Prioritization of Watershed Restoration Areas Using Hydrologic, Environmental, and Socioeconomic Criteria: Model Development and Field Validation

Abstract

The selection of restoration areas in large river basins is a technical and socioeconomic challenge due to the environmental and socioeconomic complexities involved, particularly in data-scarce tropical basins. To address this challenge, multi-criteria models (MCMs) and multi-criteria decision analyses (MCDAs), coupled with GIS, are used by managers and decision-makers as integrated and objective prioritization alternatives. In the present study, an MCM was developed integrating hydrological, environmental (runoff, sediment yield, and degraded lands), and socioeconomic (stakeholder prioritization) variables to screen priority restoration areas in the Urucuia Basin, a large watershed in the Brazilian savanna, leading to the final selection of 2000 hectares for the Urucuia’s Proaguas Project. Runoff and erosion were estimated using the calibrated InVEST Seasonal Water Yield (SWY) and Sediment Delivery Ratio (SDR) models, respectively. Integrated with a Geographic Information System (GIS) and supported by off-the-shelf watershed data, the MCM identified 31,455 hectares of priority areas, spanning all basin landscapes. These areas were validated by ground-truthing and by a corresponding confusion matrix, as well as by a validation workshop with local stakeholders. The prioritization map had an accuracy of 90% and was consistent with local stakeholder knowledge and preferences. The method could be applied to similar basins after appropriate calibration and validation.

1. Introduction

Tropical watersheds are fundamental landscape units that integrate climatic, hydrological, ecological, and socioeconomic processes across multiple spatial scales [1]. However, the accelerated degradation of these systems—driven by land-use change, deforestation, agricultural expansion, urbanization, and climate change—compromises essential ecosystem services, including water regulation [2], sediment retention [3], nutrient cycling [4], and flood control [5], reducing watershed sustainability.

In tropical watersheds, the technical, financial, and human resources available for environmental restoration are limited relative to the extent of degraded areas [6]. Consequently, restoring all degraded sites within a watershed is often technically, ecologically, and economically unfeasible. Therefore, prioritizing restoration areas has become a major challenge in watershed management, restoration ecology, and landscape planning [7].

The prioritization process seeks to maximize the environmental return of restoration efforts by achieving the greatest possible benefit per unit of invested resource. This process requires balancing multiple and often competing objectives [8], including erosion and sediment control [9,10,11], biodiversity conservation, habitat connectivity [12], carbon sequestration, and climate change adaptation [13,14].

Several conceptual and methodological approaches have been developed to address this multi-objective and spatial prioritization problem. Among the most widely used are spatially explicit conservation planning and trade-off analysis frameworks [15], spatial multi-criteria decision analysis (MCDA) with or without the Analytic Hierarchy Process (AHP) [16], benefit–cost analysis [17], multi-criteria models [18], production frontier analysis [19], and process-based hydrological modeling coupled with optimization algorithms [20].

Multi-criteria decision analysis (MCDA) is a structured approach for supporting decisions by identifying relevant criteria, measuring the performance of alternatives against given criteria and constraints, and aggregating this information to compare or rank alternatives [21]. On the other hand, multi-criteria models (MCMs) are mathematical or computational models that represent, integrate, or concomitantly evaluate multiple criteria to characterize alternatives or scenarios or map key spatial units [18]. Both tools have been used for decades to support decision-making processes, with mixed results [22].

An extensive review and comparison of dozens of MCDA and MCM studies indicated that, although they utilize different approaches and techniques, the recommended course of action did not vary significantly with the model utilized [23]. In the past, prioritization approaches commonly relied on weighted linear combinations of biophysical variables, whereas recent frameworks integrate ecological, hydrological, socioeconomic, and feasibility criteria within a single spatial decision-support system [7].

More recently, high-resolution remote-sensing products and cloud platforms have substantially improved the spatial and temporal representation of land cover, land degradation, fire occurrence, hydrological processes, and landscape connectivity [24], enhancing prioritization analyses.

Another important advance was the incorporation of multiple restoration benefits, such as biodiversity conservation, carbon sequestration, water regulation, erosion control, and connectivity, while simultaneously considering restoration costs and opportunity costs [25,26]. Recent prioritization models also incorporate natural regeneration potential, ecological connectivity, climate refugia, hydrological impacts, and restoration feasibility, moving beyond simple degradation-based identification [7].

