Multiscale Remote Sensing Data Integration for Gully Erosion Monitoring in Southern Brazil: Case Study

Abstract

The degradation and loss of arable soils pose significant challenges to global food security, requiring advanced mapping and monitoring techniques to improve soil and crop management. This study evaluates the integration of Unmanned Aerial Vehicles (UAVs) and orbital sensor data for monitoring and quantifying gullies with low-cost data. The research focuses on a gully in southern Brazil, utilizing high-spatial-resolution imagery to analyze its evolution over a 25-year period (2000–2024). Photointerpretation and manual delineation procedures were adopted to define gully shoulder lines, based on low-cost and multiple-spatial-resolution data from Google Earth Pro (GEP), UAVs and conventional aerial photographs. Planimetric, volumetric, climatic, and pedological parameters were assessed and evaluated over time. Field inspections supported our interpretations. The results show that gully expansion can be effectively mapped and monitored by combining high-spatial-resolution GEP data with aerial imagery. The gully area has increased by more than 50% over the past two decades, based on GEP data, which were corroborated by submeter-resolution UAV data. The findings indicate that the erosive process remains active, progressing toward the base level. These results provide critical insights for land managers, policymakers, and agricultural stakeholders to implement targeted soil recovery strategies and mitigate further land degradation.

1. Introduction

Advances in Brazilian agriculture are crucial to ensuring food security in various regions of the world [1], as Brazil’s agriculture sector feeds 10% of the world’s population. Furthermore, the quality of agricultural areas associated with soil degradation and loss has led to a reduction or even made crop production unfeasible, as well as, in some cases, the advancement of erosion processes [2,3,4]. Considering the five classes of soil vulnerability intensity to water erosion (mainly based on erodibility and slope), a higher concentration of vulnerable soils is observed in the western regions of the states of South Brazil and São Paulo [5]. Degraded soil results in a loss of quality and productive capacity, and, thus, its ecological, economic, and social functions become restricted.

Proper management practices and monitoring systems must be adopted to achieve sustainable production [6,7]. Understanding the dynamics of degraded soils is crucial, as their recovery is time-consuming and generates high costs [8,9]. For example, ravines and gullies are among the primary types of environmental degradation that require efforts for proper control, mitigation, and recovery [2,10,11]. However, monitoring these areas requires the use of techniques that allow for combining punctual field data with macro-scale approaches, such as remote sensing techniques [12,13] with proper methods for each scale [14].

Among the existing conventional techniques, the use of pegs and erosion pins stands out for measuring the retreat of margins and verifying erosion movement [15]. For this reason, it is relevant to seek more accurate methods for monitoring erosive processes [6,16]. In this scenario, geotechnologies become significant tools [3,17,18]. With the availability of high-spatial-resolution remote sensing data, such as products associated with the Google Earth Pro (GEP) platform (© Google LCC, Mountain View, CA, USA), the monitoring of gully areas can be carried out at different scales [19,20]. In parallel, Unmanned Aerial Vehicles (UAVs) have gained popularity and are finding niches in agricultural and environmental applications [16,21,22,23,24,25]. Despite some methodological proposals [20], the combined use of GEP, UAV, and conventional (manned aircraft) aerial photographs still requires more investigation. The challenge refers to the integration of multiple high-spatial-resolution data without precise ground control points (GCPs), across multiple dates, as well as the spectral capabilities of such sensors.

Considering the relevance of studies on erosive processes and the potential use of multiscale and multi-temporal remote sensing data, the study objective was to evaluate the potential of integrating airborne and orbital high-spatial-resolution data for gully monitoring and characterization in southern Brazil. Specifically, we focused on the following: (a) planimetric measurement of gully size; (b) volumetry of removed sediments; and (c) the time dynamics of these variables. A semi-automatic integration of UAV, conventional airborne data, and high-spatial-resolution data from Google Earth Pro (GEP) was implemented. As a case study, a gully located in the municipality of Palmeira das Missões, in the north of Rio Grande do Sul State, Brazil, was selected, considering the area’s representativeness in the region and the scarcity of studies in the area. This study focused on low-cost or freely available data [20].

