XIV WORLD FORESTRY CONGRESS, Durban, South Africa, 7-11 September 2015

Forest inventory design principles – challenges and solutions

Andrew J. Lister1, Charles T. Scott2, C. Kenneth Brewer3, Beth R. Stein4
1
[email protected]; USDA Forest Service, FIA, 11 Campus Blvd, Ste. 200 Newtown Square, PA 19073
2
[email protected]; USDA Forest Service, FIA, 11 Campus Blvd, Ste. 200 Newtown Square, PA 19073
3
[email protected]; METI, Inc. 8600 Boeing Drive El Paso, TX 79925
4
[email protected]; USDA Forest Service, FIA, 11 Campus Blvd, Ste. 200 Newtown Square, PA 19073

Abstract
Large-area forest inventories are complex and require a multidimensional approach. The National
Inventory and Monitoring Applications Center (NIMAC), a division of the United States Department of
Agriculture (USDA) Forest Service’s Forest Inventory and Analysis (FIA) program, has participated in
the design of several large-area forest resource monitoring programs in recent years. These international
efforts have been part of the US government’s interagency SilvaCarbon program, which provides
technical assistance to tropical countries in support of the United Nations’ Reducing Emissions from
Deforestation and Forest Degradation (REDD) effort. NIMAC has developed a comprehensive series of
steps and associated tools to provide technical assistance to inventory efforts. These steps can serve as a
roadmap for inventory design and implementation. In the current paper, we describe these steps and
provide observations on strengths and weaknesses of aspects of forest inventories from our work with
both FIA and other countries around the world.

Keywords: Forest inventory design, forest monitoring systems, FIA, sample design, plot design

Introduction, scope and main objectives

Over a period of several hundred years, forest inventory science and technology have evolved to the
point where countries at all stages of development have the capacity to implement national forest
inventories. However, forest inventory design is a complicated business. It is often helpful for countries
to investigate what has been done elsewhere in order to learn current techniques and avoid common
pitfalls.

The USDA Forest Service’s Forest Inventory and Analysis (FIA) program has accumulated many
decades of experience conducting the US national forest inventory. FIA’s National Inventory and
Monitoring Applications Center (NIMAC), a group of inventory techniques specialists, is sometimes
asked to share some of this accumulated knowledge with partner countries. NIMAC works closely with
the SilvaCarbon program (www.silvacarbon.org), a consortium of US government agencies and partners
that conduct capacity building and scientific partnerships with countries interested in developing or
improving systems for monitoring terrestrial carbon. Activities include informal consultations, formal
trainings, workshop participation and science partnerships.

The goal of the current paper is to describe the forest inventory design principles that NIMAC uses as
the foundation of its technology transfer efforts. These principles are based in sampling theory and the
practical experience NIMAC staff members have accumulated over many decades of helping design and
implement not only the US national forest inventory, but also those of other partner countries.
Methodology/approach
NIMAC has found that the use of the following set of roughly linear monitoring steps can help structure
inventory design and implementation projects. These steps are interrelated such that choices made
earlier will affect those made lower down in the process.

1) Information needs assessment

Inventory planners are often reluctant to invest in carefully identifying the objectives and specific
monitoring questions that motivate the establishment of a forest inventory. We therefore advocate
an approach to planning that we have captured in a Microsoft Excel-based tool called the Design
Tool for Inventory and Monitoring (DTIM) (Scott et al. 2009). The tool helps users make efficient
design choices. DTIM users first choose broad monitoring objectives, like improving forest
productivity, carbon balance or diversity. Next, the user selects monitoring questions that address
those objectives – for example, to meet the objective of forest productivity increase, a manager
might want to know the current forest growth and harvest rates. With monitoring questions chosen,
the user then chooses metrics and identifies specific tables of results that will answer the questions
that address the broad objective. DTIM contains a relational database with commonly-chosen,
linked objectives, questions and metrics (figure 1), and has functionality for choosing sample sizes
under different sample design scenarios. The DTIM approach can help groups efficiently make and
document design decisions based on the relationship between monitoring objectives, questions,
metrics, and results tables. DTIM helps address both the “why” and the “what” for the forest
inventory.

