The successful implementation of habitat preservation and management demands regular and spatially explicit monitoring of conservation status at a range of scales based on indicators. Woodland condition can be described in terms of compositional and structural attributes (e.g. overstorey, understorey, ground flora), evidence of natural turnover (e.g. deadwood and tree regeneration), andanthropogenic influences (e.g.disturbance, damage). Woodland condition assessments are currently conducted via fieldwork, which is hampered by cost, spatial coverage, objectiveness and repeatability. This project evaluates the ability of airborne remote sensing (RS) techniques to assess woodland condition, utilising a sensor-fusion approach to survey a foreststudy site and develop condition indicators. Here condition is based on measures of structural and compositional diversity in the woodland vertical profile, with consideration of the presence of native species, deadwood, and tree regeneration. A 22 km2 study area was established in the New Forest, Hampshire, UK, which contained a variety of forest types, including managed plantation, semi-ancient coniferous and deciduous woodland. Fieldwork was conducted in 41 field plots located across this range of forest types, each with varying properties. The field plots were 30x30m in size and recorded a total of 39 forest metrics relating to individual elements of condition as identified in the literature. Airborne hyperspectral data (visible and near-infrared) and small footprint LiDAR capturing both discrete-return (DR) and full-waveform (FW) data were acquired simultaneously, under both leaf-on and leaf-off conditions in 2010. For the combined leaf-on and leaf-off datasets a total of 154 metrics were extracted from the hyperspectral data, 187 metrics from the DR LiDAR and 252 metrics from the FW LiDAR. This comprised both area-based and individual tree crown metrics. These metrics were entered into two statistical approaches, ordinary least squares and Akaike information criterion regression, in order to estimate each of the 39 field plot-level forest variables. These estimated variables were then used as inputs to six forest condition assessment approaches identified in the literature. In total, 35 of the 39 field plot-level forest variables could be estimated with a validated NRMSE value below 0.4 using RS data (23 of these models had NRMSE values below 0.3). Over half of these models involved the use of FW LiDAR data on its own or combined with hyperspectral data, demonstrating this to be single most able dataset. Due to the synoptic coverage of the RS data, each of these field plot variables could be estimated and mapped continuously over the entire study site at the 30x30m resolution (i.e. field plot-level scale). The RS estimated field variables were then used as inputs to six forest condition assessment approaches identified in the literature. Three of the derived condition indices were successful based on correspondence with field validation data and woodlandcompartment boundaries. The three successful condition assessment methods were driven primarily by tree size and tree size variation. The best technique for assessing woodland condition was a score-based method which combined seventeen inputs which relate to tree species composition, tree size and variability, deadwood, and understory components; all of whichwere shown to be derived successfully from the appropriate combination of airborne hyperspectral and LiDAR datasets. The approach demonstrated in this project therefore shows that conventional methods of assessing forest condition can be applied with RS derived inputs for woodland assessment purposes over landscape-scale areas.
Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:589976 |
Date | January 2013 |
Creators | Sumnall, M. J. |
Publisher | Bournemouth University |
Source Sets | Ethos UK |
Detected Language | English |
Type | Electronic Thesis or Dissertation |
Source | http://eprints.bournemouth.ac.uk/21078/ |
Page generated in 0.0013 seconds