Faculty Publications (BioMed Central & Faculty of Science)
Permanent URI for this collection
Articles from BioMed Central by University of Victoria, Faculty of Science authors. Also other journal articles by UVic Faculty of Science authors.
Click on this link to see Work published with BioMed Central, Chemistry Central and SpringerOpen by researchers at University of Victoria.
Click on this link to see Work published with BioMed Central, Chemistry Central and SpringerOpen by researchers at University of Victoria.
Browse
Recent Submissions
Item Host diversification may split epidemic spread into two successive fronts advancing at different speeds(Bulletin of Mathematical Biology, 2022) Hamelin, Frédéric M.; Mammeri, Youcef; Aigu, Yoann; Strelkov, Stephen E.; Lewis, Mark A.Host diversification methods such as within-field mixtures (or field mosaics, depending on the spatial scale considered) are promising methods for agroecological plant disease control. We explore disease spread in host mixtures (or field mosaics) composed of two host genotypes (susceptible and resistant). The pathogen population is composed of two genotypes (wild-type and resistance-breaking). We show that for intermediate fractions of resistant hosts, the spatial spread of the disease may be split into two successive fronts. The first front is led by the wild-type pathogen and the disease spreads faster, but at a lower prevalence, than in a resistant pure stand (or landscape). The second front is led by the resistance-breaking type, which spreads slower than in a pure resistant stand (or landscape). The wild-type and the resistance-breaking genotype coexist behind the invasion fronts, resulting in the same prevalence as in a resistant pure stand. This study shows that host diversification methods may have a twofold effect on pathogen spread compared to a resistant pure stand (or landscape): on one hand they accelerate disease spread, and on the other hand they slow down the spread of the resistance-breaking genotype. This work contributes to a better understanding of the multiple effects underlying the performance of host diversification methods in agroecology.Item A tale of two surveys: Improving biodiversity monitoring through rapid baseline assessments(University Of Victoria, 2025) Toma, Emily; Melchers, Grace; Dudas, Sarah E.; Hunt, Brian; Hessing-Lewis, Margot; Juanes, Francis; Cox, KieranBiodiversity monitoring is critical for understanding ecosystem condition and guiding conservation efforts. While the scale and scope of biodiversity data collection have expanded through novel techniques and citizen science initiatives, methods for integrating diverse datasets remain poorly developed. This has hindered our ability to leverage the full potential of modern biodiversity monitoring approaches. I address this gap by producing a framework for synthesizing data across multiple techniques, comparing rapid baseline assessments that emphasize expert identification with systematic surveys that prioritize replication over space and time. I use a multi-ecosystem approach, examining 3 distinct marine communities: soft-sediment bivalves, temperate kelp forests, and tropical coral reefs to test the broad applicability of this framework. Using species lists to develop a standardized framework for data integration, I address a fundamental challenge in biodiversity monitoring: how to effectively combine data from diverse sources to create more comprehensive and accurate biodiversity assessments. The results will inform the development of more efficient, ecosystem-specific monitoring.Item Revealing decision-making strategies of Americans in taking COVID-19 vaccination(Bulletin of Mathematical Biology, 2024) Aghaeeyan, Azadeh; Ramazi, Pouria; Lewis, Mark A.Efficient coverage for newly developed vaccines requires knowing which groups of individuals will accept the vaccine immediately and which will take longer to accept or never accept. Of those who may eventually accept the vaccine, there are two main types: success-based learners, basing their decisions on others’ satisfaction, and myopic rationalists, attending to their own immediate perceived benefit. We used COVID-19 vaccination data to fit a mechanistic model capturing the distinct effects of the two types on the vaccination progress. We estimated that 47% of Americans behaved as myopic rationalists with a high variation across the jurisdictions, from 31% in Mississippi to 76% in Vermont. The proportion was correlated with the vaccination coverage, proportion of votes in favor of Democrats in 2020 presidential election, and education score.Item Study design and parameter estimability for spatial and temporal ecological models(Ecology and Evolution, 2016) Peacock, Stephanie J.; Krkošek, Martin; Lewis, Mark A.; Lele, SubhashThe statistical tools available to ecologists are becoming increasingly sophisticated, allowing more complex, mechanistic models to be fit to ecological data. Such models have the potential to provide new insights into the processes underlying ecological patterns, but the inferences made are limited by the information in the data. Statistical nonestimability of model parameters due to insufficient information in the data is a problem too-often ignored by ecologists employing complex models. Here, we show how a new statistical computing method called data cloning can be used to inform study design by assessing the estimability of parameters under different spatial and temporal scales