2. Materials and Methods

This study’s main steps were field campaigns, the acquisition of UAV and conventional aerial data, and orbital data from the Google Earth Pro platform. The main methodological steps are presented in Figure 1.

2.1. Study Area

In Brazil, the processes of soil loss and degradation through sandification (similar to desertification [26,27,28]) are expanding, primarily due to inadequate anthropic land use in conjunction with its inherent fragility [26,27,28]. Guerra and Jorge [29] and Guerra et al. [30] highlight two erosion drivers in Brazil: the first is associated with natural conditions with emphasis on topography, soil texture, vegetation cover, and hydrological regime; the second considers anthropic actions related to the removal of vegetation cover for cultivation/pasture/urban expansion and specific interventions, such as slope cuts. The pressure exerted by anthropogenic activities on soils can be considered a significant contributor to the emergence of erosion processes [31]. Therefore, coupled effects must be considered on a local scale.

In the Rio Grande do Sul, Brazil’s southernmost state, the study of erosive processes has focused on the western region due to the natural fragility of its soils in the Pampa biome (grasslands) [32,33,34]. The western region of the state has the highest concentration of gullies, which mainly extend through the municipality of Cacequi over the Piramboia Formation. Several studies have been conducted to analyze these processes, their conditioning factors and types of evolution [35,36].

In the northern region of Rio Grande do Sul, the Tupanciretã formation extends across several municipalities and significantly contributes to the formation of gullies due to its association with erosion processes during its formation. Unlike in the west of the state, processes in this geological formation are less frequent but generate intense gullies [37,38]. Although research into gullies in the state has been ongoing for several decades, the application of modern technologies such as UAV is recent and could enhance our understanding of these processes. In this region, the occurrence of gullies has been a significant environmental problem that needs to be mitigated. This region represents a significant commodity as a production region. Interestingly, various erosive processes can be identified in the Palmeira das Missões, whose economic base is primarily livestock and agriculture. Despite this, the nature of the origin and evolution of these areas remain under discussion [26,27,28,29,30,31,32,33,34,35,36].

A regional representative gully located in the municipality of Palmeira das Missões, Rio Grande do Sul State, Brazil, was selected due to the availability of airborne, UAV, and GEP data, which allowed for evaluating the dynamics of the erosive process (Figure 2). The selected gully limits are located near the state road designated as RS-508. The base of the gully (deposition area) remains a source of construction sand extraction, which has implications for the silting of headwater streams. The surrounding area consists of agricultural land, and according to Sangiovo [39], the process began more than 40 years ago. The region’s relief is gently undulating to flat. Several other erosive processes can be found in the neighborhood, indicating the region’s fragility.

The region is characterized by a humid subtropical climate, as classified by the Köppen–Geiger climate classification: Cfa [40,41]. According to EMBRAPA [42], the predominant soils in the municipality of Palmeira das Missões are Dystrophic Red Latosol + Eutrophic Red Nitosols; Dystrophic Red Latosol; Dystrophic Red Latosol + Eutrophic Red Nitosols + Eutrophic Haplic Nitosols; and Eutrophic Litholic Neosols + Orthic Ebanic Vertisols + Eutrophic Haplic Cambisols. In the gully area, there is a predominance of Dystrophic Red Latosol. The region is characterized by cultivated commodities (soybeans, maize, and wheat) and livestock. Annual crops are usually associated with monocultures such as soybeans, corn, and wheat, in line with the state’s agricultural production incentive policy [43,44]. Based on Ruppel et al. [45] and Menegotto et al. [46], the region is located on the Tupaciretã formation.

The municipality of Palmeira das Missões is located in a transition area of contact between two phytogeographic formations of Rio Grande do Sul, composed of a Deciduous Seasonal Forest and a Mixed Ombrophilous Forest [47,48].

