1 Introduction

Hydrologically sensitive areas (HSAs) are the parts of a landscape that produce disproportionate amounts of surface runoff. HSAs can be targeted to effectively implement planning and management measures to control nonpoint source pollution (Amin et al. 2017; Qiu et al. 2019) and mitigate negative impacts of stormwater runoff (Qiu et al. 2014, Martin-Mikle et al. 2015; Kaykhosravi et al. 2019). Topographic indices (TIs) have long been used to parsimoniously delineate HSAs based on the concept of variable source area (VSA) hydrology (Walter et al. 2000; Heathwaite et al. 2005; Qiu 2009). Specifically, the TI values mimic the sequential distribution of water storage capacity of a landscape allowing identification of runoff-generating areas (Dahlke et al. 2009). As such, the locations in a landscape with TI values greater than a given threshold delineate HSAs. There are several mechanistic ways to define such a threshold for delineating HSAs in landscapes, including the controlling specific storm event method (Lyon et al. 2006) and the average saturation probability method (Agnew et al. 2006), but these methods require ancillary information about streamflow or runoff production for calibration (e.g. Lyon et al. 2004). Herron and Hairsine (1998) suggested HSAs should be 20% of watershed area with the highest saturation potential. Qiu (2009) used a static TI threshold of 10 for HSA delineation in Raritan River Basin in central New Jersey.

Identification of a proper TI threshold for HSA delineation faces two practical difficulties. The first difficulty is the use of a priori TI derivation, given the variety of different TIs available (Buchanan et al. 2014). TIs vary significantly by the physical attributes being considered and the methods being used to derive those attributes. For example, Buchanan et al. (2014) compiled and compared over 400 unique TI formulations that considered different digital elevation model (DEM) resolutions, vertical precision of the DEMs, flow direction and slope algorithms, and smoothing via low-pass filtering. They found that a soil topographic index (STI) derived using a fine-scale LiDAR DEM with additional of soil information was a better indicator to represent soil moisture conditions in central New York relative to other formulations.

The second practical difficulty is the lack of the empirical validation of the delineated HSAs. Giri et al. (2017) conducted Trellis plot analyses based on a polynomial regression model (order two to four) to analyze the relationship between TIs and the changes in soil moisture measured in two fields. They concluded that the TI threshold for delineating HSAs can range from 9 to 15 when considering a STI in Central New Jersey. Archibald et al. (2014) found good agreement between HSAs mapped using STIs and the observed saturated areas measured through groundwater table positions. However, these empirical studies looking at defining TI thresholds for delineation of HSAs were conducted at field or small-watershed scales. While large-scale patterns of saturated areas or runoff contributing areas can be estimated using vegetation patterns (Kulasova et al. 2014) or satellite remote sensing images (Temimi et al. 2010; Lei et al. 2016), such approaches have thus far not been used to corroborate TI thresholds for HAS delineation from a regional perspective in a manner meaningful to landscape managers.

There is clear need to identify proper TI thresholds for HSA delineation that are scientifically defensible and are more appealing to landscape managers. The objective of this study is thus to identify TI thresholds by comparing the spatial patterns of various delineated HSAs to the spatial pattern of floodplains. While HSAs are the hydrological hotspots in a landscape prone to generate surface runoff, floodplains are the areas that are inundated in a landscape during a storm event. Despite these definitional differences, both are closely related as the floodplains are often an important part of HSAs and result from runoff accumulation from runoff generation associated with upland areas within the HSAs. As such, the delineated HSAs could approximate the runoff-contributing areas to the floodplains and offer a meaningful spatial extent for landscape managers to direct their planning and management actions designed to mitigate flood and other water-related impacts. To our knowledge, this study is novel as it is the first study that provides a systematic approach for identifying the threshold associated with delineating HSAs for water resources management and planning at large (regional to state) scales. The procedure not only is useful to select the TI threshold, but also helps refine the spatial extent of landscape planning and management to manage flood hazard.

