Impact of Coastal Wetland Restoration Plan on the Water Balance Components of Heeia Watershed, Hawaii

: Optimal restoration and management of coastal wetland are contingent on reliable assessment of hydrological processes. In this study, we used the Soil and Water Assessment Tool (SWAT) model to assess the impacts of a proposed coastal wetland restoration plan on the water balance components of the Heeia watershed (Hawaii). There is a need to optimize between water needs for taro cultivation and accompanying cultural practices, wetland ecosystem services, and streamflow that feeds downstream coastal fishponds and reefs of the Heeia watershed. For this, we completed two land use change scenarios (conversion of an existing California grassland to a proposed taro field and mangroves to a pond in the wetland area) with several irrigation water diversion scenarios at different percent of minimum streamflow values in the reach. The irrigation water diversion scenarios aimed at achieving sustainable growth of the taro crop without compromising streamflow value, which plays a vital role in the health of a downstream fishpond and coastal environment of the watershed. Findings generally suggest that the conversion of a California grassland to a patched taro field is expected to decrease the baseflow value, which was a major source of streamflow for the study area, due to soil layer compaction, and thus decrease in groundwater recharge from the taro field. However, various taro irrigation water application and management scenarios suggested that diverting 50% of the minimum streamflow value for taro field would provide sustainable growth of taro crop without compromising streamflow value and environmental health of the coastal wetland and downstream fishponds.


Introduction
In the Hawaiian Islands, coastal wetlands represent a critical interface between terrestrial and ocean zones with a vital importance in terms of economic, cultural, and environmental values. As described by Mitsch and Gosselink [1], coastal wetlands naturally purify water from sediments and contaminants, transform nutrients, slow down the flow of freshwater from the mountains to the ocean, and provide suitable habitats both for flora and fauna, including a decrease in greenhouse emission through carbon sequestration processes and micro-climate mitigation. Coastal wetlands are also considered very attractive and agriculturally productive regions for tourists and residents. In addition, these regions play an important role against flooding, pollution, and the negative impacts of climate and land cover changes. They also act like a sponge by absorbing water during the wet season and releasing it

Study Area
The Heeia wetland is the coastal part of the Heeia watershed, located on the windward side of the northeast coast of Oahu, representing the lower drainage basin of the watershed (Figure 1). Therefore, it is considered as a reservoir of freshwater originating from the springs in the mountains as surface water supplemented with lateral flow and baseflow. In the past, the watershed's hydrologic features enabled the indigenous society to meet their food and resources needs from land and sea in a prized coastal region [17]. The elevation of the watershed ranges from 0 to 854 m above mean sea level (amsl) with an average slope of 40%, while the elevation of wetland ranges from 0 to 17 m amsl with an average slope of 5% [7].  The land use of the wetland is dominated by emergent wetland (77%), forested wetland (8%), shrub wetland (5%), evergreen (4%), and other land use (6%) (Figure 2). Currently, most of the wetland area is blanketed by the invasive California grass, while the forested wetland is covered by mangrove trees (https://coast.noaa.gov/ccapatlas/). While the proposed taro land use will cover the cultivated land, shrub wetland, and emergent wetland portion of the Heeia wetland portion, the proposed pond area will cover part of its forested wetland ( Figure 2). Oahu's historical land use is documented elsewhere (https://guides.library.manoa.hawaii.edu/c.php?g=704383&p=5000954), whereas the historical land use of the HCW, including its pre-development condition, is detailed in Kakoo Oiwi [7].

Available Data
The following data were used to construct a SWAT model and assess the WBCs of the Heeia watershed and coastal wetland: As taro land use is not included in the crop database of SWAT, the specific plant parameters were obtained from actual field measurements and literature values [18][19][20][21][22][23][24][25].