Fig. 1: Schematic diagram of an example of the DTIM process. A: Broad monitoring objectives. B:
Specific monitoring questions. C: Metrics used to create D: Specific tables of results.

2) Identification of constraints
Inventory budgets are limited, so the goal of the design process is to generate inventories with
useful results for the minimum cost. A cost optimization principle can help achieve this goal. This
principle requires the definition of “useful results”, and then leads to a design that meets, but does
not exceed, this constraint. For example, if the original motivation of the inventory was carbon
sequestration at the national level, and the specific monitoring question identified was to determine
current forest tree carbon stocks, then the planner may ask what will make a result useful. The
answer generally involves a precision requirement, often expressed as a relative confidence
interval. For example, entities that grant carbon credits are more likely to grant them if the 95%
confidence intervals of the estimates of carbon stock change are ±10%, versus a higher number. It
is critical that the precision requirement be attached to a reporting unit, or geographic area in which
relevant management decisions will be taken.
 Other constraints can include:
a) A desire for an inventory with at least some permanent plots. Remeasurement of sample
plots is generally the best method for estimating change. However, if the inventory uses pre-
stratification, it should use stratum boundaries that will not change over time.
b) Interest in statistically-valid estimates for the study region before the end of the inventory
cycle. This could lead to a spatially-balanced, panelized design, in which a percentage of the
plots are measured uniformly across the region each year, allowing for valid annual estimates
for the study area. This approach can be more costly, but it allows for more annual consistency
in the budgeting process, creates more frequent contact between government agencies and those
who live in outlying areas, and guards against the danger that there will be no national-scale
inventory estimates if priorities shift and the budget for the inventory is cut before a full cycle
has been completed.
c) Problems with accessibility and other field logistics. There are several sample and plot design
choices that can account for this. For example, multistage designs are often suitable when
moving large distances across the landscape is costly relative to measuring several plots in a
smaller area.

3) Choice of sample design

Generally, it becomes clear based on the above analysis what type of sample design is most
appropriate (Köhl et al. 2011). The design should cover the entire landscape (e.g., country) such
that all areas have a positive and known probability of selection. Sampling all land uses but
measuring trees only in forests results in accurate estimates of forest area and other characteristics
and their change over time. There are several nuances that can complicate matters. For example,
sample designs that assume a stable budget and a consistent inventory team (i.e., those that require
a complex logistical network and a lot of expert decisions) might not be appropriate if the country’s
inventory team is likely to be transient. Similarly, overly-simplified designs can waste money.
Designs that incorporate remote sensing can exploit the correlation structure between the ground
and image data and thus be much more efficient (McRoberts et al. 2013). The best strategy for
choosing a design that will meet information needs within the identified constraints is to carefully
analyse the options with all parties and document the rationale for decisions.

4) Choice of plot design

In keeping with the cost optimization principle, the optimal plot design is one that, when used with
the chosen sample design, meets but does not exceed precision requirements for each of the key
metrics. The best design would be one that, for the least cost, makes the mean value of the attribute
of interest on each plot as close as possible to that of the sample mean. This criterion emerges from
the way variance of estimates is calculated:
n __

 ( y  y)
i 1
i
2

Variance of the mean =

n(n  1)
where yi is the plot-level summary, y̅ is the mean of the sample, and n is the sample size. To create
a plot design where yi is as close as possible to y̅ for the minimum cost, one can increase the plot
area measured, thus capturing more of the variability found on the landscape on each plot.
However, measurements taken close to each other are, on average, more similar than measurements
taken further apart. For example, trees in one habitat patch tend to be more similar to their
neighbours than to those in different habitat patches. In order to avoid wasting effort on measuring
redundant information (effort that could have been spent collecting non-redundant information),
cluster plots are often used. Separation between subplots within a cluster maximizes the amount of
new, non-redundant information that is collected over the same amount of plot area sampled.

The complexity arises when having to choose what combination of the plot design parameters
(area, cluster shape, number of subplots, and separation distance between subplots) is optimal with
respect to cost. This is because every amount of effort expended on a single plot is effort that
 cannot be invested on the second way to lower the variance of the estimate in the above equation –
increasing the sample size, n. The plot design optimization process that NIMAC espouses thus
incorporates on-plot costs, between-plot costs, autocorrelation structure of the attribute of interest,
and sample design effects (Scott 1993). NIMAC has developed the Forest Resource Inventory, Edit
and Design tool (FRIED) for helping implement this optimization procedure. It allows users to
evaluate the effects of different combinations of cost structure and plot and sample design on
inventory cost and precision. Users of FRIED can assess the impacts of different choices of design
variables and make more informed decisions.

