Combining Akaike's Information Criterion and discrete return LiDAR data to estimate structural attributes of savanna woody vegetation

• Models to estimate height variables, crown dimension and biomass of woody vegetation respectively were comparable.
• Maximum height and biomass estimations were satisfactory, despite structural complexity of savanna vegetation.
• Mean height and crown diameter estimations were relatively poor, showing the uncertainty associated with these variables.
• The AIC modelling approach yields competing models that can be compared based on statistical and theoretical explanations.

This study was aimed at estimating selected plot-level structural attributes (mean height, maximum height, mean crown diameter and total aboveground biomass) of woody plants in a savanna environment using discrete return small footprint LiDAR data. A number of metrics including descriptive statistics, height percentiles and densities of LiDAR returns were extracted at the plot level. An information-theoretic approach known as Akaike's Information Criterion (AIC) was used to develop competing regression models relating field-observation and LiDAR metrics for each attribute. Comparison of five best models for each attribute showed decreasing accuracies as the number of predicting LiDAR metrics decreased. The decreases in R2 were 0.65–0.53 (mean height), 0.95–0.93 (maximum height), 0.48–0.42 (crown diameter) and 0.80–0.78 (total aboveground biomass). Analysis of variance (ANOVA) however showed that there was no significant difference among the estimates as well as between estimated and observed values for each attribute. The results show that AIC modelling approach enables the identification and subsequent comparisons of LiDAR-based models to estimate the structural attributes of interest considered in this study. Investigating such an approach should be encouraged for different savanna environments that are characterized by a great deal of structural variability at various spatial scales.