Development of a Watershed-Scale Long-Term Hydrologic Impact Assessment Model with the Asymptotic Curve Number Regression Equation

In this study, 52 asymptotic Curve Number (CN) regression equations were developed for combinations of representative land covers and hydrologic soil groups. In addition, to overcome the limitations of the original Long-term Hydrologic Impact Assessment (L-THIA) model when it is applied to larger watersheds, a watershed-scale L-THIA Asymptotic CN (ACN) regression equation model (watershed-scale L-THIA ACN model) was developed by integrating the asymptotic CN regressions and various modules for direct runoff/baseflow/channel routing. The watershed-scale L-THIA ACN model was applied to four watersheds in South Korea to evaluate the accuracy of its streamflow prediction. The coefficient of determination (R2) and Nash–Sutcliffe Efficiency (NSE) values for observed versus simulated streamflows over intervals of eight days were greater than 0.6 for all four of the watersheds. The watershed-scale L-THIA ACN model, including the asymptotic CN regression equation method, can simulate long-term streamflow sufficiently well with the ten parameters that have been added for the characterization of streamflow.


Introduction
In recent years, environmental disasters, such as droughts and floods, caused by climate changes have increased in occurrence, and various approaches to finding solutions for these issues have been suggested and investigated [1][2][3][4]. Estimation of runoff in watersheds is very important to preventing droughts and floods, preserving the ecological integrity of aquatic systems and managing water quality [5][6][7]. There are two ways to estimate the runoff in a watershed: monitoring of streamflow and use of rainfall-runoff models.
Monitoring of streamflow is more accurate than the use of computer models in estimating runoff in a watershed. However, it can be difficult to measure streamflow in all of the subbasins in a watershed without appropriate manpower and financial resources. In addition, it can be difficult to collect streamflow data during flooding and typhoon seasons [8]. Furthermore, for sustainable

LC-ACN-RE Approach to Considering Hydrologic Soil Groups
In the project "Long-term monitoring of Nonpoint Source (NPS) pollution" funded by the Ministry of Environment (MOE) of South Korea, runoff and water quality samples were collected for each representative land cover type rather than for specific Hydrologic Soil Groups (HSGs) or other soil infiltration properties. Thus, the LC-ACN-REs developed by [29] were classified as being applicable by land cover type; the effects of soil infiltration properties on direct runoff estimation cannot be analyzed with LC-ANC-RE approaches.
To overcome this limitation of the LC-CAN-REs [29], rainfall and direct runoff data from the "Long-term monitoring of Nonpoint-Source (NPS) pollution" project were analyzed to obtain the CN value for each combination of rainfall and direct runoff using Equation (1), which was proposed by Hawkins [30]. HSG information for the monitoring site was then compiled for use in estimating CN values for other HSGs. The CN values for other HSGs were estimated by multiplying the CN values obtained using Equation (1) by the ratios of the CN values for other HSGs in the NEH-4 CN table (Table 1). After obtaining the CN values for 13 land cover types and four HSG combinations, 52 asymptotic regression equations, such as the following, were obtained from regression analysis, as illustrated in Figure 1 (Equation (2) [30]).
where CN 8 is the asymptotic CN value, P is the rainfall (mm) and k is a fitting constant. Figure 1. Asymptotic CN regressions obtained in the study by Hawkins [30]. CN(P) is the Curve Number as a function of rainfall, and CN 0 = 100/(1 + P/2) defines a threshold below which no runoff occurs until the rainfall P in mm exceeds an initial abstraction of 20% of the maximum potential retention.

Development of the Watershed-Scale L-THIA ACN Model
The watershed-scale L-THIA ACN model developed in this study consists of three modules, for direct runoff, baseflow (with extended LC-ACN-REs developed as described in Section 2.1) and channel routing capabilities ( Figure 2). The model requires daily rainfall, point source data and hydrological response unit (HRU) mapping created by combining a subbasin map, a soil map and a land use map.

Development of the Watershed-Scale L-THIA ACN Model
The watershed-scale L-THIA ACN model developed in this study consists of three modules, for direct runoff, baseflow (with extended LC-ACN-REs developed as described in Section 2.1) and channel routing capabilities ( Figure 2). The model requires daily rainfall, point source data and hydrological response unit (HRU) mapping created by combining a subbasin map, a soil map and a land use map.

