- The Sage Grouse of Fort Yakima -
Dynamic modeling has been used for generating ecological models. Recent
advances in computer processing speeds and efficiencies has made it possible
to incorporate the spatial distribution of ecological components as an important
aspect of the ecosystem dynamics.
We demonstrate the state of the art in Dynamic Landscape Simulation hardware,
software, and interdisciplinary research collaboration. The Sage Grouse
model laid a foundation which has since been used to generate location-specific
ecological models that we expect to become useful tools for improving land
management.
This study also provides the starting point for the design and development
of a next generation software environment that will allow increasingly efficient
generation of location specific ecological models. Some guidance for this
next generation is provided in the conclusions.
The purpose of this study was to demonstrate the collective abilities of a diverse group of researchers to assemble computer hardware, software, and personnel to create a working dynamic spatial model of a selected ecosystem.
An interdisciplinary group of researchers used state-of-the-art hardware and software environments to design and develop a demonstration model of the impacts of military training on a threatened species. Individuals from the Construction Engineering Research Laboratories (USACERL) and the University of Illinois at Urbana-Champaign and National Center for Supercomputing Activities (NCSA) formed the research team.
Although a real problem was utilized, this study resulted in a foundation only for the design and development of dynamic spatial ecological models. This model should not yet be incorporated into actual land management policy decisions.
This work describes the result of a demonstration project which uses state of the art dynamic, spatial, ecological simulation technology. This work forms the foundation for the design of significantly more powerful and flexible computer software. In the meantime, military installation mangers can utilize the software and hardware capabilities used in this demonstration to generate location, user, and landuse specific dynamic models today.
* Page Created September 1993, Revised September 1994 and September 1998
A powerful set of hardware and software tools was assembled and, in conjunction with programming expertise, was used by a multidisciplinary group of researchers to develop a working dynamic, spatial, ecological model.
Hardware environment includes color Apple Macintosh machines, UNIX-based
workstations, and a CM-5 Connection Machine (a parallel processing computer).
University of Illinois and faculty used the Macintosh machines to design
and develop the cellular ecological model. The UNIX environments generated
system starting parameters in the form of digital maps and to developed
the CM-5 software. The CM-5 was the target machine for running the simulations.
Software environment consists of a combination of commercial, public
domain, and special purpose programs. STELLA provided
the basic modeling environment used by the multidisciplinary team. STELLA conversion
programs developed by Dr. Thomas Maxwell (1993) transformed the STELLA models
into C code that could be run in parallel processing environments, (in this
case, the CM-5). To seed the model with a starting point, the public domain
GRASS was used. Intercellular movement of information between adjacent grid
cells was done with code written in FORTRAN by Albert Cheng of the University
of Illinois National Center for Supercomputing Activities. STELLA Translator</H2>
Once the cellular model was created in STELLA, it was applied simultaneously
to all grid cells in the study area (342 cells x 342 cells = 116964 cells).
The cell size was chosen to represesent an area that was normally large enough
to hold only one female grouse at any instant. The translation from the STELLA
equations was done with software developed for this purpose (Maxwell 1993).
The output of the translator can be compiled to run in several hardware environments
including a SUN workstation, a network of SUN workstations, the parallel CM-5
machine, or a small network of transputers attached to a desktop Macintosh.
We ran the program on the CM-5 at a rate of 60 gigaflops per model year.
FORTRAN Programming- The cellular model (created through STELLA)
as applied in parallel (through Maxwell's translator) was not sufficient
to generate a complete model. Although the cellular model simulates the
interactions within each cell, it does not provide rules for exchanging
information between cells. Research programmer, Mr. Albert Cheng of the
National Center of Supercomputing Applications, developed the code that
provided the movement of relevant information (movement of individual animals)
between cells.
GRASS GIS Cellular modeling is an extension of raster Geographic
Information System (GIS) technology. As in any simulation process, the starting
state must be modeled and provided for the simulation. In the case of cellular
modeling, the initial state was represented as a series of maps; each map
provided a single-state variable for each cell. Cellular simulation output
was also in the form of digital maps which could be fed back into a raster
Geographical Information System for further analysis and display. The GIS
employed was the Geographic Resource Analysis Support System (
GRASS), an internationally used program, in the public domain.
The project's fundamental goal was to draw together and demonstrate a suite
of expertise, hardware, and software that could be applied toward the design
and development of cellular-based ecological models. The following objectives
were:
To generate additional interest within the US Army Corps of
Engineers and environmental offices at Army installations.
To take a cellular approach toward modeling as opposed to
attempting to model the ecosystem as a whole.
To run the model on the NCSA CM-5 parallel processing machine.
The Yakima Training Center (YTC) in Washington State provided the modeling problem.
The researchers chose to model the interaction between human training and sage
grouse behavior in a threatened sage grouse habitat at YTC. Based on the recommendations
of the field biologist, a set of training ranges, home to a significant percentage
of the state's threatened sage grouse population was chosen. Military training
landscapes are managed differently than contiguous and surrounding lands. While
neighboring lands are often highly managed for agriculture or human habitation,
the training land is likely to have been maintained in a more natural state.
At YTC, a small community of sage grouse remains on a remnant of an original
desert-steppe habitat that had extended over much of the Northwest. Although
the installation is more accommodating to the birds than surrounding private
lands, increased training intensity might produce a negative effect. We suspected
that training frequency and intensity could be tolerated if it were scheduled
with respect to seasonal variations in the sensitivity of the birds.
The sage grouse problem provided an excellent test ground for modeling capabilities
for the following reasons:
Spatial in nature</U> - The problem must incorporate
the movement of individual Sage Grouse across the landscape.
Human activity - Because human activity is involved, the problem
must consider management and policy decisions, making it more interesting to
communities that will benefit directly from the development of ecological models.
Real problem- Although theoretical problems can be effective
test environments, the specification of a real problem represented by real land
managers provides effective additional motivation.
Nonmovable components - This problem required the modeling
of sessile communities such as sagebrush, forbs, and grasses based on physical
qualities such as slope, elevation, aspect, and soil characteristics.
Available data - Because the research was largely unfunded,
the data had to be already collected. YTC has a good spatial database and had
recently completed documentation on the study of the Sage Grouse communities.
Such data is based on field studies which provides necessary input for model
design and calibration.
Interesting to land managers involved with problem - The Sage
Grouse provided a problem important to environmental groups and to Army installations.
There are strong interests in managing the land exclusively for the Grouse.
There are equally strong requirements to use the land exclusively for military
training. The result was an important forum for policy debate and land management
innovations.<U>
Interesting for researchers - This problem was an example
of the kind of modeling that provides significant academic challenges. It simultaneously
requires degrees of simple cause-effect modeling with spatial behavior modeling
and also involves chemical, physical, and biological processes.<U>
Interesting for students - The technology used in this project
will be used by today's students as they enter the world of land management.
One outcome of the project was to train students who might participate in future
research and application of these modeling techniques.
From a management perspective it would be impossible for the twenty person modeling
group to be responsible for every aspect of the model. For this reason, four
modeling groups (Sage Grouse, Vegetation, Human, and Physical) and a software
development group were formed. The software group would develop the software
that would allow the cellular model created by the other groups to be run on
the CM-5 Connection Machine. It would also be responsible for the design of
the algorithm that would provide for the inter-cell movement of the Sage Grouse.
The STELLA software is an excellent organizing device for the design and development
of single models, written either by individuals or by groups. In this case,
four separate groups were developing different parts of the final model. Therefore,
a shared base model was created. It provided a simplified time-series of anticipated
output from the other models within which individual submodels could be developed.
The base model contained the expected output of each of the four groups' efforts.
Each group continually updated the simplified section of the base model which
held the place for their work with the submodels they developed. This approach
provided a common ground for communication and a straight forward approach to
combining the group efforts.
The four subgroups spent the majority of the semester developing their sections
of the final model. During this time the class assembled once a week to present
the status of their individual work. Communications between the groups identified
successes, failures, and specific group needs for connecting the separate submodels.
Intercellular movement (section 6) of the Sage Grouse was conducted by research
activities outside of the class cellular modeling effort. This effort was key
to meeting the goal of modeling the affect of spatial location and distribution
of the birds with respect to training activities.
While the class developed a cellular model based on a range of expected system
starting points, the groups understood that the initializing factors for each
grid cell would come from a snapshot of the system represented by a set of digital
maps prepared within and stored by a geographical information system. Again,
these maps were developed by outside researchers.
Once the cellular model was completed, the migration algorithm designed and
created, and the initializing GIS data generated, these components were brought
together. The combination required debugging and simplification before any results
could be obtained.
The debugging process required a significant amount of model simplification.
The model generated by the group had a combination of short-term (week-oriented)
impacts and long-term (year-oriented) impacts. The extracted short-term impact
section of the model was debugged and demonstrated. This reduced model neglected
for example, the impacts on the Grouse via tracked vehicle compression of the
soil, and on the sage brush community via tracked vehicle damage (long-term
impacts).
Once a demonstrable output could be created with the reduced model, the project
reached its goals. Although this project used a real problem affecting real
people with real land management problems, the goal was to demonstrate the technical
capabilities of a set of hardware and software tools in the hands of a multidisciplinary
group of researchers. With sufficient interest, field work, and participation
by end-user land managers, such models could become a powerful tool for land
management and endangered habitat protection.
This section describes the STELLA submodel components and design and development
of the intercellular migration algorithm. Appendix C provides a full listing
of the equations described and visualized graphically in this section. Refer
to this Appendix for further details about the model.
Figure 5.1 Full model.
The complete model initially generated by the researchers is represented graphically
in Figure 5.1. Although the details are not visible, the pieces discussed in
the following subsections can be visually mapped back to this figure. Starting
from the top left and moving clockwise, the boxes contain:
(1) Base model. All other components communicate with each other through this
common section.
