Optimal sampling and modeling strategies for quantifying natural resources over large geographical regions
Date
2008
Journal Title
Journal ISSN
Volume Title
Abstract
Chapter 1 evaluates a new approach of modeling the spatial distribution of soil attributes over large geographical regions. A combination of three-stage least squares (3SLS) and multivariate regression trees (MRT) was used to model the spatial variability in soil texture. In 2006, 1427 soil samples were collected as part of a state-wide inventory and monitoring program (IMRENAT) implemented in the State of Jalisco, Mexico. A two-way nested stratified design was used to allocate samples throughout the state based on the spectral variability of land cover and climatic conditions. The final set of models described 61% of the observed variability in soil pH, 62% of the variability in sand and 56% for clay. Comparison with other interpolation techniques such as ordinary kriging, suggest that the approach used in this study is far superior in terms of the accuracy and precision. Chapter 2 evaluates three sampling designs (i.e., simple random sampling, systematic sampling and two-way nested stratified design) for modeling the spatial variability in forest tree biomass in the State of Jalisco, Mexico. Monte Carlo simulations were used to implement the three sampling designs using samples of 500 and 1100 30 m x 30 m primary sampling units. Statistically, the two-way-nested stratified design outperformed the simple random and systematic sampling design. There was no significant difference between the simple random and systematic designs. Chapter 3 evaluates the statistical properties of plot size and sample intensities in estimating forest stand characteristics in seasonal dry evergreen forests in Huai Kha Khaeng Wildlife Sanctuary, Thailand. Monte Carlo simulations were used to evaluate plotsizes (5 m x 5 m, 10 m x 10 m, 20 m x 20 m, 25 m x 25 m and 50 m x 50 m) and sample intensities (0.5%, 1%, 2%, 5%, 10%, and 15%) on a 50 ha mapped dataset. All plot sizes and sampling intensities provided unbiased estimates of the population mean and variance for tree basal area and tree density. All plot sizes and sampling intensities were biased with respect to estimating the total number of tree species on the 50 ha plot.
Description
Rights Access
Subject
multivariate regression tree
natural resources
optimal plot size
sampling
soil pH and texture
spatial statistical models
three-stage least square
forestry