2.2. Data Acquisition and Pre-Processing

2.2.1. Field Data

Two field campaigns, conducted on 18 October 2017 and 9 May 2023, were carried out to obtain UAV data. For this purpose, a Matrice 100 UAV with an X3 camera (16 MPixels) [49] and a Phantom 4 with a 12 MPixels RGB camera [50] were used. In both campaigns, autonomous planned flights and manual observation flights were conducted to acquire photographs and videos.

Complementary field activities were conducted to collect three soil samples from the topsoil of the gully, with the aim of determining the following physical parameters: soil texture and saturation (Figure 2). The determinations were made in the soil physics laboratory of the Federal University of Santa Maria—Frederico Westphalen Campus. The samples were collected at a depth of 0–10 cm. The collection site was cleaned with a hoe, and volumetric rings of 100 cm³ were placed on the bare soil. Using an iron extractor, the rings were fitted with a hammer, and soil samples were taken from three different locations in triplicate for soil density analysis. Deformed (mixed) samples were collected for texture analysis to complement the soil analysis. The soil analysis followed the Brazilian recommendation [51].

Figure 2. UAV RGB true-color composition of 2023 for the gully in the municipality of Palmeira das Missões, in the north of Rio Grande do Sul State, Brazil. Brazilian states are indicated in the top-right subfigure. Source: Vector data were acquired from IBGE [48]. Yellow dots indicate the soil sample collection points, at the border of the gully topsoils. Cross symbols indicate the geographic coordinates within the figure.

Figure 2. UAV RGB true-color composition of 2023 for the gully in the municipality of Palmeira das Missões, in the north of Rio Grande do Sul State, Brazil. Brazilian states are indicated in the top-right subfigure. Source: Vector data were acquired from IBGE [48]. Yellow dots indicate the soil sample collection points, at the border of the gully topsoils. Cross symbols indicate the geographic coordinates within the figure.

2.2.2. UAV Data

Planned flights were defined using the Pix4D^(©) 1.3.1 capture application, with 80% along and across the overlay. The photographs were taken at nadir (90°) with automatic sensitivity control (gedISO). The 2017 flights were conducted at altitudes of 80 and 120 m above the highest point in the region (the takeoff point). For 2023, flights were conducted at heights varying from 60 to 120 m. Manual flights were conducted at varied heights and geometries to highlight the “ridges” and “walls” that form the features of interest within the gully and to facilitate the post-interpretation.

For the processing of planimetric images, Agisoft Metashape software [52] was used, implementing Structure from Motion (SfM) algorithms [53,54]. All processing steps were performed using the highest-quality settings available. Thus, sequential processing was adopted: (a) alignment of photographs; (b) dense point cloud; (c) Digital Surface Model (DEM); (d) composition of an ortho-mosaic. In the first step, the number of keys and tie points was limited to 40,000 and 4000, respectively. In the second step, dense cloud creation, a mild depth filtering process was applied using ultra-high setting. In this step, we also calculated a confidence band (95%). For DEM, the point cloud was used with interpolation enabled. For the year 2021, we evaluated the vegetation classification algorithm and generated a DEM based only on the ground points. The ortho-mosaic was created using the DEM as the surface model, and hole filling was enabled. The ground sample distance (GSD) of the acquisitions ranged from 2.50 cm to 7.20 cm, depending on the chosen height. The creation of 360 photographs and videos was the purpose behind the flying of the manual flies.

Additionally, data filtering tests were conducted to attempt to remove vegetation from the DEM using Metashape’s standard parameters. Finally, the data were exported to GIS 3.40 software (QGIS [55]) for the quantification of the gully area. Volumetry was calculated within Agisoft Metashape software [52] to quantify the soil loss over time.

The Agisoft Metashape software used the “best fit” model to delineate cross-sectional profiles and calculate volumes. The profiles were drawn in defined sectors to enable the analysis of the temporal evolution of the erosive process.

