Assessing site quality of South Australian radiata pine plantations using airborne LiDAR data

© 2011 Dr. Jan RomboutsSite quality information underpins many aspects of radiata pine plantation management in South Australia. Site quality assessment is essentially a problem of assessing the spatial variation of standing volume in unthinned stands at reference age nine or ten. Data collected during three experimental trials and two operational surveys were used to research a new site quality assessment system relying on airborne LiDAR data. A two-stage method was adopted characterised by the calibration of predictive models relating forest and LiDAR variables, and the subsequent application of these models to predict the forest variable across an area of interest. The four parts of the study examined the properties and behaviour of LiDAR prediction models, the sampling design and timing of field data collection, the spatial resolution of model application and the special case of partially thinned plantations. Single-variable, linear models, fitted to high altitude/low density LiDAR data captured across 20 sites scattered over an area of 10,000 km2 , had RMSE of 10-11% and 3-4% for stand volume and predominant height respectively. Evidence of site effects in the models was inconclusive. Models fitted to four LiDAR and field datasets acquired in 2002, 2006, 2007 and 2009 had consistent structure but model parameters were sensitive to the differences in operational LiDAR campaign parameters, indicating that prediction models should be re-calibrated each survey. Fifty field plots were found to be adequate to fit a regionally applicable volume prediction model. Sample selection methods only influenced model precision when sample sizes were small (less than twenty plots). The correlation between forest and LiDAR variables remained strong when field and LiDAR data were collected several years apart, but model parameter values changed rapidly as a result of tree growth. A time lag between LiDAR and field data collection can be tolerated but field measurements, once commenced, should be concentrated in time. The LiDAR predictor variable for volume was found to be insensitive to changes in the reference plot area indicating that volume prediction models may be applied in partitions (spatial units) with areas different than those of the calibration field plots. Comparison of alternative configurations of size, shape and arrangement of partitions, coupled with one of four spatial interpolation techniques, demonstrated significant differences in precision of predicted volume surfaces, with key factors being the dimensions of the interpolation neighbourhood and LiDAR data density. Several configurations closely approximated or, in the case of low density data, exceeded the precision of the prediction models. A method for assessment of partially thinned plantations appeared effective but requires further validation. The results of this study will be used to guide the second operational LiDAR-based site quality survey in South Australia, scheduled for early 2012