, depending on the level of biodiversity conservation. The dominant shrub species in these areas include L. divaricata, M. carinatus, S. gilliesii, and C. pallida. In the southern and central depressed areas, which have higher moisture levels and finer soils, plant communities are dominated by N. flexuosa and C. pallida. In the northern saline lowlands, plant communities are dominated by halophilic species such as Geoffroea decorticans (Gillies ex Hook. and Arn.) Burkart, Allenrolfea spp. Kuntze, Suaeda spp., and Atriplex cordobensis Gand. and Stuck [50,54,55,56].

Shrub communities vary in composition and height. Mixed shrublands often exhibit two layers (shrubby and herbaceous) with some isolated trees under 5 m tall (A. quebracho-blanco and N. flexuosa). The shrubby layer is usually dominated by L. divaricata, M. carinatus, and S. gilliesii. In significantly degraded areas, the proportion of bare soil is high, while woody species diversity is low. In some locations, woody communities have been replaced by sprayed crops such as potatoes, corn, and wheat or by implanted Cenchrus ciliaris L. (buffel grass) pastures with scattered small trees [57].

2.2. Definition and Selection of Sampling Sites

Field data were collected between 2017 and 2019 from 49 sampling sites across a 300,000-hectare area in the southeastern sector of the Arid Chaco, encompassing both major landscapes: plains and foothills (Figure 1). Sampling was conducted using a stratified approach, incorporating Google Earth Pro (Google LLC, Mountain View, CA, USA; version 7.3.7) and expert knowledge, along a gradient of land use history and conservation levels. The conservation level was classified on a scale from 1 (very high) to 8 (very low), based on the extent of structural and species composition modifications within the natural ecosystem matrix. Logistical factors, such as accessibility via roads or trails and obtaining permits from public and private landowners, also influenced site selection. The sampling strategy was designed to comprehensively capture the area’s heterogeneity by considering geomorphological landscapes, vegetation types, management practices, and levels of structural degradation.

Following previous studies in the forests of the Arid Chaco and other regions [57], we defined a sample size that was statistically robust and representative [58], while ensuring that the fieldwork effort remained manageable within a limited time frame and with the available staff (see [50] for details).

A minimum distance of 1 km was maintained between sampling sites. Within each landscape (foothills and plains), we selected mature, well-preserved forests that had remained undisturbed for at least the past 30 years (e.g., mature forests without signs of grazing, browsing, or logging, and without evident soil erosion) to serve as reference sites. These reference forests included Chancaní Forest Reserve (representative of the foothill landscape; 31°21′22″ S, 65°27′29″ W) and a forest relic located on a private property, Estancia El Álamo (representative of the plain landscape; 31°43′28.54″ S, 65°24′11.78″ W).

Additional sampling sites were selected in the surroundings of the reference sites, maintaining similar soil types, topography, and potential plant communities (e.g., same ecological site sensu [59]) but differing in management practices, resulting in distinct physiognomies. These included closed forests with emergent trees, closed forests without emergent trees, low closed forests, open forests, natural grasslands, and cultivated pastures (

Table 1. Forest classes sampled in the study area including common name, landscape type, land-use management history, number of sampling sites, and the conservation level (CL: with higher conservation values close to 1 and lower than 8).

Forest Class/ Structure-Physiognomy	Landscape	Land Use History	N° Sites	CL
Mature forest of Neltuma flexuosa and/or Aspidosperma quebracho blanco. Common name: Mature forests on foothills	Foothills	Natural protected areas with limited logging or livestock grazing for at least five decades.	7	1
Mature forest of Neltuma flexuosa and/or Aspidosperma quebracho blanco. Common name: Mature forests on plains	Plains	Scarce logging or livestock grazing for at least five decades.	2	2
Closed forest with emergent trees of Neltuma flexuosa and/or Aspidosperma quebracho blanco, with understory dominated by Mimozyganthus carinatus. Common name: Closed forests on foothills	Foothills	Low to moderate livestock forestry pressure. Selective felling of large (e.g., >40 cm diameter at breast height), with intervals of more than 20 years. Annual stocking rate lower ** than 1 * Cow Equivalent (CE)/20 hectares, grazing in autumn–winter. Seasonal supplement, small plots (e.g., <3% of the farm area) with megathermal pastures.	6	3
Closed forest with emergent trees of Neltuma flexuosa and/or Aspidosperma quebracho blanco, with understory dominated by Mimozyganthus carinatus. Common name: Closed forests on plains	Plains	Low to moderate livestock forestry pressure. Selective felling of large trees (e.g., >40 cm diameter at breast height), with intervals of more than 20 years. Annual stocking rate lower ** than 1 * CE/20 ha., grazing in autumn–winter. Seasonal supplement, small plots (e.g., <3% of the farm area) with megathermal.	4	4
Closed forest of Neltuma flexuosa, Larrea divaricata, Mimozyganthus carinatus, and/or Parkinsonia praecox. Common name: Low closed forest	Foothills and plains	No mature trees remained, no resting periods or forest management. Low logging for firewood extraction. Stocking rate moderate to high (e.g., 1 CE/5 ha). Some time with fires.	10	5
Low closed forest of Neltuma flexuosa, Aspidosperma quebracho blanco, Larrea divaricata, Mimozyganthus carinatus, and/or Celtis ehrenbergiana. Common name: Shrublands	Foothills and plains	Fires or total logging of trees have occurred, followed by a high post-fire stoking rate. Moderate to heavy logging, moderate to high stocking rate of cows and/or goats (e.g., more than 1 CE/2–4 ha, sustained for more than a decade).	10	6
Open forest of Neltuma flexuosa, Mimozyganthus carinatus, and/or Celtis ehrenbergiana. Common name: Savannas	Foothills and plains	Heavy logging and high stocking rate of cows and/or goats over the past decades, with mechanical shrub removal and/or partial felling of the woody layer every 3 to 5 years (e.g., 50 to 70% of shrub cover is removed). Carrying capacity is declining due to reduced productivity caused by chronic degradation.	6	7
Natural grassland and implanted pastures with Tricloris sp. and/or Cencrus ciliaris. Isolated individuals of Neltuma flexuosa and Aspidosperma quebracho blanco. Common name: Grasslands with trees	Foothills and plains	Sites cleared totally or partially for grassland productivity and/or to establish megathermal grass pastures. High livestock density with significant anthropic input (e.g., rolling every 3 to 5 years with interplanting, removing more than 80% of woody cover). Livestock stocking rate 1 CE/ha.	4	8
TOTAL			49

