Soil and Water Bioengineering in Fire-Prone Lands: Detecting Erosive Areas Using RUSLE and Remote Sensing Methods

Abstract

Soil and water bioengineering (SWBE) measures in fire-prone areas are essential for erosion mitigation, revegetation, as well as protection of settlements against inundations and landslides. This study’s aim was to detect erosive areas at the basin scale for SWBE implementation in pre- and post-fire conditions based on a wildfire event in 2019 in southern Ecuador. The Revised Universal Soil Loss Equation (RUSLE) was used in combination with earth observation data to detect the fire-induced change in erosion behavior by adapting the cover management factor (C-factor). To understand the spatial accuracy of the predicted erosion-prone areas, high-resolution data from an Unmanned Aerial Vehicle (UAV) served for comparison and visual interpretation at the sub-basin level. As a result, the mean erosion at the basin was estimated to be 4.08 t ha⁻¹ yr⁻¹ in pre-fire conditions and 4.06 t ha⁻¹ yr⁻¹ in post-fire conditions. The decrease of 0.44% is due to the high autonomous vegetation recovery capacity of grassland in the first post-fire year. Extreme values increased by a factor of 4 in post-fire conditions, indicating the importance of post-fire erosion measures such as SWBE in vulnerable areas. The correct spatial location of highly erosive areas detected by the RUSLE was successfully verified by the UAV data. This confirms the effectivity of combining the RUSLE with very-high-resolution data in identifying areas of high erosion, suggesting potential scalability to other fire-prone regions.

1. Introduction

The global annual burned area is estimated to be 5.9 ± 0.5% of ice-free land, and it shows a slightly decreasing trend of 1.21% per year between 2001 and 2020 [1]. According to the UN-Report “Wildfires in Latin America” [2], the total affected area by wildfires in the continent averaged 33 million ha per year between 2010 and 2019, whereby there was an almost equal distribution between the land cover types forest, cropland and grass/shrubland. The effects of wildfires are various and reach from impacts on local ecosystem functions [1] to the release of plant-fixed carbon into the atmosphere [3], presenting a driving factor for climate change. Further, fire severity, damage to vegetation, as well as the changes in chemical or physical soil properties influence the erosion behavior of watersheds [4]. Generally, in sloping lands of mountains, water erosion and landslides are prominent threats that increase when slopes have been impacted by fire and/or overgrazing [5]. With changes to erosion in post-fire conditions, surface waterways can fill with sediment, leading to narrower transverse sections and therefore to a higher exposure of settlements to flooding events in the case of high precipitation [6]. Generally, erosion assessment has been modeled by numerous studies. However, only 1.4% of these studies include an assessment of wildfire effects [7]. The estimation and prediction of soil loss derived from fire events at basins is important to understanding if the implementation of measures, such as soil and water bioengineering [8,9,10,11,12], for erosion mitigation, as well as the protection of settlements is needed. With the present paper, the adaptation of existing methodology for erosion prediction shall be tested against its capability to predict suitable areas for the implementation of SWBE measures for erosion mitigation. The Revised Universal Soil Loss Equation (RUSLE) serves for the spatial prediction of erosive areas in post-fire conditions in an area in the temperate Andes of Ecuador. Its evaluation regarding the implementation of SWBE measures in these predicted, high erosive areas is carried out using optical, high-resolution data from an Unmanned Aerial Vehicle (UAV).

Generally, various publications exist treating, e.g., the effects of wildfires on physical-chemical soil properties [13] or vegetation recovery [14]. For the simulation of post-fire erosion approaches based on, e.g., the Universal Soil Loss Equation (USLE) [15], the modified USLE (MUSLE) [16], the revised USLE (RUSLE) [4], or the Soil and Water Assessment Tool (SWAT) [16] were used, where the input parameters rely partly on publicly available and generalized data. Hydrological models, such as the HEC-HMS, can also be used to explore relations between erosion-deposition events in watersheds and rainfall-induced hydrological processes [6]. A further approach is the use of process-based models [17], representing an alternative to the mentioned models. However, process-based models need detailed parametrization and site-specific data, making it difficult to apply on data-scarce areas. Erosion estimates for the whole continent of South America are rare and derive mostly from global, interpolated databases (e.g., GloSEM [18]) which lack accuracy when used at the basin or local scale. In Ecuador, the annual average number of wildfires between 2009 and 2019 was 404, affecting an area of 40,000 ha [2]. Some studies have been conducted estimating erosion behavior, e.g., at the watershed scale [19,20] and the regional scale [21]. Further, the effects of erosion on urban planning and social dynamics were estimated in peripheral areas [22]. However, official, and open-access government data regarding erosion are not available. Since 2021, a project is being developed aiming to provide spatio-temporal monitoring of soil erosion for Ecuador using satellite data [23].

