1. Introduction
Sediment is the largest volumetric nonpoint-source pollutant to surface waters [
1,
2,
3] and one of the most important water-quality problems in the United States [
4,
5,
6]. Upland watershed erosion is a serious issue for estuaries and the coastal region of the southeastern United States. During precipitation events, overland and streambank erosion increases in the watershed, often resulting in degradation to downstream resources in the associated estuary. Erosive rates are amplified in areas experiencing active land use changes with agriculture, and increasing urbanization and industrialization [
7,
8]. The influence of the growing human population and unrestricted development in coastal watersheds is proving to be very detrimental to the overall integrity of the fragile, yet highly productive estuarine ecosystems. This growth and development have increased pollution inputs, loss of habitat, and nutrients, and has led to degraded ecologic conditions [
1,
2,
3]. These trends of degraded conditions due to human influence will continue to impact estuaries, creating higher instances of eutrophication, hypoxia, and anoxia.
Coastal watersheds and their estuaries are important to the overall coastal environment, and are areas of high biologic productivity [
9,
10]. The high level of productivity is, in part, due to the transition zone created by the mixing of the upland drainage of fresh water with saline seawater; these areas are referred to as the nurseries of the sea [
11].
Modeling erosion in coastal watersheds is a complex task that involves a wide range of knowledge from several scientific and engineering disciplines. An effective understanding of coastal watersheds requires several inputs, such as coupling landscape characterization and hydrologic processes [
12,
13]. Developments in geographic information systems (GIS) and other geospatial technologies have greatly increased the quality and quantity of data available for hydrologic modeling [
13,
14,
15,
16]. The coupling of GIS with other models is an approach that is effective in the management of the resources of coastal watersheds. Numerous hydrologic, soil-erosion, and landscape-characterization models can couple with geospatial technologies such as GIS for improved data processing, analysis, and visualization [
13,
14,
15,
16,
17].
The design of soil-erosion models allows them to work in conjunction with GIS and other geospatial applications. Examples include the Water Erosion Prediction Project (WEPP), Soil and Water Assessment Tool (SWAT), and the open-source version of the Nonpoint Source Pollution and Erosion Comparison Tool (OpenNSPECT). These models are often described as traditional soil-loss models and are either mechanistic (i.e., SWAT) or empirical, such as the Revised Universal Soil-Loss Equation (RUSLE) and Modified Soil-Loss Equation [
18]. Soil erosion across the landscape has traditionally been characterized using models such as the RUSLE [
19,
20] and the WEPP [
21,
22,
23]. The combination of many of these models with GIS helps with the transition from models to decision-support and analysis. Modeling approaches are typically either classified as qualitative or quantitative [
24]. A quantitative model is data driven, and it is difficult to apply this model in data-poor regions. Additionally, a quantitative model is not enough to determine erosion potential when there are several factors influencing the erosion of the zone [
25]. On the other hand, a qualitative model has fewer data requirements. It can easily identify the primary factors that are responsible for erosion potential [
25]. Additionally, the use of quantitative models in decision-support and analysis is beneficial; however, the execution and data requirements of the models often limit updated assessments for specific management needs. These limitations are increased, as many of the managers lack the resources needed to readily execute the models.
Geospatial technologies have provided several contributions to watershed modeling through their ability to utilize large temporal datasets from monitoring/sampling locations (e.g., hydrometric, and climatic stations) [
16]. Remote sensing has created a pathway for the classification of land-use/land-cover changes in coastal watersheds, which help to visualize landscape changes arising from the increasing population and developments [
26]. These types of classifications, coupled with GIS and spatial analyses, are allowing environmental decision-makers to identify and rank land-use patterns for the implementation of best management practices for nonpoint-source pollutants and other related issues [
27]. These GIS and spatial analysis methods allow relationships to be established between sediment loading and the watershed landscape to help identify and prioritize management areas efficiently [
28,
29]. The mapping of watershed erosion potential, focused on watershed landscape characterizations, provides a needed measure of assessment and aids in the identification of sediment sources contributing to degraded conditions within a watershed and the associated estuarine environment. These characterizations are derived from land-use/land-cover changes and practices (i.e., land disturbance), terrain analyses, physical properties of soils, and other geomorphologic features such as surface drainage density. Previously, factors such as slope gradient, precipitation, NDVI (normalized difference vegetation index), land use, soil texture and slope aspect were studied for soil-erosion risk assessment [
25]; additionally, drainage density, slope, land use/cover, and runoff measurement were used for identifying potential zones for rainwater harvesting [
30], etc. Therefore, based on the availability of data in the specified zone, the number of factors may be varied as an input to the models.