* Cow Equivalent (CE) is the annual average requirements of a 400 kg cow, which gestates and raises a calf until weaning at 6 months of age, weighing 160 kg, including the forage consumed by the calf. ** Low annual stocking rate means that forage production is greater than livestock forage consumption. Moderate annual stocking rate implies that forage production is similar to livestock forage consumption. High annual stocking rate means that livestock forage consumption is greater than ecosystem forage production (i.e., exceeds carrying capacity).

).

2.3. Remote Sensing Data

For RS data, we utilized a monthly time series of freely available Sentinel-2 (S2) images, ensuring the closest possible alignment with the field data collection period (2020). Sentinel-2 imagery (European Space Agency, ESA, Paris, France) was obtained from the Copernicus Data Space repository (https://browser.dataspace.copernicus.eu, accessed on 14 December 2024). The Sentinel-2 constellation, comprising Sentinel-2A (S2A) and Sentinel-2B (S2B), is equipped with a multispectral instrument (MSI) that captures data in 13 spectral bands across the visible, near-infrared, and shortwave infrared regions of the electromagnetic spectrum. Sentinel-2 satellites provide high and medium spatial resolutions (10 m, 20 m, and 60 m) and have a high revisit frequency of five days at the equator [60].

Since the study area spans two Sentinel-2 grid tiles (T20JKL, T20HKK), we downloaded monthly S2 images for each tile (12 images per tile, totaling 24 images) at processing level 2A (Bottom-Of-Atmosphere, BOA). These images are geometrically orthorectified, co-registered, and atmospherically corrected to surface reflectance by ESA using the Sen2cor processor (version 2.11.0, European Space Agency, Paris, France) [61]. This processor applies atmospheric correction, terrain and cirrus correction, and scene classification to Sentinel-2 Level 1C (Top-Of-Atmosphere, TOA) imagery [61].

Only S2 images with low cloud coverage (<10%) were selected (S1, Table S1.1), and none of the 49 forest sample sites were affected by cloud cover in any month. All selected images were acquired at the same time (14:16–14:17), with sun zenith angles ranging from 24.361° to 61.536°. The Sen2cor processor applies bilinear interpolation using Lambert’s reflectance law to minimize errors caused by shadows affecting spectral responses (S1, Table S1.1) [61]. Temporal alignment between RS and field data was ensured, as the measured variables remain stable over short periods in the Arid Chaco [62]. These variables typically do not change significantly within a few years unless extreme events such as wildfires or deforestation occur, none of which were recorded during the study period.

We used a subset of Sentinel-2 bands to calculate spectral indices, including bands at 10 m spatial resolution (e.g., blue: 490 nm, green: 560 nm, red: 665 nm, and near-infrared [NIR]: 842 nm) and bands at 20 m spatial resolution (e.g., red edge [RE]: 705 nm, Shortwave Infrared [SWIR]: 1610 nm and 2190 nm). All bands were resampled to a 20 m spatial resolution using the nearest neighbor method.

2.4. Field Data

At each sampling site (Figure 1C), a 2500 m² plot (250 m × 10 m) was established to collect ecological variables relevant to forest composition and disturbance analysis [17,58]. A central straight-line transect of 250 m intersected the plot, with an additional perpendicular transect of 20 m placed at the beginning of the plot [63].

Vegetation and bare soil cover were recorded using the point-intercept method [64] at 1-m intervals along the 250-m transect [63,64]. The ratio of vegetation cover to bare soil is a key indicator of Chaco vegetation integrity, where mature forests exhibit high vegetation cover, while disturbed areas are characterized by an increase in bare soil [20,49].

Additionally, we recorded vegetation structure parameters related to forest composition and degradation [17,50]. Vertical space was divided into four layers: one herbaceous layer and three woody layers (low < 2 m, medium 2–8 m, and high > 8 m). Every 5 m, we measured the maximum height of species recorded in each vegetation layer and counted the number of seedlings (0–30 cm height) within a 2 m² subplot [58,63].

On the perpendicular 20-m transect, we collected a composite litter sample, consisting of 10 sub-samples (each 40 cm²) taken every 2 m. Finally, within the 2500 m² plot (250 m × 10 m), we recorded tree species density (for individuals > 5 cm in diameter) and measured their diameter at breast height (DBH) (

Table 2. Field sampling scheme and forest variables measured (adapted from [29,58,63]).

Field Sampling Scheme	Forest Variable
Central straight line (transect) 250 m	-Woody sp. cover (total; for each species and for each vegetation layer: low < 2 m, medium 2–8 m, and high >8 m) and the derived horizontal heterogeneity index.
	-Bare soil cover (m).
	-Maximum height of the woody species (m; measurement carried out every 5 m, and the derived vertical heterogeneity index.
Plot woody species seedling density (2 m2; 50 plots along the 250 m transect).	-Woody species seedling density (seedlings*ha−1).
Initial straight line (transect) 20 m	-Composite litter sample (comprising 10 sub-samples taken every 2 m).
Plot 2500 m2	-Tree density (trees*ha−1).
	-Basal area (m2*ha−1).

).

The structural and compositional variables of woody vegetation (e.g., cover and height by species, litter characteristics) are commonly used descriptors of the main vegetation types in the Chaco region [20,53], as well as for assessing their structural integrity and degree of alteration [17,29,58,65].

