Characterization of mediterranean aleppo pine forest using low-density als data

Abstract

Forest ecosystems provide environmental and economic services of great importance to the society. The characterization of these environments has been traditionally accomplished with intense field work. In comparison, the application of remote sensing tools provides a greater overview over large spatial and temporal scales while minimizing costs. Although optical data and Synthetic Aperture Radar (SAR) allow estimating forest stand variables, the development of LiDAR sensors such as Airborne Laser Scanner (ALS) have improved three-dimensional characterization of forest structure. The availability of two ALS public data coverages for the Spanish territory, provided by the National Plan for Aerial Ortophotography (PNOA), opens new research opportunities to generate useful information for forest management. This PhD Thesis used low-density ALS-PNOA data to estimate different forest variables, with support in fieldwork, in the Aleppo pine (Pinus halepensis Miller) forests of Aragón region. The addressed research is relevant mainly for two reasons: first, the examination of suitable methodologies and error sources in forest stand variables prediction at local (small area) and regional scales (large area), and second, the application of ALS data to the characterization of forest areas as a socio-economic reservoir. This PhD Thesis is a compendium of four scientific papers, which sequentially answer the objectives established. Firstly, a comparative analysis of different parametric and non-parametric models was performed to estimate biomass losses and CO2 emissions using low-density ALS and Landsat 8 data in a burnt Aleppo pine forest. Secondly, we assess the suitability of variable selection methods when estimating total biomass in Aleppo pine forest stands using low-density ALS data. In the third manuscript, the quantification and mapping of forest residual biomass in Aleppo pine forest of Aragón region and the assessment of the effect of ALS and environmental variables in model accuracy were accomplished. Finally, the temporal transferability of seven forest stands attributes modelling using multi-temporal ALS-PNOA data in Aleppo pine forest at regional scale was explored. In this case, the temporal transferability was assessed comparing two methodologies; the direct and indirect approach. The first one fits a model for one point in time and estimates the forest variable for another point in time. The indirect approach adjusts two models in different points in time to estimate the forest variables in two different dates. The results derived from this research indicated that Spearman’s rank and All Subset Selection are the most appropriate methods in the ALS metrics selection step commonly applied in modelling. The suitability of the regression methods depends on the sample size and complexity. Thus, multivariate linear regression outperformed non-parametric methods with small samples while support vector machine was the most accurate method with larger samples. Model accuracy increased with higher point density and canopy pulse penetration, while decreasing with wider scan angles. Furthermore, the presence of steep slopes and shrub reduced model performance. In the case of forest stand variables prediction using multi-temporal ALS data, although the indirect approach produced generally a higher precision, the direct approach provided similar results, constituting a suitable alternative to reduce modelling time and fieldwork costs. The fusion of ALS and passive optical data have evidenced the suitability of this information for quantifying wildfire CO2 emissions to atmosphere, constituting a good alternative when multi-temporal ALS data is not available. The estimation of forest inventory variables as well as different biomass fractions, such as total biomass and forest residual biomass, provided valuable information to characterize Mediterranean Aleppo pine forests and improve forest management.