Regional Analysis of Trout.exe
ModelPredictions Relative to Survey Estimates
Theodore J. Treska, Patrick J. Sullivan, and Clifford E. Kraft
Coldwater Fisheries Research Program
Department of Natural Resources
CornellUniversity
Ithaca, NY14853
Abstract— Population estimates derived using the Trout.exe model were compared to fall electrofishingsurveys applied to New York trout streams. Thesurvey data were gathered over the years 1985-2002. Contrasts between model predictions and survey observations were madeby region, stream type and certain stocking events in order to isolate trends, to improve model performanceand ultimatelyoptimize stocking efficiency. A variety of stream types existin the State of New York and reliable survey estimation of trout population size in the field can be challenging. Variation between survey observations and model predictions is large at present. Uncertainties are likely associated with survey population estimationas well as in assumed model parameter values.This report provides recommendations regarding how the accuracy of survey estimates and model parameter values can be improved.
Introduction
Methods
Data
Model
Influential Factors
Regional Results
Region 3
Region 4
Region 5
Region 6
Region 7
Region 8
Region 9
All Regions
Discussion
Factors Contributing to Low Survey Estimates or High Model Predictions
Parameter Value Estimation
Other Factors
Conclusions
Recommendations
References
Appendix
Introduction
The diversity of landscapes across New YorkState creates a wide array of different trout waters, from cascading brooks in the Adirondacks to meandering streams in the western part of the state. While it is difficult to identify factors that influence trout populations, examination ofregional patterns may allow inferences to be made about the factors influencing the survival of stocked trout populations.
This report is the third in a series of documents pertaining to the Catch Rate Oriented Trout Stocking(CROTS) program currently utilized by the New York State Department of Environmental Conservation (NYSDEC) to manage the state’sstocked trout streams. This approach, based on work by Engstrom-Heg (1984), has been modeled in a Visual Basic application called Trout.exe that allows biologists to modify inputs and parameters in order to simulate the effects of changes on in-streamtrout populations. Previous reports provide documentation and instruction on how to use the Trout.execomputer program (Treska Sullivan 2004a) and a sensitivity analysis that explore what types of effects might be expected in population and catch numbers with changes to inputs and parameter assumptions (Treska Sullivan2004b). This report focuses on stream stocking and fall survey data provided by NYSDECbiologists in Management Regions 3-9(Figure 1).
Figure 1. NYSDEC regional management divisions with county subdivisions.
The following sections willexamine how well Trout.exe, developed by the NYSDEC and CornellUniversity, predicts population levels on individual streamson a region by region basis. By comparing model predictions withfall electrofishingsurveyestimates, it may be possible to isolate regional trends and patternsthat mayprove useful in modifying the model, updating parameter values, and improving stream survey techniques. In addition,it allows examination how other aspects of the fisheries suchasnatural mortalityand fishing intensity levels affect predictions. These comparisons show how accurately the model is able to simulate the stocked trout populations which have direct impacts on its ability to predict overall angler catch rates which are a main focus of the CROTS program.
Methods
Estimates of stocked yearlingabundancesderived fromlate seasonelectrofishing surveysare compared to predicted trout abundances in streams modeled using Trout.exe according to parameters set for each stream type and stream specific data provided by NYSDECstaff.The Trout.exe model mechanics are documented in the report by Treska and Sullivan (2004a).The survey data consists mainly of single pass electrofishing surveys with assumed efficiencies conducted by biologists in each NYSDEC region although some are the results of a limited number of multiple pass surveys. On these multiple pass surveys, a variety of techniques are used for deriving the population estimates from the data, including the Leslie-Delury (Leslie and Davis 1939; DeLury 1947) and Zippin (1958) methods.
The survey estimatesand model predictions arethen graphically examined using the S-Plus statistical package with survey estimates on the y-axis and model predictions on the x-axis in terms of number of stocked yearling trout per acre. Traditionally, the x-axis is reserved for the component of the model that is known and the y-axis for the component that is observed with error (e.g. estimated total catch by year). In this report, however, both areestimated with error, and so variation may be expected in either the x or the y direction.This will make analysis a little more challenging, but if the bi-directional uncertainty is kept in mind then interpretation will be more direct. Aone-to-one diagonal line has been added to each plot to indicate the agreement between the two estimates. The closer a point is to the one-to-one line, the closer the survey estimate is to the Trout.exe prediction. Differences should be measured in horizontal or vertical distances from the line, not perpendicular to the one-to-one line. Note that if both estimates are off by the same amount from thetrue, but unknown, population value, by 50% say, then the point will still appear on the one-to-one line despite these errors. If only one of the estimates is off by this amount, the effect will be seen as a vertical or horizontal movement from the actual point. A visual representation of these situations is given in Figure 2.