Multi-criteria decision analysis (MCDA) methods have become very popular for spatial planning and management issues and are a significant tool for decision makers, especially for multicriteria assessment [
31]. The applications of MCDA are wide. It is applied to identify priority areas for soil-erosion risk measurement [
25], to calculate landscape deformation index [
31], to identify potential zones for rainwater harvesting [
30], in the field of transport for determining suitable management [
32], to generate the ranking of green bonds in corporate office management [
33], etc.
Hence, expanding GIS utilization for MCDA has improved decision-support models for land-based suitability evaluations. These expanding efforts have increased the need for ways to evaluate the performance of the models and tools utilized, as well as the sensitivity of the variables or layers used [
34,
35]. There are numerous procedures that are used with GIS for MCDA; examples include Boolean overlay, weighted linear combination (WLC), ordered weighted averaging (OWA), and the analytical hierarchy process (AHP) [
36]. The WLC is one of the most commonly used decision-support tools in the GIS environment [
37,
38]. Additionally, GIS coupled with the AHP [
39] is proving to be an important tool for MCDA [
40,
41]. GIS utilizing AHP is an established and credited approach to MCDA for land-resource-management decisions [
42,
43], and is an important part of sustainable land-planning approaches [
44,
45]. The AHP has been found to be a robust method for determining criteria weights based on expert input [
46] and works well with MCDA in the GIS environment. Additionally, the combination of GIS and AHP is useful for MCDA in the management of natural resources related to soil-erosion mapping [
47]. While models and tools of this type do not allow for the quantification of sediment yields or soil-loss rates due to erosion, they do offer resource managers and decision makers the necessary information to better manage and prioritize watersheds and the related resources.
Therefore, qualitative modeling approaches are often driven by MCDA, typically with expert input. This makes them very useful in the decision-making process, specifically for tasks such as vulnerability assessments and other methods [
48]. The qualitative nature of MCDA often requires nontraditional methods of uncertainty assessment. Sensitivity analysis is one of the common methods used to reduce uncertainty in the variable weights, which can assist with identifying stability in model performance with changing criteria weights [
34]. Sensitivity analysis with GIS-based MCDA can also provide insights into the spatial aspects of the changing criteria weights. Feick and Hall suggested that efforts to analyze criteria weight sensitivity can help to geographically visualize the sensitivity of the results [
49]. Another approach to reducing uncertainty in the modeling approach is the combination of different models using the average, which is termed as ensembled model [
50,
51]. The ensembled model is a useful combinational approach and can enhance the strength of prediction mapping while reducing the weakness of each source map [
51]. Studies showed better/improved prediction in clay content mapping for soil-quality assessment and decision making in land use [
50], and for combining digital soil-property maps derived from disaggregated legacy soil-class maps and scorpan-kriging (using soil pan data) [
51] from the ensembled model, respectively.
The primary aim of this research is to develop a qualitative assessment to map erosion potential using WLC, AHP, and ensembled modeling approaches for watersheds associated with the northern Gulf of Mexico. This assessment will provide a process to resource managers for the identification and prioritization of watershed management areas. The mapped erosion potential found from these models will be compared to the sediment yield from the SWAT model (
https://swat.tamu.edu/). The SWAT model is widely used across the globe in assessing soil-erosion prevention control, nonpoint-source pollution control, and regional management in watershed (
https://swatplus.gitbook.io/docs/). The SWAT was used in the development of the Weeks Bay Watershed Management plan. The model delineated the Weeks Bay watershed into 237 sub-watersheds (197 for the Fish River and 40 for the Magnolia River); these are used to produce the computational hydrologic response units (HRU’s) in SWAT. Sediment-yield results from this model were based on 2011 land use/cover, and it was reported that over half of the sediment yield was produced from about one-third of Weeks Bay watershed [
52].