There has been increasing convergence toward integrated approaches that explicitly consider landscape spatial configuration and the concept of strategic restoration, defined as the selection of interventions and locations that generate greater environmental benefits than restoring the same area randomly or in less strategic locations [8].

Participatory weighting of criteria and scenario analysis are increasingly used to incorporate stakeholder preferences and assess uncertainty associated with criterion weights [7]. A recent systematic review identified 59 spatial criteria grouped into nine major categories, illustrating both the increasing sophistication and the methodological diversity of restoration-prioritization models [27].

Although GIS-based MCM/MCDAs provide a systematic framework for identifying priority areas, the resulting spatial rankings remain model-based representations dependent on the selected criteria, spatial datasets, normalization procedures, weighting assumptions [28,29], and parochialism [30]. Therefore, ground-truthing of representative priority sites is important for assessing whether modeled priority classes correspond to actual field conditions, increasing the reliability of the resulting prioritization for in situ management applications [31].

Criteria weighting is a controversial issue within MCDA and MCM modeling. Although it provides an explicit representation of the relative importance/trade-offs among criteria [32], criteria weights create a false impression of objectivity and stability, particularly when the weights are based on subjective judgments [33].

Considering these aspects, the objective of this study was to develop a multi-criteria prioritization model to screen restoration areas in the Urucuia River Basin, optimizing the hydrological services and providing a sufficiently large pool of candidate sites to support a further selection process, leading to the 2000-hectare restoration goal.

2. Materials and Methods

2.1. Study Area

The Urucuia River Basin, the focus of this study, comprises 14 municipalities, spanning an area of 25,835 km², previously delineated by the National Water Agency [34]. Basin topography is gently sloping (Figure 1 and Figure 2) and the dominant soils are Orthox, Ochrepts, and Psamments [35] (Figure 3). The basin’s natural vegetation, covering 35.1% of the basin area, comprises different variations in tropical open savannah, with the remaining land uses consisting of pastureland and rainfed and irrigated agriculture (Figure 4). No missing data existed in the land-use map. Basin mean annual precipitation is 1050 mm and mean annual streamflow at the basin outlet is 250 m³/s [36].

Figure 1. Urucuia Basin altimetry. Map resolution: 30 × 30 m. Source: prepared by the authors, using ArcGISPro3.6.0 (ESRI, Redlands, CA, USA).

Figure 2. Slope grade map of the Urucuia Basin. Map resolution: 30 × 30 m. Source: prepared by the authors, using ArcGISPro3.6.0 (ESRI, Redlands, CA, USA).

Figure 3. Urucuia Basin soils. Map resolution: 30 × 30 m. Source: prepared by the authors, using ArcGISPro3.6.0 (ESRI, CA, USA).

Figure 4. Urucuia Basin land use and cover. Map resolution: 30 × 30 m. Source: prepared by the authors, using ArcGISPro3.6.0 (ESRI, Redlands, CA, USA).

In the last 40 years, the Urucuia River Basin has experienced significant environmental impacts due to deforestation and inadequate land use and management [37]. The impacts include soil erosion and siltation of watercourses [38] and reduced infiltration and aquifer recharge [39], which affect the quantity and quality of water available for sanitation, irrigation and cattle watering, especially during the dry season.

In order to mitigate the hydrologic impacts and improve soil and biodiversity in the basin, ‘Proaguas Urucuia’, an agri-environmental restoration project, was launched in 2025, aiming to restore 2000 hectares of degraded areas in the basin through reforestation and soil conservation practices. One of the project requirements is that the candidate areas are prioritized according to objective environmental and socioeconomic criteria, avoiding political interference and parochialism [30].

2.2. Development of the Multi-Criteria Model for Prioritization of Restoration Areas

The development of the multi-criteria model (MCM) followed the philosophy of MCM conceptualization, trying to avoid common pitfalls, such as subjectivity and lack of transparency [40]. Hence, the first selected MCM criteria were the two key objectives of the ‘Proáguas Urucuia’ project, namely (i) the reduction in surface runoff [16] and (ii) erosion and sedimentation abatement [11], both being key hydrological services generated by restoration interventions [41].