(2) Physical submodel
(3) Human submodel
(4) Vegetation submodel
(5) Grouse attractiveness submodel
(6) Female Sage Grouse model
(7) Extra CM-5 input variables
Grouse submodel
The core requirement of this exercise was to simulate the impact of military training on the Sage Grouse, so the life cycle of the Sage Grouse on the Yakima Training Center was modeled along with the factors (physical environment, vegetation, and human impact) that influence Grouse survival.
Inputs required for the processing of this model included sagebrush density from the vegetation section, the noise index from the human section, and both snow cover and temperature from the physical section.
The intitial distribution of the Sage Grouse consisted of 200 female Grouse in the 116,964 30x30m grid cell area. Because the densities of these animals was so low it was determined that modeling population densities, as opposed to individuals, was unreasonable. At the spatial resolution of 30m [2] cells, the average Grouse density is approximately 0.0017 birds per cell. These 200 female Grouse were distributed using the GRASS random command across the regions of the training area which had higher densities of sagebrush.
For this model only female Grouse were considered. It was presumed that the abundance of males was not a limiting factor within the anticipated range of conditions. In addition, this model allowed only one female Grouse, plus eggs and associated juveniles, to exist per cell. All stocks in the Grouse submodel (egg, juvenile, and adult populations) were discretized. The total average life span of the Grouse used in this model was five years .
The review presented is sectioned according to the life cycle stages of the Sage Grouse. The process begins with mating adults and fertilization of the females (Figure 5.2). Fertilization
Female Sage Grouse are called to the lek and then fertilized there by males during the mating season. A lek is an area of land where mating occurs seasonally on a consistent basis. The number of males on the lek was estimated by a graph that relates males on the lek to the week of the year and is contained in MALES_ON_LEK (Figure 5.2). This graph showed the twelve weeks when Grouse are active on the lek for mating. CV_LEK_DIST measured the distance (in meters) of Grouse from the nearest lek. This distance was important for identifying the direction of the lek from the female postions on the range and for indicating when a female is on the lek to mate. The number of female and juvenile Sage Grouse provided additional input to FERTILIZE. This input and their origin are discussed later in this section. The final equation for fertilization was [5] :
FERTILIZE = IF (CV_LEK_DIST = 0 & F_SG_PREGNANT=0 AND (5.1)
F_SG_JUVENILES=0) THEN F_SG_ADULTS * MALES_IN_LEK * 0.1
ELSE 0
The variable FERTILIZE was added to F_SG_PREGNANT to give the number of pregnant Sage Grouse. This number was added to GESTATE which was set at one DT. A DT represents the time it takes to perform all the commands in the program. For this model, the DT was set at one week. Egg Laying Capacity Since only one female was assumed to exist per cell, egg laying capacity was modeled for each individual. Sage Grouse lay between 6 to 10 eggs per clutch (Taffe-Pounds 1992). An estimate of six eggs per clutch was used in the model and was recorded in EGGS_PER_FEMALE. This was converted to EGG_LAY, which determined if hatching occurred in the cell, a condition based on the fertility condition of the resident female in that cell.
EGG_LAY= IF (FERTILIZE!= 0) THEN EGGS_PER_FEMALE ELSE 0 (5.2)
The value of EGG_LAY was then converted to a whole number in the equation presented in EGG_LAYING by randomly selecting the number of eggs to either round off low or high. The (1/DT) forced the full number of eggs to be laid for any chosen time step. The (+ 1 ) statement in the equation allowed the whole number to be rounded up to a full unit or down to the base unit.
EGG LAYING = (5.3)
IF (RANDOM (0.0,1.0) < (EGG_LAY - INT (EGG_LAY)))
THEN (1/DT) * (INT (EGG_LAY) + 1) ELSE (1/DT) * (INT (EGG_LAY))
The whole number of eggs produced was stored in the state variable, F_SG_EGGS_1, which gave the potential number of surviving eggs per hectare. The eggs that survived F_SG_EGGS_1 were randomly rounded to a whole number in the flow, EGG_SURV_1. The equation states :
EGG_SURV_1 = IF (RANDOM (0.0,1.0) < ( DT * F_SG_EGGS_1 - (5.4)
INT (DT * F_SG_EGGS_1))) THEN (1/DT * DT )) * (INT (DT *
F_SG_EGGS_1) + 1) ELSE (1/ (DT * DT)) * (INT (DT * F_SG_EGGS_1 ))
The resultant value was then converted to F_SG_EGGS_2. The sum of both F_SG_EGGS_1 and _2 equaled EGGS_TOTAL.
Egg survival depended on the mortality pressure in each of the life cycle stages. Egg Mortality Egg mortality was calculated by a factor equated in SG_EGG_DEATH (eggs dying). This factor was derived from the EGG_SURV_FRACTION and the incubation (EGG_WEEKS) . EGG_SURV_FRACTION was the percentage of eggs surviving per clutch; and is set at 38 degrees (EberharDT and Hofmann 1991). EGG_WEEKS, set at two weeks, was the time needed for incubation to occur (Dalke et al. 1963). Both of these factors entered the converter, MOD_EGG_SUR_FRAC (model egg survival fraction). Factors causing death in Sage Grouse eggs were included.
MOD_EGG_SUR_FRAC = 1/DT * ( 1 - EXP( LOGN (5.5)
(EGG_SURV_FRACTION) *(DT/ EGG_WEEKS)))
The survival fraction flowed into SG_EGG_DEATH and combined with the total number of eggs to determine egg mortality/DT.
SG_EGG_DEATH= DT *(1-MOD_EGG_SUR_FRAC) * EGG_TOTAL (5.6)
SG_DEATH included EGG_DEATH_A and EGG_DEATH_B. EGG_DEATH_A represented the
fraction of eggs dying from F_SG_EGGS_1; EGG_DEATH_B represented eggs dying
from F_SG_EGGS_2.
1. EGG_DEATH_A = IF ( EGG_TOTAL =0) THEN 0 ELSE (5.7)
( F_SG_EGGS_1 / EGG_ TOTAL) * SG_EGG_DEATH
2. EGG DEATH B = SG_EGG_DEATH - EGG_DEATH_A
EGG_DEATH_A was rounded to a whole number to give EGG_DEATH_1 and EGG_DEATH_B was rounded to get EGG_DEATH_2 (see equation 5.4). EGG_DEATH_1 counted the eggs dying from the stock of F_SG_EGGS_1; EGG_DEATH_2 counted the dead eggs from F_SG_EGG_2 .
Juvenile death was calculated by the juvenile survival fraction (JUV_SURV_FRACT). One part of this graph was derived from sagebrush cover (CVP_SAGEBRUSH, from the vegetation section), essential for the survival of juveniles. More cover means less predation. CVP_SAGEBRUSH was the percent of cover for the grid and was transformed with a graph into a survival fraction for juveniles. This fraction converted JUV_WEEKS (50 weeks to complete the first year) to give the fraction of juveniles to survive per time-step DT .
MOD JUV SUR FRAC = (EXP (LOGN ( JUV_SURV_FRAC) * (DT / (5.8)
JUV_WEEKS)))
This fraction was incorporated to determine juvenile deaths in F_JUVENILE_DEATH_1.
F_JUVENILE_DEATH_1 = DT * ( 1- ( MOD_JUV_SUR_FRAC)) * (5.9)
F_SG_JUVENILES
F_JUVENILE_DEATH_1 was converted into F_JUVENILE_DEATH (see equation 5.4), giving the number of juveniles that die each DT. The surviving number of Grouse was transferred to the F_SG_ADULTS stock.
SURVIVAL_2= IF (CV_WEEK = 0) THEN F_SG_JUVENILES/DT (5.10)
ELSE 0
Adult survival was calculated with a survival fraction, set at 0.30 per DT, that accounted for factors which affect Sage Grouse survival.
ADULT_SURV_FRACTION was combined with ADULT_WEEKS to give the MOD_ADULT_SUR_FRAC (see equation 5.8). This flowed into F_ADULT_DEATH_1 with F_SG_ADULTS, resulting in the number of deaths of adult Sage Grouse per DT (see equation 5.9). This value has been assigned into whole numbers of Grouse and subtracted from the store of F_SG_ADULTS. After the number of Grouse was calculated, the values of F_SG_ADULTS and F_SG_JUVENILES were combined to equal the stock, F_SG_POPULATION. This number of Sage Grouse were eligible for reproduction. The sum was transferred back to start the cycle again.
- Sage Grouse Attractiveness -
The attraction model provided input (LEK_ATTRACTION, GENERAL_ATTRACTION, and DESIRE_TO_MIGRATE) for the migration algorithm. If males were booming (mating ritual) on the lek, attraction to it was recorded and captured by the LEK_ATTRACTION value. GENERAL_ATTRACTION provided a composite attraction value as a function of noise, snow, and cover.
DESIRE_TO_MIGRATE measured the response of the Grouse to the factors affecting its decision to migrate to another cell. In addition to GENERAL_ATTRACTION, it was a function of BROOD_DESIRE, NESTING_DESIRE and the NOISE_FACTOR. Each factor compiled into the DESIRE_TO_MIGRATE flow will be discussed; concluding with their summation in the flow.
Three variables determined the GENERAL_ATTRACTION of a cell to Sage Grouse: snow cover, sagebrush, and noise. Each was assigned a value ranging from 0 to 1 (#1= most desirable value for Grouse, #0= least desirable). These values were plotted on to conversion graphs contained in each factor unit (see Figure 5.6). The curves that are generated from these graphs are based on preferential parameters that give points to generate curves. A conversion graph took data from the indices shown (initialized through GIS) and plotted it against another axis ranging from 0 to 1. Values near 1 were assigned to preferential parameters (for the Sage Grouse) for the factor; values that were undesirable were assigned a number near 0. Preferential parameters are discussed in each factor section following. These values were multiplied together to create the GENERAL_ATTRACTION numeral.
GENERAL_ATTRACTION= SNOW_COVER_FACTOR * (5.11)
SAGE_BRUSH_FACTOR * NOISE_FACTOR
A summary of the three factors affecting GENERAL_ATTRACTION follows as well as comments about establishing preferential parameters used to create the conversion graphs.
