Professional Documents
Culture Documents
a r t i c l e i n f o a b s t r a c t
Article history: Suitable selection of hydrological modeling tools and techniques for specific hydrological study is an
Available online 12 October 2014 essential step. Currently, hydrological simulation studies are relied on various physically based, con-
ceptual and data driven models. Though data driven model such as Adoptive Neuro Fuzzy Inference
Keywords: System (ANFIS) has been successfully applied for hydrologic modeling ranging from small watershed
El Niño Southern Oscillation (ENSO) scale to large river basin scale, its performance against physically based model has yet to be evalu-
Adaptive Neuro-Fuzzy Inference System ated to ensure that ANFIS are as capable as any physically based model for simulation study. This
(ANFIS)
study was conducted in Chickasaw Creek watershed, which is located in Mobile County of South
Loading Simulation Program C++ (LSPC)
Hydrologic simulation
Alabama. Since adequate rain gauge stations were not available near the watershed proximity, and
also the study area was affected with the El Niño Southern Oscillation (ENSO), the sea surface
temperature (SST) and sea level pressure (SLP) were additionally incorporated in the ANFIS model.
The research concluded that ANFIS model performance was equally comparable to a physically based
watershed model, Loading Simulation Program C++ (LSPC), especially when rain gauge stations were
not adequate. Additionally, the research concludes that ANFIS model performance was equally
comparable to that of LSPC no matter whether SST and SLP in ANFIS input vector was included
or not.
Published by Elsevier Ltd.
http://dx.doi.org/10.1016/j.eswa.2014.09.062
0957-4174/Published by Elsevier Ltd.
2214 S. Sharma et al. / Expert Systems with Applications 42 (2015) 2213–2223
Fig. 1. ANFIS architecture for two inputs (X and Y) with two rules and two membership functions (P1, P2 and Q1, Q2) for each rule (Sharma et al., in press).
X
2 as the center of the cluster. For a given set of data points from
O5i ¼ zi f i ¼ zi ðpi X þ qi Y þ r i Þ ð5Þ X1, X2, . . .. . . to . . . Xn, the density measure at a data point Xj is given
i¼1
by the following equation:
where zi is the output from layer 3 and [pi, qi, ri] are
the model parameters. X
n
kX j X i k2
Layer 6 In this layer, the final output is determined by com- dj ¼ e 2 ð7Þ
ra
j¼1
bining entire receiving signals 2
X
2 P2 where ra is a data cluster radius. The equation indicates that those
zi f
O6i ¼ zi f i ¼ Pi 2 i : ð6Þ data points which have high neighboring data points will be consid-
i 1 zi ered as a cluster center.
Finally, ANFIS model uses hybrid algorithm to compute the
model parameters. To effectively partition the data and minimize 2.2. LSPC model
the fuzzy rules, various clustering approach have been recom-
mended. Two widely used fuzzy clustering method such as The LSPC, a C++ version of the widely used Hydrologic Simula-
subtractive clustering algorithm (SCA) (Chiu, 1996) and Fuzzy tion Program-Fortran (HSPF) (Bicknell et al., 2001), is a watershed
C-Mean (FCM) (Dunn, 1973) for structure identification has been model for simulating streamflow and stream hydraulics. Though
described in the following section. the LSPC’s algorithms are identical to those of the HSPF, LSPC
model is more efficient, flexible and have been used for streamflow
2.1.1. Subtractive and Fuzzy C-Mean clustering and water quality simulation (e.g. Sharma, Srivastava, Kalin, Fang,
For large and good quality data sets covering the wide varia- & Elias, 2014). The model has been considered as one of the most
tions in the feature space, fuzzy systems can have better general- advanced hydrologic and watershed loading model. The hydrologic
ization due to the use of more fuzzy rules. However, for a small portion of the model is based on the Stanford Watershed Model
range of data sets, which is more common in streamflow data sets, (Crawford & Linsley, 1966). The hydrologic process in the model
large number of rules cannot be derived from the training data as it is conceptualized with the schematic diagram as shown in Fig. 2
may easily over fit the system, and rule out the possibility of (Tech & Center, 2009). The diagram shows the water budget dia-
generalization. Hence, clustering approach is required for both gram with the water interchanging process from surface to subsur-
the effective partition of the input space and the reduction of the face including three flow path; surface, interflow and groundwater
number of rules. Several clustering methods have been suggested flow.