Development of the Direct Runoff Estimation Module
As explained above, the 52 asymptotic regression equations (Equation (2)) were obtained for 13 land cover types and four HSGs. Using these equations, the CN values for a given set of daily rainfall data were computed for all land cover and HSG combinations for each watershed studied.
According to various studies on the NRCS-CN method, CN values can be adjusted based on the slope in a watershed [39], as well as for various local conditions that affect rainfall-runoff. Thus, an adjustment coefficient was added to explain the effect of the slope on the CN values (Equation (3)). In this study, the limits of the range for the adjustment coefficient for CN were set to −0.1 and +0.1 (−10% and +10%).
where CNHRU is the adjusted CN value for HRU, Adj_CNHRU, ACN is the CN value determined from the

Development of the Direct Runoff Estimation Module
As explained above, the 52 asymptotic regression equations (Equation (2)) were obtained for 13 land cover types and four HSGs. Using these equations, the CN values for a given set of daily rainfall data were computed for all land cover and HSG combinations for each watershed studied.
According to various studies on the NRCS-CN method, CN values can be adjusted based on the slope in a watershed [39], as well as for various local conditions that affect rainfall-runoff. Thus, an adjustment coefficient was added to explain the effect of the slope on the CN values (Equation (3)). In this study, the limits of the range for the adjustment coefficient for CN were set to´0.1 and +0.1 (´10% and +10%).
Adj_CN HRU " CN HRU,ACNˆA dj_CN (3) where CN HRU is the adjusted CN value for HRU, Adj_CN HRU, ACN is the CN value determined from the extended LC-ACN-REs and Adj_CN is the adjusted coefficient for CN. After adjustment of the CN values, the direct runoff of each HRU was estimated (Equation (4)).
where Q 1 DR,HRU is the amount of direct runoff generated by an HRU for each day (mm), P is the rainfall (mm), S is the potential maximum retention (mm) and Adj_CN HRU,ACN is the adjusted coefficient for CN.
In a large-scale watershed, the amount of direct runoff that occurs on a day can be lagged, and only a portion of the direct runoff will flow into a stream on a day. Thus, in this study, the direct runoff delay process was addressed in the module in the form of an exponential function of the time of concentration (TC) and the lag coefficient (DR lag ), as proposed in the SWAT model (Equation (5)) [38].
Once the direct runoff of each HRU is calculated using Equation (4), the amount of direct runoff flowing into the stream can be calculated using Equation (5).
where Q DR,HRU is the amount of direct runoff discharged to the main channel on a given day (mm), Q 1 DR,HRU is amount of direct runoff generated by the HRU on a given day (mm), Q stor is the direct runoff lagged from the previous day, DR lag is the direct runoff lag coefficient and TC is the time of concentration (h).
The value of the lag coefficient (DR lag ) ranges from 1 to 12 and should be provided by the user after investigating watershed characteristics or related documents. The time of concentration is defined as the time required for water to flow from a remote point in a watershed to a watershed outlet. The time of concentration is important in the rainfall runoff model and can be estimated from various formulas, although the variability of the estimates of the time of concentration given by various formulas can be high [40]. There are two types of time of concentration: the time required for overland flow and the time required for channel flow. These are calculated from watershed-specific information, such as the average slope (m/m), the slope length (m), the channel length from the most distant point to the subbasin outlet (km) and Manning 1 s coefficient, n, as shown in Equations (6) and (7) [38,41]. Equations (4) and (5) were added to the direct runoff module of the watershed-scale L-THIA ACN model.
In these equations, TC O is the time of concentration for overland flow (h); TC C is the time of concentration for channel flow (h); L Slope is the HRU slope length (m); n is Manning 1 s coefficient for overland flow; n C is Manning's coefficient for channel flow; Slope is the average slope of the HRU (m/m); Slope C is the channel slope (m/m); L is the channel length from the most distant point to the subbasin outlet (km); and Area is the area of the HRU (km 2 ).
In this study, the values of Manning's coefficient from Table 2 were used for the calculation of TC O , and the values of Manning's coefficient for TC C were calibrated based on the land cover and the parameter range, based on Table 2. The Slope, the Area of the HRU and the channel length L from the most distant point to the subbasin outlet (km) were calculated using a GIS tool for the purpose of computing the time of concentration for overland flow and the time of concentration for channel flow. However, because the calculation of the slope length is strongly affected by the digital elevation model (DEM) cell size, the field slope length can be overestimated when it is calculated using a GIS tool [43]. Furthermore, the DEM resolution, slope length, river networks and flow length estimation are among the major source of uncertainties in rainfall-runoff modeling [44]. Thus, in this study, the slope length of each HRU was calculated from the relationship between the field slope length and the average field slope, as proposed in [45]. This relationship between the field slope length and the average field slope (Table 3) was added to the direct runoff module in the watershed-scale L-THA ACN model. However, this relationship, described in Table 3, was obtained from measurements made in the USA. Thus, to reflect local field slope length properties, an additional parameter (SLSUB) was added to adjust the slope length. Table 3. Suggested maximum slope length for field slope for contouring [45].