The BROOD_DESIRE flow measured the desire to stay in a cell based on maternal instincts. This flow rating ranged from 0 to 1 and consisted of F_SG_EGGS_1, F_SG_EGGS_2 or F_SG_JUVENILES. With eggs present the females assumed nesting habits and stayed in the cell. The equation follows:
BROOD_DESIRE= IF (F_SG_EGGS_1 + F_SG_EGGS_2 >0) (5.12)
THEN 0.1 ELSE
IF (F_SG_JUVENILES >0) THEN 0.4 ELSE 1.0
NESTING_DESIRE was a 0 to 1 factor representing the female's desire to remain in her current position based on her pregnancy. As the pregnancy approached the egg laying stage, the desire to be mobile decreased. This factor wass calculated from F_SG_PREGNANT. The graph is made according to time of pregnancy and ability to manuever.
NESTING_DESIRE= GRAPH(F_SG_PREGNANT) (5.13)
The noise factor in this equation was the same as described in GENERAL_ATTRACTIVENESS and was applicable in mating and other scenarios.
The final equation for DESIRE_TO_MIGRATE was:
DESIRE_TO_MIGRATE= ( 1- GENERAL_ATTRACTION) * (5.14)
MAX(0.0, NESTING_DESIRE* BROOD_DESIRE - NOISE_FACTOR)
A factor near 0 from this equation indicated no desire to migrate; a value near 1 showed the opposite.
Another factor determining attractiveness of an adjacent cell was LEK_ATTRACTION . Modeled seperately, it indicated attractiveness increased as the noise from the lek increased; which itself is simply presumed to be a linear function of the number of males on the lek and the distance to the lek. The distance from the lek was recorded in CV_LEK_DIST for each female. During the mating season, the cells closer to the lek had higher ratings; conversely, Grouse further from the leks had a lower rating. This rating measured the attraction for female Grouse to the leks for mating. MALES_ON_LEK was the actual number of males on the lek, based on the time of the year and from experimental findings. MALES_ON_LEK and CV_LEK_DIST combined, give the LEK_ATTRACTION value.
LEK_ATTRACTION = CV_LEK_DIST * MALES_ON_LEK (5.15)
A higher value attracted female Grouse to the lek and a lower one discouraged them, signifying that females were looking for nesting sites after copulation.
This section discusses submodels for the three vegetation types that occur at the Yakima Training Center: sagebrush, grasses, and forbs. The cover of these types on the study site is 53 percent of the total cover on the site. Specifically, agropyron spicatum (grass species), artemesia tridenta (sagebrush species) and forbs represent 88 percent of the vegetative cover (Yakima Training Site report, 92). Submodels represented the sagebrush, grass and forb species and output for each submodel was the percent of total cover. Sagebrush Submodel The main state variable was sagebrush cover (SB_COVER in Figure 5.7). This stock had several inputs and outputs. The initial SB_COVER was set via a raster GIS map (refer to section 7) derived from a satellite image. SB_COVER will change over time from cover added due to the annual growth of certain plants, and from cover subtracted due to plant mortality. According to McArthur and Welch (1982), the death and growth in sagebrush communities cancel each other out, resulting in stable equilibrium over time.
In this model, growth of the sagebrush community was driven by NEW_SB_COVER, which input added cover into SB_COVER. SB_COVER for each cell had a maximum capacity of 25 percent and was included in the equation for new sagebrush cover below (YTC Range Site Description, 1989).
NEW SB COVER = MIN(NEW_SB, SB_COVER - 25) (5.16)
NEW _SB was a composite of variables relating to the environment during the time of year sagebrush grows based on data collected by Daubenmire (1975a).
NEW SB = IF (CVS_TEMPERATURE >12.0) (5.17)
AND (CVS_TEMPERATURE<22.0)
AND (CVS_SOIL MOISTURE>8.0) AND (CVS_SOIL MOISTURE <10.0)
AND (CV_WEEK>15) AND (CV_WEEK<25)
THEN SB_COVER * (Aspect_Mod * .02) + Fire_Regeneration ELSE 0
Temperature was a function of a seasonal temperature and local elevation. Units were measured in degrees celsius (refer section 5.4. Equation 5.17 gave the ideal temperature at which sagebrush grows (12 to 22 degrees [[ordmasculine]]C). Soil moisture is measured at 15cm below the surface and was recorded as millimeters of water. Soil moisture values were taken from the physical section. Ideal soil moisture for sagebrush growth was modeled at a range between 8mm and 10mm of water in the top 15cm of soil. The weeks of the year representing the growing stages for sagebrush were given in the equation. Secondary growth was modeled at 2 percent per year (McArthur and Welch 1982). The variable Fire_Regeneration monitored sagebrush regrowth after a fire. Full sagebrush regrowth can take up to a year and a half after a fire (Daubenmire, 1975a Miller et al, 1986), and can only regrow in a successional pattern after seeds are distributed on site by animals. This was incorporated in equation 5.18 using a sagebrush regrowth flow.
Fire_Regeneration = DELAY ( SB_REGROWTH,78) (5.18)
This equation allowed 78 weeks to pass before regrowth could occur. Regrowth was calculated with
RE GROWTH = IF FIRE DAMAGE = 1 THEN .05*TOTAL COVER (5.19)
ELSE 0
indicating that the regrowth percent cover after a burn for sagebrush was 5 percent of the total viable vegetative cover after one year and 7 percent after two years (Humphrey 1984). In TOTAL_COVER the percent cover for the three types of vegetation was added together and a percent cover value was obtained. According to the YTC range site description (1989), total cover can reach a maximum of 59 percent of the area in each cell . Therefore, a homogenous stand can only reach 59 percent cover of the total area in any cell.
Aspect is the direction in which an object or group of objects face and was given through Aspect_Mod. Aspect_Mod had a graph that related azimuth in degrees to an index ranging from 0 to 1. Azimuth measurements came from a GIS analysis of a digital elevation model (DEM) and were conveyed into STELLA by CV_ASPECT_SB. Sagebrush grows mostly on the south and west sides of hills and at elevations of 1520m to 2150m where soil moisture is sufficient (Barker and McKell 1983; Bonham et al. 1991). These locations were captured by the index.
Figure 5.7. Outflows for the sagebrush model (identical to other submodels).
Four outflows existed from SB_ COVER (see Figure 5.7) . Each represented factors that depleted the percent cover of sagebrush in the system.
Fire in the system was regulated by FIRE_DAMAGE, which was obtained from the base model, and was generated with GRASS data on fire occurrences. In the event of a fire, FIRE_DAMAGE was equal to 0; with no fire it was equal to 1. It was assumed that all sagebrush in a cell would be reduced to ash in a fire. The equation:
SB_FIRE = IF FIRE_ DAMAGE = 1 THEN SB_ COVER ELSE 0 (5.20)
indicated that if a fire occurred (FIRE_DAMAGE = 0), all sagebrush vegetation would be burned.
The next outflow reflected the normal dying of sagebrush plants over the year.
SB_NAT_DEATH = SB_COVER *.02/52 (5.21)
Since this was known to be a relatively stable community (West et al 1979; McArthur and Welch 1982), natural death was modeled to reflect a 2 percent decrease in growth over 52 weeks in a year.
Sagebrush consumption by animals (especially Sage Grouse) was the third outflow from this system. According to Taffe-Pounds (1992), this consumption was not considered to be a significant loss . Therefore, its parameter was set at 0, although it can be changed if future data indicates significantly higher consumption.
The fourth and final outflow from the sagebrush system consisted of sagebrush loss due to human impact:
SB_HUMAN_IMPACT = IF ( CVH_VEG_DAMAGE_INDEX >0) (5.22)
THEN SB_COVER - (CVH_VEG_DAMAGE_INDEX * SB_COVER)
ELSE 0
Sagebrush cover was reduced by a factor of a vegetative damage index provided by the human section. This index related cover loss due to human activities. The Agropyron Model (Grasses) The Agropyron spicatum model (Figure 5.9) approximated the growth and death of the major grasses at the YTC. The stand was assumed to be virgin because the YTC report had no listing for Bromus tectorum, a grass invader known to be successful after a fire and potential competitor with Agropyron. This model was similar in structure to the sagebrush model, although some equation parameters were changed. For example, NEW_AG reflected growing conditions optimum for grasses:
NEW_AG = IF (CVS_TEMPERATURE>15) (5.23)
AND (CVS_TEMPERATURE<30) AND (CV_WEEK >=18) AND
(CV_WEEK<=28) THEN AG_COVER * (AG_Aspect_Mod * .02)
+ (SOIL_MOISTURE*0) + AG_FIRE COVER ELSE 0
During the growing months, water is usually not limiting for grass growth in this type of ecosystem (Miller 1986). Therefore, the soil moisture variable was set for NEW_AG at 0. Also, the average optimal temperature for growth is between 20 and 30 [[ordmasculine]]C, and growth occurs during the months of May through mid-July (Daubenmire 1972; Harris, 1967). These grasses were thought to form a stable community where growth equaled death over the year, unless there was a perturbation such as a fire or human impact. Growth was approximated at 0.5 percent per year and the maximum cover for a cell was 59 percent (Yakima Training Center report 1989). In case of a fire, all vegetation was consumed before the regrowth process began.
According to Humphrey (1984) regrowth after a fire for grass species equals 30 percent of the viable vegetative cover the first year and 33 percent the second. The initial vegetative cover was taken from GIS maps derived from satellite imagery. For cover regrowth after a fire, the equation was:
AG_FIRE_COVER = DELAY(AG_ REGROWTH, 78) (5.24)
where AG_REGROWTH was defined as:
AG_REGROWTH = IF FIRE_DAMAGE=1THEN .33*TOTAL_COVER (5.25)
ELSE 0
indicating that after a burn, grasses return to about 30-33 percent of the initial cover (Humphrey 1984).
NEW_AG and AG_COVER both fed into NEW_AG_COVER which was defined as:
NEW_AG_COVER = MIN(NEW_ AG, AG_COVER - 59) (5.26)
indicating that total cover could never be more than 59 percent. The initial cover (AG_COVER) was determined through GIS and satellite imagery.