to organize the data and construct the rules. Some of the widely Precipitation is the input to the system (watershed) which is
used approaches are grid partition (Jang & Sun, 1995), SCA, and partitioned at the first decision node for the interception and water
FCM. In this analysis, both FCM and SCA were experimented. reaching on the land surface using the parameter CEPC (Fig. 2). In
FCM is a technique in which each data point is categorized into a the next decision node, surface and lower zone storage is divided
cluster based on membership function (Thirumalaiah & Deo, using the parameter, INFILT. The surface storage contributes to
2000). It divides the data from multidimensional space into a the three components: (i) upper zone storage which partly contrib-
particular cluster number. Cluster center and the membership utes to lower zone, partly to the groundwater/inactive groundwa-
function were decided through the repeated iteration by minimiz- ter storage, and rest to the evapotranspiration (ET), (ii) interflow
ing the objective function which was defined as a distance between storage, which eventually contributes to the channel flow,
data point and cluster center. SCA is more suitable when users do (iii) overland flow. The lower zone storage may partly contribute
not have a clear idea of cluster number required to be used for a to groundwater storage or inactive groundwater storage (deep per-
given data sets. This algorithm is based on the amount of the den- colation) or partly lost through ET. The contribution from upper
sity points in the data space, and the approach is to find areas in zone and lower zone storage to groundwater is divided into active
the data space having higher densities of data points. The algo- and inactive groundwater by using a coefficient (DEEPFR). Active
rithm tries to find the point with the highest number of neighbors groundwater storage can contribute to streamflow as base flow
2216 S. Sharma et al. / Expert Systems with Applications 42 (2015) 2213–2223
Fig. 2. Schematic diagram of the Stanford Watershed Model adapted for LSPC model. Number corresponding to each circle denotes the order of removing water to satisfy ET.
(Source: Tech & Center, 2009). INFILT, infiltration parameter; INTFW, interflow parameter; CEPSC, interception storage capacity; LZSN, lower zone storage capacity; UZSN,
upper zone storage capacity; NSUR, Manning’s surface roughness; LSUR, surface runoff length; SLSUR, surface slope; IRC, interflow recession constant; AGWETP; active
groundwater ET parameter; AGWRC; active groundwater recession; BASETP; base flow ET parameter; DEEPFR, fraction to inactive groundwater; LZETP, lower zone ET
parameter.
and losses as ET. ET can occur from the entire zone except inactive The precipitation and temperature of this region is associated
groundwater zone. The overland flow, interflow, and base flows with ENSO, characterized with fairly higher precipitation and
eventually contribute to streamflow. lower temperature in El Niño winter, and just opposite pattern in
La Niña winter (Sharma, Srivastava, Fang, & Kalin, 2012). The long
term recorded data starting since 1950 including SST, SLP in Tahiti
3. Study site and Darwin (Kessler, 2002) and trade wind index (Lukas, Hayes, &
Wyrtki, 1984), were downloaded from NOAA’s NCDC and National
This research was conducted in Chickasaw Creek watershed Center for Environment Prediction (NCEP). These data corresponds
(Fig. 3) which is located in Mobile County of Southern Alabama to the Niño 3.4 region (5°N to 5°S, 120°W to 170°W) (Trenberth &
in the Mobile River Basin of Southeast USA. The watershed is pri- Stepaniak, 2001) located in the equatorial pacific and presented in
marily in coastal region but starts at Citronelle in North, drains into Table 1.
Mobile Bay in South, and ultimately discharges to the Gulf of
Mexico. The watershed drainage area is 714 km2, and is character-
5. Methodology
ized by dense residential, forest, agricultural and industrial land
uses. The characteristics of the watershed have been dominated
5.1. Parameter estimation
by the coastal plain geology with highest elevation range of
13.11 m and lowest 0 m above the mean sea level. The region is
Using the generic model structure as described in Fig. 1, the
characterized with higher rainfall, i.e. the average annual precipita-
ANFIS model was developed considering several other inputs. The
tion in the watershed is fairly higher (1651 mm) than average
model development is borrowed from Sharma et al. (in press) and
annual precipitation of the Alabama.
