Jesus Rene Villalobos

Accurate price forecasts for fresh produce commodities are essential for farm-level decision making, particularly for small and medium-sized farmers who rely on price expectations to guide planting, harvesting, and marketing strategies. In the United States, publicly available terminal market reports provide price information for a limited number of markets. Furthermore, the set of reporting markets may vary over time which often leaves many local markets without historical records. This creates a challenging forecasting problem in which predictions must be generated for unobserved markets where no historical observations of the target variable exist. To address this challenge, this work proposes a sequential spatial-temporal framework, TT+NN, that enables forecasting at unobserved markets by leveraging information from a limited and varying number of observed markets. In the first phase, a transformer architecture models temporal dependencies and produces forecasts for observed markets. In the second phase, these forecasts are combined with contextual attributes within a neural network to learn spatial relationships and generate forecasts for unobserved market regions without relying on predefined spatial structures. The framework is applied to forecast cucumber prices for more than 100 unobserved US markets. Experimental results show that TT+NN consistently outperforms alternative approaches, achieving a MAPE of 12.23%, MdAPE of 9.50%, and a correlation of 0.7448. The resulting forecasts reveal meaningful seasonal and regional price patterns, demonstrating the potential of transformer-based spatiotemporal frameworks for agricultural decision support in data-scarce environments.

Additive manufacturing (AM) requires process-comparison tools that remain practical when sample generation and testing are costly. We propose a univariate, nonparametric workflow for comparing a candidate AM process to a stable reference process by testing distributional equivalency for a single response variable. The method discretizes the reference distribution into empirical percentile-defined bins and combines this representation with a sequential sampling protocol designed to reduce unnecessary sampling when evidence for equivalency or non-equivalency becomes sufficient. Simulation studies were used to evaluate operating characteristics across experimental settings, and a validation case study based on geometric measurements of laser based powder bed fusion plate scans correctly classified a candidate process expected to be equivalent to the reference while identifying a non-equivalent process at the first sampling step. The workflow is most appropriate for low-sample, high-cost, or throughput-constrained settings, and is best viewed as a tool for process comparability, change control, calibration, and requalification support rather than as a standalone replacement for qualification standards. The full workflow is implemented in the open-source AMEquivalency package to support reproducible analysis.

Additive manufacturing (AM) creates three-dimensional objects using various approaches, typically layer-by-layer. One emerging method is laser-based powder bed fusion of metals (PBF-LB/M), which uses high-energy lasers to melt metallic powder into shape. AM processes are influenced by many factors, yet there is no standardized framework for quantifying their effects on final products. This guide introduces key principles of experimental design and statistics, outlining a roadmap for conducting rigorous experiments. We review the literature on AM generally and PBF-LB/M specifically to assess how well current practices align with standardized methodologies. In addition, we compare the evolution of experimental techniques in PBF-LB/M to those in a more regulated industry to explore potential cross-pollination. Our analysis reveals that most studies do not adhere to best practices in experimental design and statistical analysis. For example, randomization of run order is rarely mentioned, and statistical model assumptions are often unchecked. Even in tightly regulated fields, experimental designs and statistical methods remain basic and lack sophistication. To improve research quality, we provide recommendations for establishing standardized experimental and reporting practices in AM.

The cold chain—the system of refrigerated storage and transport that provides fresh produce or other essentials to be maintained at desired temperatures and environmental conditions—is responsible for substantial energy consumption and greenhouse gas (GHG) emissions, and failures in the cold chain lead to food and energy waste. Here, we introduce the mini container concept as an alternative to conventional reefers, particularly for small growers. Mini containers are relatively small, insulated boxes, with environmental conditions controlled by an electric-powered central driving unit, which can be aggregated as needed and transported by non-refrigerated trucks and trailers. We analyze the energy consumption and GHG emissions for the transport of tomatoes in two cities representing contrasting climates, Phoenix, Arizona, and Chicago, Illinois, for conventional reefers and the proposed mini containers. These two cities provide the opportunity to compare the energy consumption and GHG emissions for the proposed mini containers versus conventional refrigerated transport under extremely different climate conditions. The results show that, as expected in both cases, as the ambient air temperature increases, the energy consumption and GHG emissions also increase. For partial reefer loads less than 72% and 85% for Phoenix and Chicago, respectively, the use of the mini containers reduces energy consumption and GHG emissions because of the reduced volume requiring refrigeration. In general, since the mini containers are fully electrified, their corresponding GHG emissions can be dramatically reduced, and since the fresh produce can be pre-cooled with renewable energy, GHG emissions can even be eliminated.

One of the key elements in horizontal collaboration in supply chains is the determination of an agreement that spells out the terms of the agreement or contract. This is particularly important in the supply chains of some agricultural products such as fresh produce which are subject to high levels of perishability and yield and demand variability. A critical part in the SC articulation process is the issue of transforming the market demand information into a contract between a buyer and the supplier of the product or producer. In this paper a general coordination framework is developed to facilitate the negotiation process resulting in a contract between potential buyers and a coalition of growers, who are represented by a coordination agent called the Supply Chain Articulator (SCA). The contract is based on the concept of real options, which consists of fixed and optional amounts to be delivered by the farmers to the buyer. We use a newsvendor-based approach to model the profit functions for each negotiation agent. By comparing the optimal decisions for both the buyer and the coordination agent, and the overall supply chain, we find the optimal option pricing combination (with respect to the purchase and exercise price) that results in an alignment in each party’s profit function in such a way that their optimal decisions correspond to the one that maximizes the SC’s profit. Furthermore, we find multiple configurations that could result in the alignment of incentives for both parties, and with the inclusion of an additional consideration, such as a minimum expected profit level for each party, we can determine a subset of possible combinations that will be initially acceptable for both parties. The resulting information forms the basis for a final negotiation process to get a contract between the growers and the buyer.