Abstract
This study demonstrates that green infrastructure (GI) landscape patterns provide stormwater runoff benefits in the Alabama and Mississippi coastal region; the study also provides insights for creating more resilient coastal communities using GI strategies and illuminates the role of GI and land development patterns in regulating floodplains within a regional setting. Landscape patterns across wetlands, vegetation, barren lands, and developed areas are analyzed to identify the types of landscapes that should be protected to mediate the impacts of local flooding. The methodology involved the application of Fragstats, ArcGIS, and the Soil and Water Assessment Tool (SWAT), followed by an analysis of correlations between variations in the runoff and landscape and development patterns using ordinary least square (OLS) regression. The results demonstrate that larger proportion and size, as well as more connected vegetation and wetlands at the regional scale, reduce peak runoff, thus confirming that effective GI conservation planning should focus on wetlands and wooded vegetation. Further, the study found that more connected developed lands are associated with reduced runoff, strengthening the case for compact development in regional land use planning from a flood mitigation perspective.
Introduction
In recent years, the practice of green infrastructure (GI) planning and conservation has emerged as an effective tool for mitigating the negative impacts of development on ecosystems that are increasingly at risk (Bu et al., 2023; Parmesan, 2006; Soni et al., 2025; Zhu & Newman, 2021). Conversion of large areas of undeveloped land into impervious surfaces leads to degradation of water quality and ecosystem services (Arnold & Gibbons, 1996; Srinivasan et al., 2013), accelerates fragmentation of landscapes and habitats (Kim et al., 2018), and increases flood risks and the amount and velocity of stormwater runoff (Kim & Park, 2016; Liu et al., 2020; Nguyen et al., 2024; Shuster et al., 2005; Zhang et al., 2015; Zhu et al., 2022). Uncontrolled growth in coastal cities makes them particularly vulnerable to natural disasters such as flooding and hurricanes and decreases resiliency (Gill et al., 2007; Kim et al., 2014; Liu et al., 2014; Miller et al., 2020; Reguero et al., 2018).
The Water Infrastructure Improvement Act of 2009 defines GI as “the range of measures that use plant or soil systems, permeable pavement or other permeable surfaces or substrates, stormwater harvest and reuse, or landscaping to store, infiltrate, or evapotranspirate stormwater and reduce flows to sewer systems or to surface waters” (Water Infrastructure Improvement Act, 2019). GI practices range from regional, through connecting greenways and greenspace nodes, protecting wetlands and other open spaces, to local site‐specific design and engineering strategies such as construction of rain gardens, bio‐retention ponds, and green roofs (Benedict & McMahon, 2012). This study focuses on regional land use planning and policy to protect and conserve large areas of land with vegetation. It does not focus on local, site‐specific strategies to complement or replace gray infrastructure.
In practice, regional‐scale GI landscape strategies such as protecting large open and natural spaces, riparian areas, wetlands, forests, or greening hillsides are implemented by local and state governments with the goal of creating resilient communities that can respond effectively to flooding events (Benedict & McMahon, 2012; Kim & Park, 2016; Lindell & Prater, 2003; Lynch, 2016; United States Environmental Protection Agency [USEPA] a). The Environmental Protection Agency (EPA) promotes GI as a cost‐effective strategy to create resilient communities, reduce flood risks, and improve water quality, providing a stormwater planning toolkit and other resources for green infrastructure planning and implementation (Kim & Park, 2016; Lynch, 2016; USEPA a). When GI is protected and enhanced across regional watersheds, the results have shown significant environmental, social, and economic benefits, including cleaner air and water (Amini Parsa et al., 2019; Jiang et al., 2008; Neto & Sarmento, 2019; Taha, 2015), flood protection for the community (Maragno et al., 2018), enhanced diversity of habitats and beautiful green spaces (Butt et al., 2018; Connop et al., 2016; Li & Mooney, 2019), reduced pollutants discharged into waterways (Pennino et al., 2016; USEPA a; Zhu & Newman, 2021), as well as urban cooling (Azmeer et al., 2024).
As climate change and coastal development intensify, GI strategies can help with environmental restoration, adaptation, and flood risk reduction (Reguero et al., 2018). The ecological functioning of urban GI also depends on the size and integrity of plant communities and ecosystems. While all GI is regarded as beneficial, isolated GI systems have reduced capacity that is exacerbated during larger storm events. The integrated configuration and scaling‐up of GI interventions is more effective in reducing runoff and dealing with larger storms (Liu et al., 2014; Liu et al., 2020; Zellner et al., 2016). GI adaptation measures, such as wetland and oyster reef restoration, are found to be particularly attractive due to their cost‐effectiveness (Reguero et al., 2018).