The only outflow different from the equation modeled for Sagebrush was natural death, defined as:
AG_NATURAL_DEATH = AG_COVER *.005/52 (5.27)
because, according to Treshow and Harper (1974), grass mortality is approximately 0.5 percent annually. The Forbs Model This model was similar in structure to the sagebrush submodel (see Figure 5.10). Parameters that reflect growing conditions for the grass model were the same for the forbs model. Optimum growing conditions for forbs in this area were not represented in the data collected and therefore could not be incorporated. Data that signified optimum growing conditions for sagebrush were assumed to be indicative for all plant growth. (This specific data is available for inclusion into a future version of the model.)
The equation for new forbs (NEW_FB) was the same as that for sagebrush, with the same optimums set for time and conditions (temperature, moisture, and season). Some parameters that fed into NEW _FB were tailored specifically for the forbs model. For example, RE_GROWTH_5, which was a part of the fire succession model was defined as:
RE_GROWTH_5 = IF FIRE_DAMAGE = 1 THEN .6* TOTAL_COVER (5.28)
ELSE 0
reflecting the forbs ability to return to 60 percent of the initial cover after a burn (Humphrey 1984). The converter NEW _FB_COVER was similar since a maximum of 59 percent of the total cover could be forbs.
NEW_FB_COVER = MIN(NEW_FB, FB _COVER -59) (5.29)
The only outflow that changed in this model, as compared to the sagebrush model, was FB_NAT_DEATH. According to Treshow and Harper (1974), the mortality rate for forbs was approximately 10 percent each year due to natural death. This was defined in the equation :
FB_NAT_DEATH = COVER 4*.1/52 (5.30)
The factor 0.1/52 should be changed if the DT is not set at one week. For example, if the DT is set at one day then this factor should read 0.1/365.
Figure 5.8. Sagebrush submodel.
Figure 5.9. The grass submodel.
Figure 5.10. The Forbs submodel.
The central requirement of this model was to drive the weather affecting other submodels and specifically to drive the soil moisture cycle contained within the abiotic model. This physical process model reflected the measurable effects of the abiotic processes contained within the entire model.
Data from many sources was used to model this section. To include soil moisture, data on soil characteristics such as the maximum water holding capacity, the soil transmissivity (permeability), measurements of density or compactability, and the location of the various soil associations or series within the chosen study area were used. Spatial GIS maps provided soil characteristics such as slope, aspect, elevation, land use, and ground cover information .
GIS maps helped create a digital elevation model (DEM) that calculated change in precipitation and temperature over time. Temperature and precipitation data on average monthly values recorded in Yakima, Washington (Mather 1965) were given in continous graphs, and could be sampled over any time interval to estimate the amount of precipitation or average temperature. The empirical relationships of how these parameters changed given different slopes, elevations, and aspects were combined and implemented.
A wetness index of a given cell based on its slope and upslope area was also calculated from this data. This wetness index was used as a measure of water flowing through a given cell following rain (page 4). To model the relationship between vegetation and the soil moisture cycle, a cover factor was developed to reflect the relative amounts of grasses (Agropyron), forbs, and sagebrush present in a given cell.
Figure 5.11 is an image taken from STELLA of the entire abiotic model. A description of the functions will follow. The model focused on the dynamics of AVAILABLE_SOIL_MOISTURE. Controlled principally by precipitation, temperature, and cover, this stock reflected the amount of available soil moisture (given in millimeters of water) which is within the rooting zone and available for plant use. A depth of 15.0cm represented the rooting zone. The model is described in terms of input and output. The input is precipitation (SM_ACTUAL_INFIL) and the output is SM_DECREASE, which was comprised of runoff and evapotranspiration. Each of these flows will be discussed in following sections.
AVAILABLE_SOIL_MOISTURE was a function of SM_ACTUAL_INFILTRATION, which in turn was derived from the combination of SM_POTENTIAL_INFILTRATION and SM_INFILTRATION_RATE.
The calculations for the potential infiltration were derived from many factors, one of which was precipitation. Before analyzing the equations regarding precipitation, an explanation of assumptions must be given. Precipitation is the amount of water, liquid or frozen, that falls on the ground. Water supply, however, is the the amount of liquid water that reaches the soil surface in a given period, and it consists of both rainfall and meltwater. In this sense, snow is precipitation at the time it falls (unavailable to soil), but it is not considered a part of the water supply until it melts. At Yakima, the main water supply input comes from summer rains and from the melting of the snow that falls in the winter. For this reason, the snow pac was another important input to the water supply and will be discussed later.
Data for the average monthly precipitation and temperature (SM_TEMPERATURE) was taken from Mather (1965). The original precipitation data was provided in SM_RAINFALL_MM (Figure 2), using a graph of average monthly precipitation vs time. Precipitation data was made usable in weekly time-steps with SM_RAINFALL_CONVERSION.
SM RAINFALL CONVERSION= {SM_RAINFALL_MM/ 4.3} (5.31)
+ {RANDOM (-0.65, 0.65)} * 0.
The average monthly precipitation values were divided by 4.3 (the average number of weeks in a month) to permit calculating precipitation levels on a weekly rather than monthly basis. In addition, the equation allowed for the calculation of random variation. This process could be shut off by multiplying by 0 (shown in the equation) or turned on by multiplying by 1.
The amount of water that actually soaked into the ground was not equal to the total amount of precipitation. First, topography affects water flow, a factor incorporated through the use of the wetness index calculated by GRASS, labeled CV_WETNESS_INDEX in the model. The wetness index was a value (0.0 to 1.0 after normalization) that represented the actual amount of water flowing through a given cell following rain, and was based on the slope and location of the cell. A high wetness index meant more water was flowing through a cell during rain.
To completely calculate a value for SM_POTENTIAL_INFILTRATION, any effects of human activity (i.e., troop movements, tank travel, etc.) on the water-holding capacity of the soil must also be included. In this case, CVH_COMPACTION from the Human Impact Section (5.5) was used to reduce the soil's transmissivity, labeled as SM_SOIL_TRANSMISSIVITY_INDEX. Both the SOIL_COMPACT_INDEX and the SM_TRANS_INDEX ranged from 0.0 to 1.0 (human section). The transmissivity was also affected by the soil permeability, CV_SOIL_PERM. Soil permeability must be derived from a GIS soils map. The equation that gave soil transmissivity was:
SM_SOIL_TRANSMISSIVITY_INDEX= (1-CVH_COMPACTION) (5.32)
* CV_SOIL_ PERM.
Soil transmissivity could be found by computing [ ln (Te/Ti)] where Te was the average transmissivity and Ti was the transmissivity of the specific cell. This would be both spatially variable across cells, due to soil type and textural qualities, and temporally variable due to varied occurrences and intensities of human disturbance.
The final index input into potential infiltration was SM_COVER_INDEX an index derived from the vegetative cover in the cell. It allowed cover to be both spatially and temporally variable. An increase in cover lowered the infiltration, a decrease raised it.
When calculating the potential infiltration, the researchers assumed that water would only infiltrate into the soil when the temperature was above freezing. This was included in the equation for potential infiltration, calculated as the product of the precipitation, wetness index, and transmissivity index when the temperature was greater than 0 deg.C.
SM_POTENTIAL_INFILTRATION = SM_TEMPERATURE (5.33)
SM_COVER_INDEX * CV_WETNESS_INDEX * SM_NEW_WATER
SM_RAINFALL_CONVERSION * SM_SOIL_TRANSMISSIVITY_ INDEX
The next section determined the infiltration rate which was calculated using the available water in the soil (SM_AVAILABLE_SOIL_WATER). The current water level in the soil was represented as a percentage of the maximum water-holding capacity. This value was converted by a graph into the infiltration rate (SM_INFILTRATION_RATE). This function allowed for the rate of water infiltrating into the soil to decrease when the soil became more than half-full. The rate of infiltration continued to decrease rapidly until it stopped completely when the available soil moisture was equal to the maximum water-holding capacity of the soil. This causes the model to work in such a way that the soil could more readily absorb water when it was dry rather than wet.
The SM_ACTUAL-INFILTRATION was based on (1) what might potentially infiltrate into the soil, and (2) the current level of water in the soil. The index for the potential infiltration rate and the infiltration rate were multiplied to give the actual infiltration.
SM ACTUAL INFILTRATION = SM_POTENTIAL_INFILTRATION (5.33)
* SM_INFILTRATION_RATE.SM_SNOW_PAC
Figure 5.13. Snow fall and snow melt.
Figure 5.13 describes the processes of snow accumulation and snowmelt. The accumulation of snow in the model was recorded in the stock labeled SM_SNOW_PAC and was regulated by temperature and rainfall.
When precipitation occurs and the temperature is below 0[[ordmasculine]]C, snow starts to accumulate. If the temperature rises above 0[[ordmasculine]]C, then snow will melt and enter the soil as water. The amount of snow accumulation (SM_SNOW_PAC) was based on the standard that 1mm of rain equals 1mm of snow. Rainfall was taken from the precipitation data and the same conversions were used that calculated rainfall in mm/week (SM_RAINFALL_CONVERSION _4).
Like temperature and rainfall, the depth of snow was assumed to be constant over the entire area of a cell. The accumulation and depth of snow would be most strongly affected by the elevation of a cell because a decrease in temperature would accompany a rise in elevation. For this reason, the first freeze comes sooner and lasts longer at higher elevations.
When the temperature climbs above 0deg. C, accumulated snow begins to melt, amount of which was determined by the relationship described in the parameter SM_MELT_VS_TEMP as a graph. This graph showed that as the temperature climbed from 0 to 10deg.C, the percentage of accumulated snow that became melt water increased exponentially from 0 to 1. This means that no snow remained on the ground if the temperature climbed over 10deg.C. The amount of water that melted was determined by the equation in SM_MELT.