5.
Due to lack of hydro-meteorological data within the watershed, we utilized various approaches, including interpolation, rescaling, and estimation based on the observed data and contour maps. For instance, fifteen virtual stations ( Figure 3) were created within the watershed based on the spatial variability of rainfall. Rainfall values were generated for each station using the closest rain gauge station and isohyets of the Rainfall Atlas of Hawaii [26]. To fill the other missing variables (temperature, wind speed, solar radiation, and relative humidity), a method proposed by Leta et al. [27] was used. 6.
Also used in the study is the daily streamflow data recorded at the Haiku station (U.S. Geological Survey (USGS) gauging station: 16275000) and others measured by this study at the coastal plain and estuary at the wetland flow sampling station for the period from 2012 to 2013. For the coastal plain, long-term and continuous streamflows were estimated based on the method developed by Leta et al. [27].
mangrove trees (https://coast.noaa.gov/ccapatlas/). While the proposed taro land use will cover the cultivated land, shrub wetland, and emergent wetland portion of the Heeia wetland portion, the proposed pond area will cover part of its forested wetland ( Figure 2). Oahu's historical land use is documented elsewhere (https://guides.library.manoa.hawaii.edu/c.php?g=704383&p=5000954), whereas the historical land use of the HCW, including its pre-development condition, is detailed in Kakoo Oiwi [7].
The pre-development (top, left) and current land use (top, right and bottom) maps of the Heeia wetland. Please note that palustrine scrub shrub, emergent, forested, and estuarine wetlands were reclassified as "wetland" on the wetland area map for clarity.

Available Data
The following data were used to construct a SWAT model and assess the WBCs of the Heeia watershed and coastal wetland:  Please note that palustrine scrub shrub, emergent, forested, and estuarine wetlands were reclassified as "wetland" on the wetland area map for clarity.

SWAT Model Description and Setup
SWAT is a physically based and semi-distributed hydrologic model that works at a basin scale and (sub-) daily time scale [16]. The applicability of SWAT has been widely proven in different hydrologic conditions, scales, and continents of the globe [28,29]. The model uses soil water balance concept and equations that consider precipitation as input and surface runoff, actual evapotranspiration, lateral flow, base flow, and deep groundwater loss as output at hydrological response units (HRUs) [30]. HRUs are the smallest spatial scale of the model representing a unique and homogeneous combination of soil, land use, and slope characteristics within a sub-basin, which is the second spatial scale of the model. Detailed approaches and equations used in SWAT for estimating the aforementioned water balance components at the HRUs level are provided in Reference [30].
We built up the SWAT model of the Heeia watershed based on the geo-spatial and hydro-meteorological data of the study area. By using the DEM data, we divided the watershed into 22 sub-basins. We captured the high spatial (topographic) variability of the watershed (Figures 1  and 2) by using lower than the SWAT default threshold value (minimum drainage area) during sub-basin delineation as streamflow routing occurs at this level. We further sub-divided sub-basins into 984 HRUs based on similar combinations of land use, soil type, and slope. As this study focused on land use change impact assessments, we used zero threshold values for HRUs classification. We set up the model for the period from 1/1/2000 to 12/31/2014. We used the period from 2000 to 2001 as In this study, we used the Soil Conservation Service Curve Number (SCS-CN) method for surface runoff simulations. We also used the Penman-Monteith method [31] option of SWAT for Potential Evapotranspiration estimation as that method is recommended for use in Hawaii's climatic condition [32]. Finally, we used the variable storage routing method [33,34] option of SWAT for the daily simulated streamflow routing.

Model Sensitivity, Calibration, Validation, and Uncertainty Analysis
SWAT has been widely used to perform sensitivity, calibration, and uncertainty analysis (see, e.g., Reference [27]). We used the Sequential Uncertainty Fitting (SUFI-2) algorithm [35] for model calibration and uncertainty analysis. SUFI-2 also has an ability to account for all sources of uncertainties esteemed from model parameters, driving variables (e.g., rainfall), model structure, and calibration data (e.g., observed streamflow) [36]. We evaluated the total model uncertainty by using P and R factors [37]. The P-factor measures the percentage of measured data bracketed at a 95% prediction uncertainty band (95PPU), while the R-factor evaluates the average thickness of the 95PPU band divided by the standard deviation of the measured data (e.g., observed streamflow). The values of the P-factor and R-factor range between 0 to 1 and 0 to ∞, respectively. A P-factor of 1 and R-factor close to zero mean that the simulated values are exactly matching the observed values [37]. Finally, we performed a manual calibration to fine-tune the calibrated parameter values and obtain a reasonable agreement between observed and simulated WBCs [38]. Such an approach substantially reduces the time-consuming manual calibration, including easy quantitative and qualitative comparisons [39].