Therefore, the comparisons will be limited to basic observations between the qualitative and quantitative output of the data. The comparisons will show a general visual alignment in sub-basins of increasing development and headland drainage areas in the study area. The comparison against the output of the SWAT model will help the resource managers to look at scenarios or management priorities without the understanding and execution of more complex soil-loss models. The ensembled model will aid in the visualization of the priorities of the management areas. The models will serve as a base for the multi-criteria decision analysis (MCDA) of erosion potential by decision makers and resource managers.
4. Discussion
The erosion potential for a watershed along the northern Gulf of Mexico was mapped qualitatively using layers representative of physical erodibility, land sensitivity, and 30-year precipitation for the Weeks Bay watershed. The criteria used to define these layers were based on regional availability and input from watershed managers and stakeholders on erosional trends in this area. The approach used a WLC model and was set up with criteria similar to numerical models such as RUSLE [
19]. Other approaches to soil-erosion mapping include empirical, conceptual, physically based, and hybrid models [
60]. Some hydrodynamic numerical models, i.e., 1-D (one-dimensional) and 2-D hydrodynamic models showed better efficiency in urban-flood-risk mapping [
61]. Compared to the conceptual models, i.e., the Hydrologic Simulation Program (HSPF) and SWAT [
60], this study proposed a WLC model which is also applicable to larger areas with less complexity. Additionally, the AHP model in the study area considered the factors’ importance in terms of being responsible for regional watershed erosion. However, this study did not show any physically based model, which could be a future demand for better estimation of erosion risk at a large scale. In addition, factors such as surface hydrology, slope aspect, and storm events could be used in 2-D physically based simulation models such as GSSHA (Grided Surface/Subsurface Hydrologic Analysis), DWSM (Dynamic Watershed Simulation Models), etc.
In this study, the northern headland and the southern agricultural regions of the watershed had the highest erosion potential, as expected [
1,
28]. Overall erosion potential in the Weeks Bay watershed tends to be lower in densely vegetated riparian and marsh areas. Many of these areas, especially in the southern area near the bay, are part of the Weeks Bay National Estuarine Research Reserve. Areas in the watershed with higher erosion potential are more associated with transitional-type lands that appear to be more agricultural or dynamic in terms of land practices. The southeast region of the watershed reflects this, as it is an area dominated by agricultural practices such as cultivated crops and turf-grass farms.
Land sensitivity accessed by soil erodibility (as defined by the soil K-factor and soil exposure or brightness) was the criterion that had the most influence on mapped erosion potential with the WLC model. The K-factor is used in USLE and RUSLE applications that represents soil texture and composition [
19,
20,
24], and was the most influential of the layers. Soil brightness is indicative of disruptive land uses and increases in erosion potential from this variable are apparent in agriculture-dominated areas of the watershed. The physical erodibility (topographic) criteria for slope and stream density moderately influenced the WLC-modeled erosion potential. These areas of increased potential are indicative of higher concentrations of stream reaches, with more surface interaction with runoff waters and lower soil infiltration rates [
55], and proximity to active stream and river channels.
The assessment was enhanced with expert input through the AHP model, allowing experts to rank the criteria. The ranking of criteria quantified the weights based on their relative importance in the study area. The physical erodibility criterion of slope was identified as the most important by the experts. In this, the AHP model differs from the traditional soil-erosion models, with the experts’ minimal emphasis on land cover or land sensitivity [
19,
21]. This difference is also concerning due to the increased erosion rates due to development in the watersheds in this region of the Gulf of Mexico [
52]. The AHP model proved to be beneficial in mapping areas of increased erosion potential, as defined by the upper ranks in the classified data. These shifts of the mapped erosion cells from the WLC model align with similar approaches using MCDA techniques [
42,
55]. The AHP model is not a typical numerical model and can be better-suited to qualitative geospatial assessment for mapping erosion potential with MCDA.
The variations in the AHP weights are normally used to identify shifts (increasing and decreasing) in the mapped erosion-potential cells. Data outliers were used for the identification of areas of model alignment for the high ranks of erosion potential. This approach allowed for an analysis (management) mask to be generated for these areas that were consistently high, irrespective of the variation in criteria weights. The areas identified were generally associated with higher slopes, which were associated with stream and channel networks. The outlier analysis was successful in helping to identify management areas; however, a suitable model approach would allow for a quantification shift between ranks of erosion potential [
34,
39,
45].
