Evaluating Modelling Strategies For A Complex Water Resource System

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 14

Hydrology for the Water Management of Large Hiver Basins (Proceedings of the

Vienna Symposium, August 1991). IAHS Publ. no. 201, 1991.

EVALUATING MODELLING STRATEGIES FOR


A COMPLEX WATER RESOURCE SYSTEM

D.M. FRICK
Resource Consultants, Inc., 402 W. Mountain Ave., Fort Collins, CO 80521,USA
J.D. SALAS
Department of Civil Engineering, Colorado State University, Fort
Collins, CO 80523, USA

ABSTRACT The use of stochastic time series modelling for


generating long-term hydrologie records to evaluate water resources
projects has become more popular in recent years. In hydrologie
time series modelling there are various uncertainties that, if not
properly evaluated, can lead to significant consequences in planning
and operational decisions for a water resource project. There have
been few studies that have determined the effects of stochastic
modelling strategies and parameter uncertainty on the operational
characteristics of complex water resource systems. The paper
summarizes the results of research focusing on the uncertainties of
stochastic modelling, in particular, those uncertainties related to
modelling strategy and parameter estimation and their effects on the
water resource system simulation. The research evaluated current
modelling strategies and their effects on generation of hydrologie
time series data. The research also investigated methods for deter-
mining the effects of uncertainty in parameter estimation on the
output of a complex water resource system simulation. The research
selected the Colorado - Big Thompson project as a case study. This
system is operated in five river basins and provides water supplies
to most of northeastern Colorado. This case study demonstrated that
the procedures developed provide a systematic method for selecting
the most appropriate modelling strategy. Also, the case study
demonstrated the simplified method for evaluating the sensitivity of
the system output to the stochastic model parameters that were
selected.

INTRODUCTION

The management of scarce water supplies in arid regions is of extreme import-


ance especially during severe droughts. Evaluation of water resource systems is
generally done utilizing historic hydrologie data. In many cases this data is
limited and does not always contain the extreme drought or flood events. The
use of stochastic time series models for generating long-term hydrologie records

105
D. M. Ftick & J. D. Salas 106

to evaluate water resource projects has become more popular during recent years.
Hydrologie time series data is usually generated for use in simulating
operation of a water resource system. The generated data is input to the operatio-
nal model for simulating operation of the various system components such as
reservoirs, diversion systems, power generation facilities, irrigation facilities, or
water treatment facilities. This paper is a summary of research completed by
Frick (1990).
Generally, a stochastic model is one which simulates a known process
using a random component to represent the unknown variability of the process.
Stochastic time series hydrologie models are generally used for generating
annual, seasonal or monthly streamflows. These streamflows are then used to
evaluate the likely performance of the water resource system. The development
of a stochastic streamflow model may involve the following basic steps: (a)
Identify the system and operational model, (b) Obtain historic streamflow
characteristics, (c) Selection of the time series modelling strategy, (d) Estimation
of time series model parameters, (e) Model validation and goodness of fit, and (f)
Water resource system modelling.
When modelling a hydrologie time series for water resource system's
operations a number of uncertainties must be considered. Salas & Smith (1980)
suggest that these uncertainties arise from: (a) characteristics of the hydrologie
time series, (b) modelling strategies and (c) modelling techniques including
parameter estimation.

Model strategies

The development of stochastic model strategies has included numerous model


types and structures. The application of a stochastic techniques is complicated by
the range in type of models, the forms of the models as well as the parameters of
the models selected. One must decide upon the type of models to be used, the
form of the models, and the model parameters needed.
The general approach to stochastic modelling is to develop a model that
reproduces the historic statistics. Generally, the system to be modeled includes
several streams and tributaries which require multi-site analysis. Also, one may
want the models to reproduce historic statistics at both the annual and seasonal
levels. The statistics compared for both annual and seasonal levels include mean
flows, variance, skewness, autocorrelation, and cross-correlation. However, often
statistics such as storage and drought related statistics may be compared as well.
The use of seasonal models, or models with periodic parameters such as
that proposed by Thomas & Fiering (1962), is able to reproduce the seasonal
statistical characteristics, but does not adequately reproduce the annual statistics
as suggested by Valencia and Schaake (1973). In order to develop a model that
can reproduce statistics at both annual and seasonal levels many authors have
suggested the development of annual models for a single site, then using disag-
gregation models for spatial and seasonal disaggregation. These models have
become quite popular for water resource system simulation.
107 Evaluating modelling strategies for a complex system