Rapid urban area expansion in coastal communities is leading to the conversion of undeveloped lands into impervious surfaces. Community planners are adopting GI mitigation strategies such as planned networks of open and natural areas and effective stormwater management for creating resilient coastal communities that are adequately protected against flooding (Benedict & McMahon, 2012; Cameron & Blanuša, 2016; Dylewski et al., 2014; Eaton, 2018; Heim LaFrombois et al., 2023; Liu et al., 2020; Lynch, 2016; Rouse & Bunster‐Ossa, 2018). Between 1960 and 2008, the population density in coastal Alabama (AL) increased by 59.8 %, and in coastal Mississippi (MS), it increased by 84.8% (Wilson & Fischetti, 2010). During 2010–2020, Gulf Shores grew by 54.13%, the Daphne‐Fairhope‐Foley metropolitan statistical area (MSA) grew by 25.8%, and Baldwin County grew by 27%, making it the second‐fastest‐growing county (Alabama Regional Economic Analysis Project). The Gulfport‐Biloxi MSA was the fastest growing area in MS during 2010–2020, increasing by 12%, even as the state’s population declined slightly (Mississippi Regional Economic Analysis Project). This rapid urban growth along the AL‐MS coast makes it an ideal region for studying the effect of landscape patterns on stormwater runoff.
Landscape metrics describe the spatial structure of patches, classes of patches, or entire landscapes; they are measured at the landscape level, class level, and/or patch level and provide useful information about the composition and configuration of a landscape, such as the proportion of each land cover type present or the size/shape of different landscape elements (Brody et al., 2017; Deng et al., 2023; Liu et al., 2020; Rasoulzadeh et al., 2023). The impact of development and landscape patterns on hydrological factors and flooding has been studied previously with varying and sometimes contradictory results. A highly connected GI landscape can either increase (at the microscale) or decrease (at the macroscale) peak runoff, depending on the scale of the study (Rasoulzadeh et al., 2023). Kim and Park (2016) investigated the effects of landscape patterns on variations in mean annual peak runoff and found that larger, more connected GI landscapes are likely to reduce runoff and larger and more clustered developed lands are likely to increase runoff. According to their findings, the impervious rate has a positive relationship with the peak runoff. In contrast, Brody et al. (2013) found that more connected and concentrated development is associated with reduced flood losses. Brody et al. (2017) determined that large, expansive, and naturally occurring open spaces that are continuous provide the largest benefit in reducing insured flood losses. Shuster et al. (2005) reviewed the literature and found that impervious surfaces have a major effect on watershed hydrology, but it is difficult to identify a single threshold imperviousness level that could reliably predict hydrologic and environmental impacts. Zhang et al. (2015) found that in different regions in Beijing, regional differences in the runoff rate were closely related to changes in the largest patch and aggregation indices of urban green spaces. Newman et al. (2017), Zhu and Newman (2021), and Zhu et al. (2022) observed that the repurposing of vacant land to facilitate the creation of landscape corridor systems enhanced habitat values and retained and distributed floodwater to increase flood attenuation. Miller et al. (2020) reported that connectivity and location of urban surfaces were more valuable than impervious areas alone in attributing timing, validating findings from distributed hydrological modeling studies.
The main objective of this study was to demonstrate that GI landscape patterns provide stormwater runoff benefits by determining what types of landscape metrics are correlated with reduced stormwater runoff and therefore more effective at managing stormwater. Our hypothesis was that larger‐sized, clustered, less fragmented, and highly connected GI landscapes consisting of wetlands and vegetation would decrease peak runoff and are therefore more desirable for managing stormwater. In addition, we hypothesized that landscape patterns characterized by larger patches, more connected patches and corridors, and heterogeneous natural areas would be more effective in reducing runoff. We also anticipated finding that proportionally large and clustered developed lands and impervious surfaces would correspond with increased runoff.
Methodology And Approach
We analyzed regional and local landscape patterns for eight watersheds, consisting of 1291 sub‐basins in the coastal AL and MS regions, using a range of GIS tools. Specifically, the methodology included: (1) delineation of the sub‐basins and assessing environmental characteristics within each, (2) calculation of stormwater runoff values within each sub‐basin, (3) landscape metrics pattern analysis within each sub‐basin, and (4) ordinary least square (OLS) regression analysis of correlations between runoff and landscape patterns. This study built upon and expanded the methodology provided by Kim and Park’s (2016) study analyzing landscape patterns in terms of size, shape, isolation, and connectedness in the four largest metropolitan areas of Texas (TX). Ungauged basins were excluded from the Kim and Park (2016) study, whereas in this study we estimated stormwater runoff from ungauged basins to further develop this methodological approach. We apply the Soil and Water Assessment Tool (SWAT), a public‐domain, river basin‐scale model developed to simulate the impact of land management practices on water resources and to assess the environmental effects of land use in complex watersheds (USEPA b) in an innovative manner. While the Kim and Park (2016) study covered four MSAs, this study covers eight watersheds, making it truly regional in scale. Extensive data were collected for regional and local landscape patterns analysis for the 1291 sub‐basins using ArcGIS, SWAT, and Fragstats tools.