Within this research, a pre- and post-fire erosion study on a wildfire area in the temperate Andes (Loja, Canton Quilanga, Ecuador) was conducted at the basin scale. The fire event appeared in 2019 during the dry season and affected 8000 ha of grass and shrubland in the southern mountainous Sierra region. For the present study, the El Saco basin has been chosen, as no post-fire measures for erosion mitigation have been implemented within the first post-fire year. To understand the erosion behavior of basins in post-fire conditions, the RUSLE [24] was used, as it is easily applicable globally and widely used. It delivers an assumed annual erosion value (A) based on the product of various input parameters: rainfall erosivity (R), soil erodibility (K), topography expressed by slope length and steepness (LS), the (plant-)cover management factor (C), as well as support practices (P) [24,25]. Depending on the scale of the area of interest, information to develop these input parameters is gained, if available, from local time series data, such as long-term mean annual precipitation, or global, spatially interpolated databases. The RUSLE has been used for post-fire erosion estimation, e.g., in the Mediterranean area [4,26]. However, with the increasing importance of remote sensing and the rising availability of open-access earth observation data, there were new options for the calculation of input parameters for the RUSLE [27]. This fact leads further to a significant advantage for wildfire investigation, as one main issue is the representation of the fire-induced change. In particular, satellite data enable the retrospective interpretation of the area in question under pre-fire conditions. The frequent acquisition time of Sentinel 2 (S2) data at an area of interest is 5 days [28], permitting the use of multi-temporal data, as well as the comparison of different time slots and change detection. As post-fire vegetation recovery plays a major role in erosion behavior, the elaboration of vegetation indices [14,29,30,31] contains valuable information regarding the interpretation of vegetation development, enabling the monitoring of the vegetation layer in post-fire conditions and therefore showing the ability of autonomous recovery [14]. This information is important, as plants directly influence the intensity of soil erosion. The combination of an established erosion estimation method, such as the RUSLE, with the relatively new approach of using multi-temporal, S2-derived NDVI values for C-factor calculation contributes to a better understanding of basin-scaled post-fire erosion behavior while containing valuable information for the spatial distribution of SWBE measures.

Therefore, the objectives of the present research were:

• To estimate water-induced pre- and post-fire soil erosion in the temperate Andes using RUSLE and S2 data;
• To locate areas of high erosion for potential implementation of SWBE measures;
• To compare these located areas with optical, high-resolution UAV data.

2. Materials and Methods

2.1. Soil and Water Bioengineering in Fire-Prone Areas

SWBE as a nature-based solution for civil and hydraulic engineering uses biological components (plants, roots, cuttings, wooden logs, etc.) to protect slopes, riverbanks, or coasts [9,32,33]. It can be implemented in fire-prone areas before and after a fire. While preventive measures are applied before a possible fire to avoid its spread, post-fire measures serve to protect settlements and infrastructure securing unstable and erosive slopes [34,35]. When planning and implementing fire protection, erosion protection and reforestation measures, priority is given to those areas that have been most damaged and therefore pose the greatest risk to public safety from a hydrogeological point of view [36]. The main objective after vegetation fires is to mitigate the enhanced hydrological and erosive response in the area [37] and to prevent long-term soil erosion and land degradation. Reduced resistance from burned vegetation means a higher rate of precipitation runoff compared to pre-fire conditions. Implementing SWBE measures to mitigate erosion after a fire event offers several advantages, as most of the material used can derive from the fire area itself or the surrounding area, depending on site conditions. Basically, two categories of SWBE measures are distinguished in fire-prone areas: (a) establishing a vegetation layer and (b) reducing the erosive energy of the runoff by barriers that slow down the runoff and store eroded sediment on slopes [38]. The vegetation layer can be established using various measures such as dry seeding, hydroseeding or mulch seeding. In addition, seeding mats, sod slabs or topsoil can be applied to the degraded site. The application of cuttings can be used to fix structures and to break through hydrophobic layers, thereby increasing the infiltration of water into the soil. A common method for reducing the erosive energy of the runoff is the establishment of contour-felled logs (Figure 1a) or vegetation fascines. In addition, gullies can be secured with tree spurs, inert branches or greened trenches to retain sediment. Pile walls (Figure 1b), crib walls, palisades, sills or weirs are used as transverse structures to stabilize channels and reduce sediment yield. Drainage systems filled with stones or inert fascines can be used for targeted drainage, as well as drainage of slope water.