Annual models

The first level in stochastic modelling is to consider a single site annual hydrolo-
gie time series model. Several types of models are usually considered. These
include autoregressive (AR) models, moving average (MA) models and autore-
gressive moving average (ARMA) models.
One of the most widely used models is the autoregressive, or Markovian,
model. Various forms of the autoregressive (AR) models have been used in the
field of stochastic hydrology. For instance, the stationary AR model with p
coefficients is given by Box & Jenkins (1976), Salas, et al, (1980) as shown in
Equation 1.

yt = v- + 4>i (y,-i - M) + ••• + <£P (yt-P - M) + 6b (i)

where y is the time dependent series and e is the time independent series with
mean zero and variance <r\ . The parameter set of the model is [/t, <£i,...., <£p,
cr^J. Other types of models include the autoregressive moving average (ARMA)
and multi-site models of various forms including AR and ARMA structure.

Disaggregation models

The concept of disaggregation modelling was originally proposed by Valencia &


Schaake (1973). Mejia & Rousselle (1976) and Lane (1979) proposed further
refinements of their original concept. More recently, Stedinger & Vogel (1984),
Santos (1983) and Santos & Salas (1983) have suggested modifications to the
original model concept. Disaggregation modelling is the process by which time
series are generated from a previously generated time series. Typically a time
series, such as an annual data time series, is generated and disaggregated into
monthly data. A temporal example is the disaggregation of annual time series into
a seasonal time series. Table 1 shows the form of several of the models that have
been developed.
Several approaches to disaggregation have been developed to preserve
statistical properties of time series data at more than one level. The largest
difficulty is in the development of schemes which do not require a large number
of parameters but preserve the desired statistics.

Selecting a modelling strategy

The process of selecting the modelling strategy is one of the most important steps
in the stochastic modelling process. If the wrong modelling strategy is selected it
can affect the predicted operation of the water resource system which may result
in design for non-optimal performance.
The general approach to the modelling strategy selection is to determine
the data necessary for simulating operation of the water resource system applica-
tion. The second step is to determine the important statistical properties of the
D.M.Frick&J.D.Salas 108

Table 1 Forms of disaggregation models used in the literature.

Form of Disaggregation Parameters Reference Eq.


Model

Y = AX+ Be Y.A.B Valencia and (2)


Schaake (1973)

Y = AX + Be + CZ A,B, C Mejia and Rousselle (3)


(1976)

Yr = AJC, + B,t + C,Yr-, Ar, B„ C, Lane (1979) (4)

Y„ =A& + B,Y„_, A„ Br C„ Hr Grygier & Stedinger (5)


(1987a and 1988)
P T-l
+ cr
. 1-1 u-l
+HTen

Y = A% + B*e + CZ1 A', É,C Santos (1983),San- (6)


tos & Salas (1983)

historic data to preserve. Finaliy, the appropriate modelling strategy is selected


which will generate the necessary data and preserve the statistical properties that
are important. Fig. 1 shows the decision tree for selection of the modelling
strategy. Described below is the strategy selection procedure. The decision points
are highlighted and information is given to aid in the selection of the appropriate
decision.
In general, previous researchers have approached the development of a
stochastic model for a hydrologie system by first identifying the statistical
properties of the of historic time series. Then a model is developed to reproduce
only those statistically significant characteristics. This approach will reproduce
those statistics, but will not necessarily preserve the statistical properties that are
important for the water resource system operation. The approach proposed here
is reversed from the general approach in that first, statistical properties important
to water resource system operation are determined, then the statistics of historic
data are evaluated and the stochastic modelling strategy selected to match historic
statistical properties necessary to system operation.

