The study outlines a methodology to extrapolate LiDAR-measured tree height to areas sampled using only multispectral IKONOS imagery in support of forest management activities. This cost reducing sampling scheme was developed using geostatistical procedures and expert knowledge in order to address the high cost of LiDAR surveys. The scheme sampled 2.25 ha of all LiDAR canopy returns in an area covering approximately 1,000 ha, which equates to 0.25 % of the study area. The premise of the study was that the transect type sampling scheme would reduce the cost of the LiDAR survey and thus make incorporation of the sensor more cost effective within an operational scenario. However, statistical results did not fully support the extrapolation approach proposed here, i. e., using spectral reflectance and age as predictor variables. While the LiDAR accurately measures tree heights, the RMSE values associated with validation data were outside the acceptable error margins employed in operational scenarios (±5 %). Over-prediction of tree heights suggested that the proposed methodology was not perfectly attuned to accurate estimation of tree height for areas external to the LiDAR collection framework. However, given

(i) the fact that goodness-of-flt statistics were promising,

(ii) the documented problems associated with field enumeration techniques,

(iii) the exacerbating factors related the spherical shape of Eucalyptus crowns, it is proposed that future research address the sampling design itself, while also investigating the use of off-nadir (multi-angular) imagery. Such multi-angular imagery has been shown to be more amenable to extraction of vegetation struc­ture than traditional nadir-only imagery. The study concluded that, even though the precision metrics reported here were unsatisfactory for truly operational purposes, specific metrics related to the correlation between LiDAR-derived tree height and extrapolated tree height (multi-spectral imagery) were encour­aging. We believe that studies such as this are necessary if we are to eventually provide large scale, operational structural assessments of our forest resources using currently expensive, but accurate LiDAR surveying approaches.

2.6 Conclusions

We believe that future research on biomass inventory should focus on multi­phase inventory with sensor fusion. Research is progressing on methodologies that combine LiDAR and high resolution UAV photography, as well as many other sensor combinations. The synergy of combining sensors like multispectral, LiDAR, and RADAR likely will hold the key to containing cost and errors, as well as covering much larger areas of standing biomass. Treuhaft et al. (2004), for instance, provide a useful summary of RADAR use for forest biomass estimation, as well as fusion of RADAR and optical remote sensing sensor data.

Multi-phase inventories combine the different strengths of different methods and sensors in localisation, as well as quantifying not only the biomass, but also the components of the biomass in terms of leaves, branch wood, and stems above ground. No direct measure of belowground biomass (BGB) is currently possible. Estimates of BGB are possible once proper models exist for an ecotype or species, most typically through an established AGB-BGB conversion factor.

Finally, a table is provided to summarize current forest biomass (inventory) technologies in terms of the level of training required to implement, the cost of execution, and the scale for which it is suitable (Table 2.1). Both the lower limit as well as upper limit of the method is indicated, where small (S) indicate community and stand level, medium (M) indicate farm to catchment level, and large (L) indicates regional to global scale inventory. An almost invariable truth holds: The larger the area required to assess with terrestrial sampling, the higher the cost, while with remote sensing methods, the inverse is true in that the unit cost of assessing a small area is higher than assessment of catchment and larger areas.

Remote sensing is not a golden solution or panacea to all of our forest inventory challenges, whether those be inventory related or focused on the condition (nutrient, moisture) of the resource. But the technologies described above do offer viable solutions to operational assessment needs in the forestry and ecological communities. Some of these technologies are closer to operational implementation

than other, e. g., airborne LiDAR vs. either space borne or ground-based LiDAR, but in all cases certain commonalities should be observed:

1. Field work will always form an essential component of any inventory, whether for calibration of biomass models developed elsewhere, or for validation of existing models for different sites or seasons.

2. Implementation costs will decrease as new markets develop, vendors’ numbers increase, and practitioners adopt novel approaches. However, we should always strive to balance these costs with associated accuracy and precision trade-offs.

3. Finally, fusion of multiple approaches or modalities likely will be key to a successful and comprehensive inventory system — some sensors are best suited to structural assessment (inventory) and others to spectral assessment (species, nutrients, moisture status), while ground-based efforts will always occupy a necessary component in the inventory chain.

There may be other such essential truths, but we trust that this chapter has provided the reader with an overview of what is possible using remote sensing. Research in this field is ongoing, and there is ample evidence that the future of technologically advanced approaches to forest resource assessment is bright.