The third MCM criterion consisted of the group of land-use classes eligible for restoration under the Urucuia Project, including pastureland, agricultural land, and non-vegetated areas, where restoration measures would be cost-effective [21].

The fourth MCM criterion was the degree of land degradation severity within basin municipalities [42]. The latter was applied to the MCM after the overlay (map algebra) of the three previous criteria in the GIS (ArcGIS Pro V.3.6.0^®), balancing environmental and social heterogeneity and improving project implementation capacity [43] (Figure 5).

Figure 5. Conceptual MCM for prioritization of restoration areas. Source: prepared by the authors, using MS Powerpoint V. 2604 (Microsoft, Seattle, WA, USA).

To allow the spatial overlay of the MCM criteria within the GIS, the quantitative and qualitative criteria datasets were previously standardized into common prioritization classes and scores [44]. Hence, quantitative variables (basin surface runoff and soil loss) were classified according to their statistical distributions, whereas qualitative variables (eligible land use and priority municipalities) were classified according to their restoration priority levels. Each criterion class was assigned a numerical score, as follows:

• Surface runoff: three classes (low = 1, medium = 2, high = 3);
• Soil loss: three classes (low = 1, medium = 2, high = 3);
• Land use: two classes (low = 1, high = 2);
• Municipal priority: four classes (low = 1, medium = 2, high = 3, very high = 4).

The four raster layers corresponding to the above criteria were subsequently combined by raster multiplication within the GIS environment [28]. To obtain the final standardized prioritization map, the four criteria and their corresponding scores were normalized in the GIS to a common scale, i.e., by the maximum product of criterion scores [44], producing four overall restoration-priority classes: low (1), medium (2), high (3), and very high (4). Because criterion normalization by the maximum score resulted in equal ranges, an equal-weight approach was maintained [1].

Since the proposed MCM aimed at the maximization of the hydrological services related to water quality and quantity, established by the Proaguas Project objectives, other potential environmental benefits, such as biodiversity improvement, effective land recovery, and connectivity with water receptors, were not explicitly assessed but were not totally excluded.

2.2.1. Surface Runoff Criterion

Annual surface runoff was estimated using the InVEST Seasonal Water Yield (SWY) model [45], with corresponding soil, land use, precipitation, and topography datasets. Simulations were performed using data from 2023 because it represented the mean basin climatology of the last decade (Figure 6) and was the latest year for which both land-use [46] and spatially distributed precipitation data [47] were simultaneously available.

Figure 6. Observed annual precipitation of decade 2014–2023, with the latter year representing the mean decadal precipitation (dotted line). Source: prepared by the authors, using MS-Excel V. 2609 (Microsoft, Seattle, WA, USA).

Monthly precipitation and potential evapotranspiration were obtained from the Climate Hazards Center InfraRed Precipitation with Station Data-CHIRPS dataset, with a grid cell size of 4750 × 4750 m [48]. To improve the reliability of the gridded precipitation data, consistency analysis and bias correction were conducted by pointwise comparing them to observed rainfall from selected rain gauges in the basin. Bias correction was achieved by a parametric equation [49]:

P_c = α + β P_g

(1)

where P_c = bias-corrected precipitation (mm); P_g = gridded precipitation (mm); and α and β are adjustment coefficients obtained from the regression of observed and gridded data. The resulting α and β were 0.85 and 0.04, respectively, with R² = 0.92. Table 1 presents the monthly precipitation, monthly potential evapotranspiration, and number of rain events in the representative year 2023, used in the simulation. The record had no missing data.

Table 1. Monthly precipitation, monthly potential evapotranspiration, and number of rain events in 2023.

Month	P (mm)	ET0 (mm)	Rain Events
jan	215.6	92.6	8
feb	98.9	76.6	17
mar	86.7	83.1	6
apr	131.0	75.3	5
may	9.0	61.0	5
jun	4.0	64.0	1
jul	4.4	49.3	0
aug	18.6	49.2	0
sep	38.0	48.0	0
oct	124.6	60.8	0
nov	95.9	60.1	12
dec	223.4	70.2	19
Total	1050.0	790.1	73

Crop coefficients for the basin land uses were obtained from the literature [50], and curve-number (CN) values, representing the basin’s soil–land-use combinations, were adopted from previous experimental studies in the Brazilian savannah [51,52]. Table 2 presents the mean annual crop coefficients and mean CN values for the Basin land use types.