2.2.3. Conventional Aerial Data

The airborne data were obtained from the collection of the Map Library of the 1 Centro de Geoinformação (1 CGEO) of the Brazilian Army’s Directory of Geographic Service (DSG). A photograph associated with the CONDOR topographic map (CONDOR Sheet—SH-22-V-A-II-1—Index map code 2916-1), dated 1995 at 1:50,000 scale (standard error < 15 m, according to the PEC), was used. The photograph was georeferenced using large topographic and anthropic landmarks as a reference; however, due to georeferencing constraints, a direct match with the other data (UAV and GEP) was not possible. Thus, only planimetric measurements were extracted from this product. Complementarily, an old version of the topographic chart was used (CONDOR Sheet—SH-22-V-A-II-1—Index map code 2916-1) prepared by the Brazilian Directory of Geographic Service (DSG).

2.2.4. Satellite Data—Google Earth Pro (GEP)

For this study, aerial and orbital images were acquired (Figure 2). The orbital data were obtained from the Google Earth Pro platform (GEP, 2024, Google LLC). To standardize the study area, a bounding rectangle was created using the high-spatial-resolution image from 2024, as a reference obtained via the Quick Map Service plugin within QGIS software [55]. To extract the other high-spatial-resolution images from the GEP platform, a vector rectangle was created and exported to Google Earth Pro. Subsequently, the images were exported at the highest-possible spatial resolution (8192 × 4712 pixels) in JPG format. They were then georeferenced based on the 2024 image corners and reference points. These images underwent adjustments to enhance the contrast, brightness, and saturation of the scenes. Despite the georeferencing process, it was not possible to provide positional error data due to the absence of permanent ground control points (GCPs) on all dates. Still, another study found GEP images geolocation average error of 4.38 m, ranging from 2 to 10 m according to the region [56]. Thus, high-spatial-resolution images of GEP were obtained for the years 2000, 2004, 2012, and 2024.

2.2.5. Ancillary Data

Additionally, precipitation data for the years under evaluation were obtained to investigate any possible qualitative relationship between rainfall and the gully evolution. A direct cross-correlation was not possible because of the absence of compatible temporal UAV coverage. This information was obtained from the automatic weather station of Palmeira das Missões [57]. To increase temporal coverage, data from the GPM GPM_3IMERGM v07 product (Time Series, Area-Averaged of Merged Satellite-Gauge Precipitation Eimate—Final Run-in mm/month) [58] were used. Furthermore, due to gaps in the time series, monthly reanalysis data from the Climate Hazards Group InfraRed Precipitation with station data were also integrated [59].

2.3. Data Analysis Procedures

Initially, the gully expansion was visually interpreted and mapped, supported by orbital, airborne, and UAV images. This step involves identifying important elements within the gully, such as shape, soil strata, land use, and land cover (anthropic use). Thus, color compositions of the RGB orbital images were created, complemented with field photographs and videos.

Because of the absence of GCPs, a careful manual procedure was adopted to define the gully shoulder lines for each date. The same specialist carried out the process using the GEP true-color RGB composition. Additionally, some pseudo-invariant targets were used as references (e.g., trees and fences). The lower stream section was used as a reference based on the triangulation of the invariant target position. Due to the absence of GCPs throughout the entire period for precise georeferencing of the images, the gully boundaries were used cautiously, respecting the possible displacements inherent to the positioning method of orbital and aerial images (e.g., viewing and illumination geometry). This careful manual procedure allowed for a reduction in viewing and illumination effects on the gully shoulder lines, resulting in a more precise area and volume calculation.

Regression equations (linear, 2nd-order polynomial, and logistic models) were developed based on the gully area to identify the evolution pattern from 1995 to 2024. All available images (orbital, aerial, and UAV) were used. With the result, the R² value and the RMSE were calculated. Considering the potential for projection into the past, the 95% confidence band and the entire residual analysis of the model were calculated. Additionally, gully area information was obtained from the original photography of the 1995 topographic chart. Finally, the first topographic chart allowed us to evaluate the relative position of the road in 1976.

