This study utilized the Soil and Water Assessment Tool+ (SWAT+) to evaluate the impact of changes in agricultural land cover on streamflow and sediment output. The watershed has seen substantial changes in land use, most notably the conversion of open forest into agricultural land. In 2010, 2015, and 2020, land cover data indicated a 59.05% decrease in open forest areas over the decade, while annual and perennial agricultural areas experienced significant increases. Daily streamflow data from 2007 to 2018 were used to calibrate and validate SWAT+, and the model performed well (R² ranged from 0.75 to 0.83; NSE ranged from 0.76 to 0.78; PBIAS ranged from 24.63 to 18.89; RSR ranged from 0.64 to 0.51). Simulations of sediment yield showed an increasing tendency, increasing by 63% from 20.564 t/ha/yr in 2010 to 33.524 t/ha/yr in 2020. A study of Landscape Units (LSUs) revealed localized increases in sediment yield. For example, LSU 2210 saw a rise from 12.2 t/ha/year to 17.7 t/ha/year, and LSU 2160 saw an increase from 5.28 to 8.96 t/ha/year within the same period. Significantly, in several LSUs, scenarios that simulated the conversion of forests to perennial crops yielded larger sediment outputs than those that simulated the conversion to annual crops. This was because the land location in these scenarios was on steeper slopes. As the Department of Agriculture–Western Visayas indicated that Aklan was the leading producer of cassava, eggplant, and irrigated palay, with corresponding growth rates of 15.71%, 13.21%, and 7.13%, the selection of BMPs was in line with regional agricultural priorities. The study demonstrated the effectiveness of Best Management Practices (BMPs) in reducing sediment
yield. In particular, mulch tillage of eggplant (BMP 2) decreased sediment yield by 33.106 t/ha/year, whereas conservative tillage of cassava (BMP 1) decreased by 32.172 t/ha/year. Dense grass waterways (BMP 4) accomplished the most notable reduction, lowering sediment yield down by 29.294 t/ha/yr, while medium grass waterways (BMP 3) also decreased sediment yield by 30.156 t/ha/yr. These BMPs can be incorporated into regional agricultural extension initiatives and landuse planning. This study confirms the use of SWAT+ as a decision-support tool for sustainable watershed planning and demonstrates that land cover change has a considerable impact on sediment output.

Kolmogorov-Arnold Networks (KANs) replace fixed activation functions with learnable spline-based functions on network edges, enabling flexible, interpretable function approximation inspired by the Kolmogorov-Arnold representation theorem. This work explores KANs for class-incremental learning (CIL), where models learn new classes sequentially without access to prior data. We propose a dynamic KAN architecture that grows and prunes both network structure and spline grids to optimize generalization across incremental tasks. To en hance stability, we introduce magnitude-based regularization that limits large changes in spline coefficients, and apply a gradual early-stopping strategy to reduce overfit ting. Experiments show our dynamic KAN performs comparably to exemplar-based state-of-the-art methods, and outperforms others in exemplar-free settings. An abla tion study confirms the effectiveness of the grow-and-prune method, which supports knowledge retention via subnetwork freezing. While magnitude-based regularization can further improve results, it is sensitive to hyperparameter settings. Results also confirm a strong correlation between exemplar quantity and model accuracy, underscoring the challenge of catastrophic forgetting with limited exem plars. Overall, KANs demonstrate strong potential as a flexible, adaptive framework for continual learning.