The structural and compositional variables measured during fieldwork (2017–2019), such as tree height, basal area, and woody species cover, remain relatively stable over short timescales in arid ecosystems like the Arid Chaco [62]. This stability ensures a reliable temporal alignment between the field data and the selected RS images from 2020. Moreover, these variables typically do not undergo significant changes within a single year, unless extreme events such as wildfires or severe deforestation occur—none of which were recorded during the study period.

2.5. Data Processing

RS Spectral Indices

For each month, we calculated a series of 22 spectral indices useful for remote sensing (RS)-based ecosystem descriptions (for the complete list of calculated indices (see S1, Table S1.2) within the 49 forest sample sites. The analyzed spectral indices serve as indicators of photosynthetic activity and biomass, canopy water content and moisture, water stress, soil characteristics, chlorophyll and carotenoid content, and fire-induced stress.

Since many spectral indices exhibit high correlation with one another, we conducted a Variance Inflation Factor (VIF) analysis to eliminate redundant variables, ensuring that the retained indices were statistically independent. Accordingly, we selected three RS variables with VIF values lower than 5 for further analysis [66]. These selected indices (

Table 3. Spectral indices describing forest characteristics derived from Sentinel-2 imagery, retained for further analysis due to low multicollinearity (VIF < 5).

Acronym	Name	Formula	Proxy	Reference
MCARISent	Modified Chlorophyll Absorption in Reflectance Index	$\left(\left(RED EDGE-RED\right)-0.2\left(RED EDGE-GREEN\right)\right)\ast RED EDGE/RED$	Leaf chlorophyll content	[67]
NDWIGao	Normalized Difference Water Index	${\displaystyle \frac{NIR-SWIR}{NIR+SWIR}}$	Leaf water content	[68]
BI2	Brightness Index 2	$\sqrt{{\displaystyle \frac{{RED}^2+{GREEN}^2+{NIR}^2}{3}}}$	Presence of bare surfaces	[69]

) include the following: Modified Chlorophyll Absorption in Reflectance Index (MCARI_Sent)—an indicator of leaf chlorophyll content [67]; Normalized Difference Water Index (NDWI_Gao)—an indicator of leaf water content [68]; Brightness Index 2 (BI₂)—an indicator of surface presence and soil moisture [69].

The MCARI_Sent index combines red edge, red, and green bands to estimate the depth of chlorophyll absorption (Table 3). It is highly sensitive to leaf chlorophyll content and strongly correlates with the Leaf Area Index. Additionally, MCARI_Sent values are not affected by illumination conditions, soil background reflectance, or other non-photosynthetic materials [67].

The NDWI_Gao index estimates leaf water content through a normalized difference of NIR and SWIR bands (Table 3). It is sensitive to changes in canopy water content and useful for estimating water stress levels. The typical NDWI_Gao range for green vegetation is −0.1 to 0.4, with 0.4 indicating high leaf water content [68].

The BI₂ index is sensitive to soil brightness and quantifies bare surfaces by calculating the square root of brightness for each pixel (Table 3). It effectively differentiates bare soil from vegetation in arid environments. The BI₂ index ranges from 0 (indicating no bare surfaces) to higher positive values, which correspond to increasing percentages of bare surfaces [69].

We focused on monthly data, as they are particularly effective for tracking fine-scale ecological temporal variations, which are especially pronounced in subtropical semiarid systems such as the Arid Chaco, where vegetation undergoes strong seasonal fluctuations. The high temporal resolution of our analysis captures phenological differences across forest types and management regimes, offering a detailed view of forest condition dynamics. This approach is well-suited to detecting intra-annual variations, which are essential for assessing forest degradation processes that do not follow a uniform yearly pattern.

2.6. Data Analysis

2.6.1. Phenological Analysis of RS Variables

We utilized linear mixed models (LMMs) to capture the spatial and temporal variations of spectral indices across the eight levels of forest conservation (Table 1). The LMM approach was chosen for its ability to explicitly handle both fixed and random effects, making it well-suited for addressing temporal heterogeneity across semi-arid seasonal ecosystems. Specifically, to account for spatio-temporal heterogeneity in LMMs, we included forest type as a fixed effect, as it represents the primary factor influencing spectral variation and conservation status, while transect ID and month were included as random effects, accounting for variability at different spatial locations and across time. The interaction term (Transect ID × Month) was introduced to account for site-specific seasonal patterns, ensuring that temporal changes were analyzed within each site rather than being treated as uniform across the study area. This allowed us to explicitly model seasonal variation at each site, ensuring that different forests could exhibit distinct temporal dynamics.

As for seasonal and geospatial differences, the random effect for Transect ID captures the inherent spatial variability among sampling sites, while the random effect for Month accounts for differences across time points, allowing us to model phenological changes in spectral indices. The inclusion of these random effects improves model transparency and prevents pseudo-replication, ensuring that repeated measurements from the same sites are not treated as independent observations. The denominator degrees of freedom were calculated using the Kenward–Roger approach. The intra-class correlation coefficient (ICC) was presented as an indicator of the extent of random variance in the models. The quality of fit for mixed models was assessed using conditional and marginal determination coefficients (R²c and R²m) to quantify unbiased assessments of variance explained by fixed effects alone and by fixed plus random effects, respectively [70].

After analyzing the models, we computed the marginal mean and the accompanying 95% confidence interval for each spectral index. We then conducted a Tukey test with Bonferroni correction to determine whether there were statistically significant differences between the forest classes. We used the R programming language (version 4.3.2 [71]) for all analyses, implementing the following packages: ‘ggplot2’ (version 3.5.1) [72] for data visualization, ‘lme4’ [73] for mixed-effects modeling, ‘rstatix’ (version 0.7.2) [74] for statistical tests, ‘multcomp’ (version 1.4-28) [75] for multiple comparisons, ‘sjPlot’ (version 2.8.17) [76] for visualizing model outputs, and ‘usdm’ (version 2.1-7) [77] for addressing multicollinearity in ecological data.

2.6.2. Structural Alteration Index Across Forest Classes

Based on field data, we estimated the Structural Alteration Index (SAI) for each site (see [65,78] for details), a well-established proxy for ecosystem degradation previously tested in other forests in Argentina (e.g., [29,58]). The SAI is a multivariate index that summarizes the structural and compositional deviation of the sampled forest relative to a non-disturbed “reference” site, previously defined for the study area [65,78].

