Search

Wildfires are one of the major disasters among many and are responsible for more than 6 million acres burned in the United States alone every year. Accurate, insightful, and timely wildfire detection is needed to help authorities mitigate and prevent further destruction. Uncertainty quantification is always a crucial part of the detection of natural disasters, such as wildfires, and modeling products can be misinterpreted without proper uncertainty quantification. In this study, we propose a supervised deep generative machine-learning model that generates stochastic wildfire detection, allowing fast and comprehensive uncertainty quantification for individual and collective events. In the proposed approach, we also aim to address the patchy and discontinuous Moderate Resolution Imaging Spectroradiometer (MODIS) wildfire product by training the proposed model with MODIS raw and combined bands to detect fire. This approach allows us to generate diverse but plausible segmentations to represent the disagreements regarding the delineation of wildfire boundaries by subject matter experts. The proposed approach generates stochastic segmentation via two model streams in which one learns meaningful stochastic latent distributions, and the other learns the visual features. Two model branches join eventually to become a supervised stochastic image-to-image wildfire detection model. The model is compared to two baseline stochastic machine-learning models: (1) with permanent dropout in training and test phases and (2) with Stochastic ReLU activations. The visual and statistical metrics demonstrate better agreements between the ground truth and the proposed model segmentations. Furthermore, we used multiple scenarios to evaluate the model comprehension, and the proposed Probabilistic U-Net model demonstrates a better understanding of the underlying physical dynamics of wildfires compared to the baselines. [ABSTRACT FROM AUTHOR]

The judges for the 2022 Daniel H. Wagner Prize for Excellence in the Practice of Advanced Analytics and Operations Research selected the four finalist papers featured in this special issue of the INFORMS Journal on Applied Analytics (IJAA). The prestigious Wagner Prize--awarded for achievement in implemented operations research, management science, and advanced analytics--emphasizes the quality and originality of mathematical models along with clarity of written and oral exposition. This year's winning application describes the design and deployment of a generalized synthetic control, a powerful and innovative statistical method for identifying, in a noisy environment, retailing innovations that produce a small percentage improvement in a large volume of sales for Anheuser Busch Inbev. The remaining three papers describe an inverse control approach to allocating lung transplants that best meets targeted outcomes and has been implemented as the national lung allocation policy on March 9, 2023, across the United States; a human-centric, optimized parcel delivery system developed for Deutsche Post that saves money while meeting constraints learned dynamically from driver behavior; and an AI-based system developed for Alibaba that learns supplier behavior to improve replenishment ordering and inventory control. [ABSTRACT FROM AUTHOR]

Blood pressure (BP) is a significant controllable risk factor for cardiovascular disease (CVD), the leading cause of death worldwide. BP comprises two interrelated measurements: systolic and diastolic. CVD risk is minimized at intermediate BP values, a notion known as the J-curve effect. The J-curve effect imposes fundamental trade-offs in simultaneous management of systolic and diastolic BP; however, assessing a comprehensive set of joint systolic/diastolic BP treatment thresholds while explicitly considering the J-curve effect via randomized controlled trials (RCTs) is not feasible because of the time and cost-prohibitive nature of RCTs. In this study, we propose an analytics approach to identify promising joint systolic/diastolic BP threshold levels for antihypertensive treatment. More specifically, using one of the largest longitudinal BP progression data sets, we first build and fit Brownian motion processes to capture simultaneous progression of systolic/diastolic BP at the population level and externally validate our BP progression model on unseen data. We then analytically characterize the hazard ratio, which enables us to compute the optimal treatment decisions. Finally, building upon the optimal joint BP treatment thresholds, we devise a practical and easily implementable approximate policy. We estimate the potential impact of our findings through a simulation study, which indicates that the impact of explicitly considering the J-curve effect and joint systolic/diastolic BP in treatment decisions could be substantial. Specifically, we estimate that between approximately 3,000 and 9,000 premature deaths from cardiovascular disease in the United States could be prevented annually, a finding that could be tested empirically in randomized trials. This paper was accepted by Stefan Scholtes, healthcare management. [ABSTRACT FROM AUTHOR]