Model Performance Evaluation
We evaluated the performance of SWAT for daily streamflow simulation by using five evaluation criteria. These include the Nash-Sutcliffe efficiency (NSE) [40], the percent bias (PBIAS), the ratio of the root mean square error (RMSE) to the standard deviation of measured data (RSR), the root mean square error (RMSE) [41], the Mean Bias Error (MBE) [42], and the correlation coefficient (r) [43].

Land Cover Change Scenario
The HCWR plan ( Figure 2) includes conversion of the California grass to organic wetland taro and the existing wetland mangrove forest to a pond as a native habitat for aquatic species. Based on the land use map data, the perennial California grassland mainly exists in the coastal wetland ( Figure 2). It covers approximately 7% of the modeled area (8.5 km 2 ). In addition, eight hectares of wetland mangrove forest (1% of the modeled area) is located around the Heeia stream estuary. The estuarine forested wetland in the C-CAP land use map of 2011 was treated as water during the land cover change conversion, while the California grassland was converted to taro cultivation. Taro was added to the SWAT crop database with the relevant properties that are summarized in Table 1. These parameters were created based on the literature values [25] and field measurements. Also, some selected existing variables of herbaceous land use from the SWAT database were used for wetland taro because taro was classified as an herbaceous perennial tropical crop [44,45]. The taro crop is chosen in the restoration plan because it is an important staple food and spiritual plant in Hawaiian cultural heritage. Moreover, until the 1940s, the HCW was actively cultivated with taro [46] and the restoration plan is to revert the land to its original state.

Wetland Taro Management
The traditional system of producing flooded or wetland taro in Hawaii requires the crop to be flooded with water for 11 months. The main source of water is the stream, which is diverted to channels and individual taro patches. The farmers build a dam from soil and stone across the stream to create enough head for diverting water to the taro patches [19,49]. In order to reproduce this scheme in SWAT and to allow the inflow and outflow of water from the taro patches, we assumed and added a pothole to the taro's management files of the SWAT model to simulate the HCW as a depressional water body. This enables to control the amount of water in the ditches of the taro field [50]. A pothole is a type of waterbody that obtains water from a sub-basin's reach and releases it through overflow via tile drainage. The water balance for a pothole is defined by Neitsch et al. [30] as: where V is the volume of stored water in the pothole at the end of the day (m 3 ), V stored is the volume of initial stored water in the pothole at the beginning of the day (m 3 ), V f lowin is the volume of entered water to the pothole during the day (m 3 ), V f lowout is the volume of water flowing out of the water body during the day (m 3 ), V pcp is the volume of precipitation falling on the pothole during the day (m 3 ), V evap is the volume of water lost from the pothole by evaporation during the day (m 3 ), and V seep is the volume of water lost from the pothole by seepage (m 3 ). The sources of water entering the pothole are from the sub-basin's streamflow diversion to irrigate a given HRU within the sub-basin. Therefore, the inflow of water from reach to the pothole is calculated as: where V f lowin is the volume of water flowing into the pothole during the day (m 3 ), irr is the amount of water irrigation diversion during the day (m 3 ), n is the number of HRUs contributing water to the pothole, f r pot , hru is the fraction of the HRU area draining into pothole, Q sur f , hru is the water surface runoff from the HRU on a given day (mm), Q gw, hru is groundwater flow generated in the pothole on a given day (mm), Q lat, hru is the water lateral flow generated in the pothole on a given day (mm), and area hru is the HRU area (ha).
We assumed the whole HRUs of taro land use to be represented by a pothole. We also assumed that a maximum volume of water stored in a pothole is 40 mm (depth) over the entire HRU, with an initial volume of 10 mm (depth) and depth to impervious layer of 250 mm to cause water ponding for taro cultivation. For irrigation application, we defined a water diversion from a reach and irrigation water schedule in the management file of the taro land use. The necessary input parameters are summarized in Table 2. We controlled the amount of water diverted from a reach to a taro field by setting a minimum flow value in the reach. For instance, if the minimum flow value in the reach is set to high, the amount of diverted water to the taro field is low. Therefore, we started the irrigation water diversion scenario (S) with an initial high value (S1) and then decreased by 50% (S2), 75% (S3), and 90% (S4) of the minimum flow (Qmin) in the reach, respectively ( Table 2).