of sampling. A case study of parasite transmission from farmed to wild salmon highlights that assessing the estimability of ecologically relevant parameters should be a key step when designing studies in which fitting complex mechanistic models is the end goal.Item Wild salmon sustain the effectiveness of parasite control on salmon farms: Conservation implications from an evolutionary ecosystem service(Conservation Letters, 2017) Kreitzman, Maayan; Ashander, Jaime; Driscoll, John; Baterman, Andrew W.; Chan, Kai M.A.; Lewis, Mark A.; Krkošek, MartinRapid evolution can increase or maintain the provision of ecosystem services, motivating the conservation of wild species and communities. We detail one such contemporary evosystem service by synthesizing theoretical evidence that rapid evolution can sustain parasiticide efficacy in salmon aquaculture, thus creating an added incentive for salmon conservation. Globally, wild and farmed salmon share native parasites: sea lice. In most major salmon farming areas sea lice have evolved resistance to parasiticides, but in the North Pacific, where farmed salmon coexist with large wild salmon populations, resistance has not emerged. We present a model to show that flow of susceptible genes from lice hosted on wild salmon to those hosted on farmed salmon can delay or preclude resistance. This theoretical and observational data suggests that wild salmon (both oceanic populations that function as a refuge and local migratory populations that connect this refuge to domesticated environments) provide an evosystem service by prolonging parasiticide efficacy. To preserve this service, aquaculture managers could avoid production quantities that exceed wild salmon abundances, and sustain wild salmon populations through regional and oceanic scale conservation. The evosystem service of resistance mitigation is one example of how a contemporary evolutionary process that benefits people can strengthen the case for conservation of intrinsically important wild species.Item Territory surveillance and prey management: Wolves keep track of space and time(Ecology and Evolution, 2017) Schlägel, Ulrike E.; Merrill, Evelyn H.; Lewis, Mark A.Identifying behavioral mechanisms that underlie observed movement patterns is difficult when animals employ sophisticated cognitive-based strategies. Such strategies may arise when timing of return visits is important, for instance to allow for resource renewal or territorial patrolling. We fitted spatially explicit random-walk models to GPS movement data of six wolves (Canis lupus; Linnaeus, 1758) from Alberta, Canada to investigate the importance of the following: (1) territorial surveillance likely related to renewal of scent marks along territorial edges, to reduce intraspecific risk among packs, and (2) delay in return to recently hunted areas, which may be related to anti-predator responses of prey under varying prey densities. The movement models incorporated the spatiotemporal variable “time since last visit,” which acts as a wolf's memory index of its travel history and is integrated into the movement decision along with its position in relation to territory boundaries and information on local prey densities. We used a model selection framework to test hypotheses about the combined importance of these variables in wolf movement strategies. Time-dependent movement for territory surveillance was supported by all wolf movement tracks. Wolves generally avoided territory edges, but this avoidance was reduced as time since last visit increased. Time-dependent prey management was weak except in one wolf. This wolf selected locations with longer time since last visit and lower prey density, which led to a longer delay in revisiting high prey density sites. Our study shows that we can use spatially explicit random walks to identify behavioral strategies that merge environmental information and explicit spatiotemporal information on past movements (i.e., "when" and "where") to make movement decisions. The approach allows us to better understand cognition-based movement in relation to dynamic environments and resources.Item Conservation Reserve Program is a key element for managing white-tailed deer populations at multiple spatial scales(Journal of Environmental Management, 2019) Nagy-Reis, Mariana; Lewis, Mark A.; Jensen, William F.; Boyce, Mark S.Understanding the underlying mechanisms driving population demographics such as species-habitat relationships and the spatial scale in which these relationships occur is essential for developing optimal management strategies. Here we evaluated how landscape characteristics and winter severity measured at three spatial scales (1 km2, 9 km2, and hunting unit) influenced white-tailed deer occurrence and abundance across North Dakota by using 10 years of winter aerial survey data and generalized linear mixed effects models. In general, forest, wetland, and Conservation Reserve Program (CRP) lands were the main drivers of deer occurrence and abundance in most of the spatial scales analyzed. However, the effects of habitat features vary between the home-range scale (9 km2) and the finer spatial scale (1 km2; i.e., within home ranges). While escape cover was the main factor driving white-tailed deer occurrence and abundance at broad spatial scales, at a fine spatial scale deer also selected for food (mainly residual winter cropland). With CRP appearing in nearly all top models, here we had strong evidence that this type