Land Slope (%)
Maximum Length (m) After calculation of the direct runoff for each HRU released to the stream, the direct runoff of each subbasin was calculated by summing the direct runoff from all HRUs within each subbasin (Equation (8)): where Q DR,sub is the amount of direct runoff generated in the subbasin on a given day (mm).

Development of Baseflow Module
The NRCS-CN method and LC-ACN-REs are both used for direct runoff estimation, but not for baseflow computation for a watershed. The baseflow component was developed and linked to the watershed-scale L-THIA ACN model for use in watershed hydrology studies, as well as evaluation of the water quality of a watershed.
According to Dingman [46], the aquifer in a watershed is composed of two aquifers, an unconfined aquifer and a confined aquifer. Water recharged into an unconfined aquifer contributes to flow in the main channel and influences the amount of streamflow, while water recharge into a confined aquifer is assumed to flow somewhere outside of the watershed [47].
Based on the user-defined fraction of infiltrated water flowing into the confined aquifer from each HRU, the amount of infiltrated water flowing into the unconfined aquifer can be estimated. The baseflow module, which was developed and integrated into the watershed-scale L-THIA ACN model, simulates these processes to account for the baseflow contribution to streamflow in a watershed. The water balance in an unconfined aquifer is calculated according to Equation (9): where aqf HRU,i is the amount of water stored in an unconfined aquifer on a given day (mm), aqf HRU,i-1 is the amount of water stored in the unconfined aquifer on the previous day (mm), ω unconf,HRU is the amount of recharge entering the unconfined aquifer on that day (mm) and Q BF,HRU is the amount of baseflow into the main channel (mm). The amount of water recharged into both aquifers (confined and unconfined) is estimated using the exponential decay weighting function (Equation (10)) proposed by Venetis [48] and used by Sangrey et al. [49] in their precipitation-groundwater response model and by Neitsch et al. [38] in the SWAT model: where ω rcharg,HRU,i is the amount of recharge entering both aquifers on a given day (mm), BF delay is the delay time in aquifer recharge once the water infiltrates from the surface (days), F HRU,i is the amount of infiltration on the given day (mm) and ω rcharg"HRU,i-1 is the amount of recharge that enters the aquifers on the previous day (mm). The amount of infiltration on a given day is calculated using Equation (11), which is modified from that used in the NRCS-CN method [22] and was used by Kim et al. [50] to estimate the CN-based infiltration and baseflow: where F HRU,i is the amount of infiltration on a given day (mm), S is the is the potential maximum retention (mm), P is the rainfall (mm), Adj_CN HRU , ACN is the CN value determined from the extended LC-ACN-REs and I a is the initial abstraction (mm).
In the baseflow module, only a fraction of the infiltrated water is assumed to flow into the unconfined aquifer, based on the user-defined fraction of infiltrated water flowing into the confined aquifer (Equations (12) and (13)): ω con f ,HRU " Fr con fˆωrchrg,HRU (12) ω uncon f ,HRU " ω rchrg,HRU´ωcon f ,HRU where ω conf,HRU is the amount of infiltrated water flowing into a confined aquifer on a given day (mm), Fr conf is the fraction of water flowing into the confined aquifer and ω unconf,HRU is the amount of recharge entering the unconfined aquifer on that day (mm). The amount of water flowing into an unconfined aquifer contributes to baseflow only if the amount of water in the unconfined aquifer exceeds a threshold value specified by the user that depends on the aquifer 1 s characteristics. The steady-state response of the baseflow is expressed by Equation (14) [51]: where Q BF,HRU is the amount of baseflow into the main channel (mm), k sat is the hydraulic conductivity of the aquifer (mm/day), L BF is the distance from the ridge or subbasin divide for the baseflow to the main channel (m) and h wtbl is the water where dh wtbl dt is the change in the elevation of the water table (mm/day), ω uncon f ,HRU is the amount of recharge entering the unconfined aquifer on a given day (mm), Q BF,HRU is the baseflow into the main channel on that day (mm) and µ is the specific yield of the unconfined aquifer (m/m).
Combining Equations (14) and (15) where Q BF,HRU is the baseflow into the main channel on a given day (mm), k sat is the hydraulic conductivity of the aquifer (mm/day), L BF is the distance from the ridge or subbasin divide for the baseflow to the main channel (m), µ is the specific yield of the unconfined aquifer (m/m), ω unconf,HRU is the amount of recharge entering the unconfined aquifer on that day (mm) and α BF is the baseflow recession constant. After integrating and rearranging Equation (14), Q BF , HRU can be expressed by Equation (17) where Q BF,HRU,i is the baseflow into the main channel on a given day (mm), Q BF,HRU,i-1 is the baseflow into the main channel on the previous day, α BF is the baseflow recession constant, ω unconf,HRU is the amount of recharge entering the unconfined aquifer on the given day (mm), ∆t is the time step (one day), aqf is the amount of water stored in the unconfined aquifer on the given day (mm) and aqf thr is the threshold water level in the unconfined aquifer for baseflow contribution to the main channel to occur (mm). The baseflow recession constant (α BF ) in Equation (17) reflects the baseflow response to the amount of recharge [52]. Values between 0.1 and 0.3 represent slow response conditions in a watershed, and values between 0.9 and 1.0 represent rapid response conditions [38,52]. Two options (simple long-term daily average and daily time series point source capabilities) were enabled in the direct runoff module to simulate the effects of discharge from Waste Water Plants (WWP) or other point sources on watershed hydrology and water quality. routing module was integrated into the watershed-scale L-THIA ACN model for simulation of flow routing using the Muskingum routing method [53]. The Muskingum routing method estimates the storage volume in a channel length as a combination of wedge and prism storage [53]. The concept of the Muskingum routing method is illustrated in Figure 3. The first case represents the storage in the river during the rising limb of a hydrograph; the second case represents uniform flow; and the third case represents the storage during the falling limb of the hydrograph. This hysteresis might cause different flood wave speeds during the rising and falling limbs of the hydrograph [54].
(simple long-term daily average and daily time series point source capabilities) were enabled in the direct runoff module to simulate the effects of discharge from Waste Water Plants (WWP) or other point sources on watershed hydrology and water quality.

