Remote sensing estimates and measures of uncertainty for forest variables at different aggregation levels

Abstract

Keywords: small area estimation; EBLUP; MSE estimator; LiDAR; estimation of natural resources SMALL AREA ESTIMATION OF FOREST ATTRIBUTESForest management involves decision-making problems and, to evaluate the different management alternatives, up-to-date information about wood stocking, structure, and health status is needed. Decisions may affect the whole forest, and also subdivisions of the forest called stands or management units (MUs) (Packalen et al., 2011;Finley et al., 2014), sub-stands, or even smaller parts such as treatment units or harvest areas. Thus, information needs to be provided at all these divisions of the forest to assist with its management.Forest areas are typically so large that obtaining field measurement of the variables of interest over all the forest is impossible in practice. Thus, information to aid decision-making processes is typically based on descriptive statistics for each area of interest (AOI). Different authors have recognized the importance of incorporating the uncertainty or lack of knowledge in forest planning decision processes (e.g., Pasalodos et al., 2013) by supplementing the estimates associated to each AOI with suitable reliability measures. Furthermore, forest inventories are sometimes enforced, either by law (Spanish Ministry of Agriculture 1971, Castilla y Leon Regional Goverment 1999) or by contractual agreements, to provide estimates for certain MUs with relative errors below specific thresholds.Traditional forest inventories rely on probability-based sampling designs to estimate means or totals for different variables of interest. During the last decades, remotely sensed auxiliary information has been included in forest inventories to increase their efficiency. The most extended methodology for this purpose is the area-based approach (ABA) (Naesset, 2002). In this methodology, remotely sensed covariate grids covering the whole study area are typically regarded as a census, and pixels play the role of population units for which auxiliary information is available. Field measurements are also conducted in a number of plots of size similar to pixels. Each field plot is also considered as a population unit, and measurements of the variables of interest as well as auxiliary information are available for them.Design-based direct estimators such as regression estimators (Cochran, 1977) or generalized regression estimators (Deville and Sarndal, 1992) make use of highly correlated auxiliary information and have been used under the ABA setup with remotely sensed auxiliary information to increase the efficiency of forest inventories (Breidenbach and Astrup, 2012). But field surveys are expensive, and the number of field plots (or sample units) that can be allocated within a specific AOI decreases with its area. Thus, for AOIs of reduced

Effects of plot size, stand density, and scan density on the relationship between airborne laser scanning metrics and the Gini coefficient of tree size inequality

“…Structural diversity depends on the spatial resolution at which an index is evaluated (Lexerød and Eid, 2006b). Varying the scale of observation may therefore distort the information retrieved from an indicator (Chen and Crawford, 2012;Mauro et al, 2016). As plot size increases, estimates may be more reliable, but also fundamentally different stand conditions may aggregate (Coomes and Allen, 2007).…”

Analysis of spatial correlation in predictive models of forest variables that use LiDAR auxiliary information

“…Sampling designs typically use grids of plots where D r a f t the field observations are too far to detect the spatial correlation. In those cases (Woods et al 2011, Mauro et al 2016, the assumed independence for the residuals is more a consequence of the inability to observe spatial correlation patterns than an empirically tested result of the model.…”

“…For example, field plot locations are often determined before knowing the LiDAR grid (e.g. Finley et al, 2014;Mauro et al, 2016), which would likely result in misalignment. The same would happen for systematic sampling designs, if the separation between plots is not a multiple of the pixel size (i.e.…”

Estimation of forest biomass components using airborne LiDAR and multispectral sensors

“…Many forest inventories currently focus mainly on timber volume for forest industries and they usually lack the quality characteristics demanded by regional managers and renewable biomass energy industry (Riaño et al 2004, García et al 2010, Hauglin et al 2013. Some of those characteristics are the stock of detailed tree biomass components and the spatial location of these resources at finer scales (Montero et al 2005, Mauro et al 2016. In this sense, the contributions of remote sensing estimates are becoming a valuable tool (Hernando et al 2012, Anderson & Mitchell 2016 to produce accurate and cost-competitive estimates at the landscape level (Zolkos et al 2013).…”