Estimating leaf area index from airborne laser scanning.
| dc.contributor.author | Pearse, Grant Dennis | |
| dc.date.accessioned | 2017-04-02T21:10:29Z | |
| dc.date.available | 2017-04-02T21:10:29Z | |
| dc.date.issued | 2017 | en |
| dc.description.abstract | Leaf area index (LAI) quantifies the amount of leaf surface area per unit ground area. LAI in forest ecosystems regulates the upper limit of possible light interception, atmospheric gas exchange, and primary production. These properties make LAI one of the most important ecophysiological variables with a wide range of potential applications. In the context of forests managed for production, knowledge of LAI offers the potential to align management activities with fundamental biophysical properties. For example, LAI offers the potential to precisely target and monitor management activities such as fertiliser application or disease control. Despite the potential benefits knowledge of LAI offers, usage is seldom seen outside of research applications. A key reason for this is the difficulty in obtaining LAI measurements over large areas, with field based optical methods largely constrained to use under uniform, diffuse sky conditions. Remote sensing of LAI offers one potential solution to obtain large-scale estimates of LAI. However, promising spectral-based approaches have been shown to have limited usefulness for forests with high LAI such as intensively managed coniferous plantations. Airborne laser scanning (ALS) data (lidar) offers enhanced ability to estimate LAI in a range of forest types with a high degree of accuracy, but the optimum methods for estimating LAI from lidar are not well established. This thesis aims to develop and demonstrate a method for estimating LAI from lidar in New Zealand’s intensively managed Pinus radiata D. Don forests. To accomplish this, two distinct areas of research are addressed. First, this thesis addresses the need for acquiring a large number of LAI field measurements covering a range of stand conditions in order to calibrate ALS-LAI models. This was accomplished by validating the use of the newly developed LAI-2200C (LI-COR Biosciences Inc., Lincoln, NE, USA). This instrument allows measurement of LAI under clear sky conditions through the application of a model to correct for the impact of scattered light on gap fraction estimates. This thesis presents the first in situ comparison of LAI measurements acquired under diffuse and clear sky conditions in a coniferous forest. These results were obtained by repeatedly measuring LAI in plots of pure P. radiata in New Zealand. In addition, the thesis presents the first assessment of the importance of acquiring accurate needle spectra to parameterise the scattering correction model. These values were acquired using newly developed methods that allow accurate spectra to be acquired from needle-leaved species via spectroradiometer. The thesis also addresses the stability of needle optical properties with respect to position in the canopy, abaxial and adaxial measurements, and variability between individual trees. The second part of this thesis used a large number of LAI measurements made possible by the new instrumentation to address key questions on the topic of estimating LAI from lidar. To date, most ALS-LAI research has been divided between establishing empirical or physical links between lidar metrics and LAI. This has resulted in a proliferation of proposed methods and lidar metrics for estimating LAI, and few studies have compared these approaches. In addition, factors known to impact ALS-LAI estimation such as the choice of plot parameters have gone relatively unexplored, as has the use of new statistical learning approaches. This thesis attempts to offer the first simultaneous assessment of the optimum combination of lidar metrics, plot parameters, and modelling approaches for estimating LAI from ALS lidar data in P. radiata forests. Results from the instrument validation suggest that the scattering correction model performs well in coniferous forests. Overall, clear sky LAI measurements were higher on average than diffuse sky measurements. However, there was evidence that this difference resulted from a reduction in erroneous readings obtained from the largest outer sensor ring under diffuse sky conditions. Traditionally, data from this part of the instrument have been error-prone and there was some evidence that clear sky LAI measurements offer increased accuracy by reducing scattering induced error across the range of zenith angles observed by the instrument. The method used to obtain spectroradiometer measurements from needle-leaved specimens was well suited to collecting accurate needle reflectance and transmittance for use in the scattering correction model. Use of these values improved agreement between clear and diffuse sky LAI measurements and reduced the magnitude of the largest differences at the extremes of the range. The results demonstrated that P. radiata spectra did not differ significantly with canopy position and were reasonably stable between trees. Measured P. radiata needle spectra are presented as part of this thesis and values are suggested for future users of the LAI-2200C scattering correction model in this forest type. Overall, use of measured spectra in combination with masking of outer ring data allowed LAI to be measured under both clear and diffuse sky conditions; however, clear sky conditions offered considerable reductions in the maximum potential measurement error resulting from changes in sky condition over time and between sensor locations. Results presented in the second part of this thesis demonstrate that LAI can be accurately estimated from lidar data in P. radiata forests. A key finding from this work was that use of standard approaches developed for use in other forest types produced some of the worst models of all those trialled, indicating that successful ALS-LAI estimation in P. radiate depends on careful selection of lidar metrics, plot parameters, and modelling approach. Specifically, results showed that (1) metrics that form a proxy for gap fraction by computing the ratio of returns above and below a chosen height threshold (ratio metrics) were key predictors of LAI; (2) choice of height threshold for ratio metrics strongly impacted model performance and P. radiata appeared to require higher thresholds than other forest types; (3) the concept of a variable height threshold was beneficial in accommodating differences in tree height across plots and led to improved estimates of LAI; (4) a larger fixed plot radius generally improved model performance; (5) use of a variable plot radius linked to instrument view distance was better than any fixed radius trialled; (6) metrics linking lidar penetration to the Beer-Lambert law were only marginally less accurate than empirical models and showed strong predictive ability. This approach may offer a means of estimating LAI without calibration by inverting the Poisson model using gap fraction from lidar and an empirical projection coefficient. Finally, the research found a high level of correlation present between lidar metrics, strongly emphasising the need for modelling approaches robust to these effects. Regularised regression via the elastic net was found to be a useful method for providing both variable and model selection in high-dimensional space while accounting for the presence of high correlation between metrics. Results from models produced by the random forests algorithm were similar to results from elastic net but provided some useful insights into variable importance. | en |
| dc.identifier.uri | http://hdl.handle.net/10092/13331 | |
| dc.identifier.uri | http://dx.doi.org/10.26021/2356 | |
| dc.language | English | |
| dc.language.iso | en | |
| dc.publisher | University of Canterbury | en |
| dc.rights | All Right Reserved | en |
| dc.rights.uri | https://canterbury.libguides.com/rights/theses | en |
| dc.title | Estimating leaf area index from airborne laser scanning. | en |
| dc.type | Theses / Dissertations | en |
| thesis.degree.discipline | Forestry | |
| thesis.degree.grantor | University of Canterbury | en |
| thesis.degree.level | Doctoral | en |
| thesis.degree.name | Doctor of Philosophy | en |
| uc.bibnumber | 2469593 | |
| uc.college | Faculty of Engineering | en |