of program will be fundamental to sustaining populations of white-tailed deer that can meet recreational demands. In addition, land managers should focus on ways to protect other escape covers (e.g., forest and wetland) on a broad spatial scale while encouraging landowners to supply winter resources at finer spatial scales. We therefore suggest a spatial multi-scale approach that involves partnerships among landowners and government agencies for effectively managing white-tailed deer.Item Factors governing outbreak dynamics in a forest intensively managed for mountain pine beetle(Scientific Reports, 2020) Kunegel-Lion, Mélodie; Lewis, Mark A.Mountain pine beetle (MPB) outbreaks have caused major economic losses and ecological consequences in North American pine forests. Ecological and environmental factors impacting MPB life-history and stands susceptibility can help with the detection of MPB infested trees and thereby, improve control. Temperatures, water stress, host characteristics, and beetle pressure are among those ecological and environmental factors. They play different roles on MPB population dynamics at the various stages of an outbreak and these roles can be affected by intensive management. However, to make detailed connections between ecological and environmental variables and MPB outbreak phases, a deeper quantitative analysis on local scales is needed. Here, we used logistic regressions on a highly-detailed and georeferenced data set to determine the factors driving MPB infestations for the different phases of the current isolated MPB outbreak in Cypress Hills. While we showed that the roles of ecological and environmental factors in a forest intensively controlled for MPB are consistent with the literature for uncontrolled forests, we determined how these factors shifted through onset, peak, and collapse phases of the intensively controlled forest. MPB presence mostly depends on nearby beetle pressure, notably for the outbreak peak. However additional weather and host variables are necessary to achieve high predictive ability for MPB outbreak locations. Our results can help managers make appropriate decisions on where and how to focus their effort, depending on which phase the outbreak is in.Item Dataset of mountain pine beetle outbreak dynamics and direct control in Cypress Hills, SK(Data in Brief, 2020) Kunegel-Lion, Mélodie; McIntosh, Rory L.; Lewis, Mark A.The data presented in this article are related to the research article entitled "Mountain pine beetle outbreak duration and pine mortality depend on direct control effort" [1]. This article provides presence of mountain pine beetle infested trees detected by the Saskatchewan Forest Service on a grid covering the spatial extent of the Saskatchewan portion of the Cypress Hills interprovincial park between 2006 and 2018. For each grid cell, associated ecological and environmental covariates, such as topography, weather and vegetation, are also provided. These data cover the spatio-temporal extent of an almost entire mountain pine beetle outbreak and contribute to the understanding of mountain pine beetle outbreak dynamics.Item Migratory hosts can maintain the high-dose/refuge effect in a structured host-parasite system: The case of sea lice and salmon(Evolutionary Applications, 2020) Bateman, Andrew W.; Peacock, Stephanie J.; Krkošek, Martin; Lewis, Mark A.Migration can reduce parasite burdens in migratory hosts, but it connects populations and can drive disease dynamics in domestic species. Farmed salmon are infested by sea louse parasites, often carried by migratory wild salmonids, resulting in a costly problem for industry and risk to wild populations when farms amplify louse numbers. Chemical treatment can control lice, but resistance has evolved in many salmon-farming regions. Resistance has, however, been slow to evolve in the north-east Pacific Ocean, where large wild-salmon populations harbour large sea louse populations. Using a mathematical model of host-macroparasite dynamics, we explored the roles of domestic, wild oceanic and connective migratory host populations in maintaining treatment susceptibility in associated sea lice. Our results show that a large wild salmon population, unexposed to direct infestation by lice from farms; high levels of on-farm treatment; and a healthy migratory host population are all critical to slowing or stopping the evolution of treatment resistance. Our results reproduce the "high-dose/refuge effect," from the agricultural literature, with the added requirement of a migratory host population to maintain treatment susceptibility. This work highlights the role that migratory hosts may play in shared wildlife/livestock disease, where evolution can occur in ecological time.Item A hybrid gravity and route choice model to assess vector traffic in large-scale road networks(Royal Society Open Science, 2020) Fischer, Samuel M.; Beck, Martina; Herborg, Leif-Matthias; Lewis, Mark A.Human traffic along roads can be a major vector for infectious diseases and invasive species. Though most road traffic is local, a small number of long-distance trips can suffice to move an invasion or disease front forward. Therefore, understanding how many agents travel over long distances and which routes they choose is key to successful management of diseases and invasions. Stochastic gravity models have been used to estimate the distribution of trips between origins and destinations of agents. However, in large-scale