Daily Streamflow Simulation and Uncertainty Analysis
The model evaluation criteria for the daily streamflow values at both Haiku and Heeia wetland stations are summarized in Table 3. Table 3 shows that the model performance is within the generally accepted values for daily time-scale, considering the scarcity of the watershed data. Overall, based on the recommended quantitative statistics (NSE, RSR, and PBIAS), the model simulation could be judged as satisfactory because the averages of the three criteria were 0.53, 0.66, and 5.9 respectively, which are within the acceptable ranges for daily streamflow simulation [51,52]. The results of simulated and observed daily streamflows along with the 95% prediction uncertainty (95PPU) are presented in Figure 4 for the Haiku station and Figure 5 for the wetland station. The figures generally show that the SWAT model reasonably simulated the observed daily streamflows' temporal evolution at both stations, except those simulated peak flow events when low observed flow values were recorded. The latter is most likely due to a lack of observed rainfall data within the watershed and high spatial rainfall gradient within a short distance (Figure 3). In addition, during the calibration period (2002-2008), 96% and 81% of the observed streamflow values were bracketed within the 95PPU at the Haiku and Wetland stations (Table 3), respectively. For the validation period (2009-2014), 96% of observed streamflow values were bracketed at the Haiku station, while 95% of the observed data were captured within the 95PPU at the wetland station. In addition, the R-factor values were close to 1 at both stations (Table 3), indicating that the model is reliable to simulate the Heeia watershed streamflow [53][54][55] and its applicability for future scenarios analysis.
width of the 95PPU interval.
The results of simulated and observed daily streamflows along with the 95% prediction uncertainty (95PPU) are presented in Figure 4 for the Haiku station and Figure 5 for the wetland station. The figures generally show that the SWAT model reasonably simulated the observed daily streamflows' temporal evolution at both stations, except those simulated peak flow events when low observed flow values were recorded. The latter is most likely due to a lack of observed rainfall data within the watershed and high spatial rainfall gradient within a short distance (Figure 3). In addition, during the calibration period (2002-2008), 96% and 81% of the observed streamflow values were bracketed within the 95PPU at the Haiku and Wetland stations (Table 3), respectively. For the validation period (2009-2014), 96% of observed streamflow values were bracketed at the Haiku station, while 95% of the observed data were captured within the 95PPU at the wetland station. In addition, the R-factor values were close to 1 at both stations (Table 3), indicating that the model is reliable to simulate the Heeia watershed streamflow [53][54][55] and its applicability for future scenarios analysis.

The Watershed Water Balance
While the annual average rainfall over the entire Heeia watershed is 2043 mm for the period from 2002 to 2014, the amount over the wetland area is only about 1065 mm for the same period (Table 4). This noticeable annual rainfall spatial gradient is also clearly observed in Figure 3. As expected, the rainfall was high during the wet season (Table 5) and highly correlated with recharge