briefly presented in this article having two inputs with two rules
for illustration. In fact, this model has been extended for numerous
4. Data input variables and implemented in our research. In this study,
streamflow (SF) is assumed to be a function of SST, SLP, temperature
The climatic data sets needed for model development including (T) and precipitation (PCP). That is, SF = f (SST, SLP, T, PCP). Consider-
Air temperature (T) and precipitation (PCP) were available from ing u1:u2 and v1:v2 were the corresponding membership functions
National Climatic Data Center (NCDC). Currently, there is no rain for two input variables (T and PCP) and a1:b1, a2:b2 were the asso-
gauge station with in the watershed boundary. The Mobile Regio- ciated model parameters, model output can be determined as linear
nal Airport (Coop ID-015478) is the closest NCDC weather station combinations of consequential parameters. The following equation
for the watershed, which is 32 km farther from the watershed can predict streamflow (SFi) as model output for a given time t1.
boundary. The long term observed streamflow data needed for SFi ¼ ðw1 TÞa1 þ ðw1 PCPÞb1 þ w1 r1 þ ðw2 TÞa2 þ ðw2 PCPÞb2 þ w2 r2
hydrologic modeling were available at the USGS gage (02471001)
which drains 357 km2 of the watershed. ð9Þ
S. Sharma et al. / Expert Systems with Applications 42 (2015) 2213–2223 2217
Fig. 3. Chickasaw Creek watershed in Mobile County of South Alabama, USA, showing the USGS gaging station and land use distribution within the watershed. The land use
classification is as per NLCD land use categories.
Table 1 2 3
2 3 SF 1
Data used for the study with their sources and format (Sharma et al., in press). a1 6 SF 7
6b 7 6 27
Data Source Additional 6 17 6 7
6 7 6 SF 3 7
information 6 r1 7 6 7
6 7
Land use, NLCD 2001 www.AlabamaView.org O¼6 7
6 a 7; and Z ¼ 6 : 7;
6 27 6 7
Soil Soildatamart.nrcs.usda.gov SSURGO soil 6 7 6 : 7
4 b2 5 6 7
data base 6 7
DEM www.seamless.usgs.gov, 10-m 4 : 5
r2
www.datagateway.nrcs.usda.gov resolution SF n
Weather gage station NOAA, National Climatic Data Coop ID-
Center, http://ncdc.noaa.gov 015478
Streamflow www.al.water.usgs.gov (USGS gage
02471001) AO ¼ Z ð10Þ
Hydrologic (stream www.aces.edu/waterquality/gisdata
network) data
GIS layers/tiger files www.AlabamaView.org where O is a matrix of unknown model parameters which can be
determined by solving the following equation.
1
O ¼ ðAT AÞ AT Z ð11Þ
Equations to predict the streamflow time series (SF1, SF2, . . . , SFn)
can be arranged in matrix form as follows:
2 3 where A1 and AT are the inverse and the transpose of the matrix
w1 T 1 w1 PCP 1 w1 w2 T 1 w2 PCP 1 w2
6 7 ‘‘A’’, respectively. ANFIS uses hybrid algorithm to solve this matrix.
6 w1 T 2 w1 PCP 2 w1 w2 T 2 w2 PCP 2 w2 7 Hybrid algorithm (Jang, Sun, & Mizutani, 1997) is a combination of
6 7
A¼6
6 : : : : : : 7
7 least-square method and back propagation algorithm which trains
6 7 the data fairly rapid with speedy convergence. Readers can find
4 : : : : : : 5
w1 T n w1 PCP n w1 w2 T n w2 PCP n w2 detail mathematical description of the hybrid approach in various
articles (Jang, 1993).
2218 S. Sharma et al. / Expert Systems with Applications 42 (2015) 2213–2223
5.2. Input data for baseflow simulation in daily and monthly simulation were
different.
Appropriate selections of suitable membership function with The ANFIS models were developed using 50 years of observed
proper number of fuzzy rules were essential for ANFIS model data. The data were segregated into model training (80%), valida-
development. Selection of appropriate and optimum number of tion (10%) and testing (10%). Initially, both FCM and SCA were
inputs is needed in fuzzy inference system for effective mapping experimented for clustering the data. Finally, we found that SCA
of the input and output relationship. Therefore, correlations of was more suitable to cluster the given datasets. Parameters needed
several relevant inputs with the outputs were examined. for SCA clustering including the membership function, range of
Since streamflow is a collective form of quick response (surface influence of the cluster center, acceptance and rejection ratio
runoff) and the slow response (base flow) of the watershed, it is were determined through a recurrent trial and error method by
always essential to separate the streamflow into surface runoff minimizing the root mean square error and maximizing R2.
