• Guidelines for Effective Management of Diamondback Moth in Brassica Crops

  Calvin, Wilfrid; Keith, Macey; McGrew, Brian; University of Arizona; Arizona Department of Agriculture (College of Agriculture, Life & Environmental Sciences, University of Arizona (Tucson, AZ), 2025-06)

  The diamondback moth (DBM) is a significant pest in Arizona brassica crops, presenting ongoing management challenges. This document offers practical, Arizona-specific guidelines for controlling DBM in both transplant production and field settings. It emphasizes careful transplant inspection, timely scouting, sanitation, insecticide rotation, and the integration of biological control strategies. These recommendations aim to help growers and pest control advisors achieve effective, sustainable DBM management while reducing the risk of resistance.
• 

  Shades of Identity: Examining the Interplays of Racism and Colorism on Ethnic-Racial Identity, Skin Tone Satisfaction, and Skin Tone Centrality Among Latine Youth

  Zeiders, Katharine H.; Osman, Kayla M.; Téllez, Michelle; Perez-Brena, Norma; Wilkinson-Lee, Ada M. (The University of Arizona., 2025)

  Colorism is pervasive globally and holds salience within Latine populations. Adolescence is an important developmental period when Latine youth make meaning of racialized experiences, and examining how racism and colorism shape youth development and lived experiences is needed. First, it is important to understand how racism and colorism co-occur, and research should explore how racism and colorism relate to how youth feel about their ethnic-racial identity (ERI; i.e., negative affect) and come to terms with their ERI (i.e., exploration and resolution), as well as how youth view their skin tone (i.e., skin tone satisfaction). Second, research utilizing advanced longitudinal methods is needed to understand how racialization experiences inform youth development across time. The current dissertation sought to contribute to colorism and ERI literature through these foci across two papers by using advanced quantitative methods and analyses. The first paper explored how negative racialization experiences (e.g., racial microaggressions) relate to the ERI of Latine adolescents, and how skin tone, skin tone satisfaction, and gender moderate this association from an intersectional and cultural ecological lens. The second paper examined how racial discrimination relates to skin tone satisfaction over four days among Latine adolescents, and how skin tone self-concept moderates this association. Understanding the complexity of colorism, racism, and ERI across distinct dimensions and approaches will advance our understanding of Latine youth development and well-being.
• Machine Learning for Efficient & Robust Next-Generation Communication Systems

  Tandon, Ravi; Teku, Noel; Lazos, Loukas; Djordjevic, Ivan (The University of Arizona., 2025)

  Machine learning (ML)-based techniques are increasingly being incorporated into next-generation wireless systems: both for improving fundamental building blocks (e.g., modulation classification, power allocation, channel decoding) as well as enabling new functionalities (e.g., AR/VR, autonomous vehicles). This dissertation makes the following contributions in these areas: As ML classifiers become integral to next-generation wireless systems, it is essential to ensure their predictions are delivered both reliably and with low delay—for instance, in applications like transmitting road condition assessments in vehicular networks or relaying critical health data from sensors to medical providers. In our first contribution, we analyze the fundamental information-theoretic tradeoffs between latency and end-to-end distortion when communicating the results of a classifier over a noisy communication system. We use techniques from finite blocklength channel capacity and show that lattice-based quantization of probability distributions leads to a significant reduction in latency compared to other baselines. In our second contribution, we present a new approach for using reinforcement learning (RL) to provide adaptive robustness to High Frequency (HF) channels. The HF band, which occupies the spectrum of 3 to 30 MHz, enables long-range communications by bouncing signals off the ionosphere with limited communication infrastructure. However, the turbulent nature of the channel, which causes frequent signal dropouts, has deterred the band from being used more heavily. To mitigate this challenge, we propose using RL to learn the optimal settings (e.g., tap length, step size, filter type, adaptive algorithm) of an adaptive equalizer and show that our techniques can provide better performance compared to adaptive equalizers with a fixed structure. In our third contribution, we devise an unsupervised learning-based framework to optimize cell-free networks (CFNs). CFNs deviate from the concept of having an access point (AP) be responsible for serving user equipment (UEs) within a fixed radius and instead deploy APs over a geographic region to collaboratively serve every UE [6]. In doing so, CFNs increase the probability of coverage and achieve stronger diversity gains [7]. To build on these improvements, we propose using an unsupervised neural network to learn how to split a UE’s message across different APs in a manner that minimizes the total latency of the CFN. We show that our unsupervised technique is more effective in ensuring higher probabilities of lower latencies compared to decentralized baselines. Additionally, when noisy channel state information is assumed, our unsupervised technique is more robust in achieving a high likelihood of lower latencies compared to centralized baselines. In our final contribution, we investigate a complementary problem of ensuring privacy when aligning Large Language Models (LLMs). LLMs have been investigated for various applications, due to their broad knowledge base attained via pre-training on large corpora of data. However, it has been shown that LLMs can generate socially unacceptable responses. Alignment procedures have been proposed to train LLMs, using preference data collected from humans, to reinforce which types of responses are socially acceptable. While such methods are effective in regulating an LLM's responses, this type of training could be susceptible to leaking privacy-sensitive information of the human labelers. To mitigate this, we study the problem of LLM alignment with labeler privacy while maintaining the utility of the alignment process. To accomplish this, we present a novel privacy-preserving approach, namely PROPS (PROgressively Private Self-Alignment), a multi-stage algorithm capable of ensuring preference privacy without causing a significant drop in the utility of an LLM as it undergoes alignment.
• Understanding the Removal of Atmospheric Aerosol in a Tropical Marine Environment