The methodology broadly followed several distinct steps. First, the study region was determined based on the availability of United States Geological Survey (USGS) gauge stations and was subcategorized into eight subsections, with 1,291 sub‐basins delineated within them. Each of the sub‐basins was modeled separately using the SWAT tool in ArcGIS v10.5 following the process shown in Figure 1.
Conceptual modeling of stream flow at an ungauged river basin using SWAT.
Next, data from various agencies were collected, including the USGS, the PRISM climate data (PRISM Climate Group), the National Resources Conservation Service (NRCS), and the Federal Emergency Management Agency (FEMA). Data were analyzed for several environmental characteristics, including daily runoff depth, precipitation, slope, soil permeability, floodplain, natural drainage density, wetlands, and impervious rate, using ArcGIS version 10.5 (ESRI Inc., 2015). Following this, for each sub‐basin, stormwater runoff was estimated for the three‐year period from 2014–2016 using the SWAT version 2012.10_5.21 tool (United States Department of Agriculture [USDA] and Texas A&M AgriLife, 2012). This was followed by an analysis of regional and local landscape patterns within each sub‐basin using Fragstats version 4.2 (McGarigal et al., 2012). Lastly, we analyzed the correlations between variations in the mean annual peak unit discharge and landscape metrics using OLS regression analysis to determine how landscape composition and configuration are correlated to runoff. Four distinct statistical models are developed, one for each land cover type: Wetlands, Vegetation, Barren Lands, and Development. These steps are described in the following sections.
Delineation of Study Region
The study region was determined based on regional hydrology as well as the availability of USGS gauge stations within those watersheds. All sub‐basins delineated within the eight larger watersheds are shown in Figure 2, with USGS gauge stations with daily discharge data shown as black dots. We subcategorized the region into eight watersheds (Foley 1, Foley 2, Foley 3, Mobile 1, Mobile 2, Mobile 3, MS 1, and MS 2), which are further comprised of 1,291 sub‐basins delineated within them. Complete watersheds are included in the study region, ensuring that upstream effects are captured. The number of sub‐basins is based on the number of streams in the watershed, where a stream is defined from the minimum area draining to a point that becomes the starting point of that stream. Each sub‐basin will therefore have one stream. Two subsections were in MS, represented as greens; three were in the region of Mobile, AL, represented in beige, brown, and yellow; and three were in the Daphne‐Fairhope‐Foley region of AL, represented as blues. All 1291 sub‐basins in the study region were included, irrespective of availability of their comprehensive record of mean annual peak flow from 2014 to 2016.
Study area showing watersheds, sub‐basins, and location of USGS gauges in the MS‐AL coastal region.
Environmental and Landscape Factors
Following the study area delineation, the general characteristics of the eight watersheds were estimated by collecting comprehensive environmental data and landscape patterns for each of the 1,291 sub‐basins. Various sources of data were used to develop an understanding of several environmental factors, including the mean annual peak unit discharge, daily mean annual precipitation, daily temperature, daily peak runoff, and landscape factors such as land uses, average watershed slope, average watershed soil permeability, floodplain area, natural drainage density, wetlands, and impervious rate. The criteria for selecting specific landscape and control variables were based on their inclusion in previous similar studies that have analyzed the relationship between landscape and development metrics with stormwater (Brody et al., 2013; Deng et al., 2023; Kim & Park, 2016; Liu et al., 2020; Miller et al., 2020; Rasoulzadeh et al., 2023; Shuster et al., 2005; Zhang et al., 2015; Zhu & Newman, 2021; Zhu et al., 2022). Runoff depth is a widely used hydrologic term, including by the USGS and the National Oceanic and Atmospheric Administration, that has been used in earlier studies, including Kim and Park (2016). It is the total runoff from a drainage basin divided by its total area. Peak runoff depth is the peak runoff divided by the watershed area recorded in a certain time interval. This study uses a hydrological metric, mean annual peak unit discharge, to represent the average depth of water during the maximum flow that runs off a particular drainage area during the highest runoff in a typical year. Mean annual peak unit discharge (in mm) was estimated by determining the peak annual runoff (m3/s) from monthly average runoff data, divided by the watershed area and multiplied by the time period of one year. Mean annual peak unit discharge was then averaged for a watershed across multiple years (2014–2016).