To maintain a balanced ratio between variables and the number of plots, we selected variables for SAI calculation in each environment (foothill or plain) using a multivariate approach. Specifically, we conducted a Principal Components Analysis (PCA) on site-by-structure-composition variable matrices, retaining those variables that contributed the most to data variability (e.g., explaining PCA1 and PCA2) to calculate SAI (see S2, Table S2.1; Figures S2.1 and S2.2).

The variables retained for SAI calculation included litter dry weight (ton·ha⁻¹), bare soil (%), Vertical Heterogeneity Index (VHI) [58], Horizontal Heterogeneity Index (HHI) [58,65], woody species seedling density (seedlings·ha⁻¹), tree density (trees·ha⁻¹), basal area (m²·ha⁻¹), and cover percentages of key species: G. decorticans, C. ehrenbergiana, P. praecox, N. flexuosa, and A. quebracho-blanco in the low (0–2 m) and medium (2–8 m) vegetation layers, as well as L. divaricata cover in both layers (see S2, Table S2.1; Figures S2.2 and S2.4). These variables describe the most important ecological characteristics of Arid Chaco Forest ecosystems, including vegetation physiognomy, resilience, soil composition, and key woody species [20,49].

Based on multivariate matrices (sites × selected structure and composition variables), SAI values for each site were computed using the Mahalanobis Distance (MD) [79] from the reference forest community (e.g., well-preserved forest). Sample sizes were n = 30 for plains and n = 19 for foothills. The Mahalanobis Distance accounts for variable correlation and weights each variable by its variance, making it applicable across different scales and independent of measurement units [79]. The Structural Alteration Index (SAI) was calculated as follows:

$$SAI=[\left(M{D}_i\times 100\right)\times (M{D}_{max}{)}^{-1}]SAI=[\left(M{D}_i\times 100\right)\times \left(M{D}_{max}{)}^{-1}\right]$$

(1)

where MDᵢ is the Mahalanobis Distance between the i-th site and the reference site, and MDₘₐₓ corresponds to the maximum registered MD value. Based on MD, SAI integrates structural and compositional variables, accounting for variable correlation and scaling differences. SAI captures both the deviation of degraded sites from reference conditions and the multidimensional nature of ecosystem alteration. The inclusion of physiognomic and species-specific variables indicative of degradation strengthens the ecological relevance of SAI in the context of the Arid Chaco Forest [78].

2.6.3. Exploring the Relationship Between RS Ecological Variables and Field-Measured SAI Values

We evaluated the relationship between the spectral indices and the Structural Alteration Index (SAI) using general linear models (GLMs). SAI was included as a predictor variable, while the average values (monthly and annual) of each spectral index (BI₂, MCARI_Sent, and NDWI_Gao) were included as response variables in the GLMs (13 GLMs per spectral index). The landscape (plains and foothills) was included as a categorical predictor. To assess the normality of residuals, we performed the Shapiro–Wilk normality test. Additionally, we calculated the Akaike Information Criterion (AIC) for each model and compared it with the AIC of a null model [80]. The analyses were conducted in R version 4.3.2 [71] using the packages ‘ggplot2’ [72], ‘sf’ [81], and ‘terra’ [82].

3. Results

3.1. Phenological Behavior of RS Variables Across Forest Classes

Spectral indices varied significantly between forest types with different conservation statuses (Figure 2,

Table 4. Linear mixed models (LMMs) of spectral indices by forest conservation classes. The fixed component of the model provides estimates, confidence intervals (CIs), and p-values for both the intercept and slope. The random effects are presented as explained variance (σ2), variance of the random intercept (τ00), and intra-class correlation coefficient (ICC). The goodness-of-fit is quantified using marginal and conditional R² values, which respectively capture the impact of fixed and fixed + random components in the model. Forest types and conservation levels (CL number) refer to mature forests on foothills (CL 1), mature forests on plains (CL 2), closed forests on foothills (CL 3), closed forests on plains (CL 4), low closed forests (CL 5), shrublands (CL 6), savannas (CL 7), and grasslands with trees (CL 8).

	BI2			MCARISent	NDWIGao
Predictors	Estimates	CI	Estimates	CI	Estimates	CI
Interc.	1925.15 ***	1859.50–1990.80	151.35 ***	142.83–159.87	0.19 ***	0.16–0.21
CL 2	−24.74	−163.98–114.51	0.49	−17.58–18.55	0.15 ***	0.09–0.21
CL 3	29.02	−67.60–125.65	1.88	−10.66–14.42	−0.00	−0.04–0.04
CL 4	203.88 ***	95.04–312.72	22.71 **	8.60–36.83	0.04	−0.00–0.09
CL 5	243.49 ***	157.91–329.07	15.57 **	4.47–26.67	0.03	−0.01–0.07
CL6	335.22 ***	249.63–420.81	19.02 ***	7.91–30.12	−0.06 **	−0.10–−0.02
CL 7	1516.11 ***	1419.49–1612.74	38.47 ***	25.94–51.01	−0.05 *	−0.09–−0.01
CL8	1510.42 ***	1401.56–1619.28	23.59 **	9.46–37.71	−0.10 ***	−0.14–−0.05
Random Effects
σ2	33,259.42			1316.25	0.00
τ00	93,595.66 ID_transect:Month			1561.27 ID_transect:Month	0.02 ID_transect:Month
ICC	0.74			0.54	0.91
N	49 ID_transect			49 ID_transect	49 ID_transect
	12 Month			12 Month	12 Month
Observ	33,157			33073	33120
Marg R2/Cond R2	0.706/0.923			0.047/0.564	0.126/0.923

* p < 0.05 ** p < 0.01 *** p < 0.001.

). Two out of three models, except MCARI_Sent, showed a conditional R² higher than 0.9. In some cases, the impact of the random component on the model was significant, as observed for MCARI_Sent and NDWI_Gao.