Another approach involved mapping the position of the highway near the gully (Figure 2), which had been sequentially relocated over the years. Thus, it was possible to identify the gully’s superficial expansion rate and better understand the dynamics of the processes involved (anthropic, superficial, and subsurface drivers). Despite having a topographic map from the 1970s, the mapping of the highway positions was not considered due to the large difference in mapping scale (1:50,000 on the map) and positioning errors (georeferencing without well-defined control points). These data were used as Supplementary Materials (Figure S1).

To obtain volumetric estimates of the lost soil, only 2017 and 2023, for which UAV data were available, were considered. Thus, based on the defined boundaries for each year, the volumetry was calculated for both the entire gully and its specific sectors.

Finally, UAV data generated a DEM for both 2017 and 2023. The relative altimetry information was cross-referenced with the gully shoulder line information. Furthermore, this information served as a basis for discussing the potential advancement of the erosive process toward the subsurface water level. Thus, areas with a higher probability of being affected by the process were identified. These results were discussed in consideration of the hydrodynamic process trends associated with the evolution of the erosive process.

To understand the context of the gully dynamics of the groundwater base level, longitudinal and transverse profiles of the RS-508 highway were developed. For this purpose, the DEM generated from the 2017 UAV images was utilized. The model refers to relative altitudes (without GCP correction). The profile locations aimed to highlight the highway and its potential contribution as an intensifier of the erosive process under study. The DEM of 2023 could not be used because of the presence of vegetation within the gully. Such conditions did not allow for the calculation of practical ground-level surface models.

3. Results

3.1. Gully Planimetric Area from Multiple Remote Sensing Data

The erosive process under study began before the monitored period [39]. In 2004 (Figure 3a), the analyzed fragment encompassed an area of 15,582 m². In less than a decade, the gully area increased by 4139 m² (+26.56%, reaching a total of 19,721 m²) (Figure 3b). Also, based on the orbital images of the GEP, in 2024, the gully covered approximately 23,734 m² (+20.35%) (Figure 3c). Thus, in 20 years, there was an increase of 8152 m² (52.32%).

Despite the expansion of the area, some sectors of the gully were more active than others (Figure 3d). Thus, using the high-spatial-resolution GEP images, there is a greater expansion to the gully’s north, west, and southern sectors. Also, in some sectors, the evolution of erosive activity was less intense (southwest, east, and northwest). This dynamic can be observed in Figure 4 (the first and last UAV images acquired in this study), where one of the southern sectors of the gully is highlighted. It is verified that only in the last years (2017 to 2024) was the increase in the erosive process lower in relation to the erosion in the previous years, which may be associated with positioning errors between the images (co-registration of the images) or the temporal stabilization of the erosive process.

The RS-508 was relocated thrice between 2004 and 2024 due to erosive advances (Figure 5). Among the direct impacts of the erosive process, the relocation of the road that crosses the watershed’s highest elevation stands out (Figure 6). If the position of the RS-508 in 1976 is considered, based on the topographic map at the 1:50,000 scale, the road would cross in the middle of the gully. Due to the difference in scale in relation to the spatial data considered in this study, a topographic map was added as Supplementary Figure S1.

By mapping the gully’s size (planimetric), a linear regression model was developed to adjust the gully’s area over nearly 30 years (Figure 7 and Supplementary Figure S2). The gully’s evolution pattern was linear (R² = 0.98 and RMSE = 736 m²). Other nonlinear models were tested (polynomial and logistic), which led to slight improvements in the fit and a small increase in adjustment (R² = 0.99). However, such models predicted gully erosion beginning after 1980. This disagrees with reports from informal interviews conducted with the study area’s surrounding community, who have informed us that the gully is much older. However, the exact date is still unknown. Therefore, the linear model was adopted thereafter. The boundaries of the confidence band (blue) and prediction band are shown below.