Table 2. Mean annual evapotranspiration coefficients (Kc) and mean CN values for the basin land uses (CN is the mean of the four NRCS soil hydrologic groups).

Land Use	Kc	CN
Forest Formation	1.0	59.5
Savanna Formation	1.0	66.5
Forest Plantation	1.0	62.0
Wetland	0.9	74.0
Grassland	1.0	61.0
Pasture	1.0	70.3
Land-Use Mosaics	0.7	63.0
Urban Area	0.4	85.3
Non-Vegetated Areas	0.6	86.0
Rocky Outcrop	0.0	99.0
Mining	0.4	85.0
River, Lake	1.0	99.0
Soybean	0.9	78.3
Other Temporary Crops	0.8	82.0
Coffee	1.0	62.0
Other Perennial Crops	1.0	62.0
Cotton	0.9	78.3

The SWY model parameters (α, β, and γ) were calibrated by minimizing the difference between observed and simulated annual streamflow (objective function), using the MS-Excel Solver optimization routine [53]. The lower and upper calibration bounds for the α and β parameters were 0 and 1 and 1/12 and 12/12 for the γ parameter, respectively. Observed annual streamflow data were obtained from the basin outlet gauging station [36], while simulated annual streamflow was the sum of InVEST-SWY annual modeled baseflow (Q_b) and surface runoff (Q_f), the latter being consistent with rainfall frequency.

Observed river discharge (m³ s⁻¹) was converted into water depths (mm yr⁻¹) by a discharge-to-depth conversion formula, using the basin area upstream of the gaging station (A = 23.765 km²). The flow accumulation threshold parameter of the Urucuia Basin was obtained by matching the stream heads of cartographic (map) and GIS-generated flow accumulation [54], resulting in a mean of 5000 pixels.

Following model calibration and validation, the spatial distribution of annual runoff was statistically analyzed to establish a restoration-priority threshold [55]. Areas with runoff values greater than twice the basin-wide mean were classified as priority areas [56], as they represent locations with greater hydrological services.

2.2.2. Soil Loss Criterion

Mean annual soil loss across the basin was estimated with the InVEST Sediment Delivery Ratio (SDR) model [45], using land-use and precipitation data from 2023. Model calibration and validation were performed by minimizing the difference between simulated and observed basin annual sediment yield for the years 2014–2023, using the MS-Excel Solver [53]. Observed sediment yield was obtained from the Urucuia Basin sediment-rating curve [39] in combination with the measured annual streamflow at the basin outlet [36].

The default value of 0.8 was adopted for the model SDR_max parameter [45], and a value of 5.0 was assumed for the L_max coefficient, both considering the basin slope and drainage density characteristics. The InVEST-SDR calibrated parameters were K_b (bounds: 0–1.0) and IC₀ (bounds: 0–10).

As with the surface runoff criterion, a soil-loss threshold was established to identify restoration-priority areas. Cells with annual soil loss that exceeded twice the overall basin mean were classified as priority areas [57], which generally approach the soil loss tolerance limit for tropical soils [52,58].

Rainfall Erosivity

Basin rainfall erosivity was obtained with a Fournier-type equation [59], calibrated for Central Brazil [60], using CHIRPS gridded monthly precipitation data [48].

$$\mathrm{R}=12.59{\sum }_{\mathrm{i}=1}^{12}{\left({\displaystyle \frac{{\mathrm{M}}_{\mathrm{i}}^2}{\mathrm{P}}}\right)}^{0.60}$$

(2)

where R = annual rainfall erosivity (MJ mm ha⁻¹ h⁻¹); M_i = monthly precipitation of 2023 (mm); and P = annual precipitation of 2023 (mm) (Figure 7).

Figure 7. Annual rainfall erosivity of the Urucuia Basin for the representative year 2023. Map resolution: 4750 × 4750 m. Prepared by the authors, using ArcGISPro3.6.0 (ESRI, Redlands, CA, USA).

