Skip to content

Status Estimates

One of the most frequent needs of FIA data are estimates of status variables. Status variables capture the state of a forest over a given window of time, typically specified by the temporal range of an evaluation. In this vignette, we will estimate various status variables for the state of Vermont using an evaluation that covers the first cycle of the inventory, 2003 to 2006. This inventory is included in a miniature database, and readers should be able to follow along with their own installation

We will produce estimates for three use cases:

  1. The proportion of forested land in the state.
  2. Growing stock by county.
  3. Average diameter by species.

The first two cases are examples of post-stratified estimation, while the third is involves the more complex ratio estimator. The ratio estimator is covered more thouroughly in a separate vignette.

As always, we need to specify a handler, which manages the connection to the database for a specific evaluation. We will use the 2003 to 2006 evaluation for Vermont status variables, which has evaluation code 500601.

library(fiaplyr)
library(DBI)
library(duckdb)
library(dplyr)
# Connect to the Vermont mini database
con <- dbConnect(duckdb(), fiadb_vt_mini_path())
# Create a handler for the 2003 to 2006 evaluation for Vermont status variables
handler <- eval_handler(con, 500601)
# Print the handler summary
handler
EvalHandler
----------
EVALID: 500601
Description: VERMONT 2006: 2003-2006: CURRENT AREA, CURRENT VOLUME
Plots: 757
Inventory Years: 2003 - 2006
Measure Years: 2003 - 2007

The estimation of the proportion of forested land involves the use of conditions, which are mapped portions of the plot. Further reading on conditions can be found [here, todo]. If we are interested in the estimate of the proportion of forested land, we can make a new handler that is filtered to only include conditions that are forested, which is obtained by filtering on COND_STATUS_CD == 1.

forested_handler <- handler |>
subset(cond(COND_STATUS_CD == 1))

Post-stratified estimates, among others used by FIA, can be made using the estimate method on the handler. By default, the estimate method uses the PostStratifiedEstimator() class, which implements the standard FIA post-stratified estimator used most frequently in FIA state reports and similar products. Specify the estimation target with the same scopes used throughout the package. For example, cond() estimates condition proportions, while tree(...) estimates tree-level attributes.

forested_handler |>
estimate(cond())
# A query: ?? x 3
# Database: DuckDB 1.5.5
var estimate se
<chr> <dbl> <dbl>
1 prop 0.772 0.00909

Here, we see the proportion of forested land in Vermont is 0.772, with a standard error of 0.009.

The estimation of growing stock is frequently used to assess the amount of merchantable timber within a region. Growing stock trees are identified in the tree table with TREECLCD == 2.

growing_stock_handler <- handler |>
subset(tree(TREECLCD == 2)) |>
partition(cond(COUNTYCD))
growing_stock_est <- growing_stock_handler |>
estimate(tree(VOLCFGRS))
head(growing_stock_est)
# A query: ?? x 4
# Database: DuckDB 1.5.5
COUNTYCD.cond var estimate se
<int> <chr> <dbl> <dbl>
1 1 VOLCFGRS 91.1 13.3
2 3 VOLCFGRS 137. 13.1
3 5 VOLCFGRS 87.3 8.22
4 7 VOLCFGRS 72.4 10.4
5 9 VOLCFGRS 88.2 8.20
6 11 VOLCFGRS 76.7 8.58

The above results require a certain care of interpretation. Like any domain estimate, the per-acre densities apply to the entire evaluation area, i.e., the state of Vermont. So the estimate for COUNTYCD 1 is interpreted as, “cubic feet of growing stock per acre in the state of Vermont for trees that are in COUNTYCD 1” rather than “cubic feet of growing stock per acre in COUNTYCD 1”. Hence, smaller counties will tend to have smaller estimates, because they occupy a smaller portion of the state, thereby deflating the per-acre density that applies to the entire state.

Normalizing county estimates by their area is simple, but requires the user to fetch their own data. At minimum, a data frame describing the county proportion areas is needed. Another option is to use a ratio estimator, which is covered in a separate vignette. However, this imparts unnecessary variation, because the area of counties is known with certainty, and the ratio estimator is designed to account for the variation in the denominator (i.e., an estimate of the county size).

Average diameter by species is a more complex estimate, because it requires the use of a ratio estimator. The numerator is formed by the sum of the diameters of trees of a given species, while the denominator is formed by the count of trees of that species.

avg_dia_handler <- handler |>
subset(tree(STATUSCD == 1)) |>
partition(tree(SPCD))
avg_dia_ests <- avg_dia_handler |>
estimate(
ratio(tree(DIA), tree())
) |>
filter(SPCD_n == SPCD_d) |>
arrange(desc(estimate))
head(avg_dia_ests)
# A query: ?? x 6
# Database: DuckDB 1.5.5
# Ordered by: desc(estimate)
SPCD_n SPCD_d var_n var_d estimate se
<dbl> <dbl> <chr> <chr> <dbl> <dbl>
1 922 922 DIA tree_count 26.8 0
2 125 125 DIA tree_count 16.5 0.473
3 540 540 DIA tree_count 15.8 0
4 742 742 DIA tree_count 14.4 3.81
5 837 837 DIA tree_count 13.3 0
6 832 832 DIA tree_count 12.2 0.908

Note that the domain specifications are crossed between the numerator and denominator (generating dia x density estimates for all pairs of species, a behavior that is useful for other purposes), hence we filter only to those that match. More expressive examples of ratio estimation is given in the ratio estimates vignette.