P2-5 Scaling problems in estimation of forest biophysical variables from remote sensing

PhD student: Collins Kukunda
Thesis Committee: Christoph Klein, Thomas Kneib, Steen Magnussen

Large area forest inventories require cost-efficient field work and remote sensing data acquisition. Increasingly common in forest inventory is the combination of both field measurements and remote sensing to estimate forest biophysical variables in a spatially explicit manner or the combination of the two datasets to improve point estimates. In this project, we focus on estimation of scale effects in prediction of forest stand variables. We are particularly interested (1) in the effect of field inventory factors, including (i) plot sizes and shapes, (ii) sample sizes and intensity, and (iii) sampling designs; and (2) in remote sensing factors including (i) ALS pulse density, (ii) spatial resolutions of optical imagery, and (iii) spectral resolution of multi-/hyperspectral imagery.

The guiding hypothesis is that in large area forest inventory, there exists a tradeoff between increased investment in either remote sensing or field inventory datasets and the overall accuracy and precision attained: a higher investment in finer scale remote sensing datasets should trigger lower investment in the amount of field inventory data required and vice versa.

To achieve our study goal, simulations aimed at estimation of precision and bias as indicators of scale effects on the expected value for forest volume and biomass are being set up. These experiments basically entail optimization of both field inventory and remote sensing options. The experiments include novel components in setup compared to what is already available in literature. First, our experiments put into account the effect of variable selection on estimation of the expected population values under a model-assisted design based inference framework. Second, we look at the effect of residual variability that is introduced by not measuring the priority target variable ?biomass? but by using (more or less generalized) allometric models. Most studies treat biomass and volume as if they were measured without error. Our study use basal area estimates and Diameter at Breast Height (DBH) as reality checks on the results for biomass considering that these four variables (i.e. volume, biomass, basal area and DBH) are known to be highly correlated. The two highlighted alterations referred to above are expected to contribute significantly to understanding scale effects in variance estimation of forest volume and biomass within the context of our experimental setup.


back to overview of all projects