In contrast, for the Brightness Index (BI₂), the fixed component alone accounted for a significant portion of the variance (R²m = 0.706). Specifically, BI₂, which indicates bare soil, was significantly lower in mature forests, closed forests, low forests, and shrublands (CL 1–6) compared to savannas and grasslands with trees (CL 7–8).

MCARI_Sent values, which indicate leaf chlorophyll content, were significantly lower in mature forests of foothills (CL 1), mature forests of plains (CL 2), and closed forests (CL 3) compared to savannas (CL 7) and grasslands with trees (CL 8; Figure 3). Finally, NDWI_Gao values were significantly higher in mature forests on plains and in the low closed forest of Neltuma flexuosa (CL 2 and CL 5) (Figure 2).

3.2. Structural Alteration Index Results Across Forest Classes

The structure and composition varied across the sampled sites, with the most altered sites in the plains landscape (i.e., those with high SAI levels—CL8 and CL7) characterized by a low cover of Aspidosperma quebracho-blanco and Neltuma flexuosa (Figure S3.1). These sites also exhibited a simplified vertical structure, low litter dry weight, lower seedling density of woody species, and cover and basal area of Celtis ehrenbergiana, high percentages of bare soil and high cover of N. flexuosa in the low woody layer, and of Geoffroea decorticans in all layers. In contrast, the most preserved sites (i.e., those with low SAI levels) on plains landscape (mature forests of CL 2) and partially also those of the foothills (mature forests of CL 1) exhibited opposite values for these structural variables. Sites with intermediate structural alteration (i.e., those with medium SAI levels), mainly referable to low closed forest (CL 5) and shrublands (CL 6), were characterized by high cover of A. quebracho-blanco in the low woody layer, Larrea divaricata in the lower and middle woody layers, and Parkinsonia praecox in all layers as well. These sites also showed high horizontal heterogeneity and tree density (S2, Figures S2.1 and S2.2). Furthermore, the most altered sites in the foothills, such as grasslands with trees (CL8) and savannas (CL7) (i.e., with high SAI levels), showed low cover of A. quebracho-blanco and N. flexuosa in the high woody layer, as well as low vertical heterogeneity, litter dry weight, and seedling density of woody species. These sites had high percentages of bare soil and cover of G. decorticans and L. divaricata in the low woody layer, intermediate N. flexuosa cover in the low and middle woody layers, and moderate P. praecox cover in all layers. Conversely, the sites with low SAI referable to mature forests (CL1, CL2) and close forests (CL 3 and CL4) recorded opposite values for these variables. In contrast, the sites with intermediate SAI levels (mainly the low closed forest of CL 5 and shrublands CL 6) showed high N. flexuosa cover in the lower and middle layers, and P. praecox cover in all layers (S2, Figures S2.3 and S2.4).

3.3. Relationship Between RS Variables and Field-Measured SAI Values

The relationship between RS spectral indices (annual averages of BI₂, MCARI_Sent, and NDWI_Gao) and SAI varied across forests in both the plains and foothills landscapes. Mean annual BI₂ showed a significant response to SAI across sites in both landscapes. Overall, mean annual BI₂ values increased at higher SAI levels in both landscapes (Figure 3A,B;

Table 5. Linear regressions between the Structural Alteration Index (SAI—predictor) and the spectral indices (BI₂, MCARI_Sent, and NDWI_Gao—response variables) for plain and foothill landscapes. The columns include landscape type; spectral index used as the response variable; adjusted Adj R² (model fit); sigma (residual error); F-statistic and its p-value (model significance); log likelihood and Akaike Information Criterion (AIC) for model evaluation; AIC of the null model (baseline comparison); number of observations (nobs); and p-value from the Shapiro–Wilk test assessing residual normality.

Landscape	RS ind	R2 adj	Sigma	Statistic	p	logLik	AIC	AIC Null	Nobs	p-Value Shapiro-Wilk
Plains	BI2	0.71	349.35	36.99	1.80 × 10−8	−216.6	441.341	483	30	0.97
	MCARISent	0.15	17.75	6.25	0.018	−127.8	261.679	265	30	0.16
	NDWIGao	0.71	0.047	73.58	2.50 × 10−9	50.04	−94.083	−57	30	0.38
Foothills	BI2	0.54	87.4	22.25	0.0002	−110.8	227.68	241	19	0.43
	MCARISent	0.39	6.66	6.87	0.01	−61.35	130.7	138	19	0.36
	NDWIGao	0.52	0.03	20.15	0.00032	38.52	−71.04	−58	19	0.19

). However, values recorded in foothills sites were 25% lower than those in the plains. Specifically, in the plains landscape, sites with SAI > 80% had significantly higher mean annual BI₂ values compared to sites with SAI < 80% (Adj R² = 0.71, p < 0.0001; Figure 3A). Both mean monthly and annual BI₂ values exhibited the same response pattern, with a good overall fit (Adj R² > 0.65, p < 0.0001; S3, Figure S3.1; Table S3.1). In the foothills landscape, mean annual BI₂ also increased with SAI (Adj R² = 0.54, p = 0.0002; Figure 3B). Similarly, mean monthly BI₂ responded significantly to SAI for all months except November (Figure S3.1). The fit of mean monthly BI₂ values for the foothills sites was lower than that for the plains, ranging from Adj R² = 0.14 to 0.65 (S3, Table S3.1).

The mean annual MCARI_Sent values of sites in both landscapes increased with greater SAI levels (S3, Figure S3.1; Table S3.1). However, 60% of the sites in the plains landscape fell outside the 95% confidence interval (Adj R² = 0.15, p = 0.0185; Figure 3C). The response of the mean monthly MCARI_Sent to SAI varied across seasons. From February to May, MCARI_Sent increased with higher SAI levels (S3, Figure S3.1). From December to January, MCARI_Sent values showed no significant variation with SAI (S3, Figure S3.1; Table S3.1). In contrast, from September to November, MCARI_Sent values decreased as SAI increased (S3, Figure S3.1). In the foothills landscape, mean annual MCARI_Sent values also increased with higher SAI levels. The GLM for this landscape showed greater explanatory power compared to the plains, with only 40% of sites falling outside the 95% confidence interval (Adj R² = 0.39, p = 0.0070; Figure 3D). Similar to the plains, the response of monthly MCARI_Sent values to SAI varied across months. From January to August, MCARI_Sent increased with higher SAI levels (S3, Figure S3.1). From September to December, MCARI_Sent values remained constant at increasing SAI levels (S3, Table S3.1).