2 Data and Methods

We consider two popular TIs in this study: a topographic wetness index (TWI) based on topography and a soil topographic index (STI) based on both topography and soil. Further, we target New Jersey, USA as a case study given concerns on flooding and water resource management (e.g. Lyon et al. 2018). The entire state of New Jersey can be divided geologically into northern and southern parts defined by what is commonly known as the Fall Line. North of the Fall Line are the Valley and Ridges, the Highlands, and Piedmont Physiographic Provinces. South of the Fall Line is the Coastal Plain Physiographic Province (Watt 2000). The total land mass of New Jersey (excluding the coastal waters and barrier islands) is about 2.02 million hectares. To effectively manage the state’s water resources, New Jersey Department of Environmental Protection (NJDEP) divided New Jersey into 20 different watershed management areas (WMAs). NJDEP also defines five different water regions: Lower Delaware, Upper Delaware, Raritan, Atlantic Coast, and Northeast, which represent 24%, 16%, 16%, 31%, and 12% of the land mass in New Jersey, respectively. Each water region is comprised of three to five WMAs (Fig. 1a). Most of the streams and rivers in the Upper Delaware water region and all of the streams and rivers in the Lower Delaware water region drain into the Delaware River, which drains into the Atlantic Ocean. The rivers in the Raritan, Atlantic Coast, and Northeast water regions drain directly into the Atlantic Ocean (Watt 2000).

Fig. 1
figure 1

The distribution of water regions and watershed management areas in New Jersey

Full size image

The following geospatial data were collected and processed for the study: (1) 10-m digital elevation models (DEM) for each of the 20 WMAs in New Jersey from NJDEP; (2) the digital soil survey geographic (SSURGO) databases for each of 21 counties in New Jersey from Natural Resources Conservation Service (NRCS); and (3) the 100-year floodplain map in New Jersey from Federal Emergency Management Agency (FEMA). The data in different coordinate systems were transformed to the New Jersey standard coordinate system for spatial data processing.

The HSAs delineated in this study were compared to the FEMA 100-year flood zone in New Jersey for the state as a whole and for each of the five different water regions (Lower Delaware, Upper Delaware, Raritan, Atlantic Coast and Northeast) with diverse topographic and soil conditions. The FEMA 100-year flood zone is the floodplain area estimated by FEMA to have a 1 % chance of being flooded at least once in a year. Further, the FEMA 100-year floodplain is a standardized product available for the contiguous United States for various applications, making it a viable product for the purpose of comparison and validation.

2.1 Topographic Indices

Topographic indices (TIs) measure the contribution of upslope areas and the ability to transfer water downslope at a given position in a landscape. Under variable source areas assumptions, TIs also express the likelihood that a location (or point) in a landscape generates saturation excess runoff. This study focuses on two widely used TIs: TWI and STI. The TWI was based on the following equation (Beven and Kirkby 1979):

$$ TWI=\mathit{\ln}\left(\frac{\alpha }{\mathit{\tan}\left(\beta \right)}\right) $$
(1)

where α is the upslope contributing area per unit contour length (m) and β is the local topographic slope (m/m). Both α and β are topographically dependent and were derived from available DEMs. In addition to topography, a STI considers the spatial variation in hydrologically relevant soil properties (Walter et al. 2002; Anderson et al. 2015) and was calculated using:

$$ STI=\mathit{\ln}\left(\frac{\alpha }{\mathit{\tan}\left(\beta \right){K}_sD}\right)=\mathit{\ln}\left(\frac{\alpha }{\mathit{\tan}\left(\beta \right)}\right)-\mathit{\ln}\left({K}_sD\right) $$
(2)

Where α and β are defined above, Ks is the average saturated hydraulic conductivity (m/day), and D is the depth to the restrictive layer (m). The product of average saturated hydraulic conductivity and depth to restrictive layer terms in Eq. (2) is considered soil transmissivity and measures how water can be transmitted horizontally through soils.

TWI for the entire State of New Jersey was derived from the 20 WMA DEMs at a 10-m resolution using the System for Automated Geoscientific Analyses (SAGA) geographic information systems (GIS), which is in a public domain (Conrad et al. 2015). The DEMs at a WMA level were merged to form five larger DEMs for five water regions following the delineation by NJDEP. These five water regions were then merged to form the DEM for New Jersey. To ensure a continuous flow in each grid of the DEM, depressions were filled using the GIS function fill sinks (Planchon and Darboux 2001). The least square fitted plane method (Horn 1981; Costa-Cabral and Burgess 1996) was used to calculate the slope of each grid cell in the DEMs while the multiple triangular flow direction method (Seibert and McGlynn 2007) was applied to obtain the catchment area. The resulting slope and catchment area layers were used to obtain the TWI for the state. Since the grid size in the DEMs was measured in feet, a correction factor of −1.191387, i.e. ln(0.3038), was added to the resulting TWI to get the TWI measured in meters as defined by Eq. (1).