systems, it is hard to collect the data required to fit these models, as the number of long-distance travellers is small, and origins and destinations can have multiple access points. Therefore, gravity models often provide only relative measures of the agent flow. Furthermore, gravity models yield no insights into which roads agents use. We resolve these issues by combining a stochastic gravity model with a stochastic route choice model. Our hybrid model can be fitted to survey data collected at roads that are used by many long-distance travellers. This decreases the sampling effort, allows us to obtain absolute predictions of both vector pressure and pathways, and permits rigorous model validation. After introducing our approach in general terms, we demonstrate its benefits by applying it to the potential invasion of zebra and quagga mussels (Dreissena spp.) to the Canadian province British Columbia. The model yields an R2-value of 0.73 for variance-corrected agent counts at survey locations.Item Predicting insect outbreaks using machine learning: a mountain pine beetle case study(Ecology and Evolution, 2021) Ramazi, Pouria; Kunegel-Lion, Mélodie; Greiner, Russell; Lewis, Mark A.Planning forest management relies on predicting insect outbreaks such as mountain pine beetle, particularly in the intermediate-term future, e.g., 5-year. Machine-learning algorithms are potential solutions to this challenging problem due to their many successes across a variety of prediction tasks. However, there are many subtle challenges in applying them: identifying the best learning models and the best subset of available covariates (including time lags) and properly evaluating the models to avoid misleading performance-measures. We systematically address these issues in predicting the chance of a mountain pine beetle outbreak in the Cypress Hills area and seek models with the best performance at predicting future 1-, 3-, 5- and 7-year infestations. We train nine machine-learning models, including two generalized boosted regression trees (GBM) that predict future 1- and 3-year infestations with 92% and 88% AUC, and two novel mixed models that predict future 5- and 7-year infestations with 86% and 84% AUC, respectively. We also consider forming the train and test datasets by splitting the original dataset randomly rather than using the appropriate year-based approach and show that this may obtain models that score high on the test dataset but low in practice, resulting in inaccurate performance evaluations. For example, a k-nearest neighbor model with the actual performance of 68% AUC, scores the misleadingly high 78% on a test dataset obtained from a random split, but the more accurate 66% on a year-based split. We then investigate how the prediction accuracy varies with respect to the provided history length of the covariates and find that neural network and naive Bayes, predict more accurately as history-length increases, particularly for future 1- and 3-year predictions, and roughly the same holds with GBM. Our approach is applicable to other invasive species. The resulting predictors can be used in planning forest and pest management and planning sampling locations in field studies.Item Exploiting the full potential of Bayesian networks in predictive ecology(Methods in Ecology and Evolution, 2021) Ramazi, Pouria; Kunegel-Lion, Mélodie; Greiner, Russell; Lewis, Mark A.1. Although ecological models used to make predictions from underlying covariates have a record of success, they also suffer from limitations. They are typically unable to make predictions when the value of one or more covariates is missing during the testing. Missing values can be estimated but methods are often unreliable and can result in poor accuracy. Similarly, missing values during the training can hinder parameter estimation of many ecological models. Bayesian networks can handle these and other limiting issues, such as having highly correlated covariates. However, they are rarely used to their full potential. 2. Indeed, Bayesian networks are commonly used to evaluate the knowledge of experts by constructing the network manually and often (incorrectly) interpreting the resulting network causally. We provide an approach to learn a Bayesian network fully from observed data, without relying on experts and show how to appropriately interpret the resulting network, both to identify how the variables (covariates and target) are interrelated and to answer probabilistic queries. 3. We apply this method to the case study of a mountain pine beetle infestation and find that the trained Bayesian network has a predictive accuracy of 0.88 AUC. We classify the covariates as primary and secondary in terms of contributing to the prediction and show that the predictive accuracy does not deteriorate when the secondary covariates are missing and degrades to only 0.76 when one of the primary covariates is missing. 