In AndreasChristianHill/forestinventory: Design-Based Global and Small-Area Estimations for Multiphase Forest Inventories

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  fig.path = "README-"
)

forestinventory

The R-package forestinventory addresses the current interest of combining existing forest inventory data, which are derived by field surveys, with additional information sources such as remote sensing data. The major benefit of these so-called multisource inventory methods is the potential increase of estimation precision without an increase in the number of expensive field surveys. Additionally, it also allows for deriving estimates of sufficient accuracy for spatial units where terrestrial information is scarcely available if not absent.

The aim of forestinventory is to facilitate the application of multiphase forest inventories by providing an extensive set of functions for global and small-area estimation procedures. The implementation includes all estimators for simple and cluster sampling published by Daniel Mandallaz between 2007 and 2014, providing point estimates, their external- and design-based variances as well as confidence intervals. The procedures have also been optimized for the use of remote sensing data as auxiliary information.

Quick demo

We look at the example dataset grisons which comes with our package:

library(forestinventory)

library(forestinventory)
?grisons

As the help tells us, grisons contains the data of a twophase inventory: We are provided with LiDAR canopy height metrics at 306 inventory locations, and at 67 subsamples we have the terrestrially measured timber volume values. We now want to estimate the timber volume in m^3^/ha within four subdomains A, B, C and D (small areas).

If we only use the terrestrial information within the small areas, we call the onephase-function:

op <- onephase(formula = tvol~1,
               data = grisons,
               phase_id = list(phase.col = "phase_id_2p", terrgrid.id = 2),
                 area = list(sa.col = "smallarea", areas = c("A", "B", "C", "D")))

summary(op)

We now try to increase the precision of our estimates by applying a twophase estimation method, where we use the large sample of LiDAR-metrics and a linear regression model to specify the relationship between the remote sensing derived predictor variables and the terrestrial timber volume:

sae.2p.uv<- twophase(formula = tvol ~ mean + stddev + max + q75, data = grisons,
                     phase_id = list(phase.col = "phase_id_2p", terrgrid.id = 2),
                     small_area = list(sa.col = "smallarea", areas = c("A", "B","C", "D"),
                                       unbiased = TRUE))

summary(sae.2p.uv)

We now want to compare the results and performances of the onephase and twophase method. For such issues, the package provides the estTable() function that concatenates the results from the different methods in one list:

sae.table<- estTable(est.list = list(op, sae.2p.uv), sae = TRUE)

data.frame(sae.table[c(1:6,9)])

We can already see that the estimation errors of the twophase estimation are up to 5% smaller than the onephase errors.

The function mphase.gain() can now be used to further compare the performance of the methods:

mphase.gain(sae.table)

The column gain tells us that the twophase estimation procedure here leads to a 67.9 % reduction in variance compared to the one- phase procedure". The column rel.eff specifies the relative efficiency that can be interpreted as the relative sample size of the one-phase estimator needed to achieve the variance of the multi-phase (here twophase) estimator. For small area "B" we can thus see that we would have to increase the terrestrial sample size by factor 3 in the one-phase approach in order to get the same estimation precision as the twophase extended psynth estimator.

So in our short example, we were able to considerably improve the estimation precision when combining the terrestrial data with the remote sensing data. The package forestinventory offers further estimators that can be applied to a wide range of multiphase inventory scenarios.

Installation

The package can be installed from CRAN:

install.packages("forestinventory")

AndreasChristianHill/forestinventory documentation built on Aug. 16, 2021, 2:13 p.m.