Parameter uncertainty

Numerous authors have proposed methods of considering parameter uncertainty


in stochastic modelling. Young et al. (1970) considered the impacts of parameter
uncertainty on the effects of the economic analysis for proposed water resource
systems. This approach utilized Bayesian techniques for evaluating parameter
109 Evaluating modelling strategies for a complex system

See Section 3.3.4 for


Description of
Disaggregation Methods

Fig. 1. Framework for selection of modelling strategies.

uncertainty. Stedinger and Taylor (1982b) looked at the uncertainties of parame-


ters in stochastic modelling. Their analysis looked at a fairly simple case because
it required numerous runs or the model with randomly selected parameters.
Generally authors have proposed either analytical techniques on random selection
of numerous parameter sets to consider parameter uncertainty. This research
looked at utilizing first order analysis techniques to consider parameter uncer-
tainty.

CASE STUDY: COLORADO - BIG THOMPSON SYSTEM

A case study was selected which used a complex water resources system to
demonstrate the selection of stochastic modelling strategy and testing the procedu-
re for evaluating parameter uncertainty. The case study selected was the Colora-
do-Big Thompson Project (C-BT). The C-BT system is a major water resources
project constructed during the 1940s and 1950s in the northeastern portion of
Colorado, USA.
An operational model was developed to determine long term system
performance which is driven by monthly streamflows from five river basins,
three of which control the demand of the model and two controlling the supply to
the model. The model is then driven by stochastically generated monthly flows
D. M. Frick & J. D. Salas 110

and the parameter uncertainty of these stochastically generated flows are analyzed
to determine the effects on the model output.

Description of system

The C-BT project is one of the most complex projects ever undertaken by the
Bureau of Reclamation (USER, 1967). It is composed of numerous major
structures for irrigation and power generation through the mountains and plains
area in northeastern Colorado. The project diverts water from the Colorado River
Basin on the Western Slope of the Continental Divide to the Eastern Slope for
power generation, agricultural irrigation, and municipal and industrial water
supplies. The primary purpose of the project was to provide supplemental
irrigation water to about 290 000 irrigated hectares in eastern Colorado. The
power is generated as the water is diverted across the Continental Divide and
dropped nearly 900 meters to the irrigation storage facilities on the eastern slope.
The project was initially authorized in 1937 and was completed in 1959.

System operational model

A simplified operational model was used which simulated the operation of the
major components of the C-BT system, not including the power generation
component or small regulation facilities such as Shadow Mountain Reservoir,
Lake Estes, etc. The model was supplied by Northern Colorado Water Conser-
vancy District (NCWCD) and was further modified for this research to be driven
by monthly stream flows. A schematic model is shown in Fig. 2. This figure
shows the major components of the model.
The model is driven by demands for water from the Poudre River, Big
Thompson River, and St. Vrain Creek, and its supplies are governed by flows
from the Colorado River and Willow Creek.
The model is operated on a monthly time step basis where demands for
water are taken from the system at the Big Thompson River, Poudre River, and
St. Vrain Creek. The model simulates water deliveries for irrigation and munici-
pal demands. Various users own shares of water from the project. There are
approximately 300 000 total shares.

Demand analysis

The water supplies from the C-BT system are set each year by the Northern
Colorado Water Conservancy District in terms of a quota. The quota is the
amount of water delivery per share and ranges from 600 to 1200 m3. The quotas
are based upon forecasted streamflows for the Poudre River, Big Thompson
River, St. Vrain River, and South Platte River. These quota are set such that in
dry years the maximum quota is used and in wet years the lower quotas are used.
With this operational policy, a methodology was developed to predict demands on
the C-BT system by looking at these annual and monthly streamflows for the
Ill Evaluating modelling strategies for a complex system
5i<n

'Q-S
51
s
!
I A
0<]
1*;
D. M. Frick & J.D. Salas 112

Cache la Poudre, Big Thompson, and St. Vrain rivers. Correlations of each
month and the total annual flows for each river system were developed. During
the winter months, usually the only demands for C-BT water is from municipal
users and is relatively constant. Table 2 shows the results of a model run using
generated demand data based on monthly regressions of stream flow versus
demand.

Table 2. Calibration results using generated monthly flows and demands


generated from monthly stream/lows (1964-1983).