The Watershed Water Balance
While the annual average rainfall over the entire Heeia watershed is 2043 mm for the period from 2002 to 2014, the amount over the wetland area is only about 1065 mm for the same period (Table 4). This noticeable annual rainfall spatial gradient is also clearly observed in Figure 3. As expected, the rainfall was high during the wet season (Table 5) and highly correlated with recharge (R 2 = 0.95) ( Figure 6). The percent of recharge comprised about 34% of the annual rainfall, which was consistent with previous studies in Hawaii [56,57]. The average annual water yield totaled 904 mm ( Table 4). The baseflow contributed 87% of the average annual water yield while surface runoff contributed 6% (Table 4), indicating that water yield was highly influenced by the groundwater discharge due to the geological features of the study area [58]. The contribution of the baseflow was very strong during the dry season (May-October), as summarized in Table 5. In contrast, the stream received more surface runoff during the wet season ( Table 5). The average annual potential evapotranspiration (PET) was 1412 mm whereas the actual evapotranspiration (AET) was 916 mm. AET was substantially lower than PET during the summer season because of the lack of sufficient soil moisture [59].

The Coastal Wetland Water Balance
We evaluated the impacts of the HCWR plan (conversion of California grassland to taro field and mangroves to an impoundment) on WBCs at three spatial scales, which included the hydrologic response units (HRUs), sub-basins, and watershed. Under the HRUs scale, the restoration is expected to impact the annual WBCs (Figure 7). Specifically, the recharge will decrease due to the soil layer compaction under the taro patches to maintain ponding water in taro. However, the neighboring areas of the taro patches would get more recharge due to lateral seepage from the taro patches [50]. The AET is expected to increase, which may result in a decrease of the other WBCs and an increase in evaporation from the ponding water area. At the wetland scale, the results indicate that recharge is expected to decrease at least by 41% under all irrigation diversion scenarios, which is probably due to taro cultivation and water ponding management. In contrast, the lateral flow and surface runoff would increase by about 77% and 62% respectively, when 90% of the minimum streamflow is diverted (Figure 8). For this scenario, although baseflow is expected to decrease by up to 42%, water yield is predicted to increase by 13%, due to the considerable increase in surface runoff and lateral flow (Figure 8). We also noted that most of the WBCs were more affected during the dry season as compared to the wet season (Table 6). Note: S1 = Scenario one (initial minimum streamflow); S2 = Scenario two (decrease 50% of minimum streamflow); S3 = Scenario three (decrease 75% of minimum streamflow); S4 = Scenario four (decrease 90% of minimum streamflow).   Jan  192  79  10  31  35  77  179  61  91  Feb  205  86  19  31  34  91  176  63  95  Mar  292  108  23  43  40  131  179  83  109  Apr  127  80  5  31  42  41  148  96  124  May  146  81  10  25  44  43  126  95  129  Jun  107  66  4  18  42  25  104  89  142  Jul  118  60  2  15  41  23  101  82  145  Aug  117  60  3  16  39  26  100  75  144  Sep  117  54  3  14  36  27  104  69  130  Oct  190  64  8  19  36  53  135  70  115  Nov  211  79  15  29  34  75  155  69  98  Dec  219  86  16  33  36  88  171 62 89   Finally, additional analysis on annual WBCs at the watershed scale indicated that the impact of land use change would have similar trends, but the relative percent change was low compared to the changes at sub-basin and HRU levels ( Table 6). That should be expected considering the size of the taro cultivation area, which was relatively small in comparison with the watershed size. Another aspect of the research focused on the impact of land cover change on stream outflow for different scenarios of irrigation diversions. We did not apply irrigation water diversion for the baseline case, representing the initial condition without taro cultivation and pond creation. However, we implemented several irrigation water diversion scenarios from stream reaches to taro field by modifying the irrigation management parameter (minimum streamflow, FLOWMIN) in the management input files of the SWAT model. We set the FLOWMIN value to 50%, 75%, and 90% of the initial minimum value for each sub-basin within the wetland area. We referred to these scenarios as S1, S2, S3, and S4, respectively (Table 2). Finally, additional analysis on annual WBCs at the watershed scale indicated that the impact of land use change would have similar trends, but the relative percent change was low compared to the changes at sub-basin and HRU levels ( Table 6). That should be expected considering the size of the taro cultivation area, which was relatively small in comparison with the watershed size. Another aspect of the research focused on the impact of land cover change on stream outflow for different scenarios of irrigation diversions. We did not apply irrigation water diversion for the baseline case, representing the initial condition without taro cultivation and pond creation. However, we implemented several irrigation water diversion scenarios from stream reaches to taro field by modifying the irrigation management parameter (minimum streamflow, FLOWMIN) in the management input files of the SWAT model. We set the FLOWMIN value to 50%, 75%, and 90% of the initial minimum value for each sub-basin within the wetland area. We referred to these scenarios as S1, S2, S3, and S4, respectively (Table 2). Note: S1 = Scenario one (initial minimum streamflow); S2 = Scenario two (decrease 50% of minimum streamflow); S3 = Scenario three (decrease 75% of minimum streamflow); S4 = Scenario four (decrease 90% of minimum streamflow). Figure 9 presents the monthly simulated streamflow values at the watershed outlet, just after the wetland area. The figure clearly indicates that all taro irrigation water diversion generally reduced stream outflows compared to the baseline values. However, as compared to the other scenarios, diverting 90% of minimum streamflow value (S4 scenario) significantly reduced the amount of outflows to the downstream. This highlights that excessively diverting irrigation water for the taro may negatively impact the downstream riverine ecosystem and environmental health, including the Heeia fishponds and reefs. Figure 9 further indicates that both S2 (50% minimum streamflow irrigation water diversion) and S3 (75% minimum streamflow irrigation water diversion) provide similar outflow values, which are close to S1, especially during the dry period. Although both S2 and S3 have similar impacts on the downstream outflow values, S2 outperformed S3 for two reasons: (i) S2 sufficiently supplied water for ponding and sustainably growing taro crop, and (ii) when compared to S3, S2 relatively showed a lower reduction in the downstream outflows that can play a vital role on the downstream fishponds and ecosystem services of the study area. Therefore, S2 is recommended to implement the proposed HCWR plan and achieve a sustainable growth of taro crop without compromising the coastal ecosystem role of the Heeia watershed. Figure 9. The monthly outflow of the Heeia Watershed for different scenarios of irrigation management. S1 = Scenario one (initial minimum streamflow); S2 = Scenario two (decrease 50% of minimum streamflow); S3 = Scenario three (decrease 75% of minimum streamflow); S4 = Scenario four (decrease 90% of minimum streamflow).