(SR) and baseflow (BF) components in order to adequately repre- Similarly, suitable numbers of epochs required for ANFIS model
sent the watershed characteristics. Hence, we used baseflow filter training were optimally decided using trial and error from the
program (Lim et al., 2005) to partition the streamflow. There are visual inspection of error curve for training and validation. The
many reasons of baseflow separation. For example, BF is linked overtraining of the data was eluded by evaluating its performance
with the precipitation of previous time steps whereas SR is entirely against checking data sets and controlling proper epoch number.
contributed by precipitation of the current time step. Recursive
Digital Filter method was utilized to develop the separate SR and 5.4. Performance evaluation
BF data sets and corresponding models from given streamflow
data. The input variables for SR and BF models are listed in Table 2. The performance of ANFIS model was evaluated through R2
The model inputs for SR were temperature, SST, SLP and ‘‘estimated (coefficient of determination), Nash–Sutcliffe efficiency, NSE
surface runoff’’ at different time step (Table 2), where ‘‘estimated (Nash & Sutcliffe, 1970), mass balance error (MBE), and the root
surface runoff (SRE)’’ was determined using SCS curve number mean square error (RMSE). The detail descriptions about the
equation (Bosznay, 1989). The parameter needed for SCS curve modeling protocols (Engel, Storm, White, Arnold, & Arabi, 2007)
number equation including curve number and hydrological soil statistical parameters to measure the model performance are
groups were estimated based on the 2001 land cover data set available in many articles (Moriasi et al., 2007). Additionally, the
(NLCD, 2010), and the SSURGO soil (SSURGO, 2010) database, performance of the ANFIS model was further compared with
respectively. However, the model inputs for BF model were simply another independent watershed model, LSPC (Shen, Parker, &
SST, SLP, temperature and precipitation of various time step Riverson, 2005; Tech & Center, 2009).
(Table 2).
In fact, Optimum sets of best inputs were decided after evaluat-
ing the parameter significant test in multiple linear regressions 5.5. Comparison with LSPC model
(MLR) model. Since precipitation and temperature were affected
by SST and SLP, we wanted to evaluate whether all these inputs The modeling paradigms in two models, ANFIS and LSPC, are
at various lead times were statistically significant (P-value <0.1). completely different. ANFIS is based on the learning approach,
The analysis depicted that both SST and SLP were statistically which is accomplished through model training between the known
significant for the seasons, when ENSO had a distinct signature set of association between inputs and output data sets, whereas,
with streamflows especially for winter and spring. Inputs having LSPC is based on physical processes related with rainfall input
no significant contribution for the improvement of model perfor- and its transformation into runoff at the watershed outlet. The
mance were ignored. brief description of the LSPC model application in Chickasaw Creek
watershed is explained in the following section.
The information about the land use data, soil data and USGS gage is Table 4
reported in Table 2. Performance of the ANFIS model in different modeling stages (monthly simulation
from 1952–2002) (Sharma et al. (in press)).
observed streamflow recorded from 1/1/1990 to 12/31/1996 at a RMSE (m3/s) 2.63 2.37 3.7
USGS gage station (station ID02471001). Since 2001 land cover NSE 0.8 0.85 0.74
MBE 0% 3.8% 8.8%
data sets were used, the calibrated model parameters were applied
to an independent time period (1/1/1997–12/31/2002) for model
validation. This period was selected as a validation period to
evaluate the model performance for the latest land use condition.
Table 5
The simulation was started from 1/1/1985, permitting long spin
Model simulation at daily scale and performance comparison (1990–1996).
up period for the model in order to minimize the effect of unknown
initial moisture conditions, and to stabilize the hydrological Statistics LSPC ANFIS
conditions. RMSE (m3/s) 9.9 7.0
The watershed parameters, difficult to measure, were calibrated NSE 0.49 0.68
MBE 2% 0.6%
within a physically possible range, through the repeated trial and
error procedure until the simulated streamflows closely approxi-
mated with the observed streamflows. The model parameters
adjusted during hydrologic model calibration are presented in of training, validation, and testing with reasonable accuracy
Table 3. Once the model was calibrated and validated, the model (Fig. 4), as indicated through the visual inspection of the observed
was run for another 40 years to simulate the long term streamflow. and simulated streamflow time series. Additionally, the ANFIS
LPSC was run in hourly time step, and simulated flows were model was simulated to a daily scale, separately for SR and BF.
aggregated into daily and monthly time scale for model compari- Table 5 shows the ANFIS model performance for combined SF
sons by computing corresponding statistical error parameters. simulation in a daily scale. Our analysis indicates that that SLP,
Once the LSPC model was simulated for long term, we com- SST and SST(t-1) were the sensitive inputs in daily scale as well.
pared the performance of LSPC model with ANFIS in different ENSO
6.1.2. LSPC
events. ANFIS models was compared with LSPC model in two ways.