Soil Erodibility

Since the original USLE equation for soil erodibility, developed for temperate soils [61], is generally unfit for tropical soils [62], the SDR soil erodibility factor (K) was calculated using an empirical equation based on soil texture developed for the Brazilian savannah [63], using the basin soil map (30 × 30 m pixel size) and representative soil texture mapping samples [64]:

K = −0.00043 (AF + SIL)/OC + 0.000437 AR + 0.000863 SIL

(3)

where K = soil erodibility (t ha h ha⁻¹ MJ⁻¹ mm⁻¹); AF (%) = fine sand content in the A horizon; SIL (%) = silt content in the A horizon; OC (%) = organic carbon content in the A horizon; and AR (%) = total sand content in the soil. Figure 4 shows the erodibility of the Urucuia Basin soils.

Equation (2) was obtained from runoff plot and soil textural data, and although it appears to have a dimensionally heterogeneous form, its fitted coefficients (constants) incorporate the required dimensional balance. The soil erodibility map of the Urucuia Basin is presented in Figure 8.

Figure 8. Erodibility of the basin soils. Prepared by the authors, using ArcGISPro3.6.0 (ESRI, Redlands, CA, USA).

Land Use and Land Cover

The SDR C factor for each land use type in the basin was obtained from previous studies in the Brazilian savannah [52,65] (Table 3).

Table 3. Basin land use/cover and C factor.

Land Use/Cover	Area (ha)	C Factor
Forest	93,847	0.01
Natural Savanna	969,171	0.04
Silviculture	15,773	0.06
Wetland	48,010	0.1
Grassland	280,034	0.06
Pasture	561,510	0.1
Land-Use Mosaic	251,122	0.15
Urban	2053	0.1
Non-Vegetated Areas	30,749	0.2
Rocky Outcrop	19	0.0
Mining	9	1.0
River, Lake	8832	0.0
Temporary Crops	39,340	0.2
Coffee	4158	0.1
Other Perennial Crops	641	0.06
Cotton	47	0.15
Soybeans	277,152	0.15
Total	2,582,467	-

Topographic Factor

The SDR slope-length and slope-steepness factor (LS) was computed internally by the model using the basin digital elevation model [45]. Elevation data were obtained from the 30 × 30 m Shuttle Radar Topography Mission (SRTM) dataset [66]. The support-practice factor (P) was assigned a value of 0.1 for terraced cropland and 1.0 for all other land-use classes [63].

Aggregation of Erosion Sub-Criteria

All erosion sub-criteria were spatially aggregated by the InVEST-SDR model, which uses a spatially distributed RUSLE model, incorporating a simplified sediment routing routine along the basin slopes, fully available in the model documentation [45].

2.2.3. Land-Use Priority Criterion

The prioritized land uses were those eligible under the ‘Proáguas Urucuia’ project, namely (i) pastureland, (ii) land-use mosaic, (iii) temporary crops, and (iv) non-vegetated areas. All remaining land-use classes were excluded from the prioritization process. Since natural savannah was considered a preserved land-use condition, it was not prioritized because it was not eligible for restoration within the Urucuia Project.

2.2.4. Municipality Criterion

The municipality criterion was based on an online survey conducted with 50 basin managers and stakeholders, representing all 14 basin municipalities. The survey included questions about the following aspects: (i) the main basin environmental issue; (ii) the main hydrologic issue; and (iii) the most critical municipalities with respect to observed hydro-environmental impacts. The issues and municipalities were ranked based on their relative frequency.

Water scarcity, soil erosion, and river sedimentation were the three most critical issues, and the 14 municipalities were ranked into four priority classes. Hence, the municipality criterion was classified into four restoration-priority categories according to the severity of their environmental degradation. Each priority class was subsequently assigned a corresponding numerical score (1–4) for incorporation into the multi-criteria model.