Soil transmissivity for each county in New Jersey was calculated using the soil saturated hydraulic conductivity and topsoil depth. Both were extracted from the SSURGO database using a soil data viewer, which is an add-on ArcGIS tool developed and maintained by NRCS. Using the aggregation methods within the ArcGIS soil data viewer, the topsoil depth to a restrictive layer and the geometrically weighted average soil saturated hydraulic conductivity layers were created. Both layers were multiplied to calculate the soil transmissivity value. Since the topsoil depth was given in cm and the soil saturated hydraulic conductivity in μm/s, a correction factor of 0.000864 was multiplied to the soil transmissivity values to convert the unit into m2/day as defined in Eq. (2). The second term in Eq. (2) was calculated by taking the natural logarithm of the resulting soil transmissivity values. Some polygons in the soil map were classified as water, rough broken land, shale, quarry, and pits, sand and gravel, for which there was no attribute value on soil saturated hydraulic conductivity in the soil database. Since these soil classifications did not have any soil transmissivity, the grids with these soil classifications were assigned a value of −2, which approximated the lowest value of ln(KsD) calculated in New Jersey.

The resulting soil transmissivity layers for all counties were merged into a single transmissivity layer for New Jersey. The soil transmissivity layer was added to the statewide TWI to form the STI layer for New Jersey based on Eq. (2). The TWIs and STIs for the five water regions in New Jersey were clipped from the statewide TWI and STI layers based on the boundaries of water regions to ensure consistency when comparing across the regions.

2.2 Hydrologically Sensitive Areas Delineation

Hydrologically sensitive areas (HSAs) can be defined as the areas in a landscape with TI values greater than a given threshold. To explore the impact of threshold selection, this study adopted a wide range of thresholds from 8 to 14.5 with interval of 0.5 resulting in 14 different thresholds for delineating HSAs. Given two TIs (TWI and STI) and six regional scales (five water regions plus the state), this led to 168 maps of potential HSAs delineated for comparison with the FEMA map. The FEMA 100-year floodplain map was first rasterized using the statewide DEM as a snap layer so that the resulting floodplain raster grids were aligned with the TWI and STI grids. The statewide FEMA floodplain layer was then clipped into regional FEMA floodplain layers based on the boundaries of water regions. There were six FEMA floodplain layers (five water regions plus the state). Each FEMA floodplain layer was compared to the different delineated HSAs to determine the suitability of the TIs and the various thresholds for delineating HSAs representative of the 100-year floodplain.

2.3 Spatial Pattern Comparison Indicators

An error matrix can be used to conceptualize the comparison between a delineated HSAs (Map 1) and a FEMA floodplain map (Map 2) with two categories: in the HSAs or FEMA floodplain as indicated by “yes” (positive or 1) and outside as indicated by “no” (negative or 0) (Fig. 2). In such an error matrix, the notation nij indicates the number of cells with category i in Map 1 and category j in Map 2 and i and j are either “1” or “0”; N1 and N0 are the numbers of cells within and outside the HSAs in a region; M1 and M0 are the numbers of cells within and outside the FEMA floodplains in a region; and N is the total number of cells in a region. Kuhnert et al. (2005) assess several spatial pattern comparison indicators based on such error matrix values. Four of these pattern comparison indicators were used here to assess the consistency between the delineated HSAs and the FEMA floodplain.

