Table 1.
Aggregated wheat rust prevalence in Ethiopia during years 2010–2019.
Fig 1.
Spatial patterns of wheat rust outbreaks in Ethiopia in years 2010–2019.
(left column) wheat stripe rust; (middle column) wheat stem rust; (right column) wheat leaf rust. In (A-C) all surveys of years 2010–2019 are mapped to illustrate the total spatial extent and disease patterns in point surveys. In (D-F) the number of surveys per administrative district are shown. (G-I) illustrates the proportion of positives per district (number of surveys with positive incidence values / total number of surveys). (J-L) shows hotspots with clustering of districts with high proportion of positives coloured in red and cold-spots with clustering of districts with low proportion of positives coloured in blue. Illustrated values are values of the Getis-Ord Gi* statistic. The Gi* statistic is a z-score indicating for each district if it belongs to a statistically significant cluster of districts with high (or low) proportions of positives. For example, values of Gi* > + 1.96 (or Gi* <– 1.96) indicate that the null-hypothesis of spatially random distribution of districts with high (or low) proportions of positives can be rejected at a significance level of 0.05 (1.96 standard deviations of the normal distribution). See S1–S4 Figs for a similar analysis using only moderate/high incidence scores as well as using severity scores. Maps created using R as GIS [18–22].
Fig 2.
Linear correlation between altitude and wheat rust prevalence in Ethiopia.
(A-B) wheat stripe rust; (C-D) wheat stem rust; (E-F) wheat leaf rust. The top row shows disease incidence, and the bottom row shows disease severity. The labels at the top of the x-axes show the total number of surveys per altitude bin. Prevalence is calculated as: number of surveys with disease incidence (severity) / total number of surveys per altitude bin. The stacked bar graphs show the parts of all positives with low, moderate, and high disease levels. For example, out of the total 291 surveys shown in the right bar in panel A, approximately 38% (110 surveys) are fields with high incidence, 5% (15 surveys) are fields with moderate incidence and 18% (52 surveys) are fields with low incidence. The remaining 114 surveys are negatives (not illustrated). The lines are linear fits to the mean prevalence values, separately calculated for the categories: ≥ low, ≥ moderate, and high disease. The lines are fitted to the stacked prevalence values. For example, the “linear fit to Yr ≥ low incidence” in panel A is obtained from all surveys with incidence scores “≥ low incidence”.
Fig 3.
Wheat rust disease prevalence on major wheat varieties grown in Ethiopia.
(A) wheat stripe rust; (B) wheat stem rust; (C) wheat leaf rust. (A-C) show the long-term mean disease prevalence during years 2010–2019 on key wheat varieties grown in Ethiopia (top: incidence; bottom: severity). (D) shows inter-annual variations of disease severity on the widely deployed wheat variety, Digalu (see S7 Fig for the corresponding incidence scores). The labels at the top of the x-axes show the total number of surveys per year. Prevalence is calculated as: number of surveys with disease incidence (severity) / total number of surveys per wheat variety. The years of first detection of wheat stripe rust pathogen races PstS6 [28,29] and PstS11 [30] and wheat stem rust pathogen races TKTTF [4] are illustrated at the bottom of (D).
Fig 4.
Inter-annual variations in aggregated total wheat rust prevalence in Ethiopia.
(A) wheat stripe rust; (B) wheat stem rust; (C) wheat leaf rust. The top row shows prevalence of incidence scores and the bottom row shows prevalence of severity scores. The labels at the top of the x-axis show the total number of surveys per year. Prevalence is calculated as: number of surveys with disease incidence (severity) / total number of surveys per year. At the bottom, the years of first detection of wheat stripe rust pathogen races PstS6 [28,29] and PstS11 [30] and wheat stem rust pathogen race TKTTF [4] are illustrated.
Fig 5.
Long-term average within-season wheat rust disease progress in Ethiopia.
(A) wheat stripe rust; (B) wheat stem rust; (C) wheat leaf rust. The time of the main wheat season in Ethiopia is separated into a set of bi-weekly time-intervals (x-axis). All surveys for years 2010–2019 are grouped into the bi-weekly time-intervals. The labels at the top of the x-axis show the total number of surveys per time-interval. For each time-interval, the mean disease prevalence (y-axis) is calculated (top row: incidence; bottom row: severity). The stacked bar graphs show the parts of all positives with low, moderate, and high disease levels. The lines show a logistic curve (see Eq 2) fitted to the mean prevalence data, separately calculated for the categories: ≥ low, ≥ moderate, and high disease. Note that there are substantial inter-annual variations (Fig 4) in disease prevalence, which are not shown in this illustration of the long-term average within-season disease-progress.
Fig 6.
Wheat production in Ethiopia and estimated financial losses caused by wheat rust outbreaks during years 2010–2019.
Blue: FAO data; grey: wheat stem rust; yellow: wheat stripe rust; red/brown: wheat leaf rust. (A-D) show national wheat production statistics of Ethiopia obtained from FAOSTAT [1]. (E-H) illustrate our estimates of the damage caused by wheat rusts during years 2010–2019. (E) shows the estimated area infected with wheat rusts; (F) shows the estimated fraction of yield lost due to wheat rusts; (G) shows the approximate total financial loss caused by wheat rusts; and (H) shows the approximate loss relative to the national total financial value of wheat produce at market price per year. As no FAO statistics were available for year 2019 at the time of this study (last checked on the 20th of June 2020), we use the mean of years 2010–2018 as input for our estimates of yield losses in year 2019.
Fig 7.
Performance of simple logistic models for predicting wheat rust occurrence in Ethiopia.
(A-B) wheat stripe rust; (C-D) wheat stem rust; (E-F) wheat leaf rust. Two simple logistic models were used to predict wheat rust occurrence: a temporal model (model 1, see Eq 2) predicting wheat rust occurrence as a function of the time since the start of the main wheat season and a spatiotemporal model (model 2, see Eq 3), predicting wheat rust occurrence as a function of the time since the start of the main season and the location in Ethiopia (latitude, longitude, and altitude). Model performance was tested by fitting the models to training data from all but 1 year of surveys and then conducting a ROC analysis for testing the performance of the fitted model against the data from the year not used for fitting (repeated for every year). The upper row shows the resulting AUC score of both models for each year and all rusts. The bottom row shows the corresponding ROC curves of one exemplar year. For the analysis illustrated here all survey entries with non-zero disease incidence were classified as “diseased” and all surveys with zero incidence were classified as “healthy”. The testing procedure was also conducted using an alternative dichotomization scheme classifying all surveys with moderate or high incidence values as “diseased” and all surveys with zero or low incidence as “healthy” (see S13 Fig).