2.3. Prioritization Field Validation

The restoration-priority map was validated by ground-truthing [67], with mapping accuracy estimated using a confusion matrix that compared predicted and observed classifications for 40 randomly selected priority and non-priority sites. Prioritization accuracy was calculated according to the following equation [54]:

$${\mathrm{A}}_{\mathrm{p}}={\displaystyle \frac{({\mathrm{T}}_{\mathrm{p}}+{\mathrm{T}}_{\mathrm{n}})}{({\mathrm{T}}_{\mathrm{p}}+{\mathrm{T}}_{\mathrm{n}}+{\mathrm{F}}_{\mathrm{p}}+{\mathrm{F}}_{\mathrm{n}})}}$$

(4)

where A_p = prioritization accuracy; T_p = number of true priority sites; T_n = number of true non-priority sites; F_p = number of false priority sites; and F_n = number of false priority sites. A true-positive observation occurred when a basin site identified as a restoration priority by the model was confirmed during field verification. Conversely, a true-negative observation corresponded to a location correctly classified as a non-priority area.

Therefore, 20 priority and 20 non-priority sites were independent of the eligible land-use classes, and randomly selected across the Urucuia River Basin, spanning different physiographic settings and land-use classes, encompassing the upper, middle, and lower reaches of the basin. Examples of priority and non-priority sites are shown in Figure 9.

Figure 9. Examples of validation sites in the Urucuia Basin. Left: priority area (degraded pasture); right: non-priority area (natural savannah). Source: Authors, Urucuia, Brazil, 2025.

The basin priority map was also evaluated during a stakeholder validation workshop [68], attended by 50 basin managers, extension officers, and farmers. During the workshop, participants assessed the consistency of the mapped priority areas based on their local knowledge and practical experience, with the help of large-format paper maps containing property lines and major geographic features of the basin (Figure 10).

Figure 10. Stakeholder validation workshop. Source: Authors, Urucuia, Brazil, 2025.

3. Results

3.1. Basin Runoff

The InVEST-SWY runoff model calibration and validation results, using the 2014–2023 series, are presented in Table 4, with the corresponding Nash–Sutcliffe efficiency coefficients [69] and R² values. Calibration (C) was performed for the odd years and validation (V) for the even years of the series. The calibrated model parameters were a = 1.0, b = 1.0, and g = 1/12.

Table 4. Calibration (C) and validation (V) results of the InVEST-SWY model for the annual flow in the series 2014–2023, with corresponding means, NSE, and R² accuracy coefficients.

Year	Observed Flow (mm yr−1)	Calculated Flow (mm yr−1)	Calibration (C) Validation (V)
2014	315.8	275.9	V
2015	260.1	160.1	C
2016	250.8	232.8	V
2017	272.0	295.0	C
2018	285.3	389.7	V
2019	294.6	283.4	C
2020	289.3	479.2	V
2021	311.8	487.0	C
2022	321.1	450.3	V
2023	318.5	278.1	C
Mean	291.9	333.1
NSE-C = −16.0;   NSE-V = −19.7;   R2-C = 0.42;   R2-V = 0.19

The spatial distribution of annual surface runoff, corresponding to the first MCM criterion, is presented in Figure 11. Basin mean annual runoff was 32 mm yr⁻¹, with 10.3% of the basin area (268,135 ha) exceeding the prioritization threshold of 64 mm yr⁻¹. These areas were therefore classified as priority locations for reforestation and soil conservation based on their runoff and environmental service generation potential.

Figure 11. Annual runoff in the Urucuia River Basin. Source: prepared by the authors, using ArcGISPro3.6.0 (ESRI, Redlands, CA, USA).

3.2. Basin Soil Loss and Sediment Yield

The InVEST-SDR erosion model calibration and validation results, using the 2014–2023 basin observed and calculated sediment yields, are presented in Table 5 below, with the corresponding Nash–Sutcliffe efficiency coefficients [69] and R². Calibration (C) was performed for the odd years and validation (V) for the even years of the series. Calibrated SDR model parameters were 1.0 (K_b) and 1.77 (IC₀).

Table 5. Calibration (C) and validation (V) results of the InVEST-SDR model relative to the basin annual specific sediment yield in the series 2014–2023, with corresponding means, NSE, and R² accuracy coefficients.