Fig. 2
figure 2

The error matrix for comparing the FEMA 100-year floodplain and the delineated hydrologically sensitive areas

Full size image

2.3.1 Hit Rate

The hit rate measures the percentage of cells in the FEMA floodplain map that are also identified as a part of HSAs. The hit rate indicates how successfully the HSAs would predict the floodplain. The hit rate simply indicates how well the model (the HSAs here) replicates the benchmark data (the FEMA floodplain here) without penalizing for overprediction (Sampson et al. 2015). Given the error matrix in Fig. 2, the hit rate is calculated as:

$$ {R}_{hit}=\raisebox{1ex}{${n}_{11}$}\!\left/ \!\raisebox{-1ex}{${M}_1$}\right.. $$
(3)

2.3.2 Agreement Rate

A cell-by-cell comparison takes a cell in the HSA map and matches it with the corresponding cell in the FEMA floodplain map. In such a comparison, a match (either a positive or negative match) is counted as 1, and a mismatch is 0. The agreement rate is the number of matched cells divided by the total number of cells being compared (Kuhnert et al. 2005). Based on the error matrix in Fig. 2, the agreement rate based on the cell-by-cell comparison is calculated as follows:

$$ {R}_{cbc}=\left({n}_{11}+{n}_{00}\right)/N. $$
(4)

2.3.3 Overall Comparison Rate

The overall comparison is an improvement over the cell-by-cell comparison by measuring the similarity after taking into consideration the disagreement measured by the differences in the number of cells in each category (Pontius 2002; Kuhnert et al. 2005). Given the error matrix and notations defined in Fig. 2, the overall agreement rate based on the overall comparison is calculated as follows:

$$ {R}_{oall}=1-\frac{1}{N}\sum \limits_{i=0}^1\left|{N}_i-{M}_i\right|=1-2\mid {n}_{10}-{n}_{01}\mid /N. $$
(5)

2.3.4 Kappa Value

The kappa value measures the agreement of two spatial patterns by taking away some of the agreement that likely occurred by chance (Cohen 1960) and has been widely used for spatial pattern comparison. Given the error matrix in Fig. 2, the chance agreement can be measured as

$$ {R}_c=\sum \limits_{i=0}^1\left({N}_i\cdotp {M}_i\right)/{N}^2. $$
(6)

The kappa value is then measured as the ratio of the difference between the observed cell-by-cell agreement and the chance agreement to the difference between the maximum agreement and the chance agreement calculated as follows:

$$ {R}_{kappa}=\left({R}_{cbc}-{R}_c\right)/\left(1-{R}_c\right). $$
(7)

It should be noted that a higher TI threshold would delineate fewer areas in a region as HSAs (i.e. a smaller HSA) and would consequentially result in fewer positive matches between the HSA and floodplain maps (i.e. n11). Therefore, a lower hit rate (Rhit) would be expected for a higher TI threshold. A higher TI threshold might decrease the positive matches (i.e. n11) but would increase the negative matches (i.e. n00). The change in the agreement rate based on the cell-by-cell comparison (Rcbc) depends on how much the increase in the negative matches offsets the decrease in the positive matches. The agreement rate would increase when the increase in the negative matches is faster than the decrease in the positive matches and decrease when the relationship is slower. The overall comparison rate (Roall) is very similar to the agreement rate with the goal of minimizing the mismatches (n10 – n01). The kappa value (Rkappa) is similar to the overall comparison rate with additional consideration of taking away the matches by chance. It is expected that both the overall comparison rate and the kappa value would behave similarly to the agreement rate as the TI threshold increases. The change in these rates (Rcbc, Roall, and Rkappa) from increasing to decreasing indicate that there would be peak levels in these indicators; in other words, these indicators tend to converge to their own levels across the gradient of TI thresholds.

2.4 Selecting TI Thresholds for HSA Delineation

HSAs and floodplains are different but interconnected features in a landscape. They overlap because the floodplain is the area in a landscape with sufficient accumulated runoff. The runoff-contributing areas of the HSAs vary by different storm events and consequently would result in different extents of flooded areas. Given their interconnection, a plausible explanation can be made using the occurrence of peak values in the spatial pattern comparison indicators. Namely, the spatial extents of the HSAs delineated by the specific TI threshold corresponding to the highest comparison indicator value represent the runoff-contributing areas to the FEMA 100-year floodplain.

In other words, the TI thresholds corresponding to these peak levels would delineate the HSAs that have the most consistent spatial patterns with the FEMA floodplain according to the respective spatial comparison indicators. Therefore, we can use these indicator values as objective functions such that identification of their peak levels allows for a systematic approach for selecting “best” TI thresholds and subsequently delineating management-relevant HSAs from regional to state scales. We present results of one possible approach to do that leveraging our process-based understanding of hydrology at the regional and state scales. Further, we consider potential strengths and weaknesses of this approach.

