Ranging

Machine learning

Footprint

Leverage (statistics)

Archaeology

Geography

Environmental science

Computer science

Lidar

Remote sensing

Telecommunications

Remote Sensing of Environment

Light detection and ranging (lidar) data acquired from airborne or spaceborne platforms have revolutionized measurement and mapping of forest attributes. Airborne data are often either acquired using multiple overlapped flight lines to provide complete coverage of an area of interest, or using transects to sample a given population. Spaceborne lidar datasets are unique to each sensor and are sample- or profile-based with characteristics driven by acquisition mode and orbital parameters. To leverage the wealth of accurate vegetation structural data from these lidar systems, a number of approaches have been developed to extend these observations over broader areas, from local landscapes to the globe. In this review we examine studies that have utilised modelling approaches to extend air- or space-based lidar data with the aim of communicating methods, outcomes, and accuracies, and offering guidance on linking lidar metrics and lidar-derived forest attributes with broad-area predictors. Modelling approaches are developed for a variety of applications. In some cases, generation of spatially-exhaustive layers may be useful for forest management purposes, driving management and inventory decisions over smaller focus areas or regions. In other cases, outputs are designed for monitoring at regional or global scales, and may be – due to the spatial grain of the structural estimates – insufficiently accurate or reliable for management. From the reviewed studies, we found height, aboveground biomass and volume, derived from either upper proportions of a large-footprint full-waveform lidar profiles, or statistically modelled from discrete return small-footprint lidar point clouds, to be the most commonly extended forest attributes, followed by canopy cover, basal area and stand complexity. Assessment of the accuracy and bias of the extrapolated forest attributes varied with both independent and model-derived estimates. The coefficient of determination (R2) was the most often reported, followed by absolute and relative (i.e., as a proportion of the mean) root mean square error (RMSE and RMSE% respectively). Compilation of the stated accuracies suggested that the variance explained in predictions of forest height ranged from R2 = 0.38 to 0.90 (mean = 0.64), RMSE from 2 to 6m and RMSE% from 12 to 34%. For volume, R2 ranged from 0.25 to 0.72 (mean = 0.53) and RMSE from 60 to 87 m3/ha and for aboveground biomass (AGB) R2 ranged from 0.35 to 0.78 (mean = 0.55) and RMSE from 28 to 44 Mg/ha. There was no consensus on the level of accuracy required to support successful extension over larger areas. Ultimately, the review suggests that the information need motivating the spatial extension over larger areas drives the choice of the type of lidar data, spatial datasets and related grain. We conclude by discussing future directions and the outlook for new approaches including new lidar-derived response variables, advances in modelling approaches, and assessment of change.

Modelling lidar-derived estimates of forest attributes over space and time: A review of approaches and future trends