2.2. Study Site

The investigated El Saco basin lies in the temperate zone (Cfb, Köppen-Geiger Classification) of the southern Andes in Ecuador (canton Quilanga/Loja). Its landscape is characterized by grass- and shrubland, as well as some forest patches (Figure 2). The presence of non-native tree species (eucalyptus and pine) influences the water balance of the area, as they tend to dry the soil, which favors the spread of wildfires. Coffee production, farming, and pastureland are the main components of the cultivated landscape. Annual precipitation is about 1100 mm whereby the wildfire risk increases from June to November. The months from December to May are characterized by short and heavy rainfall events [39]. Between 2012 and 2023, four major wildfires have been reported in the surroundings of Quilanga: in 2012 (Parroquia Fundochamba, sector Collingora, Quilanga); in 2016 (Parroquia Fundochamba, sector Guaguasaco, Quilanga); the investigated event in 2019 (Figure 2 and Figure 3) [40]; and the last wildfire in 2023 (Parroquia El Ingenio, sector San Antonio de las Aradas). The preparation of farmland using fire is a common tool in the area in question, causing major problems when not controlled sufficiently in the months with low precipitation. The investigated fire event appeared in September 2019, lasted for two weeks, and affected more than 8000 ha.

The El Saco Basin (Figure 3) with an area of 984 ha, and an altitude of 1520 m a.s.l. at the outlet to 2680 m a.s.l. at the highest point has been chosen to be representative for the wildfire area, as no post-fire measures regarding erosion protection were carried out. It includes one main and two micro basins, and the length of the mainstream is 4.83 km. The wildfire in 2019 affected 783 ha of the El Saco basin which amounts to 80% of the area. For the UAV survey the sub-basin located in the southeast of the basin was chosen. It has an area of 86 ha, and its altitude reaches from 1920 m a.s.l. to 2544 m a.s.l.

2.3. Work Flow

To understand the erosion behavior of the El Saco basin, in a first step, three pairs of S2 scenes were chosen according to cloud coverage. Using the NDVI, each pair represents the vitality of the vegetation in different phenological stages throughout the pre-fire and the post-fire year. To implement the information on the plant’s vitality into the RUSLE, the following step included the scaling of the NDVIs according to the approach of Colman [41] and the adapted tropical C-factor relation. To estimate the erosion behavior of the El Saco basin, and further the change in soil loss in post-fire conditions, the RUSLE was applied using the elaborated C-factors and the erosion’s delta. In addition, using the selected multi-temporal NDVI values, a pre-fire mean C and a post-fire mean C were calculated and implemented in the RUSLE. Figure 4 demonstrates the workflow of the erosion estimation graphically. For the spatial localization of highly erosive areas at the basin, a threshold at 95% confidence interval has been set to the pre-fire mean erosion and the post-fire mean erosion (using the mean C-factor). The remaining, highly erosive areas were optically compared to the high-resolution data of a sub-basin derived from an UAV survey in 2021 and the potential implementation of SWBE measures for erosion mitigation in fire-prone areas was assessed.

2.4. Revised Universal Soil Loss Equation (RUSLE)

The RUSLE is an empirically based equation to estimate soil erosion induced by water. It is used by soil conservationists worldwide [4,24,25,42] and delivers an assumed annual erosion value (A) based on the product of the input parameters, which are derived from a large mass of field data. The RUSLE multiplies different spatially distributed input layers influencing erosion and estimates long-term average annual soil loss based on sheet and rill erosion in tons per hectare per year (t ha⁻¹ yr⁻¹) [24]. It derivates from the Universal Soil Loss Equation (USLE), which was developed in the 1950s to calculate sheet and rill erosion. In 1987 the Agricultural Research Service (ARS) and the Soil Conservation Service (SCS) revised the USLE with some cooperators to the RUSLE [24]. From the literature, many methods are available for the calculation of the input parameters using open-access data. This fact is an advantage in data-scarce areas where erosion risk can be assessed anyway. When using the RUSLE as a prediction model, awareness regarding its limitations is important, as numerous calculation methods of the input factors are proposed by various authors, which leads further to a certain variability of the estimated erosion value results [43]. However, this limitation is not an indication of the overall performance of the RUSLE as it adequately represents the first-order effects of the factors that affect sheet and rill erosion [24,25]. Further, the model is used globally to estimate soil erosion, resulting in a high number of values to compare in the literature.

