OUTLINE OF IMP
The Integrated Modeling Project was initiated in September
1995 with a team meeting in Oak Ridge, TN. The project is
organized in the following three modules:
Linked Dynamic Model (LDM) for propagation of tree responses
to environmental changes through physiological, forest stand,
ecosystem, and lifecycle (plantation) scales. Responses are
propagated as frequency distributions by Latin hypercube sampling.
Response surfaces are generated for each model within the LDM
hierarchy for use in regional assessments.
LUCAS provides a regional grid for assessment of impacts of
environmental changes, landscape processes (fire, insect, disease),
and land use change on forest resources for 13 states of
southeastern USA.
GIS datalayers of soil, vegetation, and climate attributes for
the 13-state region provide variables required in model
applications within the LUCAS structure. Regional assessment
results are mapped with the GIS module with confidence intervals
derived from uncertainty analysis.
A steering committee will provide guidance on the scenarios
used in the development of the LDM and in applications of LUCAS.
LINKED DYNAMIC MODEL
The component models of the LDM hierarchy consist of:
MAESTRO Canopy processes
UTM Whole plant physiology
SPM Slash pine ecophysiology
NuCM Soil chemistry
LINKAGES Stand dynamics
PTAEDA2 Loblolly pine plantation management
Zeide Growth Curves Multi-species growth characteristics
MAESTRO, SPM, and LINKAGES codes have been installed on an ORNL
workstation and satisfactory reproduction of test case results
have been obtained. The SPM was first translated from basic to
Fortran 77. Some additional modifications were made to facilitate
sensitivity analysis with SPM. The UTM and PTAEDA2 codes are
available at ORNL from previous modeling projects. The NuCM code
has not yet been installed. A new, user-friendly version of NuCM
has been completed and is being tested by the developers. The new
version will be made available for installation at ORNL following
successful testing.
An additional code called PRISM is being linked to the first
six models in the LDM hierarchy. This will provide capability for:
(i) sensitivity analysis (identification of model inputs having
large effects on model outputs), and
(ii) uncertainty analysis (propagation of variability of inputs
through models generating frequency distributions of outputs).
An efficient Monte Carlo procedure called Latin hypercube sampling
is provided in PRISM for uncertainty analysis.
PRISM has been linked to SPM and LINKAGES, and sensitivity
analyses have been made for identification of important input
variables in these models. The sensitivity results are guiding the
selection of regional datalayers in the GIS task. PRISM was linked
to the UTM and PTAEDA2 in earlier work. Development of the LDM
framework is well underway.
Development of Zeide's Growth Types is also well underway.
Data from 67 sources, containing tree height vs age curves for
stands of southern pine species, have been assembled from Ralph
Amateis (Virginia), Clark Baldwin (Mississippi), David Lenhart
(Texas), and Terry Clason (Louisiana) as well as from Boris Zeide
(Arkansas). Data were received from other sources but were not
suitable for growth type calculations. The total number of
acceptable growth curves is nearly 400. Detailed instructions for
data analysis have been prepared. A research associate is now
using these instructions for calculation of growth types. This work
includes normalizing height data by setting tree height to 1 at a
selected base age, splitting the range of normalized tree heights
into segments, and obtaining growth types by averaging within
statistically selected growth type classes.
LUCAS
System Upgrade
The parallel version of LUCAS (pLUCAS) was originally
implemented on a network of Sun SPARCstation 5's using PVM
(parallel virtual machine) software. The network consisted of
regular ethernet (10 Mbs) with each machine having 32 Mbytes of
RAM. Recently, the lab was upgraded so that each machine now has
96 Mbytes of RAM, and the network is served by fast ethernet (100
Mbs) through a switching mechanism so that any machine can
communicate directly (i.e., point-to-point) with any other machine.
For performance purposes, pLUCAS was rerun on the upgraded
system and the results were compared to the previous setup. A
test simulation involved running 10 replicates of 20 time steps for
each of four pixel-based land-use change scenarios for a forested
watershed. Each timestep represents a 5-year period.
Parallelization across replications was used so that each node
performed the entire LUCAS evaluation for one replication of a 100
year simulation. The following table illustrates the timings for
the serial LUCAS, and pLUCAS using 4, 8, 12, 16, and 20
workstations (nodes) in parallel. The timings reflect elapsed wall-
clock time. The upgrade reduces clock time by a factor of 3 to 4.
Timing Comparisons (minutes) - Sun SPARC 5 Workstations
| NODES | BEFORE UPGRADE | AFTER UPGRADE |
|---|
| 1 | 278.78 | 91.47 |
| 2 | 76.18 | 24.33 |
| 8 | 41.13 | 12.85 |
| 12 | 33.33 | 10.28 |
| 16 | 26.40 | 8.17 |
| 20 | 25.88 | 6.40 |
Geographical Grid Structure
The current method of parallelization in LUCAS requires that
the entire GIS database be local (i.e., an independent copy) to all
nodes of the parallel machine/network. Considering the size of the
13-state region, placing the entire database of this magnitude on
every machine will be problematic due to memory and disk space
constraints. Therefore a method of geographically distributing the
GIS for the 13-state region is being developed. This requires that
the regional map be divided into "zones", and each node would
perform LUCAS calculations for one or more zones of the region.
At this point, a basic method for partitioning a GRASS raster
file into zones and assigning them to appropriate nodes has been
developed. The next step will involve performing LUCAS statistical
calculations and assembling the results from each zone back into a
map for the 13-state region. Alternative methods for determining
zones through cluster analysis of soil, vegetation, and climate
variables are being considered.
GIS DATALAYERS
Three databases were obtained from Government sources, and
gis maplayers were prepared for the 13-state region. The pixel
size in most of the maps is about 1 square kilometer. Maps were
prepared as follows:
USGS
30 arc-second digital elevation model for the 13 states.
(Other map layers may now be draped over a 3-dimensional
perspective of the 13-state region).
USDA Forest Service 1993 RPA Data
Percent forest cover
Distribution of major forest types
STATSGO (Natural Resources Conservation Service)
Major land resource areas (mlra)
Available soil water content (inch)
Soil organic matter (kg/m2)
Site index for loblolly pine at age 50 years (feet)
Loblolly pine productivity (m3 ha-1 y-1)
Site index for slash pine at age 50 years (feet)
The available soil water content and soil organic matter
maps were prepared as a result of the sensitivity analyses
conducted for the LINKAGES model.
Frequency distributions were prepared for site index for
loblolly pine by determining the number of pixels in selected site
index ranges. These distributions were somewhat skewed to the
right. IMP is based on the hypothesis that environmental changes
will induce change in the spatial distribution as well as the
frequency distribution of site index within the 13-state region.
Frequency distributions of site index for alternative modeling
scenarios will be statistically compared in future work.
CONTINUING EFFORTS
February-May 96
Linkage of MAESTRO to PRISM and sensitivity analysis
Calibrate UTM with MAESTRO
Implement NuCM and link to PRISM
Continue tree growth type development
Assemble data sets for application of LDM codes
Develop clustering approach for zone identification
Complete parallel zone structure for pLUCAS
Prepare GIS maplayers for soil and climate variables
June-September 96
Steering Committee meeting (June?)
Sensitivity analysis of NuCM
Complete first phase of growth type calculations
Develop response surface structure of LDM
Test pLUCAS with GIS maplayers
Progress report on current forest condition
Project meeting in Oak Ridge (September)