3 Results

The distribution of the FEMA 100-year floodplain, as expected, was closely associated with waterbodies, such as streams, rivers and oceans (Fig. 1a). The darker areas (Fig. 1b and c) indicate higher TWI and STI values and therefore represent HSAs more prone to generate runoff. The pattern of these darker areas not only captures the distribution of the floodplain, but also includes some HSAs outside the floodplain that likely contribute to flood waters. The STI pattern was similar to that of the TWI with some subtle variability across the state caused by the inclusion of soil transmissivity. The TWI values ranged from 2.60 to 26.20 with a mean of 9.04 and a standard deviation of 2.43. The STI values ranged from 0.92 to 28.23 with a mean of 7.94 and a standard deviation of 2.77. The deep topsoil in the Atlantic Coast region generally made STI values lower than TWI values (i.e. more dark areas in the region as shown in Fig. 1b) while the shallow topsoil and urbanized landscape in the Northeast region made STI values higher than TWI values (i.e. more dark areas in the region as shown in Fig. 1c).

3.1 Comparison at the State Scale

As expected, both HSAs and hit rate (Rhit) decreased as the selected threshold increases (Fig. 3). As shown in Fig. 3a, when using the threshold value of 8.0, for example, the TWI delineated 64% of the state as HSAs with the hit rate of 0.90 such that 90% of the FEMA 100-year floodplain was within the HSAs. As the threshold level was increased to 14.5, only 1% of the state was delineated as HSAs with a corresponding hit rate of 4%. The same pattern also held when using STI to delineate HSAs (Fig. 3b). The higher hit rate indicates a larger overlap between the FEMA floodplain and the delineated HSA. Although hit rate cannot be used to identify the TI threshold due to its monotonic change in response to the changes in threshold value, it can be used to assess which comparison indicators better identify the TI threshold that delineates a HSA that would be most consistent with the FEMA floodplain.

Fig. 3
figure 3

Spatial comparison indicators under different thresholds in New Jersey

Full size image

The other three spatial comparison indicators converged toward peak values at various TI thresholds. The cell-by-cell agreement rate (Rcbc) converged to a peak for a TWI of 11.5 and a STI of 11.0, which delineated 15% and 12% of the state as HSAs, respectively, and had corresponding hit rates of 45% and 34%, respectively. The overall comparison rate (Roall) was the highest for a TWI of 10.5 and a STI of 9.5, which delineated 23% and 23% of the state as HSAs, respectively, and had corresponding hit rates of 59% and 52%, respectively. The highest kappa values (Rkappa) were achieved for a TWI of 10.5 and a STI of 10.0, which delineated 23% and 18% of the state as HSAs, respectively, with hit rates of 59% and 46%, respectively. However, the convergence pattern (e.g. the peak of the objective function) was much clearer for Roall and Rkappa than for Rcbc. After its peak value, Rcbc only decreases slightly while Roall and Rkappa had clear optima.

Such convergence implies that the HSAs delineated using a certain threshold have the most consistent spatial pattern in comparison with the FEMA 100-year floodplain. The HSAs delineated for these given thresholds would closely approximate the runoff generating areas that contribute water to the 100-year flood. Specifically, the convergence patterns derived from the spatial comparison have helped narrow the possible values to be considered as threshold candidates from 28 to 5 and begin to outline a systematic approach to identifying management areas based on HSAs.

3.2 Comparison at the Regional Scale

In the Atlantic Coast water region (Table 1), Rcbc and Rkappa converged at a TWI of 11.0, which delineated 30% of the region as HSAs with a hit rate of 61%. Roall was at its highest level for a TWI of 10.5, delineating 36% of the Atlantic Coast region as HSAs with a hit rate of 68%. Rcbc converged to a peak for a STI of 9.5 which delineated 26% of the region as HSA with a hit rate of 54%. The highest Roall value occurred at a STI value of 8.5, which delineated 35% of the region as HSAs with the hit rate of 66%. Rkappa was the highest for a STI of 9.0 delineating 30% of the region as HSAs with a hit rate of 60%.