SM_MELT = IF SM_TEMP > 0 THEN (SM_MELT_VS_TEMP (5.35)
* SM_SNOW_PAC) ELSE 0
This equation gave the amount of meltwater from snow and transferred it to SM_NEW_WATER by the following formula:
SM_NEW_WATER= SM_RAINFALL_CONVERSION - SM_SNOW (5.36)
+ SM_MELT
This result was added to SM_POTENTIAL_INFILTRATION where it was incorporated into the total for soil moisture. Evapotranspiration The rest of the model revolves around the theories of evapotranspiration and the water balance approach. Descriptions of the potential and actual evapotranspiration sections of the model, in addition to the switching mechanism between the two will be explained within the explaination of the submodel.
The model that predicted available soil moisture was based on the dynamics of the water budget or climatic water balance. This water balance was defined as the interactions of energy and water described by the relationships among potential evapotranspiration, actual evapotranspiration, temperature and precipitation. They were the most important parameters in the water balance, and also determined the availability of moisture in the soil. Water balance parameters estimated how much usable energy and water was available to plants, how much evaporative demand was not met by available water, and how much water was unusable excess. The idea of representing the moisture availability as an energy/water index, came from the early work of Thornthwaite (1948) and was further developed in Mather (1974, 1985) and Stephenson (1990). In the next sections concerning the model parameters, a short explanation of the theory behind the equations is given. Soil Moisture Soil moisture is the result of the balance between system input (water supply, i.e., precipitation and snowmelt) and output (potential evapotranspiration, actual evapotranspiration), which vary both spatially and temporally, spatially because input and output are modified by terrain, and temporally because monthly and weekly observations are sampled from annual cycles of precipitation and temperature.
Figure 5.14. Temperature. Potential Evapotranspiration Potential evapotranspiration (PE) is related to the amount of energy in the environment. Theoretically, PE is the evaporative water loss from a site with a standard vegetation cover supplied with unlimited water. Potential evapotranspiration is a function of heat (temperature) and radiation, but it can be modified by air humidity and wind speed. For this model, PE was a direct function of air temperature only (a measure of heat) recorded as a monthly average. Its values were calculated according to the tables and methods described by Thornthwaite and Mather (1957). This method assumed that PE values could be obtained directly from air temperatures when the soil moisture retention occurred at 15 cm deep. Their units represented the variation in soil moisture in mm/month.
Air temperature of a cell was calculated based on its position in elevation relative to the base station at Yakima, Washington. Both the temperature and precipitation data were recorded at a weather station in Yakima at an elevation of approximately 365m above mean sea level and was assumed to have 0 slope (no aspect). Temperature data was contained in the graph SM_BASE_TEMP, and was recorded as a monthly average plotted over 52 weeks. This data was then converted from fahrenheit to celsius in SM_TEMP_CELSIUS.
The effects of elevation and aspect were combined in the SM_TEMPERATURE parameter according to the following relationships. For every 100m rise in elevation, the temperature fell 1deg. C (for every 1m rise in elevation, the temperature falls 0.01deg. C). The elevation was calculated by subtracting the base elevation (800m) from the actual elevation of the cell. This gave temperature differences from the base temperature value. Temperature was further modified based on the aspect of the cell. South facing slopes (225 degrees-315deg.) were 10 percent warmer than the base temperature; north facing slopes (45deg.-135deg.) were 10 percent cooler; and east or west facing slopes (0deg.-45deg., 135deg.-225deg., or 315deg.-360deg.) were left unchanged. The final equation for temperature was:
SM TEMPERATURE = SM_TEMP_CELSIUS + ((CV_ELEVATION) (5.37)
* .01) * SM_ASPECT_RECLASS
In this model the process of evapotranspiration was controlled principally by the air temperature. The actual amount of water which could potentially leave the system was calculated with an equation originally developed by Thornthwaite (1948). Aside from the temperature of the cell, the equation required two variables derived from the temperature, SM_A and SM_HEAT. Each of the following equations was taken directly from Thornthwaite and Mather (1955). The value used for SM_HEAT was calculated as follows:
SM_HEAT = (12/12) * ((118.44 / 5.0) ^1.514) (5.38)
and given in degrees celsius. Twelve is the number of months over which balancing is to occur, and the 118.44 is the sum of the average monthly temperatures. This sum was assumed to be stable enough to use year to year.
The SM_A was based on the SM_HEAT value and was calculated as follows:
SM_A = ((6.75/10.0^7.0) * SM_HEAT^3.00) - ((7.71/10.0^5.0) * (5.39)
SM_HEAT^2.00) + ((1.79/10.0^2.0) * SM_HEAT) + 0.49
Using the values calculated by the SM_A and SM_HEAT parameters, the amount of potential evapotranspiration was determined with the following equation and measured in mm/month.:
SM_POTENTIAL EVPT = IF SM_TEMPERATURE > 0.00 (5.40)
THEN 16.0 * ((8.0 * (SM_TEMPERATURE / SM_HEAT )) ^ SM_A) ELSE 0
The resulting value was converted to weekly values by dividing by 4.3 (for the same reason as was done for rainfall) in the SM_PE_CONVERSION parameter. The SM_PE_CONVERSION was the value used for potential evapotranspiration. Actual Evapotranspiration Actual evapotranspiration (AE) is a measure of the simultaneous availability of both the biologically usable energy and water in the environment. This value can be obtained only when the water retention capacity of the soil is known. It equals the evaporative water loss from a site covered with a homogeneous vegetation ( given water availability). Actual evapotranspiration equals either potential evapotranspiration or some fraction thereof determined by the relative energy demands of the system. For example, when PE exceeds available water, AE is limited by water and equals adjusted actual evapotranspiration. When available water exceeds PE, AE is limited by energy and AE equals PE.
The calculation of actual evapotranspiration was a two-stage process that began by modifying the potential evapotranspiration based on the relative amount of water available (SM_AVAILABLE_SOIL_WATER) given as a percentage of the maximum water-holding capacity of the soil or CV_WATER_HOLDING_CAPACITY and the AVAILABLE_SOIL_MOISTURE.
SM_AVAILABLE_SOIL_WATER= (AVAILABLE_SOIL_MOISTURE (5.41)
/CV_WATER_HOLDING_CAPACITY) * 100percent
This gave the percentage of potential saturation. SM_AVAILABLE_SOIL_WATER tells how saturated the soil is, and when it was less than 35 percent, the amount of the potential evapotranspiration removed from the system decreased based on the level of atmospheric demand The percentage of decrease was graphed in the parameter SM_PERCENT_LOSS_FROM_STORAGE derived from the SPAW (Saxton and McGuinnes 1982) and CREAMS models of evapotranspiration (Ritchie 1972). The percent loss from storage was multiplied by the available soil moisture to give SM_ACTUAL_EVPT.
SM_ACTUAL_EVPT= SM_PERCENT_LOSS_FROM_STORAGE (5.42)
*AVAILABLE_SOIL_MOISTURE.
The second stage of this process involved a modification of the AE based on the percent cover present in the cell. A vegetation index was calculated based on the relative amounts of each type of plant compared to their initial proportions.
The vegetation index was then converted into a Leaf Area Index (LAI), which measured the percent of leaf area covering the ground. For example, if the vegetation index rose from 20 to 30, the LAI rose proportionally from 0.5 to approximately 3.5. The LAI was incorporated in an equation taken from Hanson (1976):
SM_COVER_INDEX 3 = 0.55 * (( LEAF_AREA_INDEX ) ^ 0.5 ) (5.43)
The LAI peaked at 3.3 because the cover index reached the limit of100 percent. The final value calculated for actual evapotranspiration was labeled SM_ADJUSTED AE and was the product of the actual evapotranspiration and the cover index.
SM ADJUSTED AE= SM_COVER_INDEX_3 * SM_ACTUAL_EVPT (5.44)
This relationship states that as the amount of cover increased, the percentage of the potential evapotranspiration actually removed from the soil increased until SM_ADJUSTED_AE equaled the SM_ACTUAL_EVPT. The actual evapotranspiration was directed into the SM_DECREASE which was an outflow of the available soil moisture.
Figure 5.15. The switching mechanism between potential and actual evapotranspiration.
The switch controled the evapotranspiration when it was equal to the potential evapotranspiration (SM_PE_CONVERSION) or the actual evapotranspiration (SM_ADJUSTED_AE). To create the switch, the SM_MIN or minimum moisture content had to be determined. It was derived from the SM_CRITICAL_PERCENT and the CV_WATER_HOLDING_CAPACITY. The critical percent was the maximum storage capacity of the given soil type. If the moisture fell below this specified percentage the plants were affected by the lower soil moisture. This stress created competition for the small amount of water still left in the soil. Therefore, the actual evapotranspiration no longer equaled the potential evapotranspiration. The equation that gave for minimum moisture was:
SM_MIN= SM_CRITICAL_PERCENT * (5.45)
CV_WATER_HOLDING_CAPACITY
The SM_MIN was piped to the switch where it was compared with the available soil moisture.
SM_SWITCH= IF AVAILABLE_SOIL_MOISTURE < SM_MIN (5.46)
THEN 1 ELSE 0
When the switch equaled 1 factors compriseing the actual evapotranspiration act decreased the available soil moisture in the cell. If the switch was 0 factors comprising the potential evapotranspiration decreased the available soil moisture. This relationship was given in the equation for SM_DECREASE.
SM_DECREASE = IF (SM_SWITCH = 1) THEN SM_ADJUSTED_AE (5.47)
ELSE SM_PE_CONV.
It provided a value that decreased the available soil moisture of the cell.
This concludes the description of the actual modeling process and the components and relationships contained therein. As was previously explained, soil moisture was the result of the balance between system input (water supply) and system output (evapotranspiration). Spatial variations resulted from the varying input and output which were modified by the terrain, and temporal variations were considered within monthly and weekly observations sampled from annual cycles of precipitation and Temperature.