The recently published EAT-Lancet Commission report on dietary impacts on the environment suggested that their proposed diet could prevent more than 10 million annual premature mortalities from noncommunicable diseases globally. The report did not meet standards for transparency and replicability, nor did it fully account for statistical uncertainty. Our attempt to replicate the mortality calculations for the United States revealed flaws in the assumptions and methods used to estimate the avoided mortalities. After correcting some calculation errors and fully accounting for uncertainty in the avoided mortalities, the mortality reduction effect of the EAT-Lancet proposed diet in the USA is no greater than the impact of energy consumption changes that would prevent under-weight, over-weight, and obesity alone. As our findings call into question the global conclusions of the EAT-Lancet report, futher independent validation is needed before it can be used to inform dietary guidelines. [ABSTRACT FROM AUTHOR]

In some coastal cities there is an urgent need to decide on adaptation investments given the long periods needed to complete the relevant infrastructures. However such decisions are usually made under significant uncertainty due to local climate change and socio-economic impacts. The Intergovernmental Panel on Climate Change (IPCC) has developed four global scenarios according to different representative concentration pathways (RCPs); however, using these scenarios directly is equivalent to using incomplete information in the decision-making process because they do not provide information on what sea level rise behavior is most probable. In this study, I propose a model which assigns probabilities to IPCC scenarios with Local Sea Level Rise (LSLR) information. I obtain expected damage and risk measures for the world's 120 major coastal mega-cities at specific moments in time. That is, I consider uncertainty in both scenario selection and within each scenario. With this information it is possible to make adaptation investment decisions under uncertainty with a criterion of not exceeding certain risk levels in the future. The paper shows that in the year 2100 for the equal probability mixed scenario (25% RCP 2.6, 50% RCP 4.5 and 25 RCP 8.5) the expected damage is US$ 1,251,732 millions for New Orleans and US$ 1,196,517 millions for Guangzhou. The risk measurements show that in that year the damage in the 5% of worse cases will be US$ 2,800,756 millions for Guangzhou and US$ 1,832,466 millions for New Orleans. However, not all countries have sufficient resources to make the necessary adaptation investments, so we analyze expected damage and risk according to World Bank country income groups. The paper shows that the USA and China will need to make major adaptation investments in the future. The coastal LSLR risk in lower-income cities is also calculated. [ABSTRACT FROM AUTHOR]

Two concurrent social issues in the United States are food insecurity and food waste. The practice of gleaning offers a mechanism for combining these problems to create a synergistic solution. We develop a stochastic optimization model to determine the schedule that maximizes the volume of excess crops rescued from farm fields for the purpose of feeding food-insecure households, thus maximizing social impact. We model gleaning as a service operation where donation calls arrive randomly requesting to be scheduled within a limited time window. The feature that distinguishes gleaning operations from other service settings is that there is uncertainty in both when donations will arrive and the attendance of the gleaners who are volunteers that are not obliged to attend gleaning trips. We apply our model to the gleaning operation of the Food Bank of the Southern Tier in New York State, focusing on five major crops produced in the region. By characterizing how the gleaning operation behaves, our model allows us to optimize the gleaning schedule to maximize the expected total volume gleaned and determine under which conditions different operational strategies can be most useful for improving the performance of the gleaning operation. This in turn enables us to identify conditions under which alternative policy interventions (e.g., farm donation tax credits and government grants to strengthen operational capacity) are more effective for scaling up gleaning programs. [ABSTRACT FROM AUTHOR]