The average precipitation is the yearly rainfall that falls in a watershed in a year and is estimated by taking the average of the total precipitation recorded over a long period of time at that watershed. The average watershed slope represents the elevation changes across the watershed area and is calculated using a topographic map or digital elevation model. Average watershed soil permeability is the typical rate at which water can move through the soil within a specific watershed. Floodplain areas are lands along a river or stream that are prone to being inundated by floodwaters when the river overflows its banks. Natural drainage density indicates how densely distributed the network of streams in a watershed is; it is estimated by dividing the total length of natural streams in a watershed by its area. Wetlands are low‐lying land areas that are permanently or temporarily saturated with water. Lastly, the impervious rate indicates the proportion of a land area covered by built surfaces like concrete, asphalt, or rooftops that does not allow water to infiltrate the soil.
Stormwater Runoff Estimate Using SWAT
For the 1,291 sub‐basins in the study region, there were very few working gauges, resulting in insufficient data for completing an analysis of peak streamflow and runoff data. Several methods that exist for estimating stream flow through an ungauged basin within a watershed were considered for this research. The SWAT has been widely used for discharge modeling over the past two decades. For this study, the lack of complete peak flow data was addressed by modeling stream flow at ungauged river basins using SWAT to estimate the amount of peak stormwater runoff during the three‐year period from 2014 to 2016, with runoff data being collected for a total period of 12 years. Using SWAT advances our understanding of the measurement of runoff in landscape pattern planning for stormwater management in the absence of available runoff data. Whereas the Kim and Park (2016) study excluded areas without observed runoff data, the uniqueness of this study is that it used simulated runoff data obtained through the SWAT model to estimate runoff for areas without actual runoff readings recorded. This allowed runoff from the entire region to be analyzed, making this study truly regional.
Research in the development and application of SWAT has grown internationally in a wide range of settings and applications. Previous studies using the SWAT tool provided guidance on the model application, specification, and parameterization. Arnold et al. (2012) discussed recent trends in literature and provided a comprehensive overview of the multiple components of the SWAT model. Similarly, Douglas‐Mankin (2010) synthesized performance statistics and parameters from a review of 20 research articles to present the developments and applications of SWAT. In two case studies in the Cedar Creek watershed in TX, Cibin et al. (2014) found that by applying the SWAT model in a pseudo‐ungauged basin, the measured streamflow lies well within the estimated confidence band of predicted streamflow. Also, for two locations in TX, Santhi et al. (2001) discussed calibration and validation methodologies over several years and similarly found that predicted values matched well with the observed values. Another study in Woodlands, TX, simulated five hypothetical land use configurations in SWAT and explored the urbanization of watershed hydrology to understand the degree to which land development followed McHarg’s ecological planning approach (Yang & Li, 2011). A study by Sisay et al. (2017) in Vadodara, India, also found a good match between simulated and observed flow using the SWAT model in ungauged watersheds, indicating that it has strong predictive ability for the ungauged watersheds. Increasingly, SWAT has been used for watershed management planning and simulating various scenarios (Douglas‐Mankin 2010), making it a useful tool for analysis.
Landscape Pattern Analysis
Our Fragstats approach is adopted and expanded from the Kim and Park (2016) study. In this analysis, we have considered the National Land Cover Database (NLCD) 2016 Land Cover classifications for our study area. NLCD was reclassified the into the following land cover types: Water (NLCD code 11), Developed (NLCD Codes 21, 22, 23, and 24), Barren Land (NLCD Code 31), Vegetation (NLCD Codes 41, 42, 43, 52, and 71), Agriculture (NLCD Codes 81 and 82), and Wetland (NLCD Codes 90 and 95). Vegetation and wetlands were considered green infrastructure, and a range of landscape metrics were measured for each of the 1,291 sub‐basins. As shown in Table 1, measurements are taken for each sub‐basin as a whole (landscape level metrics) as well as individually for each of the five land cover types (class‐level metrics), producing a great amount of detailed and disaggregated information about landscape composition and configuration for each sub‐basin.
Landscape‐Level and Class‐Level Metrics Measured
Fragstats interprets NLCD data according to four levels of analysis: Cells (an individual pixel), Patches (contiguous pixels of the same cover type), Classes (the sum set of all patches of the same type within a landscape), and Landscapes (the sum of all patches within the study area). This study defined each sub‐basin as a landscape, so each of the 1291 sub‐basins was analyzed across all levels of analysis. Parameterization of the Fragstats software was limited to only the necessary calculations for selected metrics. Out of the total potential metrics available in Fragstats, landscape metrics were chosen based on their inclusion in other relevant literature. The input GIS Data for Fragstats analysis was the reclassified NLCD 2016 clipped to the target study area.
