Resumen de Sar sensitivity to burn severity and forest recovery in mediterranean environments - backscatter and interferometric coherence analysis
Worldwide, forests fires destroy annually millions of hectares from tropical to boreal region. In the European part of the Mediterranean basin alone about five hundred thousand hectares are affected each year making fires one of the most important ecological threats. Remotely sensed data can contribute to a better, cost effective, objective and time saving method to monitor and quantify location, extent and intensity of fire events. Despite the extensive archives of space borne synthetic aperture radar (SAR) data few studies were carried out having as topic forest fires. This thesis explores the use of synthetic aperture radar for post-fire studies highlighting not only its advantages but also the potential sources of error.
Regional models for burn severity assessment were developed using field samples (composite burn index - CBI) and optical based spectral indices (difference Normalized Burn Ratio - dNBR) to assess the relationship strength between these variables across five recently burned sites in northeastern Spain. Linear and non-linear models were tested and the estimation error of each model was estimated. SAR data at X-, C- and L-band were investigated to determine the relationship between radar metrics (backscatter and interferometric coherence) and forest burn severity over three burn scars. The dependency of the SAR backscatter upon local incidence angle and environmental conditions has been also analyzed. In addition, SAR data were used to analyze backscatter and interferometric coherence from regrowing forests previously affected by fire. L-band data analysis was extended to burned boreal forests.Statistical analysis was used to assess the radar metrics as a function of burn severity level or forest regrowth phase after stratifying the data by local incidence angle. Determination coefficients were used to quantify the relationship between radar data and burn severity estimates.
Fire scar differentiation using optical based indices is attainable in Aragón pine forests. Burn severity estimation errors for highly burned sites were well below 10 % whereas for low and moderate severities errors increased up to 25%. A strong linear relation was found between burn severity at plot level and understory and overstory composites. The analysisdemonstrated the model consistency at regional level and the need for new estimation methods in areas affected by low to moderate burn severities even for homogeneous forests.
SAR metrics (backscatter and coherence) were sensitive to burn severity. For co-polarized channels (HH and VV)the backscatter increased with burn severity for X- and C-band whereas for L-band it decreased. Cross-polarized (HV) backscatter decreased with burn severity for all frequencies. The sensitivity of the co-polarized backscatter to burn severity decreased for increasing local incidence angle for all frequencies except at L-band. For X- and C-band co-polarized data higher determination coefficients were observed for slopes oriented towards the sensors whereas for cross-polarized data the determination coefficients were higher for slopes oriented away from the sensor. At L-band the association strength of cross-polarized data to burn severity was high for all local incidence angles.
Co-polarized coherence increased with the increase of burn severity at X- and C-band whereas cross-polarized coherence was practically insensitive to burn severity. Higher sensitivity to burn severity was found at L-band for both co- and cross-polarized channels. The association strength between coherence and burn severity was strongest for images acquired under stable, dry environmental conditions. When the local incidence angle is accounted for, the determination coefficients increased from 0.6 to 0.9 for X- and C-band. At L-band the local incidence angle had less influence on the association strength to burn severity. L-band (both metrics) showed the highest potential for burn severity estimation in the Mediterranean environment. The small dynamic range observed for X-band data could hinder its use in forests affected by fires. At L-band, the results from boreal forest confirmed the data trends obtained in the Mediterranean basin at least for specific environmental conditions (images acquired during the growing season in relatively dry conditions).
Low sensitivity to forest regrowth was observed for X-band backscatter, the average backscatter increasing by 1-2 dB between the most recent fire scars and the unburned forest. C-band showed increased sensitivity, the backscatter difference between burned and unburned forest being around 4 dB. L-band backscatter presented the highest sensitivity to forest regrowth, the backscatter difference being approximately 8 dB. For a given frequency the sensitivity of the SAR backscatter to forest regrowth varied with local incidence angle and polarization. The interferometric coherence showed low sensitivity to forest regrowth at all SAR frequencies. For mediterranean forests five phases of forest regrowth were discerned whereas for boreal forest, up to four different regrowth phases could be discerned with L-band SAR data. In comparison, the Normalized Difference Vegetation Index (NDVI) provided reliable differentiation only for the most recent development stages, as it registers mostly information related to the vegetation cover. The results obtained were consistent in both environments.
These findings indicate that SAR systems have potential for burn severity estimation using either the backscatter coefficient or the interferometric coherence. The joint use of backscatter and coherence did not significantly improved burn severity estimation in any of the studied frequencies. Promising results were achieved for forest regrowth monitoring when using the backscatter coefficient of low frequency bands (L-). The higher frequency bands (X- and C-) showed low sensitivity to changes in forest structure, having little potential for differentiating between forest regrowth stages.