Figure 2. Representation of effects of errors on point locations when error is in survey estimate, Trout.exe estimate, or both. Blue square indicates survey/model estimate. Actual fish population is said to be 50 fish per acre (represented by a black cross).
Information for each point in the regional plots are also provided in a table at the end of the section with information indicating stream name and year, stream type, yearly effort distribution pattern,survey population estimate, Trout.exe prediction, and residual of survey estimate minus corresponding model prediction. The residuals are calculated to provide a measure of the magnitude of the difference between the two estimates. Plots are then examined for trends with respect to model parameters and survey strata such as region, stream type, or fishing intensities.
All model parameters used in the modeling for this report are taken from the early base values established for use in the CROTS program, based on work in the late 1970’s to evaluate trout regulations (Engstrom-Heg & Hulbert, 1982). While conditions have changed throughout New YorkState and its fisheries during the two decades that CROTS has been in use, these values represent a standardized starting point for exploring model performance by comparing fall survey and model population estimates.
Data
Data were gathered from NYSDECregional biologists and managers using a standardized CROTS datasheet shown below. A completed data sheetcontains information on fish stocking, fall electrofishing surveys, and general comments about the stream and survey. Stocking information (denoted in red below) incorporates details of fish stocking events, such as dates, species, average fish lengths, and number stocked per acre. The survey section (denoted in blue) allows biologists to describe the results of fall electrofishing surveysbased on different cohorts separated by age and a distinction of whether that cohort was wild or stocked. Data sheets also include a sectionwhere biologists canadd comments such as noting that stocked fish were not fin clipped making distinctions between wild and stocked difficult or containing information pertaining to the area surveyed. Information from these surveys includesestimated number per acre of that cohort, their age and mean lengths,as well as anote regarding the method of population estimation and the field biologists’ level of confidence in the estimate.
The following information will aid in defining the data gathered from the regions for this report, providing a short description of each component found on the preceding form. The stream types used to characterize each stream are broken into 2 major groups, those starting with Aand those starting with B, with type A being more favorable for sustaining trout. Type As and Asp sections are both good quality streams with unused carrying capacity that differ in having a light to moderate fishery (As) versus a heavily fished section (Asp). When a type A stream is managed under a creel regulation, it is classified as As9 if a nine inch size limit has been applied to that stream and Asnk if the section is restricted to no-kill or catch-and-release angling. The difference between a type Bs and Bp stream is that a Bs is a light to moderate fishery lacking in habitat while a Bprepresents a heavy fishery lacking the potential for holdover or wild contributions. Refer to the Appendix for full stream classification type definitions.
Table 1. Characteristics of Stream Types
Stream Type / Fishery Pressure / Holdover Contribution / Unused Carrying Capacity / Other AspectsAs / Light-Moderate / Some / Significant / Put-grow take fish, usually 2 increments
As9 / Heavy / Good / Significant / Superior growth, survival. Larger, more productive streams
Asnk / Heavy / High / Significant / No-Kill area
Asp / Heavy / Good / Significant / Good quality, late spring increment provides holdover
Bs / Light-Moderate / Some / Significant / Often small, infertile, habitat deficient
Bp / Moderate-Heavy / Small / Little / Little potential for wild, holdover fish
Yearly fishing effort distributions are based on one of two patterns, labeled 1 and 2, with the former representing a distribution with effort more evenly spread over the season while the latter represents a situation where greater effort is expended earlier in the season (Figure 3). This effort distribution is multiplied by a yearly fishing intensity to determine monthly effort values which are in hours per acre per year and are based on estimates or available creel census data.
Figure 3. Yearly fishing effort distribution patterns.
Information pertaining to stocking events includes dates, species, number per acre and average size of stocked fish, information usually collected from stocking cards. This is similar to the information for results of fall electrofishing surveys, which include species, ages, estimates of number per acre, confidence in that estimate, a wild or hatchery distinction and the average length of the age class.
The table below shows the stream data available from each region and the distribution of data over the different stream types, followed by a histogram indicating the distribution of surveys by year (Figure 4). These show that instances of stream typesAsp, As9, and Asnkwere not very common in the CROTS data analyzed for this report, and while some data is available for as far back as 1985, most of the surveys come from more recent years.
Table2. Regional distribution of data.
Stream TypeRegion / As / As9 / Asnk / Asp / Bp / Bs / Total
3 / 0 / 0 / 2 / 1 / 1 / 1 / 5
4 / 1 / 9 / 0 / 0 / 0 / 0 / 10
5 / 5 / 0 / 0 / 0 / 0 / 8 / 13
6 / 8 / 0 / 0 / 0 / 0 / 13 / 21
7 / 6 / 0 / 0 / 0 / 2 / 4 / 12
8 / 5 / 0 / 0 / 2 / 3 / 1 / 11
9 / 11 / 0 / 0 / 2 / 8 / 10 / 31
Totals / 36 / 9 / 2 / 5 / 14 / 37 / 103
Figure 4. Number of stream surveys conducted by year.