Size and Edges were observed by a Percentage of Landscape (PLAND) and Edge Density (ED). The high value of PLAND and ED indicate a landscape with larger and complex shapes. A fundamental measure of landscape composition, PLAND calculates the percentage of a particular patch type. It provides important information showing the richness of some types of patches (such as proportion of vegetation, wetlands, etc.) and therefore has frequently appeared in previous studies (e.g., proportion of vegetation or impervious rate). However, it is less useful in showing the spatial distribution and only indicates how much of the landscape is comprised of a particular type. A large value of PLAND for a GI land cover class (such as Woody Wetland) would be expected to reduce peak runoff. A large value of PLAND for an urbanized land cover class (e.g., Developed 20‐50) would be expected to increase runoff.
Edge Density (ED) standardizes edge to a per unit area basis that facilitates comparisons among landscapes of varying sizes. A large value would be expected to reduce mean and peak runoff because of the increased complexity and convolution. ED measures the ratio of the total perimeter of patches to a unit area and considers the shape and size of patches. Figure 3 illustrates high and low ED values. A high ED value for a GI land cover class is expected to reduce mean and peak runoff because of increased complexity and convolution along the edges. Shape metrics describe whether patches have convoluted shapes. Shape Index (SHAPE) and Congruity Index (CONTIG) are used to make shape measurements. SHAPE calculates the patch perimeter and area simultaneously, while CONTIG assesses patch shape based on spatial connectedness or contiguity of cells within a patch. When a patch is more convoluted in shape, its SHAPE and CONTIG values increase.
Study region showing (clockwise from top left): Slope, Soil permeability, Wetlands and Impervious rate, and Natural drainage density.
Isolation metrics help identify the tendency for patches to be isolated in space from other patches. Proximity (PROX) and mean Euclidean Nearest‐Neighbor Distance (ENN) are used to identify the level of isolation or nearness of landscapes. The Proximity Index integrates information on the size and distance of like patches from the specified “focal patch” within a designated search radius. Nearby large patches contribute greater value to PROX than nearby small patches and increase when a specific patch type is near the same type of patches. PROX measures within the search radius the distance between the focal patch and neighboring patches of the same type and size. An 800‐meter radius was employed as a search radius since previous studies frequently adopted this value (Brody et al., 2013; Rylands et al., 2006). A high value in PROX of GI patches would be expected to reduce peak runoff, but the same PROX value in urban patches would be expected to increase it. Mean ENN measures the shortest Euclidean distance from one patch to another patch of the same land cover type. Since it is the average of the straight‐line distance between the nearest neighbor patches of the same type, it reveals the relative location and arrangement of patches of a particular type. The closer together GI patches are, the lower the ENN. A high ENN value for a GI patch would be expected to increase peak runoff.
Connectivity refers to the functional and spatial connectivity among patches. The connectivity of patches is measured by cohesion (COHESION) and connectedness (CONNECT). COHESION calculates the physical connectedness of the corresponding patch types in an area. Patch cohesion index quantifies the connectivity of a land cover type. All other land cover types are conceived as background cells. Patch area and perimeter are used to calculate the percentage of connectedness for a given patch type. A high value of a GI land cover type would be expected to reduce peak runoff. CONNECT measures each pair of patches that are either connected or not within a certain search radius. The Connectance index measures the proportion of functional joining among all patches, where a pair of patches is deemed connected if they are closer than a given distance as specified by the user. A high value for CONNECT for a GI land cover type would be expected to decrease runoff.
Regression Analysis
The main goal of this study was to understand how regional landscape patterns affect stormwater runoff. The association of landscape patterns with the variations in the mean annual peak unit discharge was analyzed using OLS regression while accounting for a wide range of environmental factors. Four distinct models are developed for each of the four different types of landscapes (Barren lands, Vegetation, Wetlands, and Development) to provide insights on whether desirable landscape metrics apply uniformly to different types of regional landscapes or whether the effect of different landscape metrics varies by the type of landscape. This also helped in the analysis of how beneficial individual landscape metrics are. For instance, a metric such as CONTIG may be beneficial in terms of desirable GI land cover such as vegetation or wetland, but for barren and developed lands, CONTIG may have an opposite effect.
For OLS regression, the mean annual peak unit discharge, estimated using the SWAT model with ArcGIS, was considered the dependent variable. A range of environmental variables were measured and incorporated into the OLS models to control for their effect, including mean annual precipitation, average watershed slope, average watershed soil permeability, floodplain area, natural drainage density, proportion of wetland, and impervious rate.
Results
This section describes the outputs and results obtained for data measurement, stormwater runoff estimates, landscape pattern analysis, and OLS regressions.
Environmental Factors
Environmental factors measured include mean annual peak unit discharge, mean annual precipitation, average watershed slope, average watershed soil permeability, floodplain area, natural drainage density, wetlands, and impervious rate. Table 2 describes these variables using measurement, data sources and tools used, and descriptive statistics.