Conclusions
In this study, we used the Soil and Water Assessment Tool (SWAT) model to assess the impacts of the proposed Heeia Coastal Wetland Restoration (HCWR) plan on the water balance components (WBCs). We successfully derived the majority of the climatic data of the model from nearby watersheds by using some scaling techniques to capture the spatial variability of the climate data, especially rainfall. Using sensitive parameters identified during sensitivity analysis, we calibrated and validated the SWAT simulated streamflow values against the observed streamflow values, including model prediction uncertainty.
The SWAT model reasonably represented the temporal variability of the observed daily streamflow hydrographs, with an acceptable performance and satisfactory statistical evaluation values under hydrologic data scarcity. The findings showed that 34% of the annual rainfall of the watershed (2043 mm) fed groundwater as recharge (699 mm), 15% of the annual rainfall went as lateral flow (307 mm), 6% of the annual rainfall went as runoff (119 mm), and actual evapotranspiration (AET) accounted for 45% the annual rainfall (917 mm). In addition, baseflow and lateral flow contributed 87% of the annual water yield. The baseflow was found to be the main component of the water yield compared with surface runoff.
The impacts of the HCWR plan on WBCs is expected to be significant for the wetland area. Additionally, the restoration plan is predicted to reduce the recharge and baseflow values, but to increase lateral flow and surface runoff values. We completed different irrigation water diversion scenarios to taro field to identify an optimal policy to achieve sustainable growth of the taro crop without compromising the streamflow values in the mainstream and at the downstream fishponds that play a vital role in the downstream coastal ecology of the study area. We concluded that an optimal management strategy for the wetland and coastal shoreline restoration of the study area is possible by sustaining streamflow as well as water needs for the taro patches. Based on the findings, the study proposed to use 50% of the minimum streamflow value for irrigation water diversion to irrigate the taro field.