The performance of the LSPC model was promising for monthly
In the first step, ANFIS model was developed using long term
simulation and satisfactory for daily simulation demonstrating
dataset starting since 1950, and its performance was compared
that hydrological parameters were able to capture the dynamics
with the same consecutive period of LSPC model calibration and
of the system. The following section further compares the perfor-
validation. In other words, ANFIS model was developed from the
mance of the LSPC model with the ANFIS Model.
long term datasets but ANFIS simulated result from 1990 to
2002, which is a period of LSPC model calibration and validation 6.2. Model comparison
was compared with LSPC model performance. In the second step,
ANFIS model was developed with same period of datasets from Model results after calibration were assessed through the
1990 to 2002, a period of LSPC model calibration and validation. comparison of simulated and observed streamflow in terms of
This was mainly because we wanted to see the response of ANFIS water budget, storm flow comparison with respect to volume
model for shorter-term datasets and make realistic comparison and peak, low flow and high flow period (Table 6), etc. The statis-
with the LSPC model using datasets of the same period. In addition, tical parameters, such as, NSE, R2 and MBE are listed in Table 7. The
we wanted to test the performance of ANFIS model without SST comparisons of observed and simulated streamflow from cali-
and SLP. For this, we removed the SST and SLP from its input brated and validated LSPC are shown in Fig. 5 and Fig. 6, respec-
vector. The model performance excluding SST and SLP was com- tively. LSPC model performance during calibration and validation
pared with the model performance including SST and SLP. period is also compared with the ANFIS model (Figs. 5 and 6).
Even though ANFIS model was developed using long term data-
6. Result and discussion sets from 1950, the two model performance was compared for
the same period of model simulation. That is, we used ANFIS sim-
6.1. Model performance ulated streamflow from 1990 to 2002 to compare with the LSPC
model calibration (1/1/1990–12/31/1996), and model validation
6.1.1. ANFIS (1/1/1997–12/31/2002) period. The model comparisons in those
Table 4 shows the ANFIS model performance for streamflow consecutive periods suggested that the performance of the ANFIS
simulation (combined SR and BF) compared with observed data. model was comparable to the performance of the LSPC model
The ANFIS model performance was consistently better in all stages (Table 7).
Table 3
The calibrated LSPC model parameters.
Table 6
Performance comparison of two models at monthly time scale (period 1990–1996).
60
Observed flows ANFIS-Simulated flows
40
20
0
Feb-52
Aug-54
Feb-57
Aug-59
Feb-62
Aug-64
Feb-67
Aug-69
Feb-72
Aug-74
Feb-77
Aug-79
Feb-82
Aug-84
Feb-87
Aug-89
Feb-92
Aug-94
Feb-97
Aug-99
Feb-02
Fig. 4. Monthly streamflow simulated for 52 years using ANFIS model compared with observed monthly streamflow.
ANFIS model for the same period was compared, and found that
Table 7
the ANFIS model is relatively better than LSPC model (Table 8).
Performance comparison of ANFIS and LSPC model during the period of LSPC model
calibration and validation (monthly simulation). In addition, we analyzed median observed and median simulated
flow for each month and compared using the box plot diagram
Statistics LSPC ANFIS
(Fig. 7). It is noteworthy to mention that the response of the two
LSPC model calibration (1990–1996) models varied significantly from season to season. The median
NSE 0.83 0.85 observed and simulated flows from ANFIS model closely matched
MBE 1.8% 3.8%
R2 0.83 0.86
from January to August, and demonstrated significant difference
for other months (September to December). However, the response
LSPC model validation (1997–2002)
NSE 0.71 0.74
of LSPC model is almost opposite leading to the significant differ-
MBE 4.0% 8.8% ence from February to June, and close resemblance with the
2
R 0.71 0.76 observed streamflow for other months (July to December). This
indicates that ANFIS model can simulate streamflow resembling
the actual observed flows for most of the year and LSPC can simu-
In the next step, the LSPC model simulation (50 years) was late better for low flow period. Because SST and SLP were directly
extended back to 1947 using 5 years of warm up period. The model implemented into a mathematical model (ANFIS), it was interest-
performance corresponding to long term simulation against the ing for us to explore how this model would simulate streamflow
40
Observed flows
Streamflows, m3/s
30
LSPC-Simulated flows
20 ANFIS-Simulated flows
10
0
Jan-94
Mar-91
Jul-91
Mar-93
Jun-94
Dec-89
May-90
Oct-90
Dec-91
May-92
Oct-92
Aug-93
Nov-94
Apr-95
Sep-95
Fig. 5. LSPC model calibration and its performance comparison with ANFIS model from 1990 to 1996 (monthly streamflow).
