Assessing Hydrological Sensitivity to Vegetation Classifications through Integrated Modeling and Sensitivity Analysis: A Case Study of the St. Charles Bay Watershed, Texas
In the evolving discourse on climate change and its hydrological implications, numerous studies have emphasized the influence of Land Use/Land Cover (LULC) derived from satellite imagery on hydrological modeling. Yet, the specific roles of individual plant communities remain largely overlooked, and the exploration of errors introduced by various vegetation classification methods is scant. This research, set in St. Charles Bay, Texas, aims to address these gaps, illuminating the complex interplay between vegetation classifications and hydrological processes. To achieve a comprehensive comparative analysis, traditional field surveys were set against digital image classifications. Employing both object-based, pixel-based, and machine-learning techniques in Chapter II, the Object-Based Image Analysis (OBIA) method stood out with an accuracy of 77% and a Cohen's kappa of 0.71. Despite this, specific plant communities, such as those with high woody structures and certain grasslands, revealed areas for classification refinement. The supervised method recorded an accuracy of 53.30% with a Cohen's kappa of 0.49. Similarly, the unsupervised method achieved an accuracy of 53.52% and a Cohen's kappa of 0.46. Both methods were challenged in distinguishing similar vegetation types due to the inherent spectral similarities often found in marsh ecosystems. Notably, the U-Net CNN method showcased superior performance, registering an accuracy of 78.48% and a Cohen's kappa of 0.69. However, it was also challenged by certain plant communities due to insufficient training data. Simulating St. Charles Bay's hydrological processes was the focal point of Chapter III. Here, the EDYS model, through simulations spanning diverse hydrologic cycles, was used to explore relationships between vegetation, climate, and soils. Weighted least squares multiple regression revealed that certain plant communities, such as Seacoast bluestem-Silver bluestem grassland and select Live Oak variations, surfaced as important drivers of hydrological dynamics such as runoff and interception. Evaporation patterns were predominantly determined by moisture availability, yet Live Oak tree communities stood out with consistently reduced evaporation, likely attributed to their inherent shading capabilities. Transpiration patterns tended to be influenced by elevational gradients and species-specific water use efficiencies. Export dynamics were heavily influenced by soil texture, largely based on the water-holding capacities of Aransas clay compared to several sandy soils. Vadose zone moisture content appeared to be more closely tied to proximity to bay water, and subsequent inundation, with communities closer to the bay experiencing elevated moisture levels compared to upland communities. Chapter IV focused on the sensitivity of hydrological dynamics to differences in vegetation classifications. Employing sensitivity analysis and root mean square error (RMSE) metrics, the chapter evaluated responses of six hydrological variables to difference classification methods, across a range of rainfall conditions. The results revealed pronounced disparities across classification techniques and rainfall scenarios. Notably, the Unsupervised classification often eclipsed the Supervised approach, especially in variables like runoff and vadose zone storage. Additionally, RMSE analysis substantiated the sensitivity analysis findings, with the OBIA method performing well in runoff predictions but less so in other hydrological variables like interception, evaporation, and transpiration. Conversely, the Unsupervised method consistently provided the most accurate predictions for evaporation, transpiration, and interception across all periods, often surpassing the Supervised method. For reliable hydrological modeling, selection of vegetation classification method is important based on the insights gained from sensitivity and RMSE analyses. No single classification method emerged as superior across all hydrological variables and rainfall cycles, indicating a nuanced interplay between classification methods, hydrological variables, and climatic conditions. As a result, this study provides a foundation for evaluating water resource management in coastal regions, like St. Charles Bay, to guide adaptive water management strategies for a sustainable future as we attempt to navigate an ever-evolving climate.
Embargo status: Restricted until 01/2027. To request the author grant access, click on the PDF link to the left.