Parameter 20 Year Simulated


Actual Data Data

Granby Reservoir Mean Annual Storage (Mm3) 401.1 400.2


Granby Spills Mean Annual Flow (Mm3) 12.6 11.3
Adams Tunnel Diversion Mean Annual Flow (Mm3) 281.7 285.2
Poudre Demand Mean Annual Delivery (Mm3) 122.8 121.8
Big Thompson Demand Mean Annual Delivery (Mm3) 62.9 62.3
St. Vrain Demand Mean Annual Delivery (Mm3) 78.9 78.4

STOCHASTIC MODEL STRATEGIES

Frick (1990) developed a method for selecting stochastic modelling strategy. The
selection of the stochastic model for this case study illustrates typical application
of stochastic modeling. The procedure for selecting the stochastic model strategy
that is demonstrated in the case study can be applied to almost any complex water
resource problem. The stochastic model developed can be used to simulate long
term system performance especially during draughts that may not have occurred
during a brief historic record.
The first step in the development of a stochastic model for this system is
to identify the data necessary for the operational model and the important
statistical properties of that data which must be preserved. The next step is to
determine the statistical properties of the data. A modelling strategy can then be
developed to preserve these important statistical properties.
The C-BT system is operated in five river basins systems. The characte-
ristics of the gages are shown in Table 3. The Hot Suphur Gauge includes both
the Willow Creek and Granby gauge and had a longer period of record.
The total flow from all basins is an important statistic to preserve since the
total water available is an indication of drought conditions within the C-BT
system and of the total water available for irrigation and municipal use. Monthly
data is important for operational modelling since the model is designed for
monthly operations. Thus monthly data at the sites are required. Cross-correlati-
ons of the monthly data is important for monthly operations but lag cross-correla-
tions are not as important at the monthly level.
Based upon the analysis of the data, it was concluded that the stochastic
113 Evaluating modelling strategies for a complex system

Table 3. Stream gauges used in the case study.

Stream

Cache la Big St. Vrain Colo. Colo. Willow


Poudre Thompson River Hot Granby Creek
Sulfur

Drainage area 2740 790 550 2030 830 270


(km2)
Gauge Elev. 1590 1620 1610 2340 2430 2510
(m) 1884-1983 1939-1983 1896-1983 1905-1983 1951-1983 1954-1983
Years of
record 57 5TI 57 57 24 24
No. Years
used
'/Records were extended from 1927 to 1939.

model should preserve not only the mean variance and autocorrelations of the
historic data, but should also account for the skewness and cross-correlations.
Also, since the total water supply is important for water supply purposes, the
total annual flows are important to preserve the statistical characteristics. Review
of the aggregated annual data, resulted in development of a AR(1), AR(2),
ARMA(1,1) and ARMA(2,1) model for comparison purposes. Next disaggre-
gation strategies were developed. Seven disaggregation strategies were utilized as
described below.
Method 1 - univariate full disaggregation: spatial then seasonal (Valencia -
Schaake & Mejia and Rousselle). The first approach was to develop spatial
disaggregation of the aggregated annual flows to single site annual flows at the
five basins being modeled. Each basin's annual flow was then disaggregated into
monthly flows for use in the operational model. This case was considered only
for non-transformed data (Model NT-1)
Method 2 - staged disaggregation: seasonal then spatial. The second
approach was to utilize seasonal then spatial disaggregation techniques using the
basic Valencia & Schaake equation. This approach was suggested by Stedinger &
Vogel (1984). This case was considered only for non-transformed data (Model
NT-2)
Method 3 - univariate step disaggregation: spatial then seasonal (Santos -
Salas). The third approach was to disaggregate spatially in two steps; first from
the univariate case to the bivariate case of east slope basins and west slope
stations; then disaggregate spatially into the five basins' annual flows. This
method was considered for both non-transformed and transformed data (Models
NT-3 and T-3 respectively)
Method 4 - univariate step disaggregation: seasonal then spatial (Santos -
Salas). The fourth approach was to utilize the seasonal step disaggregation
process presented by Santos & Salas (1983) then the spatial disaggregation. This
method was considered for both non-transformed and transformed data (Models
NT-4 and T-4 respectively)
D. M. Frick & J. D. Salas 114