S. Sharma et al. / Expert Systems with Applications 42 (2015) 2213–2223 2221
40
Observed flows
30 LSPC-Simulated flows
Streamflows, m3/s
ANFIS-Simulated flows
20
10
0
Jan-98
Jan-99
Jan-00
Jan-01
Jan-02
Jan-03
Dec-96
Fig. 6. LSPC model validation and comparison with ANFIS model during LSPC model validation phase.
50 50
Monthly Rainfall (mm)
15 15
Flow (m3/sec)
Flow (m3/sec)
100 100
10 10
150 150
5 5
200 200
0 250 0 250
1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12
Month Month
Fig. 7. Statistical analysis of monthly streamflow simulated using: ANFIS model (left panel) and LSPC model (right panel).
2222 S. Sharma et al. / Expert Systems with Applications 42 (2015) 2213–2223
60
50
40
30
20
10
0
Jun-57
Dec-51
Nov-62
May-68
Nov-73
May-79
Oct-84
Apr-90
Oct-95
Mar-01
Fig. 8. Monthly streamflow simulation using LSPC and ANFIS models corresponding to historic La Niña and El Niño events (Table 9) since 1950. Dots and triangles represent
ANFIS and LSPC simulation during La Niña and El Niño events.
S. Sharma et al. / Expert Systems with Applications 42 (2015) 2213–2223 2223
cannot be expected to significantly improve the simulation by Lukas, R., Hayes, S. P., & Wyrtki, K. (1984). Equatorial sea level response during the
1982–1983 El Niño. Journal of Geophysical Research: Oceans (1978–2012) (89),
additional forcing of SST and SLP.
10425–10430.
Future research can be conducted in ENSO affected watershed Makkeasorn, A., Chang, N.-B., & Zhou, X. (2008). Short-term streamflow forecasting
using ANFIS model, and incorporating trade wind index besides with global climate change implications – A comparative study between genetic
SST and SLP in its input vectors, as the trade wind index is also programming and neural network models. Journal of Hydrology, 352, 336–354.
McCabe, G. J., & Dettinger, M. D. (1999). Decadal variations in the strength of ENSO
responsible for ENSO phenomenon. It is to be clearly mentioned teleconnections with precipitation in the western United States. International
that, the benefit of ANFIS model over the LSPC or any other phys- Journal of Climatology, 19, 1399–1410.
ically based model, can be expected significant, especially for a Moriasi, D., Arnold, J., Van Liew, M., Bingner, R., Harmel, R., & Veith, T. (2007). Model
evaluation guidelines for systematic quantification of accuracy in watershed
watershed which is suffered from lack of adequate rain gauge sta- simulations 10.
tions and also affected by ENSO. Morid, S., Gosain, A., & Keshari, A. K. (2002). Comparison of the SWAT model and
Even though this research does not conclude the supremacy of ANN for daily simulation of runoff in snowbound ungauged catchments. In Fifth
international conference on hydroinformatics. Cardiff, UK.
ANFIS model over the LSPC model, research concludes that ANFIS Mukerji, A., Chatterjee, C., & Raghuwanshi, N. S. (2009). Flood forecasting using
can be a better choice when rain gauge stations are limited. More ANN, Neuro-Fuzzy, and Neuro-GA models. Journal of Hydrologic Engineering, 14,
importantly, ANFIS provides an opportunity to consider additional 647.
Nash, J., & Sutcliffe, J. (1970). River flow forecasting through conceptual models part
inclusions of long term SST and SLP in the equatorial Pacific in I – A discussion of principles. Journal of Hydrology, 10, 282–290.
order to explicitly utilize the ENSO pattern for hydrologic model- Pascual, M., Rodó, X., Ellner, S. P., Colwell, R., & Bouma, M. J. (2000). Cholera
ing. This might be helpful as and when spatially distributed precip- dynamics and El Niño-southern oscillation. Science, 289, 1766.