4. As a complement to the previous work on constructing Bayesian networks by hand, we show that if instead, both the structure and parameters are learned only from data, we can achieve more accurate predictions as well as generate new insights about the underlying processes.Item Aligning population models with data: Adaptive management for big game harvests(Global Ecology and Conservation, 2021) Nagy-Reis, Mariana; Reimer, Jody R.; Lewis, Mark A.; Jensen, William F.; Boyce, Mark S.Models of population dynamics are a central piece for harvest management, allowing managers to evaluate alternative strategies and to identify uncertainty. Here we present a density-dependent population dynamics model that can be used in conjunction with adaptive management to optimize big game management, designed to use data commonly collected by state and provincial wildlife agencies. We review a case study for white-tailed deer (Odocoileus virginianus) in North Dakota, USA, where we evaluate how harvest composition and monitoring frequency affect the maximum sustainable yield (MSY). Data were obtained from winter aerial surveys and hunter questionnaires over six years between 2009 and 2019. Harvest composition moderately skewed towards antlered individuals (37.5% antlerless deer and 62.5% antlered deer, i.e., antlerless:antlered harvest ratio = 0.6) resulted in a harvest rate of 0.2, which translates to a long-term harvest that is more than double that obtained if the harvest composition matched the population composition. However, given environmental uncertainty, we recommend that managers adopt a harvest strategy that is at least 10–15% lower than the maximum sustainable yield to buffer against environmental variability. Despite the fact that contrasting monitoring schemes resulted in similar optimal harvest rates, we illustrated how adopting an adaptive harvest strategy (i.e., where harvests change with population size) affords lower risks of overexploitation than a static strategy in which populations are assessed only occasionally. An adaptive harvest strategy features resilience allowing harvested populations to return to equilibrium even after substantial perturbation events.Item An asymmetric producer-scrounger game: Body size and the social foraging behavior of coho salmon(Theoretical Ecology, 2018) Phillips, Jessica A.; Peacock, Stephanie J.; Bateman, Andrew; Bartlett, Mackenzie; Lewis, Mark A.; Krkošek, MartinA tension between cooperation and conflict characterizes the behavioral dynamics of many social species. The foraging benefits of group living include increased efficiency and reduced need for vigilance, but social foraging can also encourage theft of captured prey from conspecifics. The payoffs of stealing prey from others (scrounging) versus capturing prey (producing) may depend not only on the frequency of each foraging strategy in the group but also on an individual's ability to steal. By observing the foraging behavior of juvenile coho salmon (Oncorhynchus kisutch), we found that, within a group, relatively smaller coho acted primarily as producers and took longer to handle prey, and were therefore more likely to be targeted by scroungers than relatively larger coho. Further, our observations suggest that the frequency of scrounging may be higher when groups contained individuals of different sizes. Based on these observations, we developed a model of phenotype-limited producer-scrounger dynamics, in which rates of stealing were structured by the relative size of producers and scroungers within the foraging group. Model simulations show that when the success of stealing is positively related to body size, relatively large predators should tend to be scroungers while smaller predators should be producers. Contrary to previous models, we also found that, under certain conditions, producer and scrounger strategies could coexist for both large and small phenotypes. Large scroungers tended to receive the highest payoff, suggesting that producer-scrounger dynamics may result in an uneven distribution of benefits among group members that—under the right conditions—could entrench social positions of dominance.Item Spatial memory and taxis-driven pattern formation in model ecosystems(Bulletin of Mathematical Biology, 2019) Potts, Jonathan R.; Lewis, Mark A.Mathematical models of spatial population dynamics typically focus on the interplay between dispersal events and birth/death processes. However, for many animal communities, significant arrangement in space can occur on shorter timescales, where births and deaths are negligible. This phenomenon is particularly prevalent in populations of larger, vertebrate animals who often reproduce only once per year or less. To understand spatial arrangements of animal communities on such timescales, we use a class of diffusion-taxis equations for modelling inter-population movement responses between N ? 2 populations. These systems of equations incorporate the effect on animal movement of both the current presence of other populations and the memory of past presence encoded either in the environment or in the minds of animals. We give general criteria for the spontaneous formation of both stationary and oscillatory patterns, via linear pattern formation analysis. For N = 2, we classify completely the pattern formation properties using a combination of linear analysis and nonlinear energy functionals. In this case, the only patterns that can occur asymptotically in time are stationary. However, for N ? 