Method 5 - multivariate step disaggregation: spatial and seasonal (Santos -


Salas). The fifth approach was to utilize the multivariate step disaggregation
approach proposed by Santos & Salas (1983). This method was considered for
both non-transformed and transformed data (Models NT-5 and T-5 respectively)
Method 6 - condensed disaggregation: seasonal and spatial (LAST
Package). The next method tried was the LAST method presented by Lane and
Frevert (1985). This method was considered for transformed data only
(Model T-6)
Method 7 - Condensed disaggregation: seasonal and spatial (Grygier and
Stedinger SPIGOT). Finally, the disaggregation approach presented by Grygier &
Stedinger (1988) was used. This method was considered for transformed data
only (Model T-7).
The results of the modelling strategies were compared at three levels. The
first level is the annual data in which the statistics of annual generated data is
compared to the statistics of historic flows; the second level is in the seasonal
data in which generated seasonal statistics are compared to historic seasonal
statistics. The final level of evaluation is the operational model results. The
operational characteristics of the water resource system model under generated
data for each model was compared to historic operational characteristics of the
system.
At the annual level disaggregation methods were compared for the means,
standard deviations, skewness, autocorrelation. Also, drought statistics run sum
and run length were compared for each site. Monthly statistics for each model
were compared. Means, variances, skewness and autocorrelations and cross-
correlations were compared to historic values and compared among models.
Finally model results were compared with the results of the operational models.
These results compared total deliveries, shortages (during severe droughts), and
spills that occur from Granby Reservoir.
The results of the analysis of modelling strategy indicated that the historic
statistical parameters were generally preserved. The seasonal skewness was better
simulated with the transformed models as expected. The most important factor is
the simulation of critical system output parameters. In this case, these would be
the deliveries of water, the spills from the reservoir, and the shortages in meeting
demands. These parameters can be evaluated by reviewing the Adams Tunnel
diversions, the Granby Reservoir spills and the number of years in which
shortages occur.
Fig. 3 shows a box plot of Adams Tunnel diversions. The average
diversions are shown for each modelling strategy along with the standard deviati-
ons of the means for the 50 runs. Also shown is historic Adams Tunnel deliveries
based on 24 years of records. As can be seen from the plot, the non-transformed
models tend to slightly underestimate the deliveries whereas the transformed
models tend to slightly overestimate the deliveries. However, all are within one
standard deviation. Methods NT-1, NT-5 and T-4 provide the closest estimates.
Further analysis was also conducted for spills from Granby Reservoir and
system shortages for the various modelling strategies. As a result, the selection of
T-4 appeared to be appropriate for several reasons, first this model was on one of
the best in simulating annual drought characteristics and seasonal streamflow
statistics. Secondly, the model simulates diversions closely and finally the model
115 Evaluating modelling strategies for a complex system

Legend:
+ 2 Std. Dev.

+ 1 Std. Dor.

a
A n A n A nAnn
Mean of 50 r u n s

- 1 Std. Dev.
g 830

n D u n y uytjyt^
- 2 Std. Dev.

S
Historic Mean
a 220
(24 years)

NT-I NT-S NT-3 NT-4 NT-5 T - 3 T-4 T-5 T-6 T-7


MODEL

Fig. 3. Box plot of Adams Tunnel diversions.

has a fairly good simulation of the spills and shortages, particularly the variance
of shortages is generally less than most other models. This model is the model
that would have been selected based upon the selection framework presented in
Fig. 1.

STOCHASTIC MODEL PARAMETER UNCERTAINTY

Selection of the appropriate stochastic model for the system is the first step in the
modelling process. Once the model is selected, parameters must be estimated.
Since the parameters must be estimated from a finite record of historic data,
there is uncertainty in the values selected. The parameter values may also be
affected if the period of record contains outliers which affect parameter estimati-
on. The parameters estimated can affect the results of operational modelling
significantly and therefore uncertainty in their estimates should be considered
when using stochastic modelling for determining operational performance of a
water resource system. Presented below is a simplified procedure to consider
parameter uncertainty for the annual stochastic model.
The annual AR(2) model presented above was selected to consider the
effects of parameter uncertainty. Parameter sets were then selected from the
range of possible values and were utilized to generate flows. These flows were
then used in the operational model and the results compared to the results of
model runs based upon expected values of the parameters. In this case the output
compared was deliveries, shortages and spills from the C-BT system. The first
step in this analysis was to determine the probability distribution of the parame-
ters.
The results of the operational model runs are then compared to the
parameters of the stochastic model utilizing a first order analysis. This then
results in a simplified method for relating the variance of the output based upon
the variance of the parameters, thus providing an estimate of the reliability of the
D. M. Friche. J. D. Salas 116