This section discusses the potential effects of military training exercises on the Sage Grouse population, vegetation, and soil of the study site. Factors of training excercises that affect the environment include (1) off-road vehicle use, (2) noise created by vehicles, (3) troop activities, and, (4) encampments. Appendix B describes the GIS analysis steps that were used to generate impact maps that describe the result of a single type of training activity. The Stella model components described here utilize these impact maps through a simple on-off mechanism. Note that only a single training scenario is thus available. Modifications to this model to improve realism will require the generation of a series of impact maps resulting from different training activities. Three submodels describe these effects: soil compaction, vegetative damage, and noise damage. Each submodels accounts for environmental changes affecting the Sage Grouse.
Soil compaction affects soil moisture and productivity with regard to vegetative growth. Vegetative damage is measured because of its importance as food and cover for the Sage Grouse. Noise level affects the suitability of land for Sage Grouse habitation throughout the study site, particularly during the mating and nesting season.
For each submodel section, a set of initial variables was used (see Figure 5.16) representing the effects of tracked and untracked vehicles and troops and encampments on the site.
Figure 5.16. Twelve initial parameters used for data in the human impact submodel.
Data from the sources represented extreme situations of military damage and did not contain intermediate results. When calculating damage per hour, data had to be extrapolated to include intermediate damage for a function to be graphed. The graph compared damage per hour with a coefficient between zero and one. Units were created to define the relationships. A summary of the human impact parameters and their units are presented below.
a. SOIL_COMPACT_COEF T = 0.0001: The soil compaction coefficient defined the compaction of soil caused by troops in hectares compacted per troop-hour of training.
b. SOIL_COMPACT_COEF_TV =0.15 : The soil compaction coefficient for tracked vehicles gave the hectares compacted per tracked vehicle hour of training.
c. SOIL_COMPACT_COEF_UTV =0.10 : The soil compaction coeffiecient for untracked vehicles gave the hectares compacted per untracked vehicle hour of training.
d. SOIL_COMPACT_COEF_C =0.001 : The soil compaction coefficient for encampment gave the compaction of soil due to the encampment of troops in hectares compacted per hour of encampment.
e. VEG_DAMAGE_COEF_T= 1: The vegetative damage coefficient for troops defined the vegetation destroyed per hectare per person per hour of training.
f. VEG_DAMAGE_COEF_TV= 1000 : The vegetative damage coefficient for tracked vehicles gave the relative amount of vegetation destroyed per hectare per hour of vehicle use. By definition, tracked vehicles have 1000 times more impact on vegetation than troops.
g. VEG_DAMAGE_COEF_UTV=500: The vegetative damage coefficient for untracked vehicles measured the relative amount of vegetation destroyed per hectare per hour of vehicle use. This estimate means that untracked vehicles were estimated to have 500 times more impact on vegetation than troops.
h. VEG_DAMAGE_COEF_C= 1: The vegetation damage coefficient was created from encampments. By definition, for every troop in an encampment there was one unit of vegetation destroyed per hectare per person per hour of training.
i. NOISE_COEF_T=1: The noise coefficient for damage caused by troops is defined as noise units per troop hour of training. The units are termed troop-hour noise units.
j. NOISE_COEF_TV=100 : The noise coefficient for damage caused by tracked vehicles was defined in troop-hour noise units. It was estimated that tracked vehicles cause 100 times the noise output of a soldier.
k. NOISE_COEF_UTV=50 : The noise coefficient for damage caused by untracked vehicles was defined in troop-hour noise units. It was estimated that untracked vehicles cause 50 times the noise output of a soldier.
l. NOISE_COEF_C= .5 : The noise coefficient for damage caused by a single soldier during encampments was defined in troop-hour noise units. It was estimated that encampments cause 0.5 times the noise output of a soldier.
This information was specific to the site under study, and parameters for training excercises and damage will have to be determined. Sources that provided data for these coefficients include: Bailey and Burt 1988, Bailey et. al. 1988, Grassman et al. 1989, Griggs and Walsh 1981, Johnson and Burt 1990, Pollack et al. 1986, and Smith and Dickson 1990.
A training schedule, was the input designed to trigger the running of the human impact model. It expressed that a training exercise was in progress at a given time. A time series graph was constructed to define how often an excercise took place in the model. It gave an output such that a 1 signified a full training exercise and a 0 signified no training. A training excercise was comprised of troops accompanied by tracked and untracked vehicles manuevering through a landscape off and on-road. The training schedule was transformed into the number of tracked and untracked vehicles hours, the number of troops, and the number of troop encampment hours per 100 hectares per cell in the cellular model. The noise intensity related to these activities was also provided for each cell.
Soldiers must train at various times during the year under a variety of climatic conditions. These exercises affect the desert ecosystem through soil compaction, vegetative damage, and noise, can be minimized by manipulating the timing of the training. The spatial distribution of a training exercise was determined outside the model and was fixed. Updates to this model should accommodate different spatial training locations and types of training activities.
Through the use of the initial variables and other factors, three submodels that evaluate compaction, vegetative damage, and noise damage were created. The soil compaction submodel given in Figure 5.17 is first discussed.
Figure 5.17. Soil compaction.
The soil compaction (indicated with "SOIL_COM") submodel was divided into four sections which created multipliers (indicated by "MULT"): SOIL_COM_C_MULT (encampments), SOIL_COM_T_MULT (troops), SOIL_COM_TV_MULT (tracked vehicles), and SOIL_COM_UTV_MULT (untracked vehicles). Each multiplier had two factors: the soil compaction coefficient of the impact type, and the corresponding number of training hours. Since the logic is identical for all of the multipliers, only the SOIL_COM_TV_MULT is discussed in detail.
Figure 5.18. Soil compaction multiplier for tracked vehicles.
This SOIL_COM_TV_MULT was used to determine the level of tracked vehicle activity occuring in a cell. It captured the activity level by multiplying the coefficient, which represented the potential damage of one vehicle-hour, by the number of tracked vehicle-hours given in hours of vehicle use per 100 hectares. The result was a number that represented the complete effect of track vehicle use on the site. The following equation calculated the muliplier and gave the units in hectare/hours.
SOIL_COM_TV_MULT = SOIL_COMPACT_COEF_TV * (5.48)
CV_TRAK_VEH_HR/100
Because the logic was exactly the same for the other multipliers, their formulas in STELLA were very similar. The only difference was that each multiplier corresponded to its own soil compact coefficient value and amount of training hours. The formulas for the remaining three multipliers were:
SOIL_COM_C_MULT = CV_ENCAMP_HR * (5.49)
SOIL_COMPACT_COEF_C/100
SOIL_COM_T_MULT = SOIL_COMPACT_COEF_T* (5.50)
CV_TROOP_HR/100
SOIL_COM_UTV_MULT = SOIL_COMPACT_COEF_UTV * (5.51)
CV_UTRAK_VEH_HR/100.
The four multipliers were used as input to calculate the total compaction value (COMPACTION in Figure 5.17). COMPACTION passed the sum of its input values through an internal graph which yielded a value between 0 and 1 (y-axis). The x-axis of the graph was described as:
Training_Schedule*(SOIL_COM_TV_MULT + (5.52)
SOIL_COM_UTV_MULT+ SOIL_COM_C_MULT +
SOIL_COM_T_MULT)/1000
This graph is represented in Figure 5.18.
The construction of the NOISE_INDEX and VEG_DAMAGE models was the same as the COMPACTION model up to this point, except that they used their own corresponding inputs for each model (i.e., NOISE_INDEX depended on the noise multipliers and noise coefficients). Further modifications to the soil compaction model are shown in Figure 5.19.
Figure 5.19.
The soil compaction index portion of the model.
The value of COMPACTION represented the percentage of uncompacted land that was compacted as a result of a training event. Once the compaction index was calculated, its output was directed into NEW_COMPACTION. This computed the amount of currently uncompacted land that would then be compacted.
NEW_COMPACTION = COMPACTION * (5.53)
(1 - SOIL_COMPACTION_INDEX)
The resulting value was entered into the SOIL_COMPACT_INDEX. The SOIL_COMPACT_INDEX represented the percentage of land compacted. The index generated UNCOMPACTION, the amount of land uncompacted every DT. It was calculated as follows by subtracting a percentage of the index:
UNCOMPACTION = SOIL_COMPACT_INDEX * .01 (5.54)
To bring to life the cellular grouse model, constructed using the STELLA
II software for the Macintosh, it was necessary to design an algorithm which
could utilize the information generated within STELLA. In addition, it was
important that these algorithms could be implemented on either a serial
or parallel processing environment. For research purposes, the CM-5 served
as the parallel environment. The code was written in FORTRAN, with each
section having a parallel and serial description of the algorithm.
The main function of the program was to calculate the movement of the female
sage grouse from cell to cell. The assumption was made that when mating
season is in effect, there were enough male Sage Grouse to mate with the
females that arrived on the lek. The model allowed for movement in only
the four cardinal directions. The algorithm, however, was constructed to
allow for future adaptations.
The movement of a grouse was influenced by the relative attractiveness of
the neighboring cells versus the current cell varies based on grouse population,
noise levels, environmental factors (i.e.snow cover and sage brush cover),
and proximity to the lek during the mating season. Furthermore, several
rules affected how the grouse could and would move. Both a "general_attraction"
and a "lek_attraction" index were calculated within STELLA for
use in the algorithm. During most of the year, the attractiveness of a neighboring
cell and the current cell was based on the general_attraction index. However,
during the mating season, from weeks 4 to 16 in the STELLA model, the level
of the lek_attraction index determined the attractiveness of a cell. When
lek_attraction began to rise (week 4) the non-pregnant female grouse responded
solely to this index, but once the grouse was pregnant, the operative index
switched back to general_attraction.
A third index considered after the evaluation of the relative attractiveness
of the neighboring cells was the grouse's level of motivation or "desire_to_migrate"
(as it was named in the STELLA model). The specific functioning of these
indices are described in detail in the following sections. In addition,
it was required that only one adult grouse could occupy any cell at one
time. This rule was temporary suspended on January 1 when all of the juveniles
traveling with the mother change status from juvenile to adult. Then the
new adults in the cell with the mother scattered in all four directions
to reduce the number of adult grouse in each cell to one as quickly as possible.
