| SPIGOT is a Fortran package for the generation of stochastic, synthetic streamflow
series, comparable to the SAMS package developed by CADSWES and the U.S. Bureau of Reclamation. The
initial purpose of the package was for long-term reservoir system simulation, but it has
also been used to generate groundwater recharge, water supply demands, monthly
precipitation, and net inflow to the Great Lakes.
SPIGOT has several univariate and multivariate annual streamflow generators, and univariate and multivariate options for disaggregating annual flows to monthly (or weekly) values. These monthly (weekly) values can be further disaggregated to other sites within a basin maintaining their covariances and an approximation to their lag-one correlations at each site within the basin. Procedures are included for automatic or trial-and-error selection and estimation of marginal distributions to describe the flows. Finally, given historical series of daily flows, the weekly generated flows can be disaggregated to daily values using a bootstrap approach, where scaled pieces of the daily record are assigned to each weekly flow. The package uses disaggregation models for generating synthetic streamflows and streamflow forecasts. A univariate or multivariate annual ARMA model can generate annual streamflows at one or more sites. These can then be disaggregated to seasonal values with parsimonious condensed disaggregation models which reproduce the season-season correlations, the seasonal-annual correlations, and the covariances among concurrent seasonal flows. An additional spatial model is included for further disaggregating a seasonal flow among several sites reproducing the concurrent covariances and the persistence at each site. Model parameters are estimated based on method-of-moment (ARMA(1,1)) or maximum-likelihood techniques (all other parameters); parameters can be chosen to come in or out of the model automatically, based on the Akaike Information Criterion, or manually overridden. The package includes four modules for: display of preliminary statistics; parameter estimation; flow generation; and validation using drought or flood statistics. The streamflow generation module reproduces mean, variance, annual to seasonal correlations, site-to-site correlations and persistence of streamflows; and reproduces mean, variance, persistence and explanatory power (R2) of forecasts. It reproduces marginal distributions of flows using logarithmic (2 or 3-parameter), Box-Cox, or gamma transformations of the historical data. The number of seasons per year is user-definable; 12 months or 52 weeks are most common. The initial version of the program was based on models published by Stedinger, Cohn, Pei, Vogel and Grygier in Water Resources Research 24(10), 345, 1989; 21(5), 665, 1985; 20(1), 47, 1984. Subsequent revisions have extended the package to broaden its applicability and improve its modeling of hydrologic variables at various timescales. These include generating seasonal flows from a long record of "index" annual flows (e.g. tree-ring data) or from simulated flows at other sites (e.g. from a climate-change model); improved modeling of weekly flows, including expanding the record length by shifting forward or back by up to two weeks; optional smoothing of noisy weekly data; improved flood modeling by basing choice of marginal distribution on upper half of flow distribution; reduced adjustments required to make monthly/weekly flows sum to annuals by resampling. Full documentation is provided - a technical description and an extensive user's manual, as well as a readme file describing recent upgrades to the software. SYSTEM REQUIREMENTS AND CAPABILITIES The PC version can handle 80 sites and 15 basins using 1.5MB of extended memory. Executable modules are produced using the Lahey compiler. The program can be optimized for various Pentium (I, II, III) or 486 processors. It will also run on a Unix machine if you have a Fortran 77 compiler. COST AND ORDERING INFORMATION The daily parameter estimation/generation module, which also includes validation and graphing routines written in the Matlab programming language, is $100 extra. Members of registered educational institutions who will use the software for research (as opposed to paid consulting) receive a $100 discount. FOR FURTHER INFORMATION CONTACT: Jery Stedinger |