The backward projection of gully evolution creates an indicative date of origin of the erosive process, assuming a linear evolution. The results suggest that the beginning was centered in 1962, with a margin of error of ~±6 years (Figure 8). It should be noted that this process is old, with an unknown initial date. Still, it was visible in 1995, based on the grayscale aerial image (Figure 8), which is the oldest aerial image accessible in this study (Figure 8).

3.2. Volumetry and Advanced Lines of Gully Erosion Processes

Understanding erosive processes requires evaluating rates in conjunction with information on gully dynamics. The average monthly rainfall data from 2000 to 2024 were calculated, showing that October is the rainiest month, while February and August tend to present lower rates (Figure 9a). The monthly accumulated rainfall was generated, covering the UAV and GEP availability data. Rainfall data indicate monthly averages exceeding 150 mm (Figure 9b). The high standard deviation is associated with the irregular rainfall distribution, either due to the absence of measurements or due to stormy months (Figure 9b). Over the 25 years analyzed, several monthly precipitation peaks were identified; however, a cross-correlation with gully area and soil loss was not possible due to the lack of dense image availability. Considering the satellite-derived precipitation images, between 2023 and 2024, the monthly accumulated rainfall exceeded 500 mm.

Considering that extreme rainfall events can intensify the erosive process, it is worth noting that several precipitation peaks occurred before each high-spatial-resolution image was obtained (Figure 9b). It should also be noted that for some months, rainfall data were not obtained, possibly due to failures of the GPM GPM_3IMERGM v07 product, even in the gauge station data.

Although the gully area measurements were measured individually based on manual edge delineation, the visual inspection shows that in specific regions of the gully, significant differences were found between the area measured by the UAV (Figure 10a) and the GEP imagery (Figure 10b), in a short time interval (11.3 months). This was considered in the analysis and discussion. Comparing the UAV data (Figure 10a) to the 2024 GEP data (Figure 10b), it is apparent that the much finer detail that the UAV camera can capture increases the certainty of the gully’s shoulder line delineation.

Considering the sector indicated in Figure 10a, the gully had grown to 6249.7 m³ by 2017, based on an estimation of the volume lost, measured using only UAV data. Making the same measurement in 2023, the volume lost was approximately 7078.3 m³. This relationship indicates that in the evaluated sector, the gully increased by 828.6 m³ over a six-year period, equivalent to a 13.28% increase. It is essential to note that erosion has occurred unevenly among the branches.

The volume of growth (as an increase in void or loss of mass) was calculated across the entire gully using UAV data. The loss rate from 2017 to 2023 (114,092.1 m³ and 108,616.4 m³, respectively) decreased by approximately 5475.7 m³ (4.8%), which is lower than the rates identified in some sectors.

The gully presents lines of advancement towards the area of greater elevation to the north and west (Figure 11). It was verified that two small parts of the gully had already crossed the RS-508 road and coincided with the watershed’s ridge. To the west (Figure 11), the area is covered by vegetation (typical of the region). It has not advanced over the years, despite being surrounded by agricultural plots (smallest white arrow direction). On the other hand, the area to the east (gully) has been continuously advancing towards the west (as indicated by the longest white arrow).

The advancement of the gully from east to west, as indicated by the yellow arrow in Figure 12, reaching the highway (RS-508) reached the surface watershed ridge, as illustrated by the contour lines with relative elevation levels (Figure 12). Observing the variation in the longitudinal altitude of the road, the unevenness leads to large slope indices (Figure 13a). However, when analyzing the cross-sectional profile connecting the two drainage areas, it is verified that the base of the gully is at the same relative altitude as the drainage area on the other side of the RS-508 (Figure 13b).

