| CRiSP1.6 Theory & Calibration Manual: III.1 - Calibration Overview | |
III.1 - Calibration Overview
CRiSP.1 is a composite of individual, integrated, process submodels that jointly determine smolt migration and survival.
The model has many parameters which must be determined. The parameters with ecological meaning can often be determined from data sets from other related studies and systems. For the empirical parameters, the model or a submodel are calibrated to lab and field data using a variety of mathematical (optimization) fitting methods. The end result is that through the parameter determination and calibration process, diverse theories and data sets are synthesized into a consistent picture of the process of fish migration and survival through the river system.
Environmental variables describe the observable state of the environment in which fish live. These variables have been determined from historical records dating back as far as 1937 for some of the variables and dating back to 1970 for all variables. Future values of these variables are assessed from runs of hydromodels and management-derived scenarios of river operations. The environmental variable sets must be determined before the model can be calibrated.
Fish passage observations involve a variety of data, extending back several decades, on the passage timing and survival of fish through various segments of the river and hydrosystem. The observations range from relatively small-scale information on the passage of individual groups of fish at individual dams to system-wide estimates of passage and survival of species over specific years. Observations include brand release studies conducted during the 1970's and 1980's and PIT-tag studies starting in the late 1980's. These data sets yield two levels of information. The direct observations provide passage numbers and timing at individual dams as well as returns of adults to dams and collection points. These raw numbers can be further reduced to estimates of migration rates and fish survival between points in the river and in some cases collection efficiencies at dams.
After all possible variables and parameters have been determined and after any submodels which can be calibrated externally to the model have been calibrated, the parameters related to reservoir passage survival and travel time are calibrated within the model. That is, the model-predicted survivals and travel times are calibrated to National Marine Fisheries Service (NMFS) survival estimates and to PIT-tag passage data (courtesy of Pacific States Marine Fisheries Commission). In this way, the whole model is ultimately calibrated to data.
The CRiSP.1 model contains a number of different theoretical constructs that can be selected at run time. The selection of which construct to use depends on the available information, the effect of the feature on the calibration, and its ecological soundness. Any calibration of the model is only specific to a particular choice of theoretical constructs.
III.1.1 - Parameter Determination and Calibration Techniques
Ecological model parameters are determined (estimated) from both field observations and laboratory studies. Estimates made from field observations (such as fish passage timing or mortality rates) are used with the corresponding environmental variables. Estimates made from laboratory experiments are analyzed assuming the corresponding laboratory conditions and are used to infer the relevant ecological parameters. For example, the estimation of mortality from gas bubble disease is made based upon laboratory experiments.
Parameter determination involves mixing results from laboratory experiments, isolated field studies on aspects of migration, and system-wide studies of survival and timing. Parameters are determined directly from studies where possible. Then the calibration proceeds in a hierarchy of steps where submodels are calibrated first (where possible) and finally the migration (travel time parameters) and survival (predation parameters) submodels are calibrated within the model. The calibration sequence is: river and environmental description, flow processes, dam processes, and finally migration processes and predation mortality. The final two steps are in part connected (e.g., in the model, slower migration can result in higher predation mortality), so they are calibrated iteratively until both converge.
Goodness-of-fit
In calibration, the parameters are adjusted so that the model (or a submodel) prediction best fits the observations according to statistical criteria and within ecological constraints. A variety of goodness-of-fit measures are applied in the calibrations. The choice of method depends on the type and quantity of data and the dimensions of the data being fit. Where possible graphical examples are given along with statistical measures of the goodness-of-fit. The following approaches are used.
- Least Squares, 2 dimensional regressions (Press et al. 1992) used for
- Nonlinear regression using the Gauss-Newton algorithm to minimize sums of squares (Statistical Sciences, Inc. 1991) used for
- Hyperbolic "amoeba routine" (Press et al. 1992) used for
- Fourier series analysis (Statistical Sciences, Inc. 1991) used for
- Maximum likelihood estimators via a Marquardt method or a Conjugate Gradient method (Press et al. 1992) are used for
In cases with limited data, statistical techniques might not converge to a unique best fit solution. In this case the calibration is assisted by selecting one of the parameters within its range inferred from ecological constraints, and then calibrating the remaining parameters.
III.1.2 - Parameter Determination and Calibration Status
The calibration process involves fitting the submodels to data using goodness-of-fit measures. Environmental condition variables are ascribed and the ecological parameters are calibrated in a hierarchy that can be organized according to categories of similarity and interdependency.
Status by Type
Environmental variables and ecological parameters are listed below along with a description of the state of their calibration.