Development of Channel Routing Module
Streamflow flows downward and meets flow from other upper streams in channel networks. The amount of streamflow in a watershed is affected by various mechanisms. In this study, a channel routing module was integrated into the watershed-scale L-THIA ACN model for simulation of flow routing using the Muskingum routing method [53]. The Muskingum routing method estimates the storage volume in a channel length as a combination of wedge and prism storage [53]. The concept of the Muskingum routing method is illustrated in Figure 3. The first case represents the storage in the river during the rising limb of a hydrograph; the second case represents uniform flow; and the third case represents the storage during the falling limb of the hydrograph. This hysteresis might cause different flood wave speeds during the rising and falling limbs of the hydrograph [54]. The effects of these variables and the reach storage-discharge relationship are expressed by the equation used in the Muskingum routing method to estimate the reach storage volume, Vstor: The effects of these variables and the reach storage-discharge relationship are expressed by the equation used in the Muskingum routing method to estimate the reach storage volume, V stor : where V stor is the reach storage volume (m 3 /s), K is the storage time constant for the reach (s), X is the weighting factor, q in is the inflow rate (m 3 /s) and q out is the outflow rate (m 3 /s). Equations (18) and (19), proposed by Williams [55], can be combined and simplified as Equation (20): q out,∆t " C 1 q in,∆t`C2 q in`C3 q out (20) where q in is the inflow rate at the beginning of the time step (m 3 /s), q in,∆t is the inflow rate at the end of the time step (m 3 /s), q out is the outflow rate at the beginning of the time step (m 3 /s), q out,∆t is the outflow rate at the end of the time step (m 3 /s) and C1, C2 and C3 are expressed by Equations (21)-(23), respectively.
The value for the weighting factor, X, is a user input. The value of the storage time constant, K, is calculated using Equation (24): K " Mk1ˆbank f ull`Mk2ˆbank f ull 0.1 (24) where K is the storage time constant for a reach segment (s), Mk1 and Mk2 are weighting factors input by the user, bankfull is the storage time constant estimated for the reach segment with bankfull flows (s) and bankfull 0.1 is the storage time constant estimated for the reach segment with one tenth of the bankfull depth (s). The value of bankfull can be calculated using Equation (25) [38,53]: where bankfull is the storage time constant estimated for the reach segment with bankfull flows (s), L reach is the stream length (m) and c is the celerity corresponding to the flow for the specified depth (m/s). The value of c can be determined from Equation (26), using the cross-sectional area of the stream and Manning's equation [38]: where A is the cross-sectional area of flow in the stream (m 2 ), q ch is the flow rate in the stream (m 3 /s), R is the hydraulic radius for a given depth of flow (m), I is the slope (m/m) and n is Manning's coefficient for the channel.
To calculate V c in Equation (26), the metacenter of the streams should be evaluated, and the actual shapes of all of the streams in each subbasin should be measured. However, it would be difficult to measure the cross-sections of all of the streams. In this study, all streams in subbasins were assumed to be trapezoidal channels with side slopes of 0.5, and the slope of the flood plain was assumed to be 0.25. These assumptions are similar to those made in the SWAT model [38].