The RUSLE is written as follows:

$$A=R\times K\times LS\times P\times C$$

(1)

where R = rainfall erosivity, K = soil erodibility, LS = topography expressed by slope length and steepness, P = support practices, and C = the (plant-)cover management factor.

At the wildfire area of the El Saco basin, different sources from open-access supply, field or laboratory work were used to prepare the spatial input variables used for the pre-and post-fire erosion estimation.

2.4.1. Rainfall Erosivity (R)

The R-factor (in MJ·mm·ha⁻¹·h⁻¹·yr⁻¹) was calculated from the average monthly cumulated rainfall. For its elaboration at the study site, existing data from the Catamayo watershed [44], which is located near the El Saco basin, was interpolated using the Inverse Distance Weighting (IDW) method [45] (Figure 5a).

2.4.2. Soil Erodibility (K)

For the calculation of the soil erodibility K (in t·ha·h·ha⁻¹·MJ⁻¹·mm⁻¹), the soil texture as the percentage (sand, silt, clay) and the organic carbon content as the percentage are used. At the El Saco basin, a field and laboratory study was undertaken in November 2019, one month after the fire event. Soil samples from 18 points within the basin were air dried, sieved to the fine earth part of 2 mm and analyzed. The hydrometer method [46] served to determine the particle size and therefore the physical proportions of soil particles, which is not influenced by the fire event. To obtain the spatial distribution of the sand, silt, and clay percentage at the basin, the respective values at the sample points were interpolated using the IDW method [45]. The organic carbon content was acquired from ISRIC World Soil Information [47]. For the calculation of the K-factor, the method suggested by Williams [48] was carried out. The K-factor is shown in Figure 5b.

2.4.3. Slope Length and Steepness (LS)

The slope length and steepness factor (LS—unitless) (Figure 5c) was derived by a Digital Elevation Model (DEM) with a resolution of 30 × 30 m² provided by the IRD (Institut de Recherche pour le Développement) [49]. For the calculation of the LS factor, the method by Desmet [50] was applied.

2.4.4. Support Practices (P)

The support practice factor P (unitless) was taken to be 1 for both areas in pre- and post-fire conditions (Figure 5d).

2.4.5. The (Plant-)Cover Management Factor (C)

As soil erosion is directly related to the presence of vegetation, the cover management factor C (unitless) is the most influencing variable when calculating the amount of erosion using the RUSLE [4,19]. This factor relates soil loss from land with a specific vegetation cover to soil loss that would result from clean-tilled, continuous fallow land. For the estimation of erosion changes in pre- and post-fire conditions this factor is the most important one, as the fire event influences the vegetation layer which leads to a higher C-value. In the literature, different approaches are described to adapt the C-factor to burned areas [51,52,53]. When calculating the NDVI-based C-factor in Europe, the relation after van der Knijff [54] is often used. However, this relation tends to overestimate the C-values and is not applicable to other regions in the world. Therefore, in 2014, Durigon et al. [55] developed an NDVI-based C-factor relation for tropical regions, which was adapted by Colman in 2018 as a tenfold systematic bias was found when comparing the remote sensing approach with experimental field studies [41]. The evaluation of the Colman approach against C-factor values obtained in field experiments was carried out by Almagro et al. in 2019 [41]. According to this study, the erosion estimation using the tropical C-factor values by Colman biased 13% to the measured sediment yield, which outperformed the values derived from the literature (20% bias), as well as the ones by van der Knijff (486% bias). Therefore, in the present study, the approach after Colman was applied for the C-factor estimation using multi-temporal S2 data in the pre- and the post-fire year (

Table 1. Summary of selected Sentinel-2 scenes in pre- and post-fire conditions; Granule: T17MPR.

S2 Pair	S2 Satellite	Date	Sun Zenith Angle	Sun Azimuth Angle
1	B	24 October 2018	20.85	113.03
2	A	17 May 2019	32.33	42.67
3	A	25 August 2019	28.14	58.03
Fire Event
1	B	18 November 2019	24.54	129.68
2	A	11 May 2020	31.46	44.37
3	A	9 August 2020	31.50	50.73

). The NDVIs for the different dates in pre- and post-fire conditions are calculated according to Equation (2):

$$NDVI={\displaystyle \frac{NIR-RED}{NIR+RED}}={\displaystyle \frac{B8-B4}{B8+B4}}$$

(2)

where NIR and B8 is the near-infrared band and RED and B4 is the red band of S2.

The C-factor values are inferred using Equation (3):

$$C=0.1\left({\displaystyle \frac{-NDVI+1}{2}}\right)$$

(3)

Satellite data have been chosen according to the availability of cloud-free S2 scenes, as well as phenological development throughout a year:

• At the beginning of the rainy season in November (S2 Pair 1),
• At the end of the rainy season in May (S2 Pair 2), and
• During the dry season in August (S2 Pair 3).