The WLC and AHP erosion-potential models were run as an ensemble to improve the reliability of the output. The outputs (from models) were compared with the SWAT sediment yield; the comparisons were used to help identify if there were any visual alignments or trends between the qualitative and quantitative outputs. The comparisons were very similar, with the primary difference being the focus of higher erosion-potential values in the WLC and AHP models. The alignment of the data was most apparent in areas of transition with expanding development and agricultural land practices. The AHP model, however, produced some areas that were more focused on topographic features because of the experts’ input placing more emphasis on physical erodibility (slope and other terrain measures). This was most apparent in the headland area of the watershed where the landscape has more dissection. The comparison of prioritization and ranking of the mapped erosion-potential cells for Weeks Bay Watershed management areas displayed alignment for select areas of higher erosion potential.
These management areas include the Pensacola Branch, Waterhole Branch, and Turkey Branch basins of the Weeks Bay Watershed. Each of these management areas have experienced increased erosion due to expanding land development associated with the surrounding communities. The management areas with the lowest ranks were also in agreement. This, coupled with the alignment in the upper ranks, indicates a generalized agreement between the qualitative and quantitative assessments for the prioritization of management areas. The approach has limitations in management areas that would be considered to be of intermediate concern with regard to erosion. This is, in part, due to various reasons, including the added emphasis by the experts on terrain characteristics and temporal generalizations in layers used in the WLC/AHP models and the SWAT model.
However, these areas of upper and lower erosion ranks are indicative of the agreement between the qualitative and quantitative modeling approaches, supporting the use of MCDA for management decisions and improved applications for watershed managers.
5. Conclusions
The watersheds that drain areas along the northern Gulf of Mexico have dynamic landscapes that experience erosion and contribute volumes of sediment to the associated estuaries. These watersheds and their estuaries are important for both their services and the resources they provide. To maintain the function and value of these services and resources, these areas require proper management in terms of soil-loss and erosion. Those involved in the management of these watersheds and the associated resources need to have information that is both accurate and timely. This study took an approach that focused on a qualitative assessment with GIS and MCDA for mapping erosion potential, to facilitate the prioritization of watershed management areas for improved management decisions. This aim of this approach was to provide watershed managers with a process to quickly map erosion potential and prioritize areas for management. This will allow for better and more efficient allocation of resources to be utilized in watershed management for areas such as the Weeks Bay watershed.
Field measurements and numerical models are critical to accurately measure and estimate sediment yield and erosion, as they provide quantitative information to facilitate management plans and decisions. Qualitative assessments can produce similar results and help in the management process. Qualitative assessments do not provide numerical sediment-yield information, but often provide more rapid assessments with a more simplistic approach and execution. The design of these assessments needs to be similar to numerical modeling approaches, using similar criteria and inputs. Their design can also allow for expert input for situation-specific applications due to their understanding of the processes and issues unique to the watershed of interest. A WLC model and an AHP model were used to map erosion potential based on terrain slope, geomorphology, land cover, soil erodibility, and long-term precipitation trends.
The WLC and AHP models developed for this study mapped erosion potential to cells as defined by the input layers for the Weeks Bay watershed. The mapped erosion potential aligned with the erosion trends described in the Weeks Bay Watershed Management Plan, with increased erosion occurring in areas associated with agricultural practices and expanding areas of urban and suburban development. The MCDA with the AHP model mapped areas that are most susceptible to erosion, as evidenced by the shifts of cells in the upper ranks. This, coupled with the analysis mask generated by the criteria weighting variations, identified areas of alignment or commonality in erosion potential in the Weeks Bay watershed. These areas, in both the WLC and AHP models, were in agreement with the SWAT sediment-yield data from the Weeks Bay Watershed Management Plan. The qualitative approach was effective in prioritizing management areas in the Weeks Bay watershed and offers a simplified approach to mapping erosion potential. This simplified approach provides watershed managers with the means to define and prioritize management actions. This does not replace the need for numerical modeling for quantitative soil-erosion metrics; it does, however, provide management alternatives when needed. There are numerous pathways for future research for this work. Refinement and adaptation of the qualitative approach will continue to improve reliability as compared with numerical model outputs such as sediment yield. This will involve more interactions with modelers and watershed managers (and stakeholders). Scaling the approach to a regional level in future efforts could be beneficial in prioritizing modeling needs and guidance in watershed management plan needs. This, coupled with more development of geospatial tools, would help to transition these types of approaches to a more operational environment, allowing for enhanced planning and management-scenario development.