5) Preparation for field work

This step generally involves activities like compiling a field guide, field logistics and staffing,
training and quality assurance protocol development, and other administrative activities. A key
recommendation is the use of a field computer to collect data. Experience has shown that the
success of an inventory rests with the quality of the data collected. Data recorders can help ensure
that complete, logically consistent data are verified before leaving the plot. The act of programming
edit checking procedures in the data recorder makes the post-field data cleansing step, described
below, much easier because it forces the inventory planners to consider all of the scenarios that
might lead to errors or confusion in the field before the crews actually begin their training and
fieldwork.

6) Field work
A formal quality assurance (QA) procedure is often overlooked or not given adequate attention. We
recommend a QA procedure that combines “hot” checks (real time supervision of crews by experts,
with immediate feedback), “cold” checks (experts randomly visit completed plots, and, in the field,
compare the previous data with their own measurements), and “blind” checks (plots are randomly
selected for remeasurement by a second crew, without access to the previous crews’ data). Hot and cold
checks are quality control procedures, whereas blind checks assess repeatability of measurement
techniques. Quality assurance reports are for transparency and should be published so inventory
data consumers can assess the potential impacts of measurement error on the interpretability of
results.

7) Field data cleansing and data processing

The data must either be uploaded from the field data recorders or entered from paper records into a
database that is regularly backed up. Post-field data cleansing and processing are often complicated
by the lack of high quality data coming from the field. These include data that do not conform to
the original data model (missing plots or other measurements), or data with illogical information
that was not caught in the field by a data recorder. It is wise to assemble a standardized checklist of
data quality assessments and protocols for correcting errors, and institutionalize this procedure so
that the inventory team can consistently repeat the process as new data come in. If this is not done,
it is often difficult to discern how discrepancies were handled, making it difficult to interpret the
quality of the results.

8) Calculation of inventory estimates

There are numerous software options for calculating inventory estimates with standard methods.
However, mistakes are sometimes made when implementing the equations, or when coping with
incomplete information. For example, if inaccessible plots are simply removed from the analysis,
the user of the inventory data must accept the assumption (if he or she is even aware of what was
done) that the inaccessible areas are, on average, the same as the accessible areas with respect to
forest status and trends. However, it is likely that inaccessible areas are different than the norm due
to the same factors that make them inaccessible, like topography or human intervention.

It is advisable to use estimation software that allows for transparency, repeatability, and
implementation by members of the inventory team. Shareable software that allows for interactive
construction of results tables, including confidence intervals, is ideal. In short, biometric and
programming expertise is critical for an effective team.
9) Analysis of inventory estimates and transfer of summarized results to decision makers
Knowledge of ecology, statistics, forest management, databases and GIS software are all desirable
attributes for an analyst. Analysts should be able to identify patterns and trends in the information,
and provide interpretable summaries for decision makers. Finally, they must be able to perform the
critical last step of quality assurance – assessment of the compiled estimates to see if there are any
systemic anomalies that were not apparent when checking the raw data.

10) Making and implementing management decisions, with feedback to the process.
The ultimate goal of an inventory is to provide information that will inform decisions that will lead
to better forest management. Impacts of management decisions should be assessed and a
mechanism for giving feedback to the inventory process should exist. For example, in the United
States, continuous improvement of the FIA program is partially brought about by periodic meetings
of regional and national user-groups. At these meetings, users provide feedback and propose
changes that will make the program more responsive to their needs. If the users are satisfied with
FIA products, they provide positive feedback to legislators who will in turn continue to support the
program.