Table 1 shows the selection of the S2 scenes (atmospherically corrected level 2A products) in pre- and post-fire conditions.

Table 2. Statistical summary of the NDVI-derived C-factor values at the El Saco basin in pre- and post-fire conditions.

C-Factor Values at El Saco Basin
S2 Pair	C—Date	Min	1st Qu.	Median	Mean	3rd Qu.	Max	St. Dev.
1	24 October 2018	0.0038	0.0252	0.0306	0.0288	0.0341	0.0488	0.0073
2	17 May 2019	0.0007	0.0113	0.0153	0.0151	0.0187	0.0462	0.0052
3	25 August 2019	0.0031	0.0235	0.0286	0.0278	0.0322	0.0483	0.0067
Fire Event
1	18 November 2019	0.0039	0.0274	0.0336	0.0317	0.0380	0.0542	0.0087
2	11 May 2020	0.0041	0.0128	0.0163	0.0167	0.0199	0.0504	0.0055
3	9 August 2020	0.0024	0.0193	0.0237	0.0230	0.0271	0.0504	0.0060

and Figure 6 present the statistical, as well as graphical summary of the NDVI-derived C-factor values at the El Saco basin at different dates. The statistical and graphical summaries of the NDVIs are shown in Appendix A (

Table A1. Statistical summary of the S2-derived NDVI values at the El Saco basin in pre- and post-fire conditions.

NDVI Values at El Saco Basin
S2 Pair	NDVI—Date	Min	1st Qu.	Median	Mean	3rd Qu.	Max	St. Dev.
1	24 October 2018	0.0239	0.3188	0.3880	0.4241	0.4961	0.9238	0.1453
2	17 May 2019	0.0758	0.6250	0.6931	0.6972	0.7740	0.9863	0.1048
3	25 August 2019	0.0344	0.3555	0.4271	0.4545	0.5296	0.9378	0.1336
Fire Event
1	18 November 2019	−0.0836	0.2404	0.3286	0.3670	0.4517	0.9216	0.1738
2	11 May 2020	−0.0073	0.6014	0.6736	0.6651	0.7447	0.9174	0.1105
3	9 August 2020	−0.0087	0.4570	0.5261	0.5393	0.6136	0.9521	0.1207

and Figure A1).

2.5. Spatial Definition of Highly Erosive Areas for Potential Implementation of SWBE

For the spatial definition of highly erosive areas, the multi-temporal, S2-derived C-factor was used for pre-fire and post-fire erosion modeling. Thresholds were set at the pre-fire and post-fire scenarios using the corresponding confidence interval of 95%.

2.6. Optical Validation of the Estimated Erosion Areas

For optical validation regarding the spatial distribution of the estimated erosion, high-resolution, RGB data have been acquired from a sub-basin of El Saco using an UAV (Figure 3). The survey was taken out on 9 October 2021, using Pix4DCapture (Pix4D SA) for flight planning in combination with a DJI Mavic 2 Pro (SZ DJI Technology Co., Ltd.), equipped with a Hasselblad L1D-20c camera (1′′ CMOS sensor, 20 Megapixels). To map the sub-basin, RGB images were taken by the UAV every 2 s at a flight height of 120 m related to the home point, while following a pre-defined flight plan across the slope. Longitudinal, as well as side overlap between adjacent images, were set to be 70%, resulting in 535 high-resolution images. For the photogrammetric elaboration and the following creation of the orthomosaic DJITerra was used. The ground sampling distance of the orthomosaic resulted in 6.9 cm.

For the optical validation of the calculated erosion models the high-resolution orthomosaic, as well as the 3D model of the sub-basin was examined against visible signs of erosion. Noticeable erosion areas were delimited with a polygon and compared to the estimated pre-fire, as well as post-fire raster containing the highly erosive areas.

3. Results

3.1. Erosion Estimation in Pre- and Post-Fire Conditions Using the RUSLE

When estimating the erosion at the El Saco basin using different S2 acquisition dates for the C-factor development, the fire event, as well as the phenological development of the vegetation layer influence the outcome. As shown in

Table 3. Comparison of the estimated erosion using the RUSLE with C-factor values derived from different NDVI dates in pre- and post-fire conditions at the El Saco basin, Ecuador.