- Environmental conditions (define river condition)
- River description parameters relating geometry of river and dams. These parameters are fairly well described and no further improvements of these parameters are expected at this time.
- Headwater parameters define the river environment flow and temperatures. Flow data exist for years from 1960 through 1999. Temperature data in headwaters exists from 1966 through 1999. These parameters are fairly well described and no improvements are expected at this time (other than adding new data for each new year).
- Ecological parameters (characterize ecological interactions)
- Total dissolved gas supersaturation parameters relate the buildup of gas as function of spill, flow, and temperature. These have been calibrated with data current through 1999.
- Age at smoltification initiation (
smolt_onset) and completion (smolt_finish) which are release-specific and also may depend on release date itself. Release information along with the predicted passage information at dams and reaches comprises the passage data in the model. These parameters are critical to survival estimates and are under further study.- Dam parameters describing passage mortality at dams and fish guidance efficiency have been derived from two decades of studies including results obtained from recent PIT tag studies.
- Transportation mortality calibration depends on the transport benefit ratio and in-river survival estimates. Although initial estimates have been obtained, both of these factors are under further analysis.
- Relative predator densities have been derived for each river zone in each reach of the Snake and Columbia rivers and tributaries. These densities were derived from CPUE data where available or converted from predation index data otherwise. For reaches with no data, the densities were assumed to be the same as for nearby reaches. The density information includes base densities for 1990 and prior as well as yearly updates to account for the effects of the pikeminnow removal program.
- Migration rate parameters have been calibrated for spring/summer and fall chinook and steelhead using data from PIT-tag studies.
- Predator activity has been derived from pikeminnow consumption information from John Day reservoir for spring and fall chinook and steelhead.
- Predator temperature response parameters have been calibrated for spring and fall chinook and steelhead using NMFS survival estimates.
Status by Submodel
The CRiSP.1 submodels have been calibrated individually or within the model. Data sources are mentioned in the following list. See also the relevant sections in Chapter 2 as well as the following sections on calibration of gas supersaturation and calibration of migration and predation rate parameters.
Travel Time (Migration Rate)
The travel time submodel was calibrated for fall chinook, spring chinook, and steelhead using tagging data from the entire river system and over the entire migration season. Two separate calibration steps were applied: one to measure the spread of fish as they move through the reservoir, and the other to measure the change in relative migration velocity with fish age. The first used marked, individual stock releases over a short period of time, and the second used marked and recaptured fish over entire seasons.
Predation Survival (Predation Rate)
The predator densities were derived from predation studies in John Day Reservoir and information on the CPUE or the predation index for each of the major reservoirs. The densities were adjusted after 1990 to account for the pikeminnow removal program.
Predator-prey interactions including predator temperature response were calibrated to NMFS survival estimates for fall chinook (1995-1998), spring chinook (1993-1998), and steelhead (1994-1998). Predator activities in the forebay and main reservoir were set to the ratio of smolt consumption by pikeminnow in those zones.
Gas Bubble Disease
The rate of mortality was calibrated from dose-response studies conducted in both field and laboratory conditions.
Dam Passage
Fish guidance efficiency and spill efficiency were calibrated from a number of studies at a variety of dams. Mortalities in dam passage were determined from mark-recapture studies at dams, and we used the values produced by PATH.
Transportation Passage
Separation of large and small fish in transportation was applied from general information on separation criterion for each transportation facility compiled from various sources on the juvenile fishery operations and transportation plans and studies. A transportation mortality was estimated for each species. In addition, time to transport fish through the river system was specified.
Total Dissolved Gas Supersaturation
Total dissolved gas (TDG) supersaturation models were calibrated with data from the U.S. Army Corps of Engineers. The data includes information collected in the 1992 drawdown study in Lower Granite Reservoir and Little Goose Reservoir (Wik et al. 1993), and total dissolved measurements from basin-wide gas monitoring stations from 1994 to 1999.
Flow
Headwater flows in the Scenario Mode were calibrated from information on stream flows provided by the U.S. Geological Survey. In Monte Carlo Mode, modulators of the period average hydroregulation model flows were calibrated with daily flow records at dams.
Water Velocity
Water velocity requires information on reservoir and geometry. The relationship between geometry and elevation and free flowing stream velocities were determined from Lower Granite Reservoir drawdown studies.
Stochastic Processes
The ranges for variables used in the Monte Carlo Mode have been calibrated to available data in the above mentioned studies.
| CRiSP1.6 Theory & Calibration Manual: III.1 - Calibration Overview | |