Model
The stocking data and stream information from each data sheet were entered into a Microsoft Access database, which was accessed by the Trout.exe program to run simulations.The Trout.exe program generates predictions of trout population abundancedaily from the stocking day through mid-October. For this report,comparisons were made between estimates derived from late season surveys ofyearling stocked trout(8” to 10”) and a corresponding abundance prediction created by Trout.exe. The Trout.exe model is based on day to day population calculations whereas earlier versions were based on monthly calculations. This revision allows model predictionsto be compared withfall survey estimatesconducted at any given point in time.
The structure of the Trout.exe model tracks the stockingincrements (cohorts) individually through the season, calculatinga number of figures including catch rates and population levels.If a length restriction is in place for the stream and has been indicated in the season data part of the inputs, the model appropriately accounts for changes in mortalitycaused by size limit. Although few streams are restricted by length regulations, this method of keeping the componentsseparate also allows examination ofindividual stockstoevaluatethe effects ofalternativemean lengths or stocking dates on population trends and catch rates. The model determines daily rates of fishing mortality and release mortality for each stocked component based on user inputs. Natural mortality is added to this mortality to determine the total mortality and resulting survivorship rates. Rates directly related to fishing are dependent on a variety of month specific parameters such asa monthly proportion of the total annual fishing intensityand a monthly catchability,which portrays how the efficiency of fishing changes throughout the angling season. These monthly rates are divided into daily increments for application during model calculations. For a full explanation of the model mechanics, refer the to the Trout.exe Users Manual (Treska Sullivan 2004a)
Influential Factors
There are many factors that lead to differences between model and survey estimates, and although model-survey comparisons may not supplydefinitive answers, this report along with the sensitivity analysis can be used to broadly identify what factors are influencing trout abundance in these streams. The next section outlines three different possible sources of variation in the data:(1) the model, (2) the fall surveys, and (3) the stream habitat and its characteristics. Further discussion of these factors that influence each component will follow in later sections.
Model Factors
While the individual streams can be highly variable, it is the precision of theinput parameters that will ultimately influence the population estimates generated by Trout.exe. All of the model parameters used in the modeling for this report are taken from the early Trout4x4 spreadsheetsbased on work in the late 1970’s to evaluate trout regulations (Engstrom-Heg Hulbert 1982). While much has changed in the state and its fisheries over the two decades sinceCROTS has been in use, these values represent a standardized starting point to use in exploring the model performance by comparing fall survey and model population estimates. Estimates for values such as natural mortality, fishing intensity, proportion of legal catch kept (PKL), and catchability can be difficult to ascertain, but the degree of their precision can have compounding effects on population estimates.
Survey Factors
In addition to variation stemming from model parameter errors, there is also variation inherent inthe fall electrofishing survey estimates. These errors can arise from the method used to generate the population estimate, the accuracy of the assumed efficiency in single pass surveys, or the selection of the sites used to generate the estimate. Data indicates that a variety of methods are used to produce estimates and that in some cases, only the best habitat was surveyed, both of which could lead to bias in the estimates.
Stream Factors
Individual stream factors and characteristics can have differing effects on the agreement of the survey and model estimates. The geographical location affects regional affiliation, hatchery origin of stocked fish, along with a host of climatic and geologic attributes that can affect the overall stream productivity and dynamics. While data pertaining to the climatic and geologic factors are difficult to assign to individual streams, it is possible to evaluate regional and time dependent effectsin addition to statewide occurrences such as the drought of 2001. As with many aspects of nature there are a host of variables that cannot be accounted for that may lead to unknown biases in the model output. Events such as floods, extremely high temperature periods, and low water conditions are just a few of the natural occurrences that cannot be accounted for by the model with availableparameters and collected data. These events are possible sources of bias that may cause survey estimates and model predictions to be substantially different from each other.
Regional Results
A total of 103 stream survey estimates and model predictions were analyzed. When shown together, and not separated by region, stream type, and various model assumptions, significant differences are observed between the fall survey estimates and model predictions with no evident pattern, although closer inspection indicates that the majority of points appear to be below the one-to-one line used to show model and survey agreement (Figure 5). This predominance of points in the lower half the figure indicates that model predictions of abundance are greater than survey estimates, raising the question regarding whether the model is the better estimate and the surveys are too low or whether the surveys provide the more reliable estimate and the model is too high, or some combination of the two. This will be addressed below. A region-by-region breakdown of the data is now presented to help stratify the analysis by broad geographic characteristics of the regions and associated differences that may occur with regard to fishing activity. Following this regional analysis,a statewide analysis will be conducted with respect to different aspects of stream types and characteristics.
Figure 5. Statewide survey-model comparisons, identified by stream type. Survey estimates on y axis, model predictions on x axis, units are in number of stocked yearling trout per acre.