Calculated Erosion [A in t ha−1 yr−1] Using C-Factors from Different S2 Dates
	S2 Pair	C-Date	Min	1st Qu.	Median	Mean	3rd Qu.	Max	St. Dev.
Pre-fire	1	24 October 2018	0.62	3.94	4.78	4.94	5.69	24.21	1.79
	2	17 May 2019	0.09	1.91	2.48	2.58	3.08	23.83	1.08
	3	25 August 2019	0.55	3.72	4.50	4.71	5.39	23.93	1.78
	4	multi-temp—mean C	0.61	3.24	3.94	4.08	4.68	21.49	1.48
Fire Event
Post-fire	1	18 November 2019	0.61	4.24	5.24	5.39	6.34	26.18	1.91
	2	11 May 2020	0.61	2.14	2.69	2.84	3.34	23.64	1.08
	3	9 August 2020	0.39	3.11	3.79	3.94	4.54	24.48	1.43
	4	multi-temp—mean C	0.61	3.28	3.94	4.06	4.66	23.40	1.34

, the highest mean value (5.39 t ha⁻¹ yr⁻¹), as well as the highest maximum value (26.18 t ha⁻¹ yr⁻¹) was estimated one month after the fire event on 18 November 2019. The lowest mean erosion value (2.58 t ha⁻¹ yr⁻¹) and the lowest minimum value (0.09 t ha⁻¹ yr⁻¹) were estimated in pre-fire conditions at the end of the rainy season on 17 May 2019. The comparative analysis between the S2 pairs in pre- and post-fire conditions shows an influence of the vegetation’s phenological stage on the calculated erosion value. S2 Pair 1, representing the beginning of the rainy season in November, estimates a mean erosion of 4.94 t ha⁻¹ yr⁻¹ in pre-fire conditions, as well as 5.39 t ha⁻¹ yr⁻¹ in post-fire conditions, indicating therefore a post-fire increase of 9.14%. The estimated mean erosion of S2 Pair 2 at the end of the rainy season indicates 2.58 t ha⁻¹ yr⁻¹ in pre-fire conditions and 2.84 t ha⁻¹ yr⁻¹ in post-fire conditions, resulting in erosion increase of 10.32%. However, S2 Pair 3, representing the dry season, shows different erosion values as in pre-fire conditions a higher mean value (4.71 t ha⁻¹ yr⁻¹) was predicted, resulting in a decrease of 16.26% in post-fire conditions (3.94 t ha⁻¹ yr⁻¹). The higher pre-fire mean erosion value is explained by the drought before the fire event, resulting in lower NDVI values as shown in Figure A1e. For the erosion estimation of Pair 4, the mean C-factor of all three chosen acquisition dates was used in pre- and post-fire conditions. While the pre-fire model estimates 4.08 t ha⁻¹ yr⁻¹, the post-fire calculation results in 4.06 t ha⁻¹ yr⁻¹, indicating, therefore, a slight erosion decrease of 0.44% (

Table 4. Estimated change in post-fire erosion using the RUSLE with C factor values derived from different NDVI dates in pre- and post-fire conditions at the El Saco basin, Ecuador.

S2 Pair	Date	Mean Delta [t ha−1 yr−1]	Post-Fire Change [%]
1	24 October 2018–18 November 2019	0.45	9.14
2	17 May 2019–11 May 2020	0.27	10.32
3	25 August 2019–9 August 2020	−0.77	−16.26
4	Post-fire mean C–Pre-fire mean C	−0.02	−0.44

). These results show that erosion estimation in post-fire conditions using an individual scene of satellite data for the representation of vegetation indicates therefore a current snapshot in time. The use of multi-temporal S2 data is an alternative to display the phenological change in time properly. Figure 7 shows the estimated erosion spatially.

3.2. Area of High Erosion for Potential Implementation of Soil and Water Bioengineering

To locate areas of high erosion for potential implementation of SWBE measures thresholds were set at a confidence interval of 95% at the El Saco basin in the pre-fire and the post-fire scenario (Pair 4, mean C). As a result, in pre-fire conditions values above 6.35 t ha⁻¹ yr⁻¹ were selected. In post-fire conditions the threshold was set at 6.15 t ha⁻¹ yr⁻¹. Figure 8a shows the highly erosive areas in pre-fire conditions between 6.35 t ha⁻¹ yr⁻¹ and 21.49 t ha⁻¹ yr⁻¹. The erosive areas in post-fire conditions were defined from 6.15 t ha⁻¹ yr⁻¹ to 23.40 t ha⁻¹ yr⁻¹ (Figure 8b). Both scenarios show a similar erosion pattern, where the highest values are located along the river courses.