Data Sources and Measurements
Model for Stormwater Runoff Estimation
Each of the 1,291 sub‐basins was modeled separately using the SWAT tool in ArcGIS version 10.5 following the process shown in Figure 3. The input data required for the model included the digital elevation model (DEM) of the watershed area, land use and land cover dataset, soil type dataset and climate data (specifically daily maximum and minimum temperatures as well as daily precipitation values). These datasets are easily available through the Geospatial Data Gateway (USDA) and the PRISM database (PRISM Climate Group). Other than these data, the discharge data needed for model calibration and validation was obtained from the USGS National Water Information System (USGS). The statistical parameters that were used to check model performance included R2, percent bias (Pbias), and Nash Sutcliffe Efficiency (NSE). The SWAT model was simulated for a period of 12 years from 2005 to 2016 at the monthly time step. This 12‐year simulation period was divided into three time periods: the Initial five years (2005–2010 data) were used as the warm‐up period but not in the calibration‐validation process, the Middle years (2010–2014 data) were used for model calibration, and the Last three years (2014–2016 data) were used for validation period and considered for analysis. Table 3 shows the SWAT model results.
SWAT Model Simulation
The NSE values obtained in the model range from 0.81 to 1.00. NSE values can range from 0 to 1, with higher values indicating better results. pBIAS values obtained in the model ranged from −7.27 to 9.17. pBIAS values can range from −100 to 100 and in an ideal case will be 0. Obtained R2 values in the model ranged from 0.63 to 0.91. The simulation results are considered acceptable if R2 is more than 0.5 (Moriasi et al., 2007), the absolute value pBIAS is less than 25 (Moriasi et al., 2007), and the NSE value is greater than 0.5 (Hallouz et al., 2018).
Landscape Pattern Analysis
The landscape pattern analysis utilized the following metrics: Percentage of Landscape, Edge Density, Shape Index, Contiguity Index, Proximity index, Euclidean Nearest‐Neighbor, Patch Cohesion Index, and Connectance Index. Additional landscape metrics were also examined, such as the Shannon’s Diversity Index, a tool that measures landscape composition or diversity of land cover types in a region. Table 4 describes a few landscape measurements computed from Fragstats, and a few illustrations of landscape patterns are shown in Figure 4.
Illustrations of Edge Density output (top two images), Proximity Index output (middle two images), and Shannon Diversity Index output (bottom two images). Low values are on the left and high values are on right.
Landscape Measurements
Regression Analysis
Four separate models were developed, one each for Barren lands, Vegetation, Wetlands, and Development. The results explain some of the correlations between different green infrastructure land uses in terms of their size, shape, isolation, clustering, and connectivity with local flooding or peak stormwater runoff while controlling for several environmental conditions. The dependent variable is the mean annual peak unit discharge, measured as the average maximum daily runoff for the period 2014–2016 by water year (natural log‐transformed). As expected, in each of the four models, average precipitation (p value = 0.0000) and natural drainage density (p value = 0.0000) were found to be the strongest predictors of runoff. The higher the rainfall, the higher the peak runoff, and increased natural drainage also increases runoff. Soil permeability and floodplain area were not found to have a significant association with runoff in any of the four models. Similarly, slope was also not a significant variable in any of the models; however, they show a positive relationship between slope and runoff. In three of the four models, the percentage of wetlands was found to have a significant negative association with runoff. As expected, the higher the percentage of wetlands in the sub‐basin, the lower the expected runoff. Percentage of impervious areas was found to be significant only in the Vegetation model but had the expected positive relationship with runoff in all models. The regression output for specific landscape metrics is given in Table 5.
Summary of Regression Outputs
The results of the four models help identify the type of landscape patterns that should be protected and the type of development that should be promoted at the regional scale to mediate the impacts of flooding. The Wetlands model shows that an increased proportional size of wetlands and more connected wetlands is significantly associated with a reduction in runoff. Increase in percentage of landscape (wet_PLAND) for wetlands has a significant reduction impact on runoff (β = −0.0008, p value = 0.1009), meaning that a greater share of wetlands in the landscape is associated with reduced runoff. Better connected wetlands (wet_CONNECT), measured at the landscape level, are also significantly associated with a reduction in runoff (β = −0.0003, p value = 0.1378). This is an expected finding as highly clustered and abundant wetlands provide significant flood mitigation benefits.
The Vegetation model shows that increased proportional size of vegetation and more connected vegetation are highly significant and associated with reduced runoff. Proximity Index Mean, a landscape level metric for vegetation, is also significant and associated with reduced runoff. Percentage of landscape for vegetation (Veg_PLAND) is highly significant and is associated with reduced runoff (β = −0.0011, p value = 0.0147). Proximity Index Mean for vegetation (Veg_PROX_MN, β = −0.0000, p value = 0.1183) and connected vegetation (veg_CONNECT, β = −0.0004, p value = 0.1108) are both highly significant, indicating that highly connected vegetation is associated with reduced runoff. These are expected findings as highly clustered, abundant, and connected wooded vegetation is expected to reduce runoff.