Considering the topsoil in the ridges (Figure 2 and Figure 12), the analysis of the soil samples revealed an average texture of 34% clay, 56% sand, and 10% silt, characterizing the soil as Sandy Clay Loam. As for the structure, averages of 0.30 m³m³ of micropores, 0.08 m³m³ of macropores, and 1.64 g/cm³ for density were obtained, and total porosity was estimated at 0.38. Such results can be associated with the Tupaciretã formation, which is susceptible to erosion processes [45,46].

The evaluation of the longitudinal (Figure 13a) and transverse (Figure 13b) altimetric profiles of the relative elevation of the road showed smaller longitudinal variations along the road (common on Brazilian unpaved roads) and greater amplitudes in the transversal profile. The results show that on both sides of the road, there is a similar water base level of the spring in each watershed (Figure 11 and Figure 13). Additionally, the longitudinal profile highlights the presence of an elevated valley, which facilitates water flow in the roadside gullies up to the outlet.

4. Discussion

4.1. Planimetric and Volumetric Variables and Gully Evolution

A gully located in southern Brazil was selected to evaluate the potential of combining multi-source and multi-temporal field, UAV, conventional aerial, and Google Earth Pro (GEP) orbital data to assess the advances of erosive processes and their delineation [60]. Since the study utilized low-cost or freely available data, a relatively low-cost protocol is proposed, which may facilitate its adoption by the scientific community. The type of data source (aerial, UAV, or GEP) demonstrates different strengths. UAV data proved more effective for delineating shoulder lines, consistent with the findings of Zhang et al. [20]. GEP data offers significant potential for retrieving historical imagery with high spatial resolution, while UAV data excels in capturing fine details and providing accurate erosion assessments. However, for GEP data, gully features smaller than five meters may be less accurately represented [20]. Overall, a good agreement was observed between UAV and GEP data, aligning with the results reported by Borelli et al. [61].

An almost linear increase of more than 50% in the gully area was observed, although it is not the usual [62]. This pattern can be associated with the period of data availability and is consistent with the results presented by other authors [63,64]. This pattern enables reasonable forecasting of gully evolution and optimizes management practices [2,11], reducing costs [8,9].

The evolution of the gully is associated with the type of soil in the region, which is characterized as sandy, dry, and acidic [33,63], despite the limited number of samples. Anthropic actions have likely favored the expansion of the erosion process [17,20,30,31]. Downstream of the gully, construction sand is manually extracted. This process may have been initiated by ancient locals, which has contributed to the gully’s formation in association with the area’s natural fragility [39]. The road also intensifies the physical processes acting on the soil structure and influences the water flow [17].

Results presented by Sangiovo [39] indicate that the process began before the 1970s, in concordance with the proposed linear model (Figure 7). However, logistic and second-order polynomials indicated a turning point after 1980. Caution is necessary for the use of the model, given the lack of validation of the model, as well as the linearity of the proposed model [62]. When evaluating the topographic chart of the 1970s, the spring is located at the most downstream point of the study area, and no evidence of erosive processes is presented, given the 1:50,000 scale of the official mapping (Supplementary Figure S1). The relative stagnation in the evolution of the gully at certain points in the time series may be related to the degree of disturbance specific to anthropogenic actions, the level of soil fragility, and extreme rainfall events.

UAV and GEP data effectively monitored the gully area and the depth of cross-sectional profiles and sectors, even without the use of control points [61,65]. However, the effectiveness of the volume soil loss estimates with the UAV DTM data depended on the land cover (low vegetation within the gully). In areas without vegetation cover, the volumes of eroded soil can be estimated confidently. When the volume of soil lost from the gully was estimated for the entire region, several problems arose, leading to unexpected results, including a reduction in the volume of soil lost. This can be partially explained by the relatively low positional accuracy of our UAV data. Future studies should consider using a more accurate georeferencing approach, either by using Ground Control Points (GCPs) or survey-grade UAVs capable of generating photogrammetric products with sub-decimetre GSD and positional accuracies.