At the sub-basin of El Saco, where the UAV survey was carried out, an area of 2.45 ha (2.83% of the sub-basin) was predicted to be higher than 6 t ha⁻¹ yr⁻¹ in pre-fire conditions (Figure 9a). The maximum erosion value was predicted to be 15.27 t ha⁻¹ yr⁻¹. In post-fire conditions 1.78 ha (2.05% of the sub-basin) counted values above 6 t ha⁻¹ yr⁻¹, whereby the maximum value was predicted to be 15.76 t ha⁻¹ yr⁻¹ (Figure 9b). Generally, the erosion model predicted two intense erosion areas next to the outlet, as well as the lower area of the sub-basin. While the number of erosion values above 6 t ha⁻¹ yr⁻¹ decreased due to increased vegetation development in the first post-fire year [14], the number of values indicating high erosion, increased. According to the model prediction, the fire event intensified; therefore, so did the erosion behavior in these two, highly erosive areas of the sub-basin. The number of values above 15 t ha⁻¹ yr⁻¹ increased by a factor of 4 in the first post-fire year.

3.3. Comparison of Located Areas with Optical, High-Resolution UAV Images

To verify the significance of the RUSLE regarding the spatial detection of highly erosive areas for the potential implementation of SWBE measures in fire-prone areas, the two developed raster products, predicting high erosion in pre- and post-fire conditions, were compared to the orthomosaic, as well as the 3D model derived from the high-resolution RGB images of the UAV survey (Figure 10).

When comparing the two developed erosion model rasters (pixel size 9 m) with the high-resolution UAV data (pixel size 6.9 cm), there was a difference in spatial accuracy. At the sub-basin, the models estimated most of the erosion along river sections and few areas on the slopes. Further, an erosion ditch was clearly detected. The models were able to detect highly erosive areas, which are potentially interesting for the implementation of SWBE measures. In general, densely vegetated areas were not defined as erosive zones. However, due to the lower resolution of the erosion model, some edges of vegetated riverbanks were classified with (lower) erosion (5–7 t ha⁻¹ yr⁻¹). At the highest predicted values (>15 t ha⁻¹ yr⁻¹) erosion areas could be clearly detected on the UAV-derived orthomosaic and the 3D model in the area of the outlet (Figure 11a,b), as well as the lower area of the sub-basin (Figure 11c,d). At the higher area, spatially larger erosion extents (>1500 m²) were detected correctly by the model, while smaller extents—as identified with the high-resolution UAV data—were not displayed (Figure 11e,f).

4. Discussion

The estimation of erosion in the fire-prone study site of the temperate Andes using the RUSLE and S2 data resulted in 4.08 t ha⁻¹ yr⁻¹ in pre-fire, and 4.06 t ha⁻¹ yr⁻¹ in post-fire conditions. For South America, the Global Soil Erosion Modelling platform (GloSEM) predicted an average soil erosion rate of 3.53 t ha⁻¹ yr⁻¹ in 2001 [51]. In Ecuador, this database estimated a mean of 5.56 t ha⁻¹ yr⁻¹ in 2001, as well as 5.32 t ha⁻¹ yr⁻¹ in 2012. These values are comparable to the estimated means derived from the present study at the El Saco basin. However, a study at the neighboring canton Macará showed higher erosion values when using the RUSLE for erosion estimation [56]. While the estimated erosion values were between 0 and 10 t ha⁻¹ yr⁻¹ in 40.5% of the canton, extreme values (>200 t ha⁻¹ yr⁻¹) were predicted at 15.8% of the area. A previous study at the El Saco basin [12], using S2 data and the relation according to van der Knijff [54] for the C-factor calculation estimated a mean erosion of 116.14 t ha⁻¹ yr⁻¹. As the mentioned NDVI–C-factor relation tends to overestimate erosion [41], this mean value can be considered as overrated. Future experimental field studies at the basin scale would give an insight into the accuracy of the NDVI-C-factor relation, as well as the values of the erosion estimation at the area. The slight decrease in the mean erosion (0.44%) in the first post-fire year is explained by the main vegetation type at the area, and its fast recovery. A previous study [14] analyzed the fire severity of the fire event in 2019, as well as the following post-fire vegetation recovery at the El Saco basin. It showed the fast recuperation of the grassland due to the presence of vegetation types with a short life cycle using vegetation indices from multi-temporal S2 data. The vegetations’ vitality in low severity and moderate high severity (according to the Relativized Burn Ratio—RBR [57]) appeared to be at the pre-fire level after one year. Further, comparing the NDVIs (Figure A1e,f) in the dry season shows the low vitality of vegetation a few weeks before the fire event. When selecting S2 data in the area in question, one challenge is the steady presence of clouds, leading—especially in rainy season—to a limited availability for time series of a certain extent. Calculating the C-factor from one satellite scene can lead to a biased result of estimated erosion, depending on the vegetations’ development at this particular moment. Generally, the choice of the calculation method for each input parameter of the RUSLE influences the outcome and therefore the predicted erosion. However, the main purpose of the present study was to locate areas of high erosion for potential implementation of SWBE measures at the El Saco basin. Therefore, the spatial distribution of the erosion prediction was the most important output. According to the interpretation of the high-resolution UAV data, the presented method was capable of detecting areas of high erosion, which supports the assumption that it is scalable and applicable in other (fire-prone) regions. As the UAV survey was undertaken in 2021 and compared with the predicted erosion in 2020, an uncertainty due to temporal difference must be considered. An UAV survey some weeks after the fire event could increase the number of detected erosive areas. The presence of low vegetation and its fast recovery at the basin can influence the optical detection of erosive areas and the phenological development must be considered when interpreting the data.