In the Barren Lands model, only Euclidian Nearest Neighbor (Brn_ENN) seems to have any association with runoff; however, its effect seems to be quite negligible (β = −0.0000, p value = 0.0862). No other landscape measure for barren lands was found to have a significant effect on runoff, indicating a lack of association between the shape, size, and configuration of barren lands with runoff. Since barren lands are devoid of vegetation, they have minimal potential for reducing runoff.
In the Development model, proximity of development, as measured by the Mean Proximity Index (Dev_PROX_MN), shows a significant positive association with runoff (β = 0.0001, p value = 0.0998). Further, connectivity of development (Dev_CONNECT) also showed a significant negative association with runoff (β = −0.0004, p value = 0.0544). The results indicate that proximity to development increases runoff, possibly also increasing the risks to life and property. More connected development was found to be negatively associated with runoff as compared to scattered development pattern. This is an unexpected finding as more connected development is expected to increase runoff as it contains larger impervious surfaces that are expected to result in increased stormwater runoff. In contrast, scattered development with more natural vegetation is expected to increase ground water infiltration and reduce runoff.
Discussion
Green or natural infrastructure is increasingly seen as a cost‐effective strategy to strengthen coastal defense, thus reducing the negative impact of storm surge and waves on coastal areas. Wetland restoration and rich coastal ecosystems like coral reefs, oyster reefs, mangroves, and salt marshes not only protect the coast but are dynamic ecosystems that also adapt and grow with their changing environment (Reguero et al., 2018). Covering more land areas with wooded vegetation such as trees and shrubs and implementing wetland conservation programs will help increase the infiltration and storage of rainwater locally and help in reducing runoff and local flooding. It is evident that regional landscapes that are more connected and less fragmented, specifically consisting of vegetation and wetlands, are likely to reduce runoff. In contrast, scattered and fragmented landscapes will have less capacity to store rainwater and may have higher potential to result in local flooding (Brody et al., 2017).
Results obtained for GI landscapes such as vegetation and wetlands are broadly consistent with the Kim and Park (2016) results. Their study found that the size, connectivity, and convoluted patterns directly affect peak runoff, strengthening the argument for the conservation and enhancement of complex and regional landscapes. Our results suggest that more connected developments are associated with reduced runoff. This finding is contrary to what Kim and Park (2016) found, that more clustered and connected development was likely to augment runoff. This may be because the Kim and Park (2016) study focused on the metropolitan scale (and excluded ungauged basins), but our research covered eight large watersheds, making it truly regional in scale. At the regional scale, compact development consumes less land and preserves vast open and natural areas that are known to absorb stormwater and reduce surface runoff. On the other hand, sprawling development results in fragmented open spaces, possibly increasing flood losses (Brody et al., 2017). Miller et al. (2020) also stress the importance of connectivity and location of urban surfaces in hydrological modeling studies. Our results are consistent with the Brody et al. (2013) study, which found that more connected and concentrated development was associated with reduced observed flood losses, but they also warn that compact development should occur away from the floodplains as increased concentration of impervious areas may increase runoff in the floodplain.
These findings illustrate that local, fragmented approaches to wetlands and wooded vegetation conservation will be less effective compared to solutions developed at the regional or macro level. Thus, planning for GI needs to extend beyond local jurisdictions and local plans to consider overall networking and clustering at the regional scale. Regional planning agencies should collaborate with local agencies to optimize landscape size, connectivity, configuration, and composition across the region (Liu et al., 2020). Regional and state‐level policies that consider the role of GI in protecting coastal areas from catastrophic environmental events will be more effective than locally enforced policies. Therefore, regional planning agencies should take the lead in coordinating with local governments to develop and implement overall networking and clustering plans for creating sustainable GI. This will help identify targeted areas where GI should be strategically preserved for creating and maintaining larger and connected wetlands and vegetation. Plans can be dovetailed into local and regional comprehensive land use plans for implementation. For effective GI planning, scaling up efforts to integrate GI into existing planning processes is recommended through the use of: planning tools, such as GI‐specific plans and comprehensive plans; regulatory tools, such as zoning and building codes and stormwater ordinances; incentive‐based tools, such as grants, subsidies, and stormwater fee adjustments; and government operations, such as efforts involving public infrastructure, land, or facilities (Georgetown Climate Center).