Also, due to the one grouse per cell rule, if two grouse wanted to move
into the same cell at the same time, neither was permitted to move into
the cell, and both stayed in their current cells. Lastly, no grouse were
permitted to enter or leave the study area.
The functioning of the migration algorithm was a two-stage process. Two
passes were made over the data array to generate a new map of where the
grouse had moved following each iteration. The first pass calculated the
"intent" of the grouse. Based on the relative attractiveness of
the neighboring cells and the level of motivation a grouse had to leave
its current cell, a new array was generated. Each cell in this new "whereto"
array contained the following information: (1) how many grouse wished to
move into a cell, and (2) from which direction a grouse wanted to move.
The second pass was intended to actually "move" a grouse from
its current cell to the desired cell, all rules permitting.
Dr. James Westervelt is the principal investigator.
Dr. Bruce Hannon is a professor in the Department of Geography at the University of Illinois, Urbana-Champaign.
Albert Cheng is a research programmer with the National Center for Supercomputing Applications at the University of Illinois.
Louis Iverson, Lynne Gildensoph, are research scientists at the University of Illinois.
Helena Mitasova and Kevin Seel are research scientist at the US Army Construction Engineering Research Laboratories (USACERL).
Pervaze Sheikh, Alicia Nugteren, Cory Rubin, Bruce Dvorak, Eric Lambert, and Kenneth Pabich are all graduate students at the University of Illinois.
Michael Shapiro is a research programmer at USACERL. The work was monitored by the Environmental Compliance Division (EC) at USACERL.
Mr. William Goran is the acting Chief, USACERL-EC. Technical editing was provided by Meredith McCarthy, USACERL Information Management Office.
COL Daniel Waldo is the Commander and Director of USACERL, and Dr. L. R. Shaffer is the Technical Director.
- Grouse Model sample output movies -
The movies listed below provide mpeg representations of the movement of mathematically
modeled grouse under three simulated land-use schedules. "Land use"
is simulated military training.
In the movies, female grouse are represented by spheres that move across the
landscape. The spheres are color coded as follows:
Yellow - not associated with offspring
Purple - fertilized
Red - nesting with eggs
Green - associated with hatchlings
These movies dispense with the oblique 3-d view and compare the movements of the simulated grouse between two different experiments.
-
Conclusions
and Future Directions -
Limitations of the Model
Model requirements, in general
Future Research in modeling
The first part is a review of the conclusions and recommendations oriented toward
the Sage Grouse. Some components might be added to the model and some components
need refinement. Then, we illustrate some general conclusions regarding the
potential of the technology itself. While this effort demonstrates a powerful
set of capabilities involving hardware, software, and human resource, potential
exists for further development and refinement of all aspects of the modeling
effort.
Specific Conclusions about the Sage Grouse Model
Several limitations of the Sage Grouse Model render it unrealistic for actual
land management activity. Its use as a preliminary test of current hardware,
software, and research methods is valuable, however. We hope that the lessons
learned during its conception can be carried over to assist in future landscape
simulation development.
The customer must be the driving force behind model design
and development if it is to have any potential as a management tool.
This modeling of the training activity presumes a single type
of training exercise that is distributed across the landscape (which consists
of a training area and an impact area) with probabilities based on a fixed tradeoff
between travel time and level of training. Installation personnel will be invaluable
in the generation of a series of much more realistic training activities and
alternatives.
Training schedules are similarly unrealistic in this model.
While Yakima personnel were kind enough to provide a tape containing historical
training, the software to decipher the data was unavailable. Without significant
assistance the researchers anticipated that the data would probably be misinterpreted.
The grouse behavioral model expressed in the migration algorithm
is completely untested and unverified. With significant input from the people
that understand the details of Grouse behavior, the existing migration model
should be able to provide a good starting point.
We feel especially confident in the precipitation and soil
moisture models. With minimal parameterization this section of the model should
perform quite satisfactorily. These models do not provide a simulation of overland
water flow during storm events. This omission may not be significant, but should
be considered. Such models are being created through other CERL based research
efforts.
The sage brush, agropyron, and forb growth submodels are defensible
but require significant research into the growth parameters. The primary impact
modeled in the working model (which was a subset of the full model) relies on
the impact of noise on the female grouse - especially during their nesting and
mothering phases. Parameters that turn raw noise into grouse annoyance levels
are necessary and must be determined either through field studies or through
an interrogation of installation personnel familiar with grouse behavior.
While documentation identifies that grouse behavior is affected
by noise, it is unclear 1) how much acclimation to noise is possible, and 2)
how the noise affects stress and survival. Apparently the primary impact of
noise on grouse is the frightening of grouse away from their nests which sometimes
results in nest abandonment.
Fire was considered in the model, but parameters for regrowth
potential are needed.
Like all solutions offered to an end user, that software attains
its useful potential only when the end user provides the goals and participates
in ensuring a satisfactory solution.
General Conclusions
This modeling exercise demonstrated a powerful modeling capability that combines
the strengths and power of research personnel associated with multiple research
organizations, a set of software packages, and an array of hardware environments.
Such combinations of capabilities will be repeatedly asked to generate location
specific dynamic spatial ecological models over the coming decades. Traditional
ecological models have focussed on modeling the ecosystem as a homogeneous whole.
With clear demonstrations that the spatial distribution of ecosystem components
is dramatically important and the development of cost effective parallel processing
environments, it becomes time for land managers to demand the design and development
of location specific models of the land they manage. To review briefly, the
important components of the enterprises that will be able to create these models
which were demonstrated in this exercise are:
Modeling personnel - Ecological models are multidisciplinary
exercises which require the coordination and cooperation of specialists from
such diverse fields as ecology, biology, ethology, chemistry, agronomy, economics,
landscape architecture, geography, geology, urban and regional planning, and
civil engineering. The environment within which models are developed must be
appropriate for multidisciplinary collaborative efforts.
Modeling process - A modeling process must be established
and well understood by the modeling participants to allow collaboration. This
process must allow for the model to be broken down into components, often in
a hierarchical fashion, that allows for individuals or small modeling sub-groups
to focus on a portion of the larger model in a way that allows the components
to easily fit together after development. In this demonstration that was accomplished
through the generation of a "base model" which basically established
specific system output without any modeling. Sub-models were then developed
within exact copies of this test (or template) environment. When it was time
to plug the sub-models together the integration was accomplished with minimal
difficulties.
Modeling software - It is the software environment that makes
it possible to collaborate effectively. For this exercise, that software environment
was the Stella software package. All participants had access to this software
and were able to learn the software relatively quickly. People who had never
programmed a computer were able to participate fully in the process of model
design and development.
Recommendations for future DLS research
This exercise fully demonstrates that DLS models can be developed by a multi-disciplinary
research group and run on cost-effective desktop workstations. Using the software,
hardware, and modeling techniques employed for this exercise it is possible
for numerous location specific models to be created. There is a highly important
component of the modeling effort that is missing from this exercise: the end
user. For this model Ms. Margaret Taaffe-Pounds from Yakima Training Center
was invited to present a real land management problem to the research group.
This brought a significant and motivating degree of realism to the work. Ms.
Taaffe-Pounds provided further feedback during model development as well. To
complete this particular model in a proper fashion it will be necessary to fully
integrate the expertise, knowledge, interest, and ambitions of key personnel
that are involved with the landscape in question.
While this effort has demonstrated powerful capabilities and techniques, there
is always room for improvement. Those that became obvious in this exercise are
briefly presented here:
Object oriented design and development environment - Model
components need to be encapsulated to allow for easier interdisciplinary team
efforts. Submodels that did not communicate with other submodels under development
were very easy to generate with the software used in this effort. Submodels,
if turned into distinct objects would force the modelers to be more carefully
in understanding and identifying inputs required from other model components
and information about their components that is available to the rest of the
model.
Multiple dynamic time steps - Dynamic modeling, as used here,
operated with a fixed time step. That is, the model operator must choose a fixed
value (typically a week for this model) which represented the amount of time
that passes between two time steps. Because 1) the activities occuring within
the system may shift between fast and slow activities and 2) the individual
activities have innately different activity speeds, it becomes important to
allow dynamically changing time steps as well as different internal time steps.
The software in use for this exercise utilized a pre-determined fixed time step
which has the additional benefit of being straight forward and simple in concept.
Inter-cell interaction modeling - While Stella was used to
generate the cellular model, Fortran programming by a trained computer scientist
was necessary to effect the generation of the inter-cellular model. This effectively
removes the system modeler from the model and requires close interaction and
effort with the programmer. It is all too easy for the modeler to lose track
of what the program is actually doing. In the future the modeler should have
a more hands-on opportunity for direct ownership of the intercell models.
Object libraries - All components of the models developed
for this report were created from nothing. While it is true that the final model
is probably appropriate for the application for which it was designed, components
of the model could be reused if developed within a modeling paradigm and language
that stores and retrieves system components. This will likely involve object-oriented
modeling with stand-alone objects which can be shared between similar ecological
models.
Probability/error computations - Models such as this derive
their equations and rules from the results of experiments that have been conducted
within certain ranges of parameters. There is a fundamental error potential
in these results based on statistical analysis of the experiments that provide
the base data. Also, there is often an additional error associated with the
proximity of the system state to the limits of the experimental conditions.
It is inappropriate to extrapolate experimental results without recognizing
an increasing error potential. Finally, as the inherent errors interreact with
each other in the model, the error of the final output should interest the modeler.
Units management - The Stella modeling environment provides
a powerful environment for writing equations that describe the change in state
from one time-step to the next. The software does not check the interaction
of the units of measurements associated with the equations and hence leaves
tremendous room for modeler error. Automatic checking and combination of errors
would provide a beneficial capability to any such modeling environment.