Mean annual values of NDWI_Gao decreased with increasing SAI levels in both landscapes (Figure 3E,F). In the plains landscape, sites with high SAI levels recorded significantly lower NDWI_Gao values (Adj R² = 0.71, p < 0.0001; Figure 3E). This trend was consistent across monthly GLMs (S3, Figure S3.1), except for September, where NDWI_Gao did not vary with SAI (S3, Table S3.1). In the foothills landscape, NDWI_Gao values also declined with increasing SAI levels (Adj R² = 0.52, p = 0.0003; Figure 3F). From January to April, mean monthly NDWI_Gao values remained relatively constant despite increasing SAI levels (S3, Figure S3.1, Table S3.1). However, from May to December, mean monthly NDWI_Gao values mirrored the annual trend, decreasing with higher SAI levels (S3, Figure S3.1).

4. Discussion

Our findings highlight the substantial potential of spectral indices derived from Sentinel-2 imagery for assessing forest characteristics. The analysis of remote sensing ecological variables in the Arid Chaco region, encompassing both plains and foothills landscapes, effectively captures differences in forest conservation status associated with distinct management regimes. These differences, identified through field data (e.g., SAI) and corroborated by existing literature, further support the effectiveness of integrating remote sensing and field-based approaches for ecosystem assessment.

4.1. RS Variables Across Sampled Sites

The set of remote sensing variables selected through Variance Inflation Factor (VIF) analysis serves as reliable surrogates for key ecological aspects in subtropical semi-arid ecosystems. The Brightness Index (BI₂), a strong proxy for bare soil (Escafadal), can indicate varying levels of forest cover and degradation [83]. The Modified Chlorophyll Absorption in Reflectance Index (MCARI_Sent; [84]), which reflects leaf chlorophyll content, may help distinguish herbaceous vegetation—characterized by short-term productivity peaks during the vegetative period—from woody vegetation, which retains its leaves for longer periods [85,86]. The Normalized Difference Water Index (NDWI_Gao; Gao), an indicator of leaf water content, can capture the seasonal behavior of species with different drought adaptations [37,40]. The effectiveness of each ecological spectral index (BI₂, MCARI_Sent, and NDWI_Gao) in differentiating various forest conservation and management categories in the Arid Chaco region varies across temporal periods and months (Figure 4).

Our results underscore the effectiveness of the Brightness Index (BI₂) in capturing vegetation phenology differences among forest classes with varying levels of resource utilization and extraction. In particular, BI₂ proved highly effective in differentiating savannas (CL 7) and grasslands with trees (CL 8), primarily used for intensified livestock raising, from formations with greater woody cover. Open formations, almost exclusively used for pastoral activities, exhibited significantly higher BI₂ values throughout the year, reflecting the prominence of bare soil surfaces. In contrast, natural formations with higher woody plant cover, such as mature forests on foothills and plains (CL 1–2), high forests (CL 3–4), and low forests and scrublands, all managed for multifunctional purposes such as timber extraction, grazing, aromatic herb collection, natural medicine production, and nature-based tourism [46], displayed significantly lower BI₂ values. These findings align with previous research conducted in the Chaco region and other semi-arid areas [37,83,87,88].

Although the application of the BI₂ index may have limitations in densely vegetated areas with closed canopies due to reduced bare soil visibility [37], it remains a valuable tool for monitoring dry forest lands, which cover over 10% of the world’s forest area [89]. Despite these challenges, integrating the Brightness Index (BI₂) with other remote sensing variables has proven to be a reliable approach for distinguishing vegetation formations with different structures and levels of use in arid and semi-arid regions. Indeed, areas with high BI₂ values correspond to forests that have undergone significant modifications in structure and species composition. This observation is corroborated by the Structural Alteration Index (SAI) calculated for the analyzed sites, particularly in areas subjected to intensive livestock use (e.g., CL 6–8, Table 1).

The Modified Chlorophyll Absorption in Reflectance Index (MCARI_Sent), which reflects leaf chlorophyll content, effectively differentiates vegetation formations with varying management regimes and conservation statuses (CL in Table 1). MCARI_Sent values were significantly higher in savannas (CL 7) and grasslands with trees (CL 8), which are characterized by leaf renewal and productivity peaks during the summer. This is linked to vegetative growth during the subtropical rainy season, which triggers a rapid response in herbaceous species-dominated ecosystems, such as savannas and grasslands in subtropical regions worldwide [90]. In contrast, mature forest formations in plains and foothills (CL 1 and CL 2), dominated by woody species adapted to drought conditions (e.g., small leaves with thick cuticles and low photosynthetic rates to reduce transpiration), exhibited lower MCARI_Sent values, consistent with previous studies in high-water-stress areas [91,92,93].

The observed pattern highlights a clear ecological distinction between mature forest formations on plains and foothill landscapes (e.g., CL 1 and 2), dominated by evergreen tree species adapted to water stress, such as Aspidosperma quebracho-blanco (drought-resistant; [94]), and deciduous species with deep root systems like Neltuma flexuosa (phrea-tophytic; [95]). In contrast, savannas and grasslands (e.g., CL 7 and 8), characterized by high grass and annual species cover, are more sensitive to variations in rainfall [96]. These findings demonstrate that significant ecological variations and management regimes within Chaco forests can be effectively detected through remote sensing, supporting the classification of analyzed landscapes into two primary groups: areas preserving natural forests for multifunctional management (e.g., within protected areas) and areas predominantly used for livestock grazing. Although this index exhibits weaker performance on an annual scale, certain months showed strong alignment across both landscape types, suggesting its potential as complementary information to the annual BI₂. This synergy enhances the ability of remote sensing ecological indices to distinguish different forest classes, thereby reinforcing structural discrimination capabilities. Previous studies have emphasized the value of using multiple remote sensing indices to improve accuracy in ecosystem characterization [97,98], underscoring the relevance of our results and their contribution to understanding structural dynamics in arid forests.