Piechota, T. C., & Dracup, J. A. (1996). Drought and regional hydrologic variation in
itation datasets are not available and the watershed is affected by
the United States: Associations with the el Niño-southern oscillation. Water
ENSO. Resources Research, 32, 1359–1373.
Pramanik, N., & Panda, R. K. (2009). Application of neural network and adaptive
References neuro-fuzzy inference systems for river flow prediction. Hydrological Sciences
Journal, 54, 247–260.
Rajagopalan, B., & Lall, U. (1998). Interannual variability in western US
Barsugli, J. J., Whitaker, J. S., Loughe, A. F., Sardeshmukh, P. D., & Toth, Z. (1999). The precipitation. Journal of Hydrology, 210, 51–67.
effect of the 1997/98 El Niño on individual large-scale weather events. Bulletin Ropelewski, C., & Halpert, M. (1986). North American precipitation and temperature
of the American Meteorological Society, 80, 1399–1411. patterns associated with the El Niño/southern oscillation (ENSO). The Monthly
Bicknell, B., Imhoff, J., Kittle, J., Jr., Jobes, T., Donigian, A., Jr., & Johanson, R. (2001). Weather Review; (United States), 114.
Hydrological simulation program–fortran: HSPF, version 12 user’s manual. Roy, S. S. (2006). The impacts of ENSO, PDO, and local SSTs on winter precipitation
Mountain View, California: AQUA TERRA Consultants. in India. Physical Geography, 27, 464–474.
Bosznay, M. (1989). Generalization of SCS curve number method. Journal of Sharma, S. (2012). Incorporating El Niño southern oscillation (ENSO)-induced
Irrigation and Drainage Engineering, 115, 139. climate variability for long-range hydrologic forecasting and stream water
Chiew, F., Piechota, T., Dracup, J., & McMahon, T. (1998). El Niño/southern quality protection. Auburn University.
oscillation and Australian rainfall, streamflow and drought: Links and Sharma, S., Srivastava, P., Fang, X., & Kalin, L. (in press). Hydrologic Simulation
potential for forecasting. Journal of Hydrology, 204, 138–149. Approach for El Niño southern oscillation (ENSO)-affected watershed with
Chiu, S. (1996). Method and software for extracting fuzzy classification rules by limited rain gauge stations. Hydrologic Sciences Journal. http://www.tandfonline.
subtractive clustering. IEEE (pp. 461–465). . com/doi/abs/10.1080/02626667.2014.952640#.VD_TYeK5eXg.
Clarke, R. (1973). A review of some mathematical models used in hydrology, with Sharma, S., Isik, S., Srivastava, P., & Kalin, L. (2012). Deriving spatially distributed
observations on their calibration and use. Journal of Hydrology, 19, 1–20. precipitation data using the artificial neural network and multilinear regression
Cochocki, A., & Unbehauen, R. (1993). Neural networks for optimization and signal models. Journal of Hydrologic Engineering, 18, 194–205.
processing. John Wiley & Sons Inc.. Sharma, S., Srivastava, P., Fang, X., & Kalin, L. (2012). Incorporating climate
Committee, A. T. (2000). Artificial neural networks in hydrology. I: Preliminary variability for point-source discharge permitting in a complex river system.
concepts. Journal of Hydrologic Engineering, 5, 115–123. Transactions of the ASABE, 55, 2213–2228.
Crawford, N. H., & Linsley, R. K. (1966). Digital simulation in hydrology’s stanford Sharma, S., Srivastava, P., Kalin, L., Fang, X., & Elias, E. (2014). Predicting total
watershed model 4. organic carbon load with El Nino southern oscillation phase using hybrid and
Dunn, J. C. (1973). A fuzzy relative of the ISODATA process and its use in detecting fuzzy logic approaches. Transactions of the ASABE, 57(4), 1071–1085.
compact well-separated clusters. Shen, J., Parker, A., & Riverson, J. (2005). A new approach for a windows-based
Engel, B., Storm, D., White, M., Arnold, J., & Arabi, M. (2007). A hydrologic/water watershed modeling system based on a database-supporting architecture.
quality model applicati11. In Wiley online library. Environmental Modelling & Software, 20, 1127–1138.
