A spatial and temporal assessment of nonresponse in the national forest inventory of the U.S.

In nearly all national forest inventories (NFI), some sample plots are unable to be measured such that nonresponse may be an issue of concern. Thus, it is of particular interest to understand the phenomenon in terms of current status and temporal change in nonresponse rates and the associated spatial distribution on the landscape. In the NFI of the USA, denial of access permission on privately owned forest land and hazardous conditions has led to an overall nonresponse rate of 9.8% with some areas exceeding 20% of plots being inaccessible. Further, it was found that nearly 50% of the areas studied were exhibiting increasing rates of nonresponse over time. Comparisons between response and nonresponse plots via remote sensing characteristics suggested there may be systematic differences in some parts of the country, which may cause bias in the sample and resulting estimates. The findings indicate that improved communication strategies with private landowners are needed to reduce nonresponse rates. Due to the unlikelihood of eliminating nonresponse entirely, methods to mitigate potential nonresponse bias should be considered for incorporation into the estimation of population parameters.

Strategies for the use of mixed-effects models in continuous forest inventories.

Forest inventory data often consists of measurements taken on field plots as well as values predicted from statistical models, e.g., tree biomass. Many of these models only include fixed-effects parameters either because at the time the models were established, mixed-effects model theory had not yet been thoroughly developed or the use of mixed models was deemed unnecessary or too complex. Over the last two decades, considerable research has been conducted on the use of mixed models in forestry, such that mixed models and their applications are generally well understood. However, most of these assessments have focused on static validation data, and mixed model applications in the context of continuous forest inventories have not been evaluated. In comparison to fixed-effects models, the results of this study showed that mixed models can provide considerable reductions in prediction bias and variance for the population and also for subpopulations therein. However, the random effects resulting from the initial model fit deteriorated rapidly over time, such that some field data is needed to effectively recalibrate the random effects for each inventory cycle. Thus, implementation of mixed models requires ongoing maintenance to reap the benefits of improved predictive behavior. Forest inventory managers must determine if this gain in predictive power outweighs the additional effort needed to employ mixed models in a temporal framework.

Precision and cost considerations for two-stage sampling in a panelized forest inventory design.

Due to the relatively high cost of measuring sample plots in forest inventories, considerable attention is given to sampling and plot designs during the forest inventory planning phase. A two-stage design can be efficient from a field work perspective as spatially proximate plots are grouped into work zones. A comparison between subsampling with units of unequal size (SUUS) and a simple random sample (SRS) design in a panelized framework assessed the statistical and economic implications of using the SUUS design for a case study in the Northeastern USA. The sampling errors for estimates of forest land area and biomass were approximately 1.5-2.2 times larger with SUUS prior to completion of the inventory cycle. Considerable sampling error reductions were realized by using the zones within a post-stratified sampling paradigm; however, post-stratification of plots in the SRS design always provided smaller sampling errors in comparison. Cost differences between the two designs indicated the SUUS design could reduce the field work expense by 2-7 %. The results also suggest the SUUS design may provide substantial economic advantage for tropical forest inventories, where remote areas, poor access, and lower wages are typically encountered.

A primer for nonresponse in the US forest inventory and analysis program.

Nonresponse caused by denied access and hazardous conditions are a concern for the USDA Forest Service, Forest Inventory and Analysis (FIA) program, whose mission is to quantify status and trends in forest resources across the USA. Any appreciable amount of nonresponse can cause bias in FIA's estimates of population parameters. This paper will quantify the magnitude of nonresponse and describe the mechanisms that result in nonresponse, describe and qualitatively evaluate FIA's assumptions regarding nonresponse, provide a recommendation concerning plot replacement strategies, and identify appropriate strategies to pursue that minimize bias. The nonresponse rates ranged from 0% to 21% and differed by land owner group; with denied access to private land the leading cause of nonresponse. Current FIA estimators assume that nonresponse occurs at random. Although in most cases this assumption appears tenable, a qualitative assessment indicates a few situations where the assumption is not tenable. In the short-term, we recommend that FIA use stratification schemes that make the missing at random assumption tenable. We recommend the examination of alternative estimation techniques that use appropriate weighting and auxiliary information to mitigate the effects of nonresponse. We recommend the replacement of nonresponse sample locations not be used.

Detecting change in advance tree regeneration using forest inventory data: the implications of type II error.

Achieving adequate and desirable forest regeneration is necessary for maintaining native tree species and forest composition. Advance tree seedling and sapling regeneration is the basis of the next stand and serves as an indicator of future composition. The Pennsylvania Regeneration Study was implemented statewide to monitor regeneration on a subset of Forest Inventory and Analysis plots measured by the U.S. Forest Service. As management techniques are implemented to improve advance regeneration, assessments of the change in the forest resource are needed. When the primary focus is on detecting change, hypothesis tests should have small type II (ß) error rates. However, most analyses are based on minimizing type I (α) error rates and type II error rates can be quite large. When type II error rates are high, actual improvements in regeneration can remain undetected and the methods that brought these improvements may be deemed ineffective. The difficulty in detecting significant change in advance regeneration when small type I error rates are given priority is illustrated. For statewide assessments, power (1-ß) to detect changes in proportion of area having adequate advance regeneration is relatively weak (≤0.5) when the change is smaller than 0.05. For evaluations conducted at smaller spatial scales, such as wildlife management units, the reduced sample size results in only marginal power even when relatively large changes (≥0.20) in area proportion occur. For fixed sample sizes, analysts can consider accepting larger type I error rates to increase the probability of detecting change (smaller type II error rates) when it occurs, such that management methods that positively affect regeneration can be identified.