The Normalized Difference Water Index (NDWI_Gao), a key indicator of leaf water content, effectively distinguishes forest classes in the Arid Chaco. Moreover, it proves valuable in identifying forest classes dominated by tree species with different adaptations to drought conditions. NDWI_Gao reveals notably higher leaf water content in plains forests dominated by N. flexuosa (CL 2 and CL 5), which are phreatophytes with well-developed root systems capable of accessing deep groundwater. In contrast, lowland forests growing on weakly developed stony soils (CL 1 and CL 3), dominated by A. quebracho-blanco, which has a shallow root system, exhibit remarkably low leaf water content [91,92,93]. Some secondary formations, such as shrublands dominated by L. divaricata (CL 6), savannas (CL 7), and grasslands (CL 8), also display low leaf water content values. This low NDWI_Gao rate is likely due to the dominance of species in these classes that are more susceptible to the drought conditions characteristic of the Arid Chaco [85,86].

4.2. RS Ecological Variables Relationship with Field Integrity Values

The relationship between remote sensing spectral indices (annual averages of BI₂, MCARI_Sent, and NDWI_Gao) and the structural alteration of forests (SAI) varied across sites in both plains and foothills landscapes. The Brightness Index (BI₂) and the Modified Chlorophyll Absorption in Reflectance Index (MCARI_Sent) tend to increase in highly altered sites (e.g., with high SAI values), while NDWI_Gao exhibited the opposite trend (Figure 3).

The Brightness Index (BI₂) showed a significant response to an increasing Structural Alteration Index (SAI) in both annual averages (Figure 3A,B; Table 5) and monthly averages. The robustness of this response may stem from the sensitivity of both indices—one derived from remote sensing and the other from field data—to the presence and extent of bare soil. In the Chaco region, bare soil tends to cover larger areas in sites with low woody cover and high levels of silvopastoral and agricultural intervention (i.e., high SAI) [69,99,100]. This mosaic of woody vegetation patches intermingled with bare soil is characteristic of many dry forests worldwide, including the Chaco. In these ecosystems, degradation is often associated with an expansion of bare soil areas, and in severely degraded conditions, the landscape becomes dominated by bare soil [89,101]. The observed behavior of BI₂ and field-measured SAI across sites with different management pressures and conservation statuses effectively represents the forests of the Arid Chaco as highly heterogeneous, with woody vegetation distributed in a mosaic interspersed with bare soil and/or gaps [50,102]. Increasing intervention in these ecosystems (e.g., to enhance forage biomass) [103] inevitably leads to a simplification of vegetation structure and an increase in bare soil. The higher average BI₂ and SAI values in some sites within the plains landscape may be attributed to specific land uses, including megathermal pastures and crops, which result in greater seasonal exposure of bare soil [50]. These results highlight the potential of BI₂ as a reliable remote sensing indicator for differentiating sites with distinct conservation statuses within the Arid Chaco, allowing for assessments based on both annual averages and monthly data (Figure 3A,B).

The relationship between the Modified Chlorophyll Absorption in Reflectance Index (MCARI_Sent, annual average) and the Structural Alteration Index (SAI) showed moderate to weak correlations across sampled sites (Figure 3C,D; Table 5), likely due to phenolo-gical differences between the forest classes considered. The MCARI_Sent index, a proxy for leaf chlorophyll concentration and canopy density, increased with SAI values during the wettest months (February to May) but showed no significant relationship during the driest months (June to December). Sites dominated by herbaceous cover (e.g., savannas and grasslands with trees, CL 7–8) exhibited lower annual chlorophyll concentrations than woody communities (e.g., mature, closed forests where evergreen and deciduous species with diverse drought stress strategies coexist, CL 1–4). However, during periods of active growth (e.g., humid summers and autumns), the simplified structure of grasslands and savannas and the minimal interference from non-photosynthetic elements (e.g., trunks and branches) allowed them to exhibit higher MCARI_Sent values. Moreover, most crops and pastures receive supplemental irrigation from spring through early autumn [104].

Sites with high structural alteration values (SAI > 70%, CL 7–8), dominated by herbaceous species with shallow root systems highly dependent on monthly rainfall cycles, exhibited pronounced seasonal variability in MCARI_Sent values [50,98,105]. This variability likely reflects the rapid physiological response of herbaceous species to rainfall, which triggers chlorophyll renewal and photosynthetic activity during wet periods [44,106]. In contrast, woody communities (CL 1–4) exhibited more stable MCARI_Sent values due to their ability to access deeper water reserves through extensive root systems, decoupling their chlorophyll dynamics from surface rainfall patterns [91,107]. These contrasting strategies highlight the ecological mechanisms underlying seasonal differences in chlorophyll dynamics, with herbaceous vegetation relying on short-term water availability [108,109] and woody vegetation sustaining photosynthesis through long-term water reserves [106,110].

As a result, altered sites showed greater variability in biomass and chlorophyll content (e.g., MCARI_Sent delta) between wet and dry seasons, with a clear relationship between MCARI_Sent and SAI during wet periods (Figure 3C,D; Table 5). These results suggest that MCARI_Sent is an effective indicator for monitoring ecosystems during wet periods, when vegetation reaches peak cover, productivity, and photosynthetic activity.