This paper develops an integrated approach to model heterogeneous household water and energy use and their linkages. The approach considers variations in behavioral and technological water and energy use factors that affect U.S. indoor residential water and energy use for toilets, showers, faucets, clothes washers, and dishwashers. The study uses a recent, large, national, disaggregated household water use data set collected from 11 cities, as well as national energy data on water heater efficiency and setpoint/intake temperatures. First, probability distributions of water and energy use factors are identified and correlated. Then, Monte Carlo simulations are used to calculate probability distributions for estimated household water and energy use. Finally, linear regressions are used to find the relative effects of water and energy factors on household energy use. Results show that water and energy distributions among households are skewed, with the largest 12% of the users consuming 21% and 24% of water and energy, respectively. Water heater setpoint temperature followed by intake temperature, heater efficiency, shower hot water percentage, household size, shower flowrate, and faucet flowrate have the highest relative effect on household energy use and should be targeted to reduce household energy use. The approach improves prior homogenous and deterministic water-energy models and can help utilities select and size cost-effective, collaborative water and energy conservation actions. [ABSTRACT FROM AUTHOR]

The National Oceanic and Atmospheric Administration's (NOAA) National Geodetic Survey (NGS) is mandated to map the national shoreline, which is depicted on NOAA nautical charts, serves as an important source in determining territorial limits, and is widely used in various coastal science and management applications. The National Geodetic Survey's primary method of mapping the national shoreline is through stereo compilation from tide-coordinated aerial photography. However, over the past decade, NGS has conducted several phases of research to develop, test, and refine light detection and ranging (LIDAR)-based shoreline mapping procedures. Although important, reliable estimates of uncertainty of these products have, unfortunately, lagged behind in development. We attempt here to outline possible solutions to this lack. Specifically, this study presents and compares two new methods of assessing the uncertainty of NGS' LIDAR-derived shoreline: an empirical (ground-based) approach and a stochastic (Monte Carlo) approach. We observe uncertainties in the horizontal position of the shorelines on the order of 1 to 6 m (95%) depending on location and, especially, beach slope. We show that appropriate adjustment for biases can reduce these to about 1 m (95%) and that the two methods of assessing the uncertainty show good agreement in our test cases. [ABSTRACT FROM AUTHOR]

Abstract: Infection elimination may be an important goal of control programs. Only in stochastic infection models can true infection elimination be observed as a fadeout. The phenomena of fadeout and variable prevalence are important in understanding the transmission dynamics of infectious diseases and these phenomena are essential to evaluate the effectiveness of control measures. To investigate the stochastic dynamics of Mycobacterium avium subsp. paratuberculosis (MAP) infection on US dairy herds with test-based culling intervention, we developed a multi-group stochastic compartmental model (a continuous time Markov chain model) with both horizontal and vertical transmission. The stochastic model predicted fadeout and within-herd prevalence to have a large variance. Although test-based culling intervention generally decreased prevalence over time, it took longer than desired by producers to eliminate the endemic MAP infection from a herd. Uncertainty analysis showed that, using annual culture test and culling of only high shedders or culling of both low and high shedders with a 12-month delay in culling of low shedders, MAP infection persisted in many herds beyond 20 years. While using semi-annual culture test and culling of low and high shedders with a 6-month delay in culling of low shedders, MAP infection in many herds would be extinct within 20 years. Sensitivity analysis of the cumulative density function of fadeout suggested that combining test-based culling intervention and reduction of transmission rates through improved management between susceptible calves and shedding animals may be more effective than either alone in eliminating endemic MAP infection. We also discussed the effects of other factors such as herd size, heifer replacement, and adult cow infection on the probability of fadeout. [Copyright &y& Elsevier]

