Mapping aboveground biomass of trees outside forests in the Taita Hills, Kenya, using airborne laser scanning and individual tree detection

Mapping trees with remote sensing data has been under focus in many studies and many methods have been developed and tested using different kinds of data. Airborne laser scanning (ALS) gives the possibility to investigate vegetation heights and has thus been recently used in many vegetation mapping studies. The objective of this study was to test feasibility of mapping aboveground biomass (AGB) of trees using ALS data and individual tree detection methods. By recognizing individual trees from the ALS data, it is possible to map the tree AGB on a detailed level. The study focused on trees outside forests in the study area, which is situated in the Taita Hills, Kenya. The agriculture and agroforestry in the study area mostly consist of small-holder farming. The purpose was to determine a method that would enable examination of tree biomass on a farm scale. The individual tree detection enables mapping at this scale. The data acquired for this study was high pulse density (9.6 pulses/m²), small footprint, discrete return ALS data. It enables the detection of individual trees based on a high-resolution canopy height model. The processing of the ALS data was done with free software which does not require expensive licensing. For estimating AGB of the individually recognized trees, a prediction model was developed. For developing the prediction model, ground measurements and biomass estimates were needed. The biomass of the ground measured trees was estimated using non-destructive method employing allometric equations. The ground measurements from 77 sample plots of 0.1 ha size were used. A total number of 554 trees were used in the analysis. The tree measurements included diameter at breast height, tree height, crown diameter and species. Mean wood densities were defined on the basis of the species from the online databases and literature. The ground measurements were gathered at the same time as the remote sensing data was acquired. The mean tree AGB for the agricultural and agroforestry areas within the study area were estimated as 23.8 ± 4.2 Mg/ha. The individual tree detection method resulted in a detection rate (correctly identified trees) of 50.1 % and produced 49.9 % omission errors and 36.8 % commission errors. A regression model was developed to estimate AGB for the ALS detected trees. The derived model produces RMSE of 163 kg and bias of 5.75 kg per tree. The correlation between the predicted biomass values and field estimates was 0.659 on the individual tree level and 0.847 when assessed on the plot-level. The individual tree detection method and the derived biomass prediction model was used to map the biomass of individual trees for an area of 9 km² within the study area. The estimated mean biomass for the area was 11.4 Mg/ha. The results show that the use of ALS data and ITD methods can provide good opportunities to map AGB within the Taita Hills agricultural and agroforestry environment on farm scale. With the method presented in this study the individual tree AGB values would be somewhat biased, but on plot-level the estimation accuracy is better. This means that for individual farms that are at least 1 ha in size, the ALS estimated AGB values would be quite good even though somewhat underestimated, and could be used as basis of payment in possible future ecosystem services payment schemes. The AGB estimates accuracies can still be improved by refining the ITD method, for example by analysis of different ALS metrics or by combining ALS data with other remote sensing data.