  Sorooshian, Armin; Hilario, Miguel Ricardo Alipuddin; Behrangi, Ali; Arellano, Avelino F., Jr.; Sullivan, Sylvia (The University of Arizona., 2025)

  Aerosols and their interactions with clouds remain the largest sources of uncertainty in our understanding of the atmosphere and climate. A major factor in this uncertainty is the wet scavenging (removal) of aerosols in global models, which negatively impacts model capabilities to capture aerosol lifetimes and, consequently, aerosol impacts on climate and air quality. This dissertation focuses on scavenging over the tropical West Pacific region and consists of three distinct approaches: (1) a ground-based study investigating factors contributing to the inter-seasonal persistence of aerosol concentrations in a tropical coastal megacity despite higher precipitation during the wet season, (2) a multi-tool study using aircraft data that determines meteorological variables relevant for scavenging during long-range transport, and (3) an aircraft-based study calculating in-cloud scavenging efficiencies of multiple aerosol species and sizes in tropical convection. In the first part of the dissertation, we analyzed size-resolved aerosol composition, aerosol optical depth, and meteorology to understand why Metro Manila, Philippines exhibits similar aerosol concentrations across seasons despite large differences in seasonal rainfall. We identified two major factors: (1) opposing seasonality of black carbon and water-soluble aerosol, and (2) inefficient scavenging by short rain events (< 1 h). We demonstrated that the presence of rain does not imply efficient wet scavenging and it is important to consider rain characteristics like duration. In a changing climate with increasing urbanization, these factors are expected to become more critical for air quality policymaking and sustainable urban development. This work was published in Environmental Science: Atmospheres (Hilario et al., 2022). In the second part, we identified meteorological variables relevant for estimating wet scavenging using trajectory modeling and a combination of aircraft, satellite, and reanalysis data. We found that the accumulated precipitation along trajectories – often interpreted as a wet scavenging indicator in the literature – does poorly when used to predict aerosol scavenging and was outperformed by the following variables: (1) upper percentiles of relative humidity (RH) along trajectories, (2) the fraction of hours along trajectories exceeding a threshold value for RH or water vapor mixing ratio, and (3) precipitation intensity along trajectories. This work was published in Atmospheric Measurement Techniques (Hilario et al., 2024). The final part of this dissertation quantified in-cloud scavenging efficiencies (SE) in tropical convection. In-cloud scavenging is the primary removal pathway for accumulation mode aerosols, but SEs have not been calculated for shallow to moderate convection. We used aircraft data to calculate SEs for three cases. Efficient scavenging was observed for sulfate (>86%) and black carbon (70 – 80%); moderate scavenging for organic aerosol (53 – 60%) and nitrate (62%); and a wide range of SEs for ammonium (53 – 87%). We also found that accumulation and coarse mode aerosol volume concentrations were nearly totally scavenged in-cloud (>92%), suggesting a preferential activation of large aerosols. Comparisons of differing cloud tops showed that SEs did not vary significantly on an aerosol mass basis with cloud top height. These results demonstrate that aerosol size and composition are more important for in-cloud SEs. This work was published in the Journal of the Atmospheric Sciences (Hilario et al., 2025). This dissertation provides an explanation for how aerosol loadings can be sustained during the wet/rainy season, which should be investigated in other developing cities due to the health risks associated with pollutant accumulation. The dissertation also provides suggestions for meteorological variables that could be considered in model scavenging parameterizations. The presented method can be repeated in different environments to identify regional differences in factors that influence scavenging. Finally, the calculated in-cloud SEs motivate improvements in chemical transport models through future observation-model comparisons.
• Characterizing Ore Particle Size for Sag Mill Feed Control

  Risso, Nathalie; Saavedra, Matias; Momayez, Moe; Tenorio, Victor (The University of Arizona., 2025)

  This thesis presents a comprehensive, data-driven study for predicting the particle size distribution (PSD) at the feed of a semi-autogenous grinding (SAG) mill. A consistent PSD is essential for efficient energy use, stable throughput, and effective downstream processing. However, feed variability, caused by ore heterogeneity, stockpile segregation, and reactive blending, makes prediction challenging. The first part of this work reviews the current state of research and industrial practice, synthesizing insights from over 45 peer-reviewed studies on ore blending, stockpile management, and the application of machine learning in mineral processing. The review identifies a gap between the growing availability of high-frequency sensor data and its limited use in real-time predictive tools, highlighting the need for systems that forecast PSD and support data-informed operational understanding. Building on these insights, the second part develops and validates a machine learning framework using two years of high-resolution operational data from a copper mining operation. The methodology integrates unsupervised clustering with Random Forest regression to forecast key PSD metrics (F10–F90 and TOPSIZE) based on variables such as feeder speeds, stockpile levels, and ore throughput. Cluster-specific models capture nonlinear, regime-dependent behaviors with high accuracy ($R^2 > 0.90$). In addition to prediction, the thesis presents a data-driven sensitivity analysis to evaluate how changes in input variables, such as individual feeder rates, influence PSD outcomes. This analysis shows that coarser PSD metrics, such as F70 and TOPSIZE, are more responsive to input changes than finer ones. A variability analysis further demonstrates the model’s ability to quantify and explain fluctuations in PSD, with predicted distributions exhibiting up to a 15% reduction in standard deviation compared to historical data. All components are integrated into a web-based application that allows users to explore model behavior, visualize PSD forecasts, and assess the impact of operating scenarios in real time. This work advances the field by demonstrating how machine learning and clustering can be effectively applied to model and interpret SAG mill feed variability. The resulting framework offers a practical approach for enhancing predictive insight in mineral processing operations.

View more