Pest risk maps can provide useful decision support in invasive species management, but most do not adequately consider the uncertainty associated with predicted risk values. This study explores how increased uncertainty in a risk model's numeric assumptions might affect the resultant risk map. We used a spatial stochastic model, integrating components for entry, establishment, and spread, to estimate the risks of invasion and their variation across a two-dimensional landscape for Sirex noctilio, a nonnative woodwasp recently detected in the United States and Canada. Here, we present a sensitivity analysis of the mapped risk estimates to variation in key model parameters. The tested parameter values were sampled from symmetric uniform distributions defined by a series of nested bounds (±5%, ... , ±40%) around the parameters' initial values. The results suggest that the maximum annual spread distance, which governs long-distance dispersal, was by far the most sensitive parameter. At ±15% or larger variability bound increments for this parameter, there were noteworthy shifts in map risk values, but no other parameter had a major effect, even at wider bounds of variation. The methodology presented here is generic and can be used to assess the impact of uncertainties on the stability of pest risk maps as well as to identify geographic areas for which management decisions can be made confidently, regardless of uncertainty. [ABSTRACT FROM AUTHOR]

Abstract: The objective of this study was to describe improvements to the spectral weather generator and its application to diverse climates within the contiguous USA. The precipitation model, based on a second-order Markov chain coupled with the mixed-exponential distribution was found to closely reproduce the observed precipitation climatology, although interannual variability was underestimated at most locations. This underestimation was found to be due primarily to the precipitation occurrence process, but also to the precipitation intensity process. For the non-precipitation variables, which included maximum and minimum daily temperatures and daily solar radiation, the spectrally generated data were found to exhibit strong agreement with observations in terms of skewness, interannual variability and extreme values. With the use of an iterative method, the model was also able to reproduce the observed correlation and autocorrelation structure. [Copyright &y& Elsevier]

Alley cropping is an agroforestry system in which annual crops are grown in alleys between rows of woody perennials for timber or other products, which can provide ecosystem services and help farmers diversify outputs. But is alley cropping a financially viable alternative to monocropping? To answer this question, a Monte Carlo model of financial risk and returns was used to understand how this diversification of outputs might help farmers in the southeast United States adapt to future scenarios in which agriculture or forestry may become more risky due to a changing climate or other factors. Traditional monocropping had the highest mean returns in the base scenario (based on current risk conditions), but the highest risk. Pine plantations had the lowest returns and lowest risk, and alley cropping was intermediate (mean soil expectation values of $5513 for monocropping, $3955 for alley cropping, and $2693 for pine plantations). The model results showed that traditional monocropping does not stochastically dominate alley cropping in any of the risk scenarios, meaning that alley cropping may have a place for risk-averse farmers. Furthermore, in the scenario in which the downside risk of annual crop production is increased – perhaps due to increased frequency of floods, droughts, etc. – alley cropping mean returns are higher and risk lower than traditional monocropping (mean soil expectation values of $2951 and $2911, respectively, and pine plantations at $2688), meaning any risk averse farmer might prefer alley cropping to monocropping. • Simulated returns from alley cropping, monocropping, and pine plantations under six potential future climate scenarios. • Risk averse land managers may prefer alley cropping. • In scenarios with an increased catastrophe probability, returns from alley cropping met or exceeded returns from monocropping. [ABSTRACT FROM AUTHOR]

Stochastic models of daily precipitation are useful both for characterizing different precipitation climates and for stochastic simulation of these climates in conjunction with agricultural, hydrological, or other response models. A simple stochastic precipitation model is used to downscale--i.e. disaggregate from area-average to individual station--precipitation statistics for 6 groups of 5 U.S. stations, in a way that is consistent with observed relationships between the area-averaged series and their constituent station series. Each group of stations is located within a General Circulation Model grid-box-sizedarea, and collectively they exhibit a broad range of precipitation climates. The downscaling procedure is validated using natural climatevariability in the observed precipitation records as an analog for climate change, by alternately considering collections of the driest and wettest seasons as 'base' and 'future' climates, and comparing the2 sets of downscaled station parameters to those fit directly to therespective withheld observations. The resulting downscaled stochastic model parameters can be readily used for local-scale simulation of climate-change impacts. [ABSTRACT FROM AUTHOR]