Temporal, spatial, spectral and polarisation characteristics of the SAR backscatter from regenerating tropical forests
The establishment of an accurate global carbon budget and the consequent ability to understand and predict future environmental change is dependent on knowing the strength of terrestrial sinks and sources of carbon. Regenerating tropical forests are one of the major terrestrial carbon sinks as they are found growing quickly and are sequestering carbon from the atmosphere. Total forest biomass (which includes above and below ground living mass of plants and litter) is a measure of terrestrial vegetation carbon content. It follows that to determine the strength of terrestrial carbon sinks we require information on the location, extent, biomass and biomass change of regenerating tropical forests. Near-constant cloud cover over the tropics and an insensitivity to biomass change at relatively low levels of biomass has limited the use of optical imagery but not Synthetic Aperture Radar (SAR) imagery for the provision of such information. The biophysical properties of regenerating tropical forests are related to the temporal, spatial, spectral and polarisation characteristics of SAR backscatter (a°) and this formed the framework for this thesis. The objectives were to (i) detect biomass accumulation using the temporal characteristics of 0°, (ii) use the spatial characteristics of a° (texture) to increase the strength of the a7biomass relationship and (ill) use the spectral and polarisation characteristics of 0° to classify a surrogate for biomass in regenerating tropical forests (optical Landsat TM data were also included to widen the spectral analysis). Although no biomass change was detectable using temporal 0°, a seasonal pattern in 0° for young regenerating forest was detected, as a result of changing water content in both vegetation and soil. The influence of recent rainfall was confirmed to be an important source of variation in a°, suggesting the use of SAR data from the dry season only. Using simulated data, seven texture measures showed potential for strengthening the a7biomass relationship. However, when applied to real SAR data only GLCM (Grey Level Co-occurrence Matrix) derived contrast strengthened the a7biomass relationship. The addition of GLCM-derived contrast to a° potentially increases the accuracy of biomass estimation and mapping. Neural networks can be used for the classification of land cover in tropical forest regions. Classification accuracy of around 80% was achieved using combined multiwavelength and multipolarisation SAR and Landsat TM bands for 4 land cover classes (pasture, mature forest, 0-5 years old regenerating forests and 6-18 years old regenerating forest). These results demonstrated that multiwavelength and multipolarisation SAR data could provide information on the location, and extent of regenerating tropical forests. However an increase in the accuracy of biomass estimation relies on the optimal use of additional information that resides within the spatial, spectral and polarisation domains of SAR data.