Table 1 The list of TI threshold candidates identified by spatial comparison in New Jersey and its five water regions
Full size table

In the Lower Delaware water region, Rcbc converged for a TWI of 11.5, delineating 15% of the region as HSAs with a hit rate of 37% while both Roall and Rkappa peaked at a TWI of 10.5 delineated 25% of the region as HSAs with the hit rate of 54%. The Rcbc was the highest for a STI of 11.0 delineating 11% of the region as HSAs with a hit rate of 29%. Roall and Rkappa converged to peaks at STIs of 9.0 and 9.5, respectively, which delineated 27% of the region as HSAs with a hit rate of 53% and 21% of the region as HSAs with a hit rate of 46%, respectively.

In the Northeast water region, Rcbc converged at a TWI of 11.0 and a STI of 11.5, which delineated 15% of the region as HSAs with a hit rate of 50% and 13% of the region as HSAs with a hit rate of 41%, respectively. With a TWI of 10.0 there was 21% of the region delineated as HSA with a hit rate of 62% corresponding to the highest values in both Roall and Rkappa. The Roall was the highest for a STI of 10.0, which delineated 21% of the region as HSAs with a hit rate of 57%. The Rkappa was the highest at a STI of 10.5 that delineated 18% of the region as HSAs with a hit rate of 52%.

For the Raritan water region, Rcbc identified the optimal TWI of 12.5 and STI of 13, which delineated 5% of the region as HSAs with a hit rate of 22% and 4% of the region as HSAs with a hit rate of 21%, respectively. Roall converged at a TWI of 10.5 and a STI of 10.5 that delineated 15% and 14% of the region as HSAs, respectively, with hit rates of 46% and 41%, respectively. The highest Rkappa values were achieved at a TWI of 11.0 and a STI of 11.0, which both delineated 11% of the region as HSAs with hit rates of 40% and 36%, respectively.

In the Upper Delaware water region, Rcbc did not identify any threshold, as it did not converge. The Roall values reached to their peak levels at a TWI of 11.0 and a STI of 10.5, which delineated roughly 8% and 9% of the region as HSAs, respectively, and both had a hit rate of 38%. The highest Rkappa identified a TWI of 10.5 and a STI of 12.0 delineating 10% and 7% of the region as HSAs, respectively, with hit rates of 43% and 34%, respectively.

3.3 TI Threshold Selection

Using three spatial comparison indicators, namely agreement rate (Rcbc), overall comparison rate (Roall) and kappa value (Rkappa), complimented with the hit rate (Rhit), the threshold candidates can be quickly narrowed from 28 to something between 4 to 6 candidates across the water regions in New Jersey as presented above. While this signifies a much smaller range of TI values to consider, it is still a daunting task to select one TI threshold for delineating HSAs for water resources management. To address this, we propose a possible two-step process to finalize the selection.

First, the agreement rate is not a reliable indicator to be used to identify the TI threshold, as it fails to identify a clear threshold in some cases. Further, even if the agreement rate does converge, it always identifies the TI thresholds that delineate the most restrictive sets of HSAs. For example, the STI threshold of 11.0, which was identified by the agreement rate, delineated 12% of the state as HSAs while STI thresholds of 10.0 and 9.5, which were identified by the kappa value and overall comparison rate, respectively, delineated 18% and 23% of the state as HSAs, respectively. Such an observation holds true for the TWI thresholds at the state scale as well as for both TIs across the five water regions in New Jersey.

Second, the effectiveness of identification can be used to further refine the selection of the TI threshold. The identification effectiveness measures how successfully the identified HSAs are with regard to their overlap with the FEMA floodplain. The identification effectiveness is defined as the ratio of the hit rate to the percentage of HSA (see last column of Table 1). The best TI threshold, thus, would be the one with the highest effectiveness. For example, the TWI threshold of 11.0 would be selected as the optimal threshold because it has the highest effectiveness ratio (2.03) among the TI threshold candidates at the state scale.

Using the two-step process, the final selection of the TI thresholds in New Jersey and its five water regions are highlighted in bold in Table 1. It is also worthy to note that all recommended TIs and their corresponding thresholds (after other considerations) are consistent with the recommendations based on the kappa value. Such congruence between the effectiveness ratio and kappa value indicates that the kappa value is an efficient spatial comparison indicator for identifying the TI threshold. Of course, this holds for delineating HSAs that approximate runoff contributing areas to the floodplain. The efficiency of the kappa value comes from considering both positive and negative matches and controlling for matches by chance.