More rapid testing of full model - Once the full model is
assembled from its Stella cellular models and Fortran encoded migration algorithms,
the results need to be tested as a whole. Debugging then becomes tedious because
the model controlled by the computer scientist who must communicate the errors
back to the modeling team. Potential fixes must be effected by the programmer
and tested. The turn-around time to the modeling team discourages efficient
changes and modifications.
These modeling system suggestions will make the software environment
easier and more efficient. The coming decade will see an explosion in the modeling
of landscapes at all levels of resolution. Cost-effective computer hardware
exists now and will only become cheaper. Dynamic, spatial, ecological modeling
is a key to better land management and will become increasingly familiar to
land managers as they, in coordination with research institutions, begin to
develop increasingly sophisticated and realistic models of local systems.
Anders, F. J., & Leatherman, S. P. (1987) Effects of Off-road Vehicles on Coastal Foredunes at Fire Island, New York, USA. Environmental Management. Vol. 11, No. 1, pp. 45-52.
Anonymous. 1989. Yakima Training Center Range site description number 008BY010WA.
Bailey, A. C. and Burt, E. C. (1988) Soil Stress States Under Various Tire Loadings. Transactions of the ASAE. Vol. 31, No. 3, Pp 672-676
Bailey, A. C., Johnson, C. E., & Schafer, R. L. (1986) A Model for Agricultural Soil Compaction. Journal of Agricultural Engineering Research. Vol. 33, No. 4, Pp. 257-262.
Bailey, A. C., Nichols, T. A., & Johnson, C. E. (1988) Soil Stress State Determination Under Wheel Loads. Transactions of the ASAE. Vol. 31, No. 5, Pp 1309-1314
Barker, J. R. and C. M. McKell. 1983. Habitat differences between basin and Wyoming big sagebrush in contiguous populations. Journal of Range Management 36: 450-454.
Barker, J. R. and C. M. McKell. 1986. Differences in big sagebrush (Artemisia tridentata) plant stature along soil-water gradients: Genetic components. Journal of Range Management 39: 147-151.
Beck, Thomas D. I. 1977. Sage Grouse Flock Characteristics and Habitat Selection in Winter. Journal of Wildlife Management. 42(1):18-26.
Berry, J. D., and R. L. Eng. 1985. Interseasonal Movements and Fidelity to Seasonal Use Areas by Female Sage Grouse. Journal of Wildlife Management. 49(1) 237-240.
Braun, C. E., T. Britt, and R. O. Wallestad. 1977. Guidelines for the Maintenance of Sage Grouse Habitats. Wildlife Society Bulletin. 5(3):99-106.
Butler, C.D. & et all. A UserÕs Guide for CAMMS Version 2.0. Department of the Army: Army Corps of Engineers, 1989.
Biondini, M. E., C. D. Bonham, and E. F. Redente. 1985. Secondary successional patterns in a sagebrush (Artemisia trdentata) community as they relate to soil disturbance and soil biological activity. Vegetatio 60:25-36.
Bonham, C. D., T. R. Cottrell, and J. E. Mitchell. 1991. Inferences for life history strategies of Artemisia tridentata subspecies. Journal of Vegetation Science 2: 339-344.
Booth, G. D., B. L. Welch, and T. L. C. Jacobson. 1990. Seedling growth rate of 3 subspecies of big sagebrush. Journal of Range Management 43: 432-436.
Connelly, J. W., W. L. Wakkinen, A. D. Apa, K. P. Reese. 1991. Sage Grouse Use of Nest Sites in Southeastern Idaho. Journal of Wildlife Management. 55(3):521-524.
Connelly J. W., H. W. Browers, and R. J. Gates. 1988. Seasonal Movements of Sage Grouse in Southeastern Idaho. Journal of Wildlife Management. 52(1):116-122.
Dalke, P. D., D. B. Pyrah, D. C. Stanton, J. E. Crawford, and E. F. Schlatterer. 1963. Ecology, Productivity, and Management of Sage Grouse in Idaho. Journal of Wildlife Management. 27(4):811-840.
Daubenmire, R. 1959. A canopy-coverage method of vegetational analysis. Northwest Science 33: 43-64.
Daubenmire, R. 1970. Steppe vegetation of Washington. Washington Agricultural Experiment Station Technical Bulletin 62.
Daubenmire, R. 1972. Annual cycles of soil moisture and temperature as related to grass development in the steppe of eastern Washington. Ecology 53: 419-424.
Daubenmire, R. 1975a. Ecology of Artemisia tridentata subsp. tridentata in the state of Washington. Northwest Science 49: 24-35.
_____ 1975b. Plant succession on abandoned fields, and fire influences, in a steppe area in southeastern Washington. Northwest Science 49: 36-48.
DePuit, E. J. and M. M. Caldwell. 1973. Seasonal pattern of net photosynthesis of Artemisia tridentata. American Journal of Botany 60: 426-435.
Donovan, L.A. and J.R. Ehlering. 1991. Ecophysiological differences among juvenile and reproductive plants of several woody species. Oecologia 86:594-597.
Eberhardt, L. E., L. A. Hoffman. 1991. Sage Grouse on the Yakima Training Center: A Summary of the Studies Conducted During 1989 and 1990. Battelle Memorial Institute.
Gassman, P. W., Erbach, D. C., & Melvin, S. W. (1989) Analysis of Track and Wheel Soil Compaction. Transactions of the ASAE. Vol. 32, No. 2, Pp 23-29
Griggs, G. B., & Walsh, B. L. (1981) The Impact, Control and Mitigation of Off-road Vehicle Activity in Hungry Valley, California. Environmental Geology. Vol. 3, No. 4, pp. 229-243.
Evans, R. D., R. A. Black, and S. O. Link. 1991. Reproductive growth durin drought in Artemisia tridentata Nutt. Functional Ecology 5: 676-683.
Harris, G. A. 1967. Some competitive relationships between Agropyron spicatum and Bromus tectorum. Ecological Monographs 37: 89-111.
Haferkamp, M. R., P. O. Currie, J. D. Volesky, and B. W. Knapp. 1992. Mortality of crested wheatgrass and Russian wildrye during drought. Journal of Range Management 45:355-357.
Humphrey, L. D. 1984. Patterns and mechanisms of plant succession after fire on Artemisia-grass sites in southeastern Idaho. Vegetatio 57:91-101.
Hupp, J. W., and C. E. Braun. 1989. Topographic Distribution of Sage grouse Foraging in Winter. Journal of Wildlife Management. 53(3):823-829.
Jenni, D. A., and J. E. Hartzler. 1978. Attendence at a Sage Grouse Lek: Implications for Spring Censuses. Journal of Wildlife Management. 42(1):46-52.
Johnsgard, Paul A. 1983. The Grouse of the World. University of Nebraska Press. pp. 112-126.
Johnson, G. D., and M. S. Boyce. 1990. Feeding Trails with Insects in the Diet of Sage Grouse Chicks. Journal of Wildlife Management. 54(1):89-91.
Johnson, C. E. and Burt, E. C. (1990) A Method of Predicting Soil Stress State Under Tires. Transactions of the ASAE. Vol. 33, No. 3, Pp 713-717
Martin, N. S. 1970. Sagebrush control related to habitat and sage grouse occurrence. Journal of Wildlife Management 34: 313-320.
McArthur, E. D. and B. L. Welch. 1982. Growth rate differences among big sagebrush [Artemisia tridentata] accessions and subspecies. Journal of Range Management 35: 396-401.
Miller, R. F., J. M. Seufert, and M. R. Haferkamp. 1986. The ecology and mangement of bluebunch wheatgrass (Agopyron spicatum): A review. Agricultural Experiment Station, Oregon State University, Corvallis, Station Bulletin 669.
Peterson, J. G. 1970. The Food Habits and Summer Distribution of Juvenile Sage Grouse in Central Montana. Journal of Wildlife Management. 34(1):147-155.
Pollock, D., Perumpral, J. V., & Kuppusamy, T. (1986) Finite Element Analysis of Multipass Effects of Vehicles on Soil Compaction. Transactions of the ASAE. Vol. 29, No. 1, pp. 45-50. Remington,T. E., and C. E. Braun. 1985. Sage Grouse Food Selection in Winter, North Park, Colorado. Journal of Wildlife Management. 49(4):1055-1061.
Sapsis, D. B. and J. B. Kauffman. 1991. Fuel consumption and fire behavior associated with prescribed fires in sagebrush ecosystems. Northwest Science 65: 173-179.
Smith, D. L. O. & Dickson, J. W. (1990) Contributions of Vehicle Weight and Ground Pressure to Soil Compaction Journal of Agricultural Engineering Research. Vol. 46, No. 1, Pp. 13-29.
Treshow, M. and K. Harper. 1974. Longevity of perennial forbs and grasses. Oikos 25: 93-96.
U. S. Army Corps of Engineers. 1986. Draft environmental impact statement supplement-Fort Lewis Military Installation. Fort Lewis and Yakima Firing Center, Washington.
U.S.D.A. Forest Service. 1937. Range Plant Handbook. United States Government Printing Office, Washington, D. C.
Uresk, D. W., W. H. Rickard, and J. F. Cline. 1980. Perennial grasses and their response to a wildfire in south-central Washington. Journal of Range Management 33: 111-114.
Wandera, J. L., J. H. Richards, and R. J. Mueller. 1992. The relationships between relative growth rate, meristematic potential, and compensatory growth of semiarid-land shrubs. Oecologia 90: 391-398.
Welch, B. L. and T. L. C. Jacobson. 1988. Root growth of Artemisia tridentata. Journal of Range Management 41: 332-334.
West, N. E., K. H. Rea, and R. O. Harniss. 1979. Plant demographic studies in sagebrush-grass communities of southeastern Idaho. Ecology 60: 376-388. Wright, H. A. and J. O. Klemmedson. 1965. Effect of fire on bunchgrasses of the sagebrush-grass region in southern Idaho. Ecology 46: 680-688.
Young, J. A. and R. A. Evans. 1978. Population dynamics after wildfires in sagebrush grasslands. Journal of Range Management 31:283-289.
© Site Created by Joel Danowitz
All Rights Reserved Updated 2000