modeling strategy selected. Again, output variables compared were deliveries,


shortages and spills from the C-BT system.
Utilizing the confidence limit presented above, stochastic model parame-
ters were estimated for the 80%, 95% and 99% confidence limit for the AR(2)
model parameters. The model was run for each parameter at each confidence
interval limit, generating 50 sets of 50 years of data, then running the operational
model run for these 2,500 years of synthetic records. In all, 24 model runs were
made, each run considering variation of one parameter at a time, although
variation of the autocorrelation coefficients did account for the covariance
between the coefficients and were adjusted accordingly.
The first order approximation of output variance is expressed in Equation
7 for the AR(2) case.
Sg var (o^)
var(g) 'var(M)+ii;i[|JvarW+ ( | (7)
S/S
where (ôg/ôx) denotes the partial derivative of g, any of the operational model
output parameters as shown in Table 1, with respect to x, any of the parameters.
The variance of the output parameters is g evaluated at n, the population value.
This then relates the output of a water resources model to the parameters of the
AR(2) model. The problem becomes determining the partial derivatives. First,
the important output variables are selected. This would include deliveries,
diversions, spills and shortages. The rate of change in output parameter (ôg/ôx) is
then evaluated for the variation of each stochastic model parameter. The resulting
values are applied to Equation 7.
For the time series of the five sites considered in the C-BT case study the
year 1983 could be considered an outlier. During this year the largest runoff
event of the record for all sites occurred and could be considered an outlier. The
effect of this event on the output of the water resource system model can be
estimated by determining the parameters of the AR(2) model without considering
this event and then using the first order analysis to determine the resulting change
in output that could be expected from the model with the revised parameters.
The results of this analysis are presented in Table 4. This shows the
estimated output using first order analysis and the results of the model runs using
the revised estimate of the parameters. This shows that the resulting output is
nearly the same as using the first order analysis and therefore additional runs of
the model are not necessary in order to evaluate the effects of outliers. If the
(ôg/ôx) over a larger range in which system limits are reached these estimates
would overestimate the shortages and spills.
Table 4. Results of model output considering outliers.

Annual Diversions Annual Deliveries Spills and Shortages


(Mm3/yr) (Mm3/yr) (Mms/yr)

Adams Willow Poudre Big T. Poudre Spill Years Amount

*g 1.09 0.27 0.27 0.53 0.39 -5.03 0.38 0.36


g 284.3 48.2 126.9 64.2 64.2 16.9 0.90 0.73

Model Run 231.3 39.3 103.3 52.2 63.5 13.6 1.0 0.72
117 Evaluating modelling strategies for a complex system

The errors in the estimate are, for all output variables, less that 1 MmVyr. This
is well within acceptable limits. Therefore, using the first order analysis tech-
nique will allow estimation of the effects of outliers in the data without rerunning
the model.
The results of this analysis indicate that, the first order analysis technique
does not only provide a simplified method to determine the confidence limits of
the output of a complex water resource system model, but also this technique can
be used to evaluate the effects of outliers or other uncertainty in parameters of
the stochastic model without additional model runs.

CONCLUSIONS

The use of stochastic time series modelling in hydrology is important when


predicting the availability of water resources during extreme events. The selection
of stochastic modelling strategy and parameters can produce significant effects in
the output and operational decisions for a water resource system. This paper
presented a methodology for selecting a modelling strategy for stochastic
modelling of complex water resource systems. The paper also presented a
methodology for estimating model parameters and to determine the effects of
uncertainty of annual stochastic model parameters on the important output of the
operational model such as deliveries, shortages or spills.
The unique problems related to the C-BT system is that water supplies
which are meant to meet demands during droughts come from adjacent basins
which are highly correlated to the basins receiving the supply. Therefore the
system performance during drought conditions is very important in meeting the
irrigation and municipal demands of the area. This approach showed a systematic
method for developing a stochastic model for simulating long term system
performance. It also showed a method for analyzing the sensitivity of stochastic
model parameters. One interesting variable is that C-BT system quotas are set
based on political decisions as much as hydrologie knowledge, therefore this is
one variable that cannot be predicted.