Hansen, J., Jones, J., Irmak, A., & Royce, F. (2001). El Niño-southern oscillation Taheri Shahraiyni, H., Ghafouri, M. R., Saghafian, B., & Bagheri Shouraki, S. (2013).
impacts on crop production in the southeast United States. ASA Special Evaluation of a novel fuzzy method and a conceptual model for a long-term
Publication, 63, 55–76. daily streamflow simulation. River Systems, 20, 249–260.
Jang, J. S. R. (1993). ANFIS: Adaptive-network-based fuzzy inference system. Talei, A., Chua, L. H. C., & Quek, C. (2010). A novel application of a neuro-fuzzy
Systems, Man and Cybernetics, IEEE Transactions on, 23, 665–685. computational technique in event-based rainfall–runoff modeling. Expert
Jang, J. S. R., & Sun, C. T. (1995). Neuro-fuzzy modeling and control. Proceedings of Systems with Applications, 37, 7456–7468.
the IEEE, 83, 378–406. Tayfur, G., Ozdemir, S., & Singh, V. P. (2003). Fuzzy logic algorithm for runoff-
Jang, J.-S. R., Sun, C.-T., & Mizutani, E. (1997). Neuro-fuzzy and soft computing-a induced sediment transport from bare soil surfaces. Advances in Water
computational approach to learning and machine intelligence [book review]. Resources, 26, 1249–1256.
Automatic Control, IEEE Transactions on, 42, 1482–1484. Tayfur, G., & Singh, V. P. (2006). ANN and fuzzy logic models for simulating event-
Jayawardena, A., Muttil, N., & Lee, J. (2006). Comparative analysis of data-driven and based rainfall–runoff. Journal of Hydraulic Engineering, 132, 1321–1330.
GIS-based conceptual rainfall–runoff model. Journal of Hydrologic Engineering, Tech, T., & Center, S. N. A. F. (2009). Loading simulation program in C++ (LSPC)
11, 1. Version 3.1 User’s Manual. Fairfax, VA.
Keener, V., Ingram, K., Jacobson, B., & Jones, J. (2007). Effects of El Niño/southern Thirumalaiah, K., & Deo, M. C. (2000). Hydrological forecasting using neural
oscillation on simulated phosphorus loading in South Florida. Transactions of the networks. Journal of Hydrologic Engineering, 5, 180–189.
ASAE, 50, 2081–2089. Tokar, A. S., & Johnson, P. A. (1999). Rainfall–runoff modeling using artificial neural
Kessler, W. S. (2002). Is ENSO a cycle or a series of events? Geophysical Research networks. Journal of Hydrologic Engineering, 4, 232–239.
Letters, 29, 40-41–40-44. Trenberth, K. E., & Shea, D. J. (1987). On the evolution of the southern oscillation.
Khalil, A. F., McKee, M., Kemblowski, M., & Asefa, T. (2005). Basin scale water Monthly Weather Review, 115, 3078–3096.
management and forecasting using artificial neural network. JAWRA Journal of Trenberth, K. E., & Stepaniak, D. P. (2001). Indices of El Niño evolution. Journal of
the American Water Resources Association, 41, 195–208. Climate, 14, 1697–1701.
Kim, S. M., Benham, B. L., Brannan, K. M., Zeckoski, R. W., & Doherty, J. (2007). Wang, P., Skahill, B., Samaitis, H., & Johnston, R. (2002). GIS-based artificial neural
Comparison of hydrologic calibration of HSPF using automatic and manual network and processed-based HSPF model for watershed runoff in Sinclair and
methods. Water Resources Research, 43, W01402. dyes inlet, WA. Proceedings of the Water Environment Federation, 2002,
Kulkarni, J. (2000). Wavelet analysis of the association between the southern 1547–1564.
oscillation and the Indian summer monsoon. International Journal of Yan, H., Zou, Z., & Wang, H. (2010). Adaptive neuro fuzzy inference system for
Climatology, 20, 89–104. classification of water quality status. Journal of Environmental Sciences, 22,
Lim, K. J., Engel, B. A., Tang, Z., Choi, J., Kim, K. S., Muthukrishnan, S., & et al. (2005). 1891–1896.
Automated web Gis based hydrograph analysis tool, WHAT1. In Wiley online Zadeh, L. A. (1965). Fuzzy sets⁄. Information and Control, 8, 338–353.
library.