A firebreak placement model for optimizing biodiversity protection at landscape scale.

A solution approach is proposed to optimize the selection of landscape cells for inclusion in firebreaks. It involves linking spatially explicit information on a landscape's ecological values, historical ignition patterns and fire spread behavior. A firebreak placement optimization model is formulated that captures the tradeoff between the direct loss of biodiversity due to the elimination of vegetation in areas designated for placement of firebreaks and the protection provided by the firebreaks from losses due to future forest fires. The optimal solution generated by the model reduced expected losses from wildfires on a biodiversity combined index due to wildfires by 30% relative to a landscape without any treatment. It also reduced expected losses by 16% compared to a randomly chosen solution. These results suggest that biodiversity loss resulting from the removal of vegetation in areas where firebreaks are placed can be offset by the reduction in biodiversity loss due to the firebreaks' protective function.

Stochastic forestry harvest planning under soil compaction conditions.

We present a study of annual forestry harvesting planning considering the risk of compaction generated by the transit of heavy forestry machinery. Soil compaction is a problem that occurs when the soil loses its natural resistance to resist the movement of machinery, causing the soil to be compacted in excess. This compaction generates unwanted effects on both the ecosystem and its economic sustainability. Therefore, when the risk of compaction is considerable, harvest operations must be stopped, complicating the annual plan and incurring in excessive costs to alleviate the situation. To incorporate the risk of compaction into the planning process, it is necessary to incorporate the analysis of the soil's hydrological balance, which combines the effect of rainfall and potential evapotranspiration. This requires analyzing the uncertainty of rainfall regimes, for which we propose a stochastic model under different scenarios. This stochastic model yields better results than the current deterministic methods used by lumber companies. Initially, the model is solved analyzing monthly scenarios. Then, we change to a biweekly model that provides a better representation of the dynamics of the system. While this improves the performance of the model, this new formulation increases the number of scenarios of the stochastic model. To address this complexity, we apply the Progressive Hedging method, which decomposes the problem in scenarios, yielding high-quality solutions in reasonable time.

Exploring the multidimensional effects of human activity and land cover on fire occurrence for territorial planning.

The strong link between climate change and increased wildfire risk suggests a paradigm change on how humans must co-exist with fire and the environment. Different studies have demonstrated that human-induced fire ignitions can account for more than 90 % of forest fires, so human co-existence with wildfires requires informed decision making via preventive policies in order to minimize risk and adapt to new conditions. In this paper, we address the multidimensional effects of three groups of drivers (human activity, geographic and topographic, and land cover) that can be managed to assist in territorial planning under fire risk. We found critical factors of strong interactions with the potential to increase the likelihood of starting a fire. Our solution approach included the application of a Machine Learning method called Random Undersampling and Boosting (RUSBoost) to assess risk (fire occurrence probability), which was subsequently accompanied by a sensitivity analysis that revealed interactions of various levels of risk. The prediction performance of the proposed model was assessed using several statistical measures such as the Receiver Operating Characteristic curve (ROC) and the Area Under the Curve (AUC). The results confirmed the high accuracy of our model, with an AUC of 0.967 and an overall accuracy over test data of 93.01 % after applying a Bayesian approach for hyper-parameter optimization. The study area to test our solution approach comprised the entire geographical territory of central Chile.

Large language models for preventing medication direction errors in online pharmacies.

Errors in pharmacy medication directions, such as incorrect instructions for dosage or frequency, can increase patient safety risk substantially by raising the chances of adverse drug events. This study explores how integrating domain knowledge with large language models (LLMs)-capable of sophisticated text interpretation and generation-can reduce these errors. We introduce MEDIC (medication direction copilot), a system that emulates the reasoning of pharmacists by prioritizing precise communication of core clinical components of a prescription, such as dosage and frequency. It fine-tunes a first-generation LLM using 1,000 expert-annotated and augmented directions from Amazon Pharmacy to extract the core components and assembles them into complete directions using pharmacy logic and safety guardrails. We compared MEDIC against two LLM-based benchmarks: one leveraging 1.5 million medication directions and the other using state-of-the-art LLMs. On 1,200 expert-reviewed prescriptions, the two benchmarks respectively recorded 1.51 (confidence interval (CI) 1.03, 2.31) and 4.38 (CI 3.13, 6.64) times more near-miss events-errors caught and corrected before reaching the patient-than MEDIC. Additionally, we tested MEDIC by deploying within the production system of an online pharmacy, and during this experimental period, it reduced near-miss events by 33% (CI 26%, 40%). This study shows that LLMs, with domain expertise and safeguards, improve the accuracy and efficiency of pharmacy operations.