ACKNOWLEDGEMENT
The contributions of the second author was partially funded by the Colorado
Experiment Station Project N. 645.

REFERENCES

Frick, D.M. (1990) Developing and evaluating modelling strategies for a complex water resource
system. Ph.D. Dissertation, Colorado State University: Fort Collins, Colorado.
Grygier, J.C. & J.R. Stedinger (1988) Condensed disaggregation procedures and conservation
corrections for stochastic hydrology. Water Resources Research. 24(10):1574-1584.
Grygier, J.C. & J.R. Stedinger (1987) SPIGOT: A Synthetic Flow Generation Software Package -
Technical Description, Version 2.0. Cornell University: Ithica, New York.
Grygier, J.C. & J.R. Stedinger (1987) SPIGOT: A Syntlietic Flow Generation Software Package -
User's Manual Version 2.0. Cornell University: Ithica, New York.
Lane, W.L. (1979) Applied Stochastic Techniques (LAST computer package) User Manual. Division
D. M. Frick & J. D. Salas 118

of planning and Technical Services, Bureau of Reclamation: Denver, Colorado.


Lane, W.L. & D.K. Frevert (1985) Applied Stochastic Techniques (LAST A Set of Generally
Applicable Computer Programs) User Manual. Fourth Revision. Division of Planning and
Technical Services, Bureau of Reclamation, U.S. Departmant of Interior: Denver, Colorado.
McLeod, A.I. & K.W. Hipel (1978) Simulation procedures for Box-Jenkins models. Water Resources
Research. i4(5):969-975.
Mejia, J.M. & J. Rousselle (1976) Disaggretion models in hydrology revisited. Water Resources
Research. 72(2)185-186.
Salas, J.D. & R.A. Smith (1980) Uncertainties in hydrologie time series analysis. Preprint No.
80-158, ASCE Convention and Exposition Portland, Oregon, April 14-18.
Salas, J.D., J.W. Delleur, V. Yevjecich & W.L. Lane (1980) Applied Modeling of Hydrologie Time
Series. Water Resources Publications.
Santos, E.G. (1983) Disaggregation Modelling of Hydrologie Time Series. Ph.D. Dissertation,
Colorado State University: Fort Collins, Colorado.
Santos, E.G. & J.D. Salas (1983) A parsimonous step disaggregation model for operational
hydrology. Paper presented at Fall Meeting of American Geophysical Union, San Fransico.
AGU Abstract, EOS, Vol 64, No.45, p 706, November.
Stedinger, J.R. & M.R. Taylor (1982) Synthetic streamflow generation 1 - model verification and
validation. Water Resources Research. iS(4):909-918.
Stedinger, J.R. & M.R. Taylor (1982) Sythentic streamflow generation 2 - effect of parameter
uncertainty. Water Resources Research. iS(4):919-924.
Stedinger, J.R. & R.M. Vogel (1984) Disaggregation proceedures for generating serially correlated
flow vectors. Water Resources Research. 20(l):47-56.
Thomas, H.A. & M.B. Fiering (1962) Mathematical synthesis of streamflow sequences for analysis of
river basins by simulation, in Design of Water Resource Systems, edited by A. Maas, M.M.
Hufschmidt, R. Dorfman, H.A. Thomas Jr., S.A. Marglin & G.M. Fair. Harvard Univ.
Press.
U.S. Bureau of Reclamation (1961) Reclamation Project Data. U.S. Government Printing Office.
Valencia, D. & J.C. Schaake Jr. (1973) Disaggreation processes in stochastic hydrology. Water
Resources Research. P(3):580-585.
Vicens, G.J., I. Rodriquez-Iturbe & J.C. Schaake Jr. (1975) Bayesian framework for use of regional
information in hydrology. Water Resources Research, ii(3):405-414.
Young, G., G.T. Orlob & L.A. Roesner (1970) Decision criteria for using stochastic hydrology.
ASCE Hydraulics Journal, P6"(HY4):911-925.

You might also like