Assessing the transferability of statistical predictive models for leaf area index between two airborne discrete return LiDAR sensor designs within multiple intensely managed Loblolly pine forest locations in the south-eastern USA

• Published in: Remote sensing of environment Vol. 176; pp. 308 - 319
• Main Authors: Sumnall, Matthew, Peduzzi, Alicia, Fox, Thomas R., Wynne, Randolph H., Thomas, Valerie A., Cook, Bruce
• Format: Journal Article
• Language: English
• Published: Goddard Space Flight Center Elsevier Inc 01.04.2016; Elsevier
• Subjects: Correlation; Earth Resources And Remote Sensing; Forest management; Leaf area index; LiDAR; Loblolly pine; Mathematical models; Optics; Pine; Regression; Remote sensing; Sensors; ANW, USA, North Carolina; ANW, USA, Virginia; Forest management; Loblolly pine; Remote sensing; LiDAR; Leaf area index
• ISSN: 0034-4257; 1879-0704
• DOI: 10.1016/j.rse.2016.02.012

Leaf area is an important forest structural variable which serves as the primary means of mass and energy exchange within vegetated ecosystems. The objective of the current study was to determine if leaf area index (LAI) could be estimated accurately and consistently in five intensively managed pine plantation forests using two multiple-return airborne LiDAR datasets. Field measurements of LAI were made using the LiCOR LAI2000 and LAI2200 instruments within 116 plots were established of varying size and within a variety of stand conditions (i.e. stand age, nutrient regime and stem density) in North Carolina and Virginia in 2008 and 2013. A number of common LiDAR return height and intensity distribution metrics were calculated (e.g. average return height), in addition to ten indices, with two additional variants, utilized in the surrounding literature which have been used to estimate LAI and fractional cover, were calculated from return heights and intensity, for each plot extent. Each of the indices was assessed for correlation with each other, and was used as independent variables in linear regression analysis with field LAI as the dependent variable. All LiDAR derived metrics were also entered into a forward stepwise linear regression. The results from each of the indices varied from an R2 of 0.33 (S.E. 0.87) to 0.89 (S.E. 0.36). Those indices calculated using ratios of all returns produced the strongest correlations, such as the Above and Below Ratio Index (ABRI) and Laser Penetration Index 1 (LPI1). The regression model produced from a combination of three metrics did not improve correlations greatly (R2 0.90; S.E. 0.35). The results indicate that LAI can be predicted over a range of intensively managed pine plantation forest environments accurately when using different LiDAR sensor designs. Those indices which incorporated counts of specific return numbers (e.g. first returns) or return intensity correlated poorly with field measurements. There were disparities between the number of different types of returns and intensity values when comparing the results from two LiDAR sensors, indicating that predictive models developed using such metrics are not transferable between datasets with different acquisition parameters. Each of the indices were significantly correlated with one another, with one exception (LAI proxy), in particular those indices calculated from all returns, which indicates similarities in information content for those indices. It can then be argued that LiDAR indices have reached a similar stage in development to those calculated from optical-spectral sensors, but which offer a number of advantages, such as the reduction or removal of saturation issues in areas of high biomass. •12 LiDAR LAI prediction indices tested–computed from return no. and intensity•Tests were performed on two different LiDAR sensors and acquisitions.•Indices computed using all returns correlated highly with field estimates of LAI•Indices computed using specific return types (e.g. first) and intensity were poor.•All-return indices appear to be transferable between acquisitions and locations.