A unified framework for land cover monitoring based on a discrete global sampling grid (GSG).

Environmental monitoring and assessment of the extent and change of land uses and their renewable natural resources over time is a key element in many international processes and one crucial basis for sustainable management. Remote sensing plays an increasingly important role in these monitoring systems, especially if the interest is in large areas. Integration of remote sensing requires comprehensive and careful preprocessing and a high level of expertise which is not always at hand in all applications. However, easy-to-implement sampling techniques based on visual interpretation are an alternative approach for utilizing remote sensing imagery, including the evolving archives of georeferenced and preprocessed data provided by virtual globes like Google Earth, Bing, and others. The goal of this paper is to propose a simple unified framework that may be used in the context of sampling studies and environmental monitoring from local to global scale. Besides the definition of a sampling design, the observation or plot design, i.e., defining how observations are to be made and recorded, has a strong influence on the precision of estimates as well as the overall efficiency of a sampling exercise. As an example, we present a simulation study focusing on the estimation of forest cover in artificial landscapes with different coverage and degree of fragmentation. The sampling units we compare are point clusters with different configuration and spatial extent.

Model errors in tree biomass estimates computed with an approximation to a missing covariance matrix.

BACKGROUND: Biomass and carbon estimation has become a priority in national and regional forest inventories. Biomass of individual trees is estimated using biomass equations. A covariance matrix for the parameters in a biomass equation is needed for the computation of an estimate of the model error in a tree level estimate of biomass. Unfortunately, many biomass equations do not provide key statistics for a direct estimation of model errors. This study proposes three new procedures for recovering missing statistics from available estimates of a coefficient of determination and sample size. They are complementary to a recently published study using a computationally intensive Monte Carlo approach. RESULTS: Our recovery approach use survey data from the population targeted for an estimation of tree biomass. Examples from Germany and Mexico illustrate and validate the methods. Applications with biomass estimation and robust recovered fit statistics gave reasonable estimates of model errors in tree level estimates of biomass. CONCLUSIONS: It is good practice to provide estimates of uncertainty to any model-dependent estimate of above ground biomass. When a direct approach to estimate uncertainty is impossible due to missing model statistics, the proposed robust procedure is a first step to good practice. Our recommended approach offers protection against inflated estimates of precision.

Permanent sample plots for natural tropical forests: a rationale with special emphasis on Central Africa.

Permanent sample plots (PSP), where trees are individually and permanently marked, have received increased interest in Central Africa as a tool to monitor vegetation changes. Although techniques for mounting PSP in tropical forests are well known, their planning still deserves attention. This study aims at defining a rationale for determining the size and number of replicates for setting up PSP in mixed tropical forests. It considers PSP as a sampling plan to estimate a target quantity with its associated margin of error. The target quantity considered here is the stock recovery rate, which is a key parameter for forest management in Central Africa. It is computed separately for each commercial species. The number of trees to monitor for each species defines the margin of error on the stock recovery rate. The size and number of replicated plots is obtained as the solution of an optimization problem that consists in minimizing the margin of error for every species while ensuring that the mounting cost remains below a given threshold. This rationale was applied using the data from the M'Baïki experimental site in the Central African Republic. It showed that the stock recovery rate is a highly variable quantity, and that the typical cost that forest managers are prone to devote to PSP leads to high margins of error. It also showed that the size and number of replicated plots is related to the spatial pattern of trees: clustered or spatially heterogeneous patterns favor many small plots, whereas regular or spatially homogeneous patterns favor few large plots.

Spatially Explicit Large Area Biomass Estimation: Three Approaches Using Forest Inventory and Remotely Sensed Imagery in a GIS.