Year	Observed Sediment Yield (t km−2 yr−1)	Calculated Sediment Yield (t km−2 yr−1)	Calibration (C) Validation (V)
2014	4.7	3.2	V
2015	3.1	3.7	C
2016	2.9	3.7	V
2017	3.4	4.3	C
2018	3.8	4.8	V
2019	4.1	4.1	C
2020	3.9	3.7	V
2021	4.6	4.9	C
2022	4.9	4.9	V
2023	4.8	3.7	C
Mean	4.0	4.1
NSE-C = −0.03;   NSE-V = −0.6;   R2-C = 0.11;   R2-V = 0.02

The map of mean annual soil loss is shown in Figure 12. Mean basin soil loss was 3.8 t ha⁻¹yr⁻¹, with 12.7% of the basin area (332,065 ha) presenting erosion rates greater than the prioritization threshold of 7.6 t ha⁻¹yr⁻¹. These areas have severe erosion and the highest potential for environmental service generation resulting from restoration. In both runoff (Figure 11) and soil loss (Figure 12) maps, the unclassified areas around the border represent only 1.5% of the total basin area, as a result of the large pixel size of the gridded CHIRPS data.

Figure 12. Annual soil loss in the Urucuia River Basin. Prepared by the authors, using ArcGISPro3.6.0 (ESRI, Redlands, CA, USA).

3.3. Priority Land-Use Classes

The eligible land-use classes for restoration occupied an area of 752,140 ha, representing 29.1% of the Urucuia River Basin area. These constituted the third criterion of the multi-criteria model and corresponded to the land uses targeted by the ‘Proáguas Urucuia’ project, excluding the natural land covers, such as natural savannah.

3.4. Municipal Priority

The municipal restoration-priority map, based on land degradation severity established by basin managers and stakeholders, is shown in Figure 13, representing the fourth criterion of the multi-criteria model. According to the local stakeholder assessment, two municipalities were classified as having very high restoration priority, three as high priority, three as medium priority, and six as low priority.

Figure 13. Municipal restoration priority from the local stakeholder survey. Source: prepared by the authors, using ArcGISPro3.6.0 (ESRI, Redlands, CA, USA).

3.5. Restoration-Priority Map

The final restoration-priority map generated by the multi-criteria model is presented in Figure 14. Prioritized areas spanned the upper, middle, and lower portions of the Urucuia River Basin and all landscape settings, encompassing 31,455 ha (1.2% of the total basin area).

Figure 14. Urucuia Basin priority areas for environmental restoration. Source: prepared by the authors, using ArcGISPro3.6.0 (ESRI, Redlands, CA, USA).

A comparison of the final prioritization map (Figure 14) with the maps of slope grade (Figure 2), soil erodibility (Figure 8), and land use (Figure 4) indicates that the prioritized areas were associated with high slope grades (S > 6%), highly erodible soils (K > 0.04), and low-density land cover, such as pastureland and agriculture.

Because the prioritization criteria focus on the principal drivers of hydrological degradation—including excessive runoff, accelerated soil erosion, and inappropriate land use—the selected restoration areas are expected to provide substantial improvements in water regulation, sediment retention, and overall watershed resilience [20]. Consequently, restoration investments could generate greater environmental returns than those achieved through non-prioritized or subjective site selections, thereby contributing to a positive feedback cycle of watershed recovery and sustainable resource management [70].

Considering that the ‘Proáguas Urucuia’ project aims to restore 2000 ha, the prioritized areas, approximately 15 times the restoration target, provide a sufficiently large pool of candidate sites to support a comprehensive selection process. This allows restoration priority to be considered alongside additional operational factors, including landowner socioeconomic conditions and project implementation logistics [8].

3.6. Validation of Prioritized Areas

The ground-truthing confusion matrix is presented in Table 6. Application of Equation (3) resulted in an overall mapping accuracy of 90%, indicating good agreement between predicted and observed verification sites.

Table 6. Confusion matrix results.

	Non-Priority
		No	Yes
Priority	No	16	0
	Yes	4	20
Accuracy = 90%

Additionally, at the stakeholder validation workshop, the majority of the 50 participants considered the prioritization method relevant and transparent, and the corresponding prioritization map to be consistent with their personal knowledge of basin environmental issues and restoration needs.