The Normalized Difference Water Index (NDWI_Gao), a proxy for vegetation water content, showed a significant inverse relationship with forest structural alteration (SAI) across sites. Beyond its ability to differentiate forests with varying alteration levels—an ability also demonstrated by BI₂ and MCARI_Sent—NDWI_Gao enables more precise differentiation of sites with high woody cover, particularly where the SAI is less than 30% (Figure 3E,F). High NDWI_Gao values in well-preserved forests (SAI < 30%) effectively indicate the amount of biomass and water stored in the canopy, which is significantly greater compared to altered sites (SAI > 70%) [68]. NDWI_Gao proved particularly effective in distinguishing forest classes with varying levels of SAI during periods of greatest water deficit. This seasonal behavior may be attributed to the ability of woody species to access deep water layers, allowing them to maintain stable foliar water levels during dry periods and decouple their water stress from seasonal rainfall [40,111]. Additionally, during this period, perennial herbaceous vegetation is dry (or in seasonal dormancy), and there is no cover of annual species, enabling a focus on structural differences in woody vegetation.

Previous studies have shown that species of the genus Neltuma exhibit remarkable adaptations for accessing deep groundwater reserves in arid and semi-arid regions. They employ a stratified root system with vertical roots extending to depths of 17 m to tap into groundwater, stabilizing growth during low-rainfall periods and mitigating water stress by relying on stable aquifers. This strategy is particularly relevant in environments such as the Arid Chaco, where surface water is scarce [112,113]. The ability of NDWI_Gao to differentiate various woody communities’ likely stems from the fact that those with adult individuals capable of accessing the water table can decouple their growth from the rainy season, unlike shallow-rooted species that rely solely on surface water. These findings highlight NDWI_Gao as a valuable indicator for identifying well-preserved forest areas (i.e., low SAI) and an essential tool for monitoring forest ecosystem degradation in semi-arid regions such as the Arid Chaco.

Monitoring vegetation changes is particularly important in arid and semi-arid regions due to the sparse and sensitive nature of vegetation cover [114]. The correlation between remote sensing ecological indices (BI₂, MCARI_Sent, and NDWI_Gao) and field-measured forest structural alteration (SAI) suggests a strong potential for using these indices complementarily in forest monitoring in semi-arid landscapes. Seasonal monitoring is crucial in such ecosystems [114], and the phenological behavior of remote sensing ecological indices further strengthens their ability to distinguish different forest types throughout the year.

While BI₂ is applicable year-round, NDWI_Gao and MCARI_Sent exhibit seasonal sensitivity in detecting differences between communities dominated by various species and growth forms. These findings emphasize the value of integrating monthly remote sensing data to capture seasonal fluctuations in spectral indices, providing critical insights into vegetation responses to different management regimes.

5. Conclusions

Our findings highlight the significant potential of spectral indices derived from Sentinel-2 imagery for assessing forest characteristics and monitoring forest degradation in semi-arid landscapes such as the Arid Chaco. The Brightness Index (BI₂), Modified Chlorophyll Absorption in Reflectance Index (MCARI_Sent), and Normalized Difference Water Index (NDWI_Gao) proved effective in capturing differences in forest conservation status and structural alterations associated with varying management regimes and silvopastoral use.

BI₂ demonstrates robust year-round applicability, distinguishing open formations with high levels of bare soil from dense forests with greater woody cover. It is particularly effective in identifying sites with high structural alteration (e.g., intensive livestock use) due to its sensitivity to bare soil exposure. On the other hand, MCARI_Sent excels during wet periods, reflecting seasonal variations in chlorophyll content and canopy density. It effectively differentiates sites with varying levels of structural alteration and highlights phenological differences among forest types, particularly in areas dominated by herbaceous species. NDWI_Gao was especially useful in distinguishing forest types during dry periods, providing reliable insights into vegetation water content and differentiating forest types based on water-use strategies. Its sensitivity to leaf water content makes it particularly valuable for identifying well-conserved forest areas and vegetation adapted to drought conditions.

The seasonal variability of MCARI_Sent and NDWI_Gao complements the year-round applicability of BI₂, allowing for a nuanced understanding of vegetation structure, water stress adaptations, and conservation status. Together, these indices effectively differentiate sites with varying levels of forest degradation, reflecting differences in vegetation structure, water-use strategies, and management practices.

The integration of remote sensing (RS) indices with field-based measurements such as the Structural Alteration Index (SAI) further underscores the value of a hybrid approach for ecosystem monitoring. The strong correlations between RS indices and SAI values validate their utility as complementary tools for detecting and monitoring forest structural alteration and conservation status.

Monitoring vegetation changes is critical in all ecosystems, but particularly in arid and semi-arid regions, where vegetation cover is sparse and highly sensitive to environmental changes. Early detection of structural changes and assessment of their extent and severity at local and regional scales are therefore essential. In summary, the combined use of Sentinel-2 RS ecological variables (BI₂, MCARI_Sent, and NDWI_Gao) enhances our ability to monitor forest degradation, assess conservation efforts, and understand the ecological dynamics of semi-arid forests, making them valuable tools for sustainable land management and biodiversity conservation.

Despite the demonstrated effectiveness of these spectral indices, further efforts are needed to improve this phenology-based assessment. Incorporating Synthetic Aperture Radar (SAR) data or implementing spectral unmixing algorithms could enhance the accuracy and applicability of these analyses. Nevertheless, our results reveal distinct spectral differences between Arid Chaco forests under different management strategies, reinforcing the potential of RS-based approaches for ecosystem monitoring.

Our findings support the hypothesis that in highly seasonal subtropical forests, the phenological patterns of remote sensing ecological indices are influenced by ecosystem characteristics and the extent of alteration. This reinforces the crucial role of RS as a valuable proxy for modeling forest attributes. Incorporating a statistically sufficient number of field samples alongside monthly Sentinel-2 RS ecological indices would enhance the accuracy of large-scale assessments while significantly reducing costs compared to traditional field inventories.

From an applied perspective, the integration of field and RS data into a phenology-based forest conservation assessment provides relevant knowledge for forest monitoring and management. Considering that the data used are free and the procedures are standardized, the proposal could also be implemented by managing authorities to monitor the progress of the management models implemented. We encourage further studies to analyze increasingly larger areas to test the proposed approach and, at the same time, provide homogeneous information for forests in other biogeographical regions.