3, oscillatory patterns can occur asymptotically, giving rise to a sequence of period-doubling bifurcations leading to patterns with no obvious regularity, a hallmark of chaos. Our study highlights the importance of understanding between-population animal movement for understanding spatial species distributions, something that is typically ignored in species distribution modelling, and so develops a new paradigm for spatial population dynamics.Item Mountain pine beetle outbreak duration and pine mortality depend on direct control effort(Journal of Environmental Management, 2020) Kunegel-Lion, Mélodie; Lewis, Mark A.The efficacy of direct control methods in bark beetle outbreaks is a disputed topic. While some studies report that control reduces tree mortality, others see little effect. Existing models, linking control rate to beetle population dynamics and tree infestations, give insights, but there is a need to take into account the environment spatial variability and its impact on beetle life cycle. Here, we use natural variability found in a carefully monitored and controlled infestation to simulate outbreak dynamics under different control effort and to explore the impact of control on outbreaks suppression and tree mortality. Our semi-empirical predictive model of the number of infested trees as a function of ecological and environmental variables is coupled to a simulation model for infestation dynamics. We show that even a little control can have a major impact on the number of infested trees after several years of sustained effort. However, a moderate control of 60% is required to reduce the beetle population on the long term. Furthermore, a control rate of 69-83% is needed to achieve outbreak suppression in under 13 years depending on the abundance of incoming flights from outside sources.Item Persistence metrics for a river population in a two-dimensional benthic-drift model(AIMS Mathematics, 2019) Jin, Yu; Huang, Qihua; Blackburn, Julia; Lewis, Mark A.The study of population persistence in river ecosystems is key for understanding population dynamics, invasions, and instream flow needs. In this paper, we extend theories of persistence measures for population models in one-dimensional rivers to a benthic-drift model in two-dimensional depth-averaged rivers. We define the fundamental niche and the source and sink metric, and establish the net reproductive rate R0 to determine global persistence of a population in a spatially heterogeneous two-dimensional river. We then couple the benthic-drift model into the two-dimensional computational river model, River2D, to study the growth and persistence of a population and its source and sink regions in a river. The theory developed in this study extends existing R0 analysis to spatially heterogeneous two-dimensional models. The River2D program provides a method to analyze the impact of river morphology on population persistence in a realistic river. The theory and program derived here can be applied to species in real rivers.Item A robust and efficient algorithm to find profile likelihood confidence intervals(Statistics and Computing, 2021) Fischer, Samuel M.; Lewis, Mark A.Profile likelihood confidence intervals are a robust alternative to Wald's method if the asymptotic properties of the maximum likelihood estimator are not met. However, the constrained optimization problem defining profile likelihood confidence intervals can be difficult to solve in these situations, because the likelihood function may exhibit unfavorable properties. As a result, existing methods may be inefficient and yield misleading results. In this paper, we address this problem by computing profile likelihood confidence intervals via a trust-region approach, where steps computed based on local approximations are constrained to regions where these approximations are sufficiently precise. As our algorithm also accounts for numerical issues arising if the likelihood function is strongly non-linear or parameters are not estimable, the method is applicable in many scenarios where earlier approaches are shown to be unreliable. To demonstrate its potential in applications, we apply our algorithm to benchmark problems and compare it with 6 existing approaches to compute profile likelihood confidence intervals. Our algorithm consistently achieved higher success rates than any competitor while also being among the quickest methods. As our algorithm can be applied to compute both confidence intervals of parameters and model predictions, it is useful in a wide range of scenarios.Item Accurate long-range forecasting of COVID-19 mortality in the USA(Scientific Reports, 2021) Ramazi, Pouria; Haratian, Arezoo; Meghdadi, Maryam; Oriyad, Arash Mari; Lewis, Mark A.; Maleki, Zeinab; Vega, Roberto; Wang, Hao; Wishart, David S.; Greiner, RussellThe need for improved models that can accurately predict COVID-19 dynamics is vital to managing the pandemic and its consequences. We use machine learning techniques to design an adaptive learner that, based on epidemiological data available at any given time, produces a model that accurately forecasts the number of reported COVID-19 deaths and cases in the United States, up to 10 weeks into the future with a mean absolute percentage error of 9%. In addition to being the most accurate long-range COVID predictor so far developed, it captures the observed periodicity in daily reported numbers. Its effectiveness is based on three design features: (1) producing different model parameters to predict the number of COVID deaths (and cases) from each time and for a given number of weeks into the future, (2) systematically searching over the available covariates and their historical values to find an effective combination, and (3) training the model using "last-fold partitioning", where each proposed model is validated on only the last instance of the training dataset, rather than being cross-validated. Assessments against many other published COVID predictors show that this predictor is 19-48% more accurate.