A notable example of integrated GI planning is Maryland’s Greenprint program, which flows from the state to the county level cohesively (Maryland Department of Natural Resources). The program identified 39% of the state’s land as Targeted Ecological Areas (TEAs), important ecological lands (large forests, aquatic biodiversity hotspots, and rare species habitats) eligible for state conservation funding. TEAs are further incorporated into multiple county plans (Green Infrastructure Network; Anne Arundel County, 2022) that detail a GI network of hubs and corridors, ensuring consistency with Maryland’s Greenprint program. Regional GI planning is also included in Virginia Beach and Chesapeake comprehensive plans, watershed management, parks and recreation, sustainability, and water quality implementation plans (Chesapeake Planning Department, 2014; City of Virginia Beach, 2016). Many cities specify GI objectives in their local plans. For instance, the Chicago ON TO 2050 Regional Plan aims to use GI in public parks and rights‐of‐way, protecting ecological cores and landscapes (Chicago Metropolitan Agency for Planning, 2018). Milwaukee’s Greenseams Flood Management Program protects over 5,000 acres of open spaces using conservation easements in flood‐prone areas and provides guidance on the ways to incorporate appropriate GI planning (The Conservation Fund). While AL and Georgia provide some guidance on low impact development for new developments to better manage runoff and prevent local flooding (Dylewski et al., 2014; Georgia Department of Natural Resources, 2014), the focus on protection of regional resources is inadequate.
The relationship between land development patterns and runoff requires further exploration as results are inconsistent across various studies. Our results indicating that more connected developed land was negatively associated with runoff is inconsistent with Kim and Park (2016) but generally consistent with Brody et al. (2013). Future research could also suggest appropriate models for regional frameworks and dovetailed local plans incorporating conservation measures. In fast‐growing coastal areas of AL and MS, forests and wetlands compete with other land uses such as development and infrastructure. At the regional scale, compact urban development that consumes a smaller share of land has the potential to make available the rest of the undeveloped lands for conservation. This points to a need for flexible land use planning approaches that accommodate new developments at different spatial settings, away from floodplains, while protecting and advancing interconnected ecosystems at a regional scale.
While the SWAT model has been widely used, this study demonstrated that also applying Fragstats expands the methodological possibilities for estimating runoff, identifying desirable landscape patterns, and simulating the impact of land management practices on stormwater management. Unlike Kim and Park (2016), whose study area of four separate metropolitan areas excludes ungauged basins, this study is truly regional in scale as it applies the SWAT tool to also estimate runoff from ungauged basins, resulting in coverage of eight complete watersheds. An ArcGIS interactive map of the partner communities’ area showing USA NLCD land cover, along with a gallery of base maps and a video tutorial, has been shared with relevant AL and MS planners and decision‐makers. The USA NLCD land cover data are available at https://aub.ie/landcover_map. These resources can be used by the communities to develop flexible plans that can accommodate new development at different spatial settings while protecting urban ecosystems for recreation and the environment.
Conclusion
This study investigated landscape patterns in the coastal regions of AL and MS and estimated the impact of these patterns on stormwater runoff. Importantly, this research makes a significant methodological contribution by innovatively applying the SWAT tool to estimate runoff from ungauged basins, advancing the methodological approaches for identifying desirable landscape patterns for stormwater management. Generally, the findings suggest that the proportion and size, fragmentation, and connectivity of GI landscapes such as vegetation and wetlands matter at the regional scale and affect peak runoff and local flooding. Proportionally larger, less fragmented, and highly connected GI landscapes are associated with decreased peak runoff, and more connected developed lands are associated with reduced runoff, strengthening the case for compact urban development that consumes less land.
While traditional land conservation and GI planning focus on environmental restoration and preservation, this research provides guidance for coastal and regional planning agencies to concentrate on the space, shape, and location of green infrastructure, paying attention as well to the relationship of new development to natural resources and amenities. It is intended to help communities become more resilient by mediating flood damage while at the same time providing opportunities for regional‐scale recreational spaces through well‐connected open spaces and compact development. Other benefits of GI protection include heat mitigation, better wildlife movement, and more connected wildlife populations between habitats. This research extends our knowledge on desirable landscape and development patterns for open spaces, greenspaces, floodplains, and impervious surfaces within a regional setting; it also provides support for interconnected and clustered GI planning for better managing stormwater.
Peer Review
Landscape Journal uses a double‐blind peer review process for research manuscripts, systematic literature reviews, and other article types.
Acknowledgments
This publication was supported by the U.S. Department of Commerce’s National Oceanic and Atmospheric Administration under NOAA Award NA18OAR4170080, the Mississippi‐Alabama Sea Grant Consortium, and Auburn University. The views expressed herein do not necessarily reflect the views of any of these organizations. The authors thank the three anonymous reviewers for providing valuable feedback and Dr. Latif Kalin of Auburn University for reviewing the hydrological aspects of this article.
This open-access article is distributed under the terms of the CC-BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0) and is freely available online at https://lj.uwpress.org.
References
Jump to section
Related Articles
Cited By...
- No citing articles found.