University of Twente Student Theses

Login

Integrating Single and Multi-Band Radar Data for Aboveground Biomass Assessment in Indonesia's Complex Tropical Landscape

Chawla, Anchal (2024) Integrating Single and Multi-Band Radar Data for Aboveground Biomass Assessment in Indonesia's Complex Tropical Landscape.

This is the latest version of this item.

[img] PDF
3MB
Abstract:Forests are huge reserves of carbon. A reliable proxy for estimating the carbon stored in a forest is the Above Ground Biomass (AGB) of the vegetation. Accurate reporting of the change in forest carbon is an essential requirement under various international agreements and also for individual targets of countries. Radar remote sensing has the ability to penetrate the forest canopy and hence has been widely used for estimating AGB. Different radar bands, each with distinct wavelengths, are sensitive to various components of vegetation. Longer wavelengths (like L- band) are more sensitive to the trunk while shorter wavelengths (like X- band) are more sensitive to the leaves and branches. The landscape in Indonesia is rapidly changing due to several reasons. Hence, the long-standing primary forests are converted into other land use categories like plantations. Therefore, It becomes essential to estimate the carbon lost during this process. This study aimed at examining and comparing individual radar X-, C- and L- bands and an integration of these bands for predicting the AGB of a complex tropical landscape in Indonesia with heterogeneous vegetation. The models were created using linear regression incorporating all field plots across the four land use categories, as well as specific subsets: forest, oil palm plantation, and a combination of both forest and oil palm plantation plots. The training R2 of the models using field plots of all four land use categories was consistently below 0.1 across all bands. When two land use categories were removed- rubber and shrubland, the R2 increased by many folds, with the highest achieved when using a combination of all three bands together and a combination of L- and C- band. For the oil palm plantations, X- band performed the best (0.44), along with its combination with C- band. For forests, L-band proved to have the highest accuracy (0.20), along with its combination with C- band and using all 3 bands together. Moreover, a comparison of these model coefficients was carried out to understand the difference in the slopes and intercepts, using ANCOVA. Between the three bands, the model slopes did not differ significantly for different land use categories, except between X- and L-band for forests. Between models for forests and for oil palm plantations, the slopes were not significantly different, except when using X- band backscatter as the continuous variable(factor). The intercepts were different when using L- and C- bands as the factor. Further analysis revealed that pooling the land use categories with statistically different intercepts might introduce a persistent bias in the models. To check the performance of the models on unseen data, validation accuracies (R2 and RMSE) were computed using the leave-one out cross validation methods. The highest validation R2 was obtained for oil palm plantation field plots when using backscatter from X- and C- band (0.29). Relative RMSE showed that X- band had the least prediction errors for oil palm plantations, whereas L-band had the least errors for forests. A number of uncertainties in the models were discussed, some of them being the complexity of using different land use categories, non-linear relationships between AGB and backscatter and presence of outliers.
Item Type:Essay (Master)
Faculty:ITC: Faculty of Geo-information Science and Earth Observation
Programme:Geoinformation Science and Earth Observation MSc (75014)
Link to this item:https://purl.utwente.nl/essays/100509
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page