Epidemics caused by long-distance dispersed pathogens result in some of the most explosive and difficult to control diseases of both plants and animals (including humans). Yet the factors influencing disease spread, especially in the early stages of the outbreak, are not well-understood. We present scaling relationships, of potentially widespread relevance, that were developed from more than 15 years of field and in silico single focus studies of wheat stripe rust spread. These relationships emerged as a consequence of accounting for a greater proportion of the fat-tailed disease gradient that may be frequently underestimated in disease spread studies. Leptokurtic dispersal gradients (highly peaked and fat-tailed) are relatively common in nature and they can be represented by power law functions. Power law scale invariance properties generate patterns that repeat over multiple spatial scales, suggesting important and predictable scaling relationships between disease levels during the first generation of disease outbreaks and subsequent epidemic spread. Experimental wheat stripe rust outbreaks and disease spread simulations support theoretical scaling relationships from power law properties and suggest that relatively straightforward scaling approximations may be useful for projecting the spread of disease caused by long-distance dispersed pathogens. Our results suggest that, when actual dispersal/disease data are lacking, an inverse power law with exponent = 2 may provide a reasonable approximation for modeling disease spread. Furthermore, our experiments and simulations strongly suggest that early control treatments with small spatial extent are likely to be more effective at suppressing an outbreak caused by a long-distance dispersed pathogen than would delayed treatment of a larger area. The scaling relationships we detail and the associated consequences for disease control may be broadly applicable to plant and animal pathogens characterized by non-exponentially bound, fat-tailed dispersal gradients.

In addition to pathogen propagule dispersal, disease spread requires successful infection of host tissue. In plant disease epidemiology, susceptibility of host tissue is often assumed to be constant. This assumption ignores changes in host phenology due to developmental stage. To examine this assumption, 3-, 4-, and 5-week-old wheat plants were inoculated with equal quantities of urediniospores of Puccinia striiformis f. sp. tritici, the causal agent of wheat stripe rust (WSR). Disease severity on each leaf was assessed and fit by mixed-effect linear model as a function of leaf position and plant age. Younger plants had significantly greater disease severity than older plants, with mean severities of 50.4, 30.1, and 12.9% on plants that were 3, 4, and 5 weeks old, respectively, at time of inoculation. This effect was greater on leaves higher on the plant. Within same-aged plants, younger leaves had significantly greater disease severity than older leaves, with mean severities of 40.2, 34.8, and 17.7% on the uppermost, second, and third leaf, respectively. These results suggest that the vertical distribution of WSR lesions in agricultural fields could be driven by differences in host susceptibility more so than propagule dispersal.

Understanding how disease foci arise from single source lesions has not been well studied. Here, single wheat leaves were inoculated with Puccinia striiformis f. sp. tritici urediniospores, and all wheat leaves within two intersecting 0.3 × 3.0 m transects were sampled in eight replicates over three years. The lesions observed on each of the top three leaves on plants within 1.5 m from the source lesion were three-dimensionally mapped. The total number of lesions within a 1.5 m radius was estimated by dividing the number of lesions observed within each 0.025 m-wide annulus by the fraction of the annulus sampled. The estimated total number of lesions produced within 1.5 m of a single source lesion ranged from 27 to 776, with a mean of 288 lesions. Eighty percent of the lesions were recorded within 0.69 m of the source infection. The proportion of total lesions observed at a given distance from the source was fitted well by the Lomax and Weibull distributions, reflecting the large proportion of lesions arising close to the source, and when fitted to an inverse-power distribution had a slope (b) of 2.5. There were more lesions produced on leaves higher in the canopy than on lower leaves, with more lesions being detected above than below the point of inoculation. Simultaneous measurement of lesion gradients and spore dispersal in the final year of the study suggests that this pattern is due to greater susceptibility of upper leaves, rather than increased dispersal to upper leaves.

Accurate infectious disease forecasting can inform efforts to prevent outbreaks and mitigate adverse impacts. This study compares the performance of statistical, machine learning (ML), and deep learning (DL) approaches in forecasting infectious disease incidences across different countries and time intervals. We forecasted three diverse diseases: campylobacteriosis, typhoid, and Q-fever, using a wide variety of features (n = 46) from public datasets, e.g., landscape, climate, and socioeconomic factors. We compared autoregressive statistical models to two tree-based ML models (extreme gradient boosted trees [XGB] and random forest [RF]) and two DL models (multi-layer perceptron and encoder–decoder model). The disease models were trained on data from seven different countries at the region-level between 2009–2017. Forecasting performance of all models was assessed using mean absolute error, root mean square error, and Poisson deviance across Australia, Israel, and the United States for the months of January through August of 2018. The overall model results were compared across diseases as well as various data splits, including country, regions with highest and lowest cases, and the forecasted months out (i.e., nowcasting, short-term, and long-term forecasting). Overall, the XGB models performed the best for all diseases and, in general, tree-based ML models performed the best when looking at data splits. There were a few instances where the statistical or DL models had minutely smaller error metrics for specific subsets of typhoid, which is a disease with very low case counts. Feature importance per disease was measured by using four tree-based ML models (i.e., XGB and RF with and without region name as a feature). The most important feature groups included previous case counts, region name, population counts and density, mortality causes of neonatal to under 5 years of age, sanitation factors, and elevation. This study demonstrates the power of ML approaches to incorporate a wide range of factors to forecast various diseases, regardless of location, more accurately than traditional statistical approaches.

Aim:Understanding how spatial scale of study affects observed dispersal patterns can provide insights to spatiotemporal population dynamics, particularly in systems with significant long-distance dispersal (LDD). We aimed to investigate the dispersal gradients of two rusts of wheat with spores of similar size, mass and shape, over multiple spatial scales. We hypothesized that a single dispersal kernel could fit the dispersal from all spatial scales well, and that it would be possible to obtain similar results in spatiotemporal increase of disease when modelling based on differing scales.Location: Central Oregon and St. Croix Island.Taxa: Puccinia striiformis f. sp. tritici, Puccinia graminis f. sp. tritici, Triticum aestivum. Methods:We compared empirically derived primary disease gradients of cereal rust across three spatial scales: local (inoculum source and sampling unit = 0.0254 m, spatial extent = 1.52 m) field-wide (inoculum source = 1.52 m, sampling unit = 0.305 m and spatial extent = 91.4 m) and regional (inoculum source and sampling unit = 152 m, spatial extent = 10.5 km). We then examined whether disease spread in spatially explicit simulations depended upon the scale at which data were collected by constructing a compartmental time-step model. Results:The three data sets could be fit well by a single power law dispersal kernel.Simulating epidemic spread at different spatial resolutions resulted in similar patterns of spatiotemporal spread. Dispersal kernel data obtained at one spatial scale can be used to represent spatiotemporal disease spread at a larger spatial scale.