4. Discussion

A total of 31,455 hectares were prioritized with the multi-criteria model in the Urucuia River Basin, with selected areas spanning all 14 municipalities of the upper, middle, and lower reaches of the watershed, encompassing different land cover typologies, such as forests, agricultural land, wetlands, and built-up areas. The prioritized areas were associated with steep slopes with little or no land cover, highly erodible soils, and zones of high runoff generation, features which were previously recognized in different tropical basins [3,71].

A key strength of the proposed method is the integration of hydrological, environmental, and socioeconomic criteria within a single decision-support framework [1]. This integrated approach enables restoration actions to be directed toward areas where environmental interventions are expected to generate the greatest overall benefits with respect to water quality and quantity, which are compatible with basin management priorities [72].

Although the relatively low number of model criteria could potentially reduce the prioritization reliability of the proposed MCM, a large number of factors would increase model complexity and the risk of model multicollinearity [73]. Also, the assignment of equal criteria weights contributed to improving the transparency and reproducibility of the methodology, since balanced models are less prone to subjectivity and bias associated with expert judgment [74]. This assumption was supported by similar basin prioritization studies elsewhere [75].

The fact that the InVEST-SWY (runoff) and SDR (erosion) models were designed for relative comparisons between land-use scenarios and not for quantitative model predictions [45] explains in part their lack of accuracy during the calibration and validation process, as shown in Table 4 and Table 5.

Indeed, previous research using both models resulted in low NSE and R² in basins around the world. In American basins, the InVEST-SWY yielded R² scores below 0.5 [76]. In Brazil, InVEST-SDR model calibration and validation accuracies were also low, with NSE < −0.3 and R² < 0.2 [77], reinforcing the fact that those models are more suitable for qualitative applications, such as basin prioritization.

Since the main objective of the present study was the screening of effective prioritized basin areas, the model efficiency indicator was the field-check accuracy (90%), above the 85% threshold adopted as a benchmark for acceptable land-use/land-cover classification [78]. Although the number of ground control points was not too large, they were spatially distributed across the study area and land-cover classes and followed principles of probability-based sampling [79].

Furthermore, the validation workshop provided additional evidence of stakeholder compliance, a key engagement indicator of institutions and communities involved in project implementation [80]. Although the field verification and the participatory mapping workshop do not span all dimensions of prioritization accuracy and representativeness, they are independent verification processes and complementary indicators of mapping effectiveness [81].

Although the proposed methodology proved to be a useful screening tool in the Urucuia River Basin, its application to different basins should be preceded by appropriate verification [82]. Finally, the proposed multi-criteria framework is intended to support—not replace—managerial decision-making, requiring that the final area selection process considers factors that cannot be fully represented in spatial models [83], including landowner willingness to participate, land tenure, project logistics, implementation costs, and institutional priorities.

5. Conclusions

The proposed method allowed for accurate identification of prioritized restoration areas in the Urucuia Basin, using a simple and balanced multi-criteria model, in which biophysical variables were aggregated by runoff and erosion models, followed by ground-truthing and stakeholder validation.

The prioritized areas spanned all basin physiographic settings, land-use classes, and municipalities, and were associated with steep slopes, low vegetation cover/density, and highly erodible soils, which are responsible for most of the runoff and sediment generation in the basin, and where hydro-environmental services of restoration could be maximized.

Another methodological innovation was the strong stakeholder involvement in the prioritization process, both in MCM development (municipality criterion) and validation (stakeholder workshop). This is a key element in successful basin management initiatives. The selected MCM criterion thresholds were sensitive enough to generate a total prioritization area equivalent to 15 times the project restoration goal (2000 ha), providing a sufficiently large pool of candidate sites to support a subsequent selection process, subject to socioeconomic and logistical constraints, and guaranteeing project sustainability.

Although the small number of MCM criteria and their equal weights of the proposed method could be considered a limitation, by reducing the complexity of real-world conditions, model simplicity and equilibrium are often preferred in situations where data uncertainty and scarcity are present, such as in tropical basins.

Although the multi-criteria model was developed for the Urucuia River Basin, its conceptual framework is sufficiently simple and flexible to be adapted to other tropical and subtropical watersheds. However, its successful application in different basins should be preceded by calibration, field verification, and stakeholder validation to account for differences in environmental and socioeconomic conditions, land-use patterns, and restoration objectives.