Input Parameters of Watershed-Scale L-THIA ACN Model
The watershed-scale L-THIA ACN model was developed to estimate streamflow using direct runoff, baseflow and channel routing modules.
The parameters of these three modules consist of CN parameters for all HRUs, two direct runoff parameters, four baseflow parameters and three channel routing parameters, as summarized in Table 4. The watershed-scale L-THIA ACN model requires a smaller number of parameters than other watershed models, such as SWAT and HSPF. Direct runoff lag coefficient 1-12 SLSUB (2) Adjustment for slope length´10-10

Applications of Watershed-Scale L-THIA ACN Model
The watershed-scale L-THIA ACN model, developed in this study, was applied to four watersheds (Goboo A, Tancheon A, Kumbon A and Pyungchang A) in South Korea, where TMDLs have been implemented to evaluate the water quality achieved with various management practices (Figure 4).
The watershed-scale L-THIA ACN model requires daily rainfall data, as well as HRU maps, which are prepared from combinations of subbasin, land cover and soil maps. Subbasin maps were delineated by 30-m resolution DEM and stream data using the ArcGIS watershed delineation geoprocessing tool. HRU maps were created by combining subbasin maps, reconnaissance soil maps and land cover maps provided by the Ministry of Environment of South Korea.
Daily streamflows for 1 January 2010-31 December 2014 were estimated for the four watersheds using the watershed-scale L-THIA ACN model and precipitation data from the Korean Meteorological Administration (KMA). The watershed-scale L-THIA ACN model was calibrated and validated by adjusting the parameters of the direct runoff and baseflow to fit the simulated daily streamflows for eight-day intervals to observed streamflow data. The calibration period was 1 January 2008-31 December 2010 and the validation period was 1 January 2011-12 December 2014.

Result of Extended LC-ACN-RE Approach for the Consideration of HSGs
In this study, thirteen land cover-based asymptotic CN regression equations (LC-ACN-REs) were extended to 52 regression equations to consider HSGs, using the ratio of CN for a given HSG in  (Table 5). Thus, the CN values for each HSG group can be estimated with the 52 extended LC-ACN-REs. These CN values were used to estimate the direct runoff, the infiltration from each HRU and the baseflow component. The 52 extended regression equations predict the lowest asymptotic CN values for high-permeability soils of the HSG A type and the highest asymptotic CN values for low-permeability soils of the HSG D type. The highest asymptotic CN value predicted by these equations was 92.0 for commercial areas, and the lowest asymptotic CN value was 23.0 for pasture.

Application of the Watershed-Scale L-THIA ACN Model
The watershed-scale L-THIA ACN model was calibrated using observed streamflow data for the four study watersheds in South Korea. The calibrated values of the ten parameters for each study watershed are shown in Table 6. Among the ten parameters, Adj_CN and DR lag ranged from´0.04-0.09 and from 3-10, respectively. As the average slope of each watershed decreased, the Adj_CN parameter value also decreased, and the DR lag value increased. Thus, Adj_ CN and DR lag should be adjusted for the average slope of a watershed before being used in the watershed-scale L-THIA ACN model. As described in the previous section, field slope lengths should be adjusted based on local conditions. For the Goboo A watershed, the SLSUB parameter, which is the adjusted slope length parameter, was found to be seven times greater than the default value.
The greater the area of a watershed is, the higher the threshold water level for baseflow (aqf thr ) in an unconfined aquifer is.
The BF delay parameter was found to exhibit the same trend as the aqf thr parameter. The baseflow recession constant (α BF ) values were found to be similar for all four of the study watersheds. The baseflow response to main streamflow in the four watersheds constituted normal response conditions, as indicated by α BF values from 0.4-0.6 (note that α BF values of 0.1-0.3 represent a slow response, and α BF values of 0.7-1.0 represent a fast response) [52].
Comparisons of the simulation results and observed eight-day interval streamflows revealed reasonable agreement for the Goboo A watershed, with a coefficient of determination (R 2 ) of 0.66 and Nash-Sutcliffe Efficiency (NSE) of 0.64. The R 2 values for Tancheon A, Kumbon A and Pyungchang A were 0.62, 0.9 and 0.62, respectively, and the NSE values were 0.61, 0.92 and 0.60, respectively ( Figure 5 and Table 7).