Dissolved oxygen (DO) is a critical indicator of water quality, as it directly affects the health and survival of aquatic
organisms, including those raised in aquaculture systems. This study investigates the feasibility of using supervised machine
learning-based models for predicting DO levels in Sampaloc Lake, San Pablo, Laguna, utilizing Filter and Wrapper methods of
feature selection. The analysis is based on a dataset comprising 104 historical records collected from 2010 to 2022, incorporating a total of 21 features. These features encompass a range of water quality, physical, and hydroclimatic parameters, which are integrated as predictors to develop reliable predictive models. Both Filter and Wrapper feature selection methods proved effective in identifying influential parameters and predicting DO levels. Among the models, XGBoost Regression (XGBR) emerged as the top performer for the Filter Method with R2 = 0.95, RMSE=0.14, and MAE = 0.10. For the Wrapper Method, Random Forest (RF) demonstrates better generalization, with higher test R2 (0.91) and lower RMSE (0.39) and MAE (0.22) values, making it the most suitable representative for the Wrapper Method. Comparing the two feature selection approaches, the Filter Method was found to be superior overall. The SHAP value analyses for both the Filter and Wrapper feature selection methods consistently identified biochemical oxygen demand (BOD), air temperature (AT), and surface pressure (SP) as the most influential predictors of DO concentration in Sampaloc Lake. Overall, this study concludes that the Filter Method with XGBR offers the most reliable and effective framework for predicting DO levels in Sampaloc Lake, contributing valuable insights for water quality management and sustainable aquaculture practices.

Hyperdimensional computing (HDC), drawing inspiration from the brain, uses extremely high-dimensional vectors to solve classification problems in a more robust and energy-efficient way than traditional machine learning methods. Unlike conventional algorithms, which often rely on complex models and computationally-intensive training, HDC offers a scalable and lightweight alternative by performing simple bitwise operations on these high-dimensional vectors, or hypervectors. State-of-the-art HDC architectures rely on binary dense hypervectors with approximately equal number of 1 and 0 elements. While these provide good classification performance, they may suffer from high energy consumption due to high switching activity within the architecture, especially with hypervectors whose dimensionality is in the thousands.

In this thesis, we explore the use of binary sparse representations, in which most of the elements are 0s, for a highly energy-efficient HDC architecture. We evaluate our approach on three classification tasks, namely language recognition, character recognition, and hand gesture recognition, targeting a 65nm CMOS process. Our results show that sparse representations achieve a 1.34-5.70x reduction in energy per inference compared to dense representations, while maintaining comparable classification performance across all three tasks, with little to no area overhead. This improvement is observed at a hypervector dimensionality of 10,000, and remains consistent for a wide range of dimensionalities, from 256 to 8,192. Furthermore, increasing the sparsity in the architecture, where as few as 1%-2\% of the elements in a hypervector are 1s, improves energy efficiency without degrading classification accuracy. Despite this higher sparsity, the proposed HDC architecture remains robust against memory errors, comparable to conventional architectures based on dense representations. These findings highlight the potential of sparse representations in enabling more energy-efficient HDC architectures, presenting a promising solution for the wider adoption of cognitive applications in energy-constrained environments, such as the Internet of Things.

Local red macroalgae in the Philippines, Betaphycus gelatinus and Eucheuma denticulatum, provide an alternative end-use towards the valorization of biomass as a renewable energy source. A previously explored method, microwave-assisted
hydrothermal carbonization (HTC), has been applied to foreign seaweed species to recover hydrochar as a solid fuel alongside process liquor containing levulinic acid (LA), a high-value platform chemical. In this study, the HTC method was applied to the aforementioned species. A design of experiments (DOE) was carried out to assess the effect of sulfuric acid (H2SO4) concentration, reaction temperature, and reaction time, and their parameter interactions on the yield of hydrochar and LA, as well as the hydrochar heating value (HV). Ultimate, proximate, FTIR, and HV analyses were performed on the hydrochar products. UV-Vis spectroscopy was performed on the process liquor to obtain the LA yield. Constructing DOE interaction plots revealed that the individual effects of H2SO4 concentration and time increase the LA yield but decrease the hydrochar yield, with an increasing magnitude. For both species, hydrochar samples produced at 200 °C, for 30 minutes, in the absence of H2SO4 had the highest carbon content (> 50 wt%) and least moisture (< 6.5 wt%), thus having the highest HV comparable to bituminous coal or peat, with the highest HV (10,057 BTU/lb) for B. gelatinus overall, but the highest energy densification (ED) of 3.68 for E. denticulatum. It was found through the combined severity factor (CSF) that yields had weak dependence on process severity, with no significant linear relationship to ED. Some recommendations to improve the study include increasing the number of experimental conditions, broadening the temperature range, and carrying out separation on the process liquor, with more studies made on the Philippine context to assess the feasibility of hydrochar and LA production through this method.