4 Discussions

This study presents a feasible approach to identify a TI threshold for delineating HSAs at a regional to state scale in a landscape where HSAs approximate the runoff-contributing areas to the FEMA 100-year floodplain. Spatial comparison indicators allowed us to effectively narrow a wide range of potential thresholds to smaller ranges of TI values; however, these indicators alone were not sufficient to identify a specific TI threshold without comparison for validation. Clearly, additional discretion and knowledge are needed to select the proper TI threshold to delineate HSAs. Specifically, for this research, we determined that the agreement rate alone was not reliable enough for selecting the TI threshold. Rather, based on our experience in this region and on the process-based connections between HSAs and floodplains, the concept of identification effectiveness needed to be introduced to finalize the selection of one “best” TI threshold for each region (Table 1).

While our two-step approach is not the only way to finalize the TI selection, we feel it does achieve the goal of bringing in knowledge of the hydrological processes in the region relevant for water resource management. It is interesting to note that the two-step process allowed TWI thresholds to be selected for the Atlantic Coast and Lower Delaware water regions and STI thresholds to be selected for the Northeast and Upper Delaware water regions (Table 1), which is consistent with our local knowledge of hydrological processes (Qiu 2009; Lyon et al. 2018). The Atlantic Coast and Lower Delaware water regions, for example, are located within the physiographic province of the Coastal Plain dominated by shallow water tables. These regions also have been experiencing extensive floods in recent years, which likely make TWI better to delineate HSAs in these regions as topographical convergence dominates flow pathway distributions. The Northeast and Upper Delaware water regions are located in the physiographic provinces of the Valley and Ridges, Highlands and Piedmont. Those regions have less extensive floods relative to the Coastal Plain province and also have various restrictive soil layers that dictate the interflow movement of water. These natural conditions likely make STI better to delineate HSAs in those regions as soil transmissivity is a limiting factor for flow pathway distributions. The only exception here is the selection of TWI for delineating HSAs in the Raritan water region. Our prior knowledge identified STI is an appropriate index to delineate HSAs in the region dominated by the Highlands and Piedmont physiographic provinces (Qiu 2009; Qiu et al. 2014). This discrepancy highlights the challenge of selecting simple techniques for mapping out complex hydrological interactions across regional scales and helps reiterates the need for interpretation based on the understanding of underlying hydrological process.

Furthermore, the two-step approach presented here was developed to eventually be use by landscape managers and/or regulatory agencies to address local water resource concerns allowing for the involvement of diverse stakeholders. Stakeholder engagement would eventually introduce other factors such as local stakeholder preferences, implementation feasibility, and availability of resources for conservation into the delineation of HSAs (Qiu 2009). Engagement of stakeholders through a participatory process is a powerful technique to implement such an approach successfully as it would allow for feedback and criterion-based voting systems to ensure buy-in on water resources planning and management actions (Lyon et al. 2018). Our approach presented here can provide a short list of TI threshold candidates that would be ideal for such a stakeholder engagement process to produce a meaningful threshold from which realistic and consensual goals can be met. As such, our approach provides a dynamic rather static basis to landscape managers for selecting a TI threshold and delineating a meaningful spatial extent to direct their planning and management actions at regional to state scales.

5 Conclusions

This study used four spatial comparison indicators to assess the consistency between the FEMA 100-year floodplain and HSAs delineated using a gradient of TI thresholds in New Jersey at the state scale and across five water regions. Convergence of indicators varied by regions. Such information was combined with other local knowledge, such as flood extent and local topographic and soil conditions, to identify the “best” TI and the corresponding optimal threshold for delineating HSAs to approximate the runoff-contributing areas to a 100-year flood. Altogether, this study provides a practical approach for HSA delineation to support efficient and effective landscape and water resources management. The assessments presented in this study offer a first-cut analysis of the complicated spatial hydrological responses of a landscape at regional to state scales. The main strength in a threshold-based approach for HSA delineation such as that developed here are the simplicity and the lack of requirement of ancillary data for training models. Depending on the landscape problems to be addressed, such as flood mitigation, nonpoint source pollution prevention or soil erosion control, the approach developed here offers transparency and flexibility in selecting slightly different TI thresholds to delineate larger or smaller HSAs for targeting specific best management practices.