Conclusions
In this study, LC-ACN-REs were improved by enabling consideration of HSG characteristics. In addition, a watershed-scale L-THIA ACN model was developed with direct runoff, baseflow and The validation results for the Goboo A watershed indicated that the R 2 and NSE values were 0.79 and 0.78, respectively. The R 2 values for Tancheon A, Kumbon A and Pyungchang A were 0.72, 0.62 and 0.80, respectively, and the NSE values were 0.70, 0.60 and 0.79, respectively ( Figure 5 and Table 7).
According to Ramanarayanan et al. [56] and Moriasi et al. [57], R 2 and NSE values that reflect satisfactory calibration of streamflow are R 2 ě 0.5 and NSE ě 0.5. By these criteria, the performance of the watershed-scale L-THIA ACN model developed in this study was acceptable, because the R 2 and NSE values obtained from the calibration and validation were greater than 0.6 for all four study watersheds considered (Table 7).
There were no significant differences in the average streamflow for the study watersheds. The differences between the observed data and simulated streamflows in the calibration were 11.9, 2.2, 2.4 and 6.9% for the Goboo A, Tancheon A, Kumbon A and Pyungchang A watersheds, respectively, and the differences in the validation were 8.5, 11.8, 6.6 and 4.4% ( Table 7). As the results of this study show, the watershed-scale L-THIA ACN model can simulate streamflow well for watersheds ranging in size from 200.0 km 2 -1756.9 km 2 .
As shown in Figure 5, the estimated peak flow during the high-flow season was lower than the observed peak flow. This can be explained by reduced infiltration, which is estimated using Equation (11). As the rainfall amount becomes greater, the CN estimated using the asymptotic CN approach becomes lower, resulting in a greater value of S (the potential maximum retention, mm). As S becomes greater, the infiltration approaches zero, especially for forest and pasture land covers. This results in a lower baseflow contribution to the total streamflow. Similar issues have been mentioned in other studies [58,59]. Although Equation (11) is simple to use in estimating the contributions of infiltration and baseflow to streamflow using the CN value, more in-depth investigation is needed to account for the lower infiltration that occurs in land cover areas with lower CN values.

Conclusions
In this study, LC-ACN-REs were improved by enabling consideration of HSG characteristics. In addition, a watershed-scale L-THIA ACN model was developed with direct runoff, baseflow and channel routing capabilities integrated together. With this new L-THIA model, users can simulate streamflow (direct runoff + baseflow) in a watershed. Daily rainfall data, land cover and HSG maps, DEM and ten additional model parameters are needed for the simulation of streamflow.
The simulated streamflow agreed well with the observed streamflow for the four study watersheds (as indicated by R 2 values in the range of 0.62-0.93 and NSE values in the range of 0.60-0.93 for both calibration and validation). These results demonstrate the predictive capability of the watershed-scale L-THIA ACN model developed in this study. Two model parameters (Adj_CN and DR lag ) were found to be closely related to the average field slope. Further in-depth investigation is needed to derive the relationships between the field slope length and these two model parameters. It should be noted that the watershed-scale L-THIA ACN model was not applied comprehensively to the watersheds for ranges of rainfall, land use, soil and topography conditions. It should also be noted that the values of the model parameters were estimated using manual calibration processes, which are affected by the subjective judgments of the model users.
For these reasons, automatic calibration using PARASOL (Parameter Solution), SUFI-2 (Sequential Uncertainty FItting algorithm), GLUE (Generalized Likelihood Uncertainty Estimation) and GA (Genetic algorithm) is needed for objective evaluation of the model parameters. With this function enabled, the relationship between the average field slope and the Adj_CN and DR lag parameters could be analyzed and utilized in the watershed-scale L-THIA ACN model for streamflow estimation in ungaged watersheds.