Abstract

The Greater Cape Floristic Region (GCFR) is a hyper-diverse region that encompasses two global biodiversity hotspots, while being threatened by habitat loss and fragmentation, invasive species, altered fire regimes, and climate change. Managing and mitigating these threats requires regularly updated, spatially-explicit information for the entire region, which is currently only feasible using satellite remote sensing. However, detecting the signal change over the spectral and structural variability of ecosystems in the region is challenging, due to the high ecosystem diversity and the confounding influence of structural variables at the leaf, stem, and whole-crown scales. This complexity is exacerbated by the fact that GCFR exhibits high plant diversity.   The primary goal of this research is to enhance our understanding of spectral diversity, and its relation to species diversity, in the hyperdiverse fynbos region of the GCFR using fynbos trait measurements and radiative transfer modeling in a biophysically- and physics-robust simulation environment, such as DIRSIG (Digital Imaging and Remote Sensing Image Generation). We aim to investigate spectral traits at various spatial scales - ranging from canopy to plot to landscape - and their relationship to biodiversity indicators. By examining these traits across different scales and burned areas in fynbos, we seek to identify optimal monitoring strategies that can effectively assess post-fire recovery in these regions and improve our understanding of interactions of light within the fynbos biophysical traits. Our study will mechanistically explore the spatial, spectral, and temporal scales at which our ability to monitor post-fire recovery trajectories and biodiversity changes starts to deteriorate. This approach will address practical limitations in current remote sensing methodologies, such as trade-offs among spatial, spectral, and scale resolution, to determine the best parameters for capturing the complexity of the fynbos ecosystem. Ultimately, this research aims to improve biodiversity conservation strategies in the GCFR by enhancing our capacity to monitor post-fire recovery and biodiversity changes, supporting sustainable management of this unique region. The pursuit of these goals will be guided by the following objectives: 1)    Assess the limit of species identification in a well-established complex forest ecosystem, such as Harvard Forest, using DIRSIG simulation as a proof-of-concept; systematically varying spatial, spectral, and scale resolution and classifying species with 1D-CNN, HybridSN, and SVM models; classification accuracy ranged from 83% at 1 m spatial resolution down to 50% at 30 m, with the best configuration (1 m spatial, 3 nm spectral resolution) reaching approximately 84% overall accuracy;   2)    Evaluate approaches based on UAS (unmanned aerial system) high spatial resolution multispectral imagery, combined with sequential forward feature selection and supervised classifiers (SVM, Logistic Regression, Decision Tree), to differentiate vegetation of different post-fire ages in the fynbos biome and predict the alpha diversity within the burned areas; the best-performing classifiers reached 97% accuracy, and alpha diversity was found to decline with time since fire at the plot scale;  3)    Develop virtual (simulated) DIRSIG scenes of the biodiverse region of the fynbos biome for post-fire plots of different ages, by integrating field-measured fynbos trait measurements and radiative transfer modeling in a biophysically- and physics-robust simulation environment; the resulting scenes, quantitatively validated against real UAS multispectral, airborne hyperspectral (AVIRIS-NG), and LiDAR data, showed strong agreement in spectral shape and magnitude (Pearson r = 0.98, RMSE = 0.045, NRMSE ≈ 15.4%) and in structural characteristics (canopy height-percentile RMSE = 0.079), improving our understanding of how fynbos species re-occupy the spatial, structural, and spectral "spaces" in the context of post-fire recovery; and 4)    Assess the influence of imaging spectrometer specifications on spectral diversity and biodiversity relationships in the fynbos biome by simulating hyperspectral imagery at five spatial resolutions (2 cm to 1 m) across a richness gradient of fynbos plots and comparing eight spectral diversity metrics, including two novel autoencoder-based measures; autoencoder-based metrics outperformed conventional PCA-based approaches at fine-to-intermediate resolutions (R² up to 0.84 vs. 0.69), the coefficient of variation proved the most resolution-robust metric, and the SWIR-1 spectral region emerged as the strongest single predictor of species richness (R² = 0.90 at fine resolution), together defining optimal imaging spectrometer specifications for effective assessment of diversity in complex ecosystems, like the GCFR.

Publication Date

12-2026

Document Type

Dissertation

Student Type

Graduate

Degree Name

Imaging Science (Ph.D.)

Department, Program, or Center

Chester F. Carlson Center for Imaging Science

College

College of Science

Advisor

Jan van Aardt

Advisor/Committee Member

Carl. Salvaggio

Advisor/Committee Member

Anthony Vodacek

Campus

RIT – Main Campus

Share

COinS