Forest inventory data often provide the required base data to enable the largearea mapping of biomass over a range of scales. However, spatially explicit estimates ofabove-ground biomass (AGB) over large areas may be limited by the spatial extent of theforest inventory relative to the area of interest (i.e., inventories not spatially exhaustive), orby the omission of inventory attributes required for biomass estimation. These spatial andattributional gaps in the forest inventory may result in an underestimation of large areaAGB. The continuous nature and synoptic coverage of remotely sensed data have led totheir increased application for AGB estimation over large areas, although the use of thesedata remains challenging in complex forest environments. In this paper, we present anapproach to generating spatially explicit estimates of large area AGB by integrating AGBestimates from multiple data sources; 1. using a lookup table of conversion factors appliedto a non-spatially exhaustive forest inventory dataset (R² = 0.64; RMSE = 16.95 t/ha), 2.applying a lookup table to unique combinations of land cover and vegetation densityoutputs derived from remotely sensed data (R² = 0.52; RMSE = 19.97 t/ha), and 3. hybridmapping by augmenting forest inventory AGB estimates with remotely sensed AGB estimates where there are spatial or attributional gaps in the forest inventory data. Over our714,852 ha study area in central Saskatchewan, Canada, the AGB estimate generated fromthe forest inventory was approximately 40 Mega tonnes (Mt); however, the inventoryestimate represents only 51% of the total study area. The AGB estimate generated from theremotely sensed outputs that overlap those made from the forest inventory based approachdiffer by only 2 %; however in total, the remotely sensed estimate is 30 % greater (58 Mt)than the estimate generated from the forest inventory when the entire study area isaccounted for. Finally, using the hybrid approach, whereby the remotely sensed inputswere used to fill spatial gaps in the forest inventory, the total AGB for the study area wasestimated at 62 Mt. In the example presented, data integration facilitates comprehensiveand spatially explicit estimation of AGB for the entire study area.

A variance estimator for constrained estimates of change in relative categorical frequencies.

Consistent estimators of change and state becomes an issue when sample data come from a mix of permanent and temporary observation units. A joint maximum likelihood estimator of state and change creates estimates of state that depend on antecedent viz. posterior survey results and may differ from estimates of state derived from a single-date analysis of the sample data. A constrained estimator of change in relative categorical frequencies that eliminates this potential inconsistency is proposed and a model based estimator of their sampling variance is developed. The performance of the constrained estimator is quantified against six criteria and a joint maximum likelihood estimator in simulated sampling from 15 populations with three combinations of permanent and temporary samples, four to six categorical class attributes, and constant size between sampling dates. Bias of the constrained estimators was negligible but larger than for joint maximum likelihood estimators. Mean absolute deviations and variances of constrained estimators were generally at par with the joint estimators. Constrained estimators of root mean square errors and achieved coverage of nominal confidence intervals of constrained estimators were occasionally better. A generalized variance function for the constrained estimates of change is provided as a computational shortcut.

Effects of inbreeding on coastal Douglas fir growth and yield in operational plantations: a model-based approach.

In advanced generation seed orchards, tradeoffs exist between genetic gain obtained by selecting the best related individuals for seed orchard populations, and potential losses due to subsequent inbreeding between these individuals. Although inbreeding depression for growth rate is strong in most forest tree species at the individual tree level, the effect of a small proportion of inbreds in seed lots on final stand yield may be less important. The effects of inbreeding on wood production of mature stands cannot be assessed empirically in the short term, thus such effects were simulated for coastal Douglas fir [ Pseudotsuga menziesii var. menziesii (Mirb.) Franco] using an individual-tree growth and yield model TASS (Tree and Stand Simulator). The simulations were based on seed set, nursery culling rates, and 10-year-old field test performance for trees resulting from crosses between unrelated individuals and for inbred trees produced through mating between half-sibs, full-sibs, parents and offspring and self-pollination. Results indicate that inclusion of a small proportion of related clones in seed orchards will have relatively low impacts on stand yields due to low probability of related individuals mating, lower probability of producing acceptable seedlings from related matings than from unrelated matings, and a greater probability of competition-induced mortality for slower growing inbred individuals than for outcrossed trees. Thus, competition reduces the losses expected due to inbreeding depression at harvest, particularly on better sites with higher planting densities and longer rotations. Slightly higher breeding values for related clones than unrelated clones would offset or exceed the effects of inbreeding resulting from related matings. Concerns regarding the maintenance of genetic diversity are more likely to limit inclusion of related clones in orchards than inbreeding depression for final stand yield.

Joint versus separate estimation of state and change in category frequencies from repeat stratified two-phase sampling with a fallible classifier.

Joint maximum likelihood estimates (JML) of category frequencies and change from repeat stratified two-phase sampling surveys with a fallible classifier are often seriously biased and have large root mean square errors when they are obtained for small populations (<5,000) with three or more categories and a moderate to small phase II sample size (<1,000). JML estimates of state also depend on antecedent or posterior data, a recipe for inconsistency. In these situations, a separate maximum likelihood estimation (SML) of category frequencies at each survey date appears preferable. SML estimates of net change are obtained as the difference in states. SML standard errors of change are obtained via an estimate of the temporal correlation and variances of state. A bivariate binary logistic model of change provided the estimate of temporal correlation. SML generally outperformed JML significantly in terms of bias and root mean square errors in eight case studies.