The soil loss between 2017 and 2023 was estimated at 1934.7 tons per year. The expectedly low soil loss value of 4.8% indicates that the method for estimating volumetry with UAV data has some limitations associated with vegetation removal in the DEM. These limitations can be attributed to the emergence of small-sized vegetation (bushes) within the gully from 2017 to 2023, thereby reducing the internal volume of the gully [66]. Methods for removing vegetation were tested, but due to their small size, they were not efficient and required further tests and adjustments [67]. Environments with higher spectral contrast between vegetation and substrate, as well as those with sparse vegetation, tend to facilitate this procedure [68].

Additionally, problems were identified with the geometry of the photographs selected by the SfM algorithm, which varied as a function of solar illumination. In this case, orthogonal lighting is desirable for UAV acquisitions. Finally, errors during the manual border delimitation can propagate into the volume estimates.

4.2. Geomorphological Analysis of the Gully

The trend of the gully advancing from east to west (Figure 12 and Figure 13) shows an ongoing process that will reach the water base level. The presence of the road (RS-508, Figure 6) is an important anthropogenic factor in accelerating the erosive process in association with extreme rainfall events (Figure 10). The greater depth of the central furrows suggests a significant movement of solid sediments, possibly due to rainwater runoff. Some sectors of the gully experience energy reduction, resulting in a deposit of sediments within the gully. The erosion process advancements are facilitated by the topsoil characteristics (Sandy Clay Loam soil) in association with the ridge position between two watersheds. A detailed visual inspection of the images (Figure 11) reveals slump features on the erosion slope, as well as the presence of ducts, indicating a process of recent evolution. The results represent a cost-effective and scalable approach to overcoming the limitations of traditional methodologies [6,16]. Such a workflow, based on manual edge delineation and automatic image processing, can be implemented and adopted for monitoring gullies in southern Brazil.

From the perspective of modeling the geomorphological process, obtaining piezometric measurements at the ridge near RS-508 can aid in understanding the position of the groundwater divide. In addition, intensifying soil characterization studies is fundamental for the future implementation of methods for controlling and containing erosive processes supported by mechanical or bioengineering methods [69].

4.3. General Constraints of the Study

Among the limitations of the process applied in the study is the need to improve the processes of co-registration of high-spatial-resolution images from multiple sensors using GCPs or image registration algorithms. A process of aligning photographs of different dates, followed by the separation of dates for generating dense cloud, DEM, and orthomosaic, has been proposed in the literature [68,69]; however, it was not effective in the present study. The results suggest an advancement over traditional methods despite their ease of installation and affordable value; however, some measurements can be uncertain, and there are not many studies that prove their accuracy [14,70]. Also, morphometric index-based approaches can be adopted and are contingent upon the quality of the input data [71].

Accurate terrain modelling was also challenging and sensitive to the filtering method for ground point extraction and DEM generation [72]. Even with sparse and small vegetation, the ground point algorithm could not adequately remove the vegetation, which impacted the estimated volume of soil loss. This problem could be even more challenging in gullies with dense vegetation, as reported in other regions of southern Brazil where eucalyptus is cultivated within the gully. Such a scenario creates a homogeneous surface that reduces the identification of matching points as well as ground point discrimination [6,16,69,72,73].

5. Conclusions

The combined use of high-spatial-resolution data of GEP and UAVs proved effective for mapping and monitoring the expansion of the erosive process in southern Brazil. GEP-based estimates indicated more than a 50% increase in the gully area over 25 years, which was corroborated by submetric UAV spatial resolution data.

UAV-derived 3D modeling enabled the precise delineation of gully shoulder lines and boundary boundaries, as well as the estimation of soil loss, although dense vegetation within the gully complicated volumetric assessments.

The erosive process exhibits a linear east-to-west progression toward the base level, suggesting potential stabilization upon reaching the watershed shoulder lines. These findings underscore the need for targeted mitigation strategies to curb further soil degradation and support land rehabilitation efforts.