However, the ability to predict highly erosive areas as demonstrated in the presented study using the combination of the erosion model RUSLE and earth observation data, provides support to planning parties in accurately assessing the first effects of fire on a basin’s erosion behavior. This desktop solution allows municipalities, planners, or stakeholders to quickly evaluate the spatial needs for erosion control in fire-prone areas, as well as to identify suitable SWBE measures. The increasing application and availability of high-resolution UAV data for engineering purposes opens further perspectives and possibilities for SWBE as a nature-based solution, that is capable of diminishing erosion by implementing site-specific measures for vegetation layer development. The interpretation of the orthomosaic from the study site showed that erosion areas can be detected visually. However, in some cases (e.g., sheet erosion), the 3D model provides an even improved visibility for erosion recognition. As the used input factors for the erosion model are developed mainly from open-access data, the applicability of the presented method in other regions is given. The collection of site-specific UAV data is supported by the general increase in low-cost drones that provide high-quality optical data.

While the vegetation at the study site recovered to a great extent within the first post-fire year, extreme values increased by a factor of 4 in post-fire conditions. Areas that are generally prone to erosion can be influenced negatively by fire events, as existing vegetation, halting soil loss under normal circumstances, is damaged. The fast implementation of adequate SWBE measures in these areas within the first weeks after the fire event, can help to support vegetation recovery and prevent further erosion. Priority is given to areas that have been most damaged and pose the greatest risk to public safety from a hydrogeological point of view [36]. For the detection of these areas the proposed workflow is an adequate measure. The additional estimation of soil volumetric erosion using GPS and UAV data (optical or lasers can) [58] from different dates can be applied into the workflow of soil conservationists to understand the necessary dimensions of the implemented SWBE measures. While the presented method can provide important information for planning parties, small erosion areas were not detected by the RUSLE. Therefore, a site visit cannot be completely substituted before planning the measures. However, the presented framework can be used as a powerful tool for an initial erosion assessment and the first spatial decision as to where to implement SWBE measures in fire-prone areas.

5. Conclusions

This study aimed to estimate erosion in the fire-prone temperate Andes using the RUSLE and S2 data found pre- and post-fire erosion rates of 4.08 t ha⁻¹ yr⁻¹ and 4.06 t ha⁻¹ yr⁻¹, respectively. Despite the slight decrease in average post-fire erosion, likely due to the rapid recovery of the area’s dominant vegetation, high erosion values quadrupled, highlighting the vulnerability of the fire-affected regions. This study underlines the importance of rapid area detection with high erosion risk. To effectively and immediately control erosion, it is important to identify areas of high erosion within a catchment. This understanding is a precondition for the implementation of post-fire SWBE measures that aim to mitigate erosion and support vegetation recovery. By focusing the measures on the most urgent areas, their effectiveness can be significantly enhanced. As SWBE techniques are often applied after the occurrence of erosion, this study’s findings can support the adaptation of a more preventive approach for erosion protection. The research methodology combining the RUSLE with earth observation data, as well as high-resolution UAV data proved effective in identifying areas of high erosion, suggesting potential scalability to other fire-prone regions. However, limitations in accuracy may occur when using single satellite images instead of multi-temporal images for C-factor calculation. Challenges in detecting small erosion areas with the RUSLE make field visits necessary for comprehensive planning. Integrating volumetric erosion data from GPS and UAV can further refine SWBE strategies. Overall, the study approach provides a robust initial assessment tool for erosion management and supports the strategic implementation of nature-based erosion control measures in at-risk landscapes.