Discussion
Common weak points exist in many inventories:
1. There is often weak coordination between the remote sensing experts and the field
inventory experts, either due to lack of awareness, competing interests, geographic
separation of technical staff, or a combination. Furthermore, technical assistance offered by
foreign aid agencies often revolves around the use of experts with one or the other technical
approach. Countries thus end up following two parallel paths that are both aimed at the same
goal of creating a forest monitoring system. It is better to take a holistic view by first creating a
vision for how these two parallel approaches can be combined by leveraging the strengths of
both. With wall-to-wall coverage and a correlation with features related to carbon density and
perhaps species composition, remote sensing can serve as a medium with which to carry the
detailed, comprehensive information found on design-based forest inventory plots across the
landscape using methods like model-based or model-assisted estimation. Similarly, remote
sensing maps can be used to impute information from known areas into small areas without
enough inventory plots for probabilistic design-based estimation.
2. Inventory staff is transient and institutional knowledge is lost. Due to the nature of
budgeting and the shifting nature of employment of natural resource professionals in many
countries, it is often difficult to institutionalize knowledge and protocols that will persist beyond
the near term. Because forest inventories have many complex, interdependent components, the
steward of the inventory must have a clear vision of what is needed for success, and have
technical staff that she or he can count on to understand the science behind the inventory. If
there is no core group that understands the process fully and has a sense of pride and ownership
in the inventory, it is difficult to maintain momentum and quality as conditions change.
Documented protocols, software tools, instruction manuals and checklists for carrying out
common tasks like field training, quality assurance, data cleansing, and estimate calculation are
critical for full transparency. Ideally, core staff will stay for long periods of time and serve as
mentors to newer team members.
3. There is not a clear understanding of the competing purposes of a forest inventory. Often,
ecologists are not satisfied with a forest inventory design that emerges from the cost
optimization approach described above. For example, biological diversity assessments or
vegetation community classification is traditionally done by establishing large plots in “typical”
conditions – in stands that represent a specific vegetation type, or represent a gradient of
 environmental conditions. However, the act of moving plots to representative areas biases
inventory results, and measuring large plots can lead to redundant information that does little to
lower the variance of the estimate compared to other options. Similarly, remote sensing
specialists will prefer large, square plots that align well with pixels, but these plots are often not
optimal in terms of inventory efficiency. It is therefore important that all partners – ecologists,
remote sensing specialists, and sampling specialists – understand the theory and assumptions
behind each other’s perspectives and communicate with the same vocabulary.

Conclusions/outlook
Sound planning is critical to the success of a forest inventory. Following a logical, comprehensive
inventory design process will lead to efficient decisions, and help ensure that project management is
more effective. Taking advantage of forest inventory experience from elsewhere can help planners avoid
common pitfalls and leverage institutional knowledge. Above all, a country’s sense of ownership and
commitment to a national forest inventory will raise the chances of success and help achieve the goal of
healthy, productive, sustainable forests that will lead to the betterment of the quality of human life for its
citizens.

The views expressed in this information product are those of the author(s) and do not
necessarily reflect the views or policies of FAO.

References
Köhl M, Lister A, Scott CT, Baldauf T, Plugge, D. 2011. Implications of sampling design and sample
size for national carbon accounting systems. Carbon Balance and Management, 6(10). Available at:
http://treesearch.fs.fed.us/pubs/46276 [accessed 09.04.15]
McRoberts R, Tomppo, EO, Vibrans, AC, de Freitas, JV. 2013. Design considerations for tropical
forest inventories. Brazilian Journal of Forestry Research, 33(74): 189-202.
Scott CT. 1993. Optimal design of a plot cluster for monitoring. pp. 233-242 in The Optimal Design
of Forest Experiments and Forest Surveys: Proceedings, IUFRO S.4.11 Conference. Rennolls, K and
Gertner, G (eds.). University of Greenwich, London, UK. Available at:
http://www.nrs.fs.fed.us/pubs/jrnl/1993/ne_1993_scott-c_001.pdf [accessed 09.04.15]
Scott CT, Bush R, Brewer K. 2009. Design Tool for Inventory and Monitoring. pp. 230-237 in Forest
Inventory and Analysis (FIA) Symposium 2008. McWilliams, W, Moisen, G, and Czaplewski, R
(eds.). USDA Forest Service, Rocky Mountain Research Station, Fort Collins, CO. RMRS-P-56CD.
Available at: http://www.fs.fed.us/rm/pubs/rmrs_p056.pdf [accessed 09.04.15]