-
Modeling Environment -
Data Sources
A combination of data including raster maps, satellite imagery, vector maps, and
point data were used to calibrate initial conditions within the model. Vegetation
coverage and tortoise density was provided by transect data from the Land Condition
Trend Analysis (LCTA) program at USACERL.
A back propagation
neural network converted the data into GIS raster maps of Fort Irwin.
Computational Structure
Computer hardware and software are the essential tools of modern ecological
modeling. This model is supported by networked UNIX and Mac workstations and
has been built with the following software: STELLA,
STELLA translator, Express (to facilitate
UNIX computer networking), Madonna, and SME.
STELLA is a desktop modeling tool that uses icons and
schematics, linked with equations to build models. Given its ease of learning
and operation, STELLA removes the barriers that often exist in traditional modeling/programming
tools, and opens up the modeling process to a wider group of participants. Within
the context of the multidisciplinary team, STELLA was an excellent facilitator
between the different disciplines and modeling backgrounds.
STELLA, however, is not equipped to manage a large landscape, the size of Fort
Irwin. To apply the model simulation across multiple cells, STELLA equations
are translated into an environment called the Spatial Modeling Environment (SME).
This program was developed by Dr. Thomas Maxwell, University of Maryland. SME
then mimics the same functions of the single cell STELLA model, but it runs
the model within each cell of the landscape, thus generating the final output
data layers.
- Model Parameters
-
Time Frame
Like weather reports generated by elaborate weather pattern models, the
longer the weather simulation runs the less confidence one can attribute
the results. The same risks apply to a spatially dynamic model: How rapidly
will the predictability of the model decay over time? The model may demonstrate
stability at a gross scale, but reveal apparently random output at a detailed
scale. In other words, the overall pattern remains the same, but details
of exactly where the pattern is located may change with different runs of
the simulation. To preserve the highest integrity of output, minimize the
computational burden, as well as accommodate the longevity of desert tortoises
(tortoises have been reported to live between 50 and 100 years) and land
use management decisions, the "results" of the model will be gleaned
from a 100-year time span. This time span will make the best of available
data in capturing short term seasonal factors without slowing down the model
and producing only overly-generalized and inaccurate results. This time
frame also extends the vision of land managers by presenting a long-term
forecast, rather than a short-term prediction. When coordinated with a seasonal
and smaller time-step, the 100-year time span maintains an efficient calculation
and running time.
Time Step Considerations
Three basic possibilities were taken into consideration to determine the
time step for the model: fixed, variable, or event driven.
- Fixed.
This is conceptually the most simple time-step, but functionally the
most limiting. The model runs with a set time-step, such as .25 or 1 which
can represent days, months, or years. A known time-step simplifies the model
because all equations are generated with respect to the same time-step.
A fixed time step, by its definition cannot accommodate variability in the
system. For example, a weekly time-step can not capture daily temperature
and moisture fluctuations, or plant growth. However, a daily time-step would
miss the affects of a flash flood that takes just seconds but causes great
devestation to vegetation.
- Variable.
One alternative to the fixed time-step is a variable time step with
two options available. First, the time step is set to a large resolution
that can be dynamically modified as changes occur and detected within the
model. For example, when a flash flood occurs the model will detect the
rapid changes occurring, and stop the simulation, back up, and rerun the
model using an appropriately smaller time step. The alternative is to set
different fixed time steps to different parts of the model. This approach
provides some computational relief while maintaining the relative simplicity
of fixed time steps.
- Event driven.
In this approach there are no time steps, rather time is counted with
a calendar that schedules events. A plant sub-model within the larger model
may execute plant growth and then schedule itself to be updated at some
later time based on its own rate of activity. The storm sub-model would
be programmed to run at a specific time, and while it runs the storm model
can interact with the plant model and schedule the plant sub-model to "grow"
faster in response to the influx of water. This approach is most attractive
for models that have limited computing resources, however, from a modeler's
perspective it is the most time consuming approach as it requires significantly
larger simulation models to be created.
Software limitations made the fixed time step the most realistic choice,
although ideally, being able to run different parts of the model such as
the tortoises and the vegetation growth at different time steps would have
been preferred. This is something to work on with future research.
A one month time step was the most practical time step for our research.
This period coordinated well with the 100 year time frame. Addtionally,
it accommodated seasonal changes within the landscape, such as weather patterns,
tortoise nesting and egg-laying seasons, and vegetation growth cycles. This
time step also generates output that is reasonable to interpret versus daily
or weekly time steps that may provide too much detail, or an annual time
step with output that is too aggregated.
Spatial Resolution
Given that the model is to be placed within individual cells across the
Fort Irwin landscape, how many of those cells (i.e., spatial resolution)
are necessary to accurately describe the changes that occur over time, and
how many cells can be handled given the computational limits of the available
hardware? A key assumption of the model is the spatial distribution of characters
and events across the landscape is critical in understanding how these entities
interact. The spatial resolution needs to conform to a fixed time step.
For example, if a predator moves 100 meters in one time step over a terrain
that is divided into 10 meter units, that predator will appear to be unaffected
by time and space constraints. In essence, it will "warp" through
space, avoiding any obstacles or opportunities in its path. Hence, a fixed
time step directly affects the resolution of the salient terrain features.
Spatial resolution schemes can be categorized as follows:
- Fixed.
The terrain is divided into a regular array of square grid cells or
hexagons. This is a conceptually simple strategy.
- Hierarchical.
Models that simulate activities at different spatial resolutions may
adopt a spatial data structure that organizes information in a hierarchical
fashion. Each cell or hexagon can be iteratively decomposed into increasingly
smaller components. Movement of large entities (e.g., weather systems, flocks
of birds, clouds of spores or pollen) are allowed to move rapidly across
the environment using relatively long time steps and large spatial patches.
Small entities (e.g., individuals or vehicles) can operate at smaller time
steps and within smaller patches. This strategy stores data at varying scales,
simultaneously.
- Variable.
Large objects that move slowly across the landscape (e.g., roaming
herds of ungulates, or populations of invading species) should be preserved
as a whole entity, but include detailed spatial structures defining the
varying extent of the entity. This type of operation requires a fine spatial
resoluation (i.e., 10-100 m). However, small objects can be simulated at
a more coarse resolution.
A fixed cell size was determined to be the most practical spatial resolution
given computer limitations. The team selected a one square kilometer grid cell
for the entire Fort Irwin landscape. Tortoises have home ranges that extend
up to 1 square kilometer and this was the primary reason for the team's decision.
However, the 1 square kilometer cell size compromises details such as variations
in slope which are important in determining the location and placement of tortoise
burrows. These assumptions will be discussed further within the tortoise sub-model
section of this paper.
Mojave
Desert showing location of Fort Irwin, California
