Irrigation can create new green bridges that promote rapid intercontinental spread of the wheat stem rust pathogen

We use a meteorology driven Lagrangian particle dispersal model (NAME; Jones et al 2007) originally adapted by Meyer et al (2017a, 2017b) for dispersal of urediniospores (hereafter referred to as 'spores') Pgt. The NAME model is driven by 3D meteorological data from the UK Met Office Numerical Weather Prediction model the Unified Model (Walters et al 2019), with approximately 10 km horizontal and 3 h temporal resolution and up to 59 vertical layers. The model simulates the release, turbulent dispersal, wet and dry deposition and UV-induced death of spores (refer to supplementary information and Meyer et al (2017a, 2017b)).

We consider spore dispersal between donor sites (from which spores are dispersed from infected wheat) and receptor sites (where spores land on susceptible wheat). Amongst donor and receptor sites we distinguish confirmed disease locations (CDLs) in which TTKTT is confirmed and representative disease locations (RDLs; see Meyer et al 2017b) where Pgt is known to occur but TTKTT was not confirmed. Each CDL and RDL is represented by nine model grid cells (90 km × 90 km) and so characterises the broad wheat production region of the area.

We use a simple SEIR epidemiological framework (hereafter the integrated model) operating on a daily timestep to classify the infection status of the sites. A site moves into the susceptible (S) class after a period of initial growth. Infected sites move into the exposed (E) class, which accounts for the delay between initial infection and the site moving into the infected (I) class when capable of infecting other sites. The site remains infectious until leaf senescence prevents further spore production when it moves into the removed (R) class. Sites that pass into the removed class are not passed back to the susceptible class within the same cropping season because we do not define the length of the infectious period and we do not allow for chemical control in the current model. A meteorology-driven, environmental suitability model (see supplementary information) is used to assess whether temperature and moisture conditions are suitable for infection when spores land on a susceptible site.

Infected receptors remain in the exposed class until a crop latency period (the time between crop receptors being infected and themselves becoming infectious donors) has passed. A default version uses a temperature-dependent latency period based on the model of Nopsa and Pfender (2014), defined as:

$$\begin{align*} &{\text{Proportion of Lp50 completed}}\nonumber\\ &\quad = {\text{hours}} \times (2.9 \times 10^{-4}) \times (T-1.8)\nonumber\\ &\qquad \times [1-e^{(T-30.9)}],\end{align*}$$

for values of T between 1.8 and 30.9 °C, where Lp50 is the time when 50% of the maximum number of pustules have erupted and T is the average temperature over the timestep. A second version using a fixed latency period of 30 d (Roelfs 1992) was examined as a sensitivity test.

A metric of the relative probability of infection was created as follows:

$$\begin{align*} P\left( I \right) = \left\{ \begin{array}{*{20}{c}} {{\text{if }}D \lt 1{\text{ spore h}}{{\text{a}}^{{{ - 1}}}}{\text{, then }}0} \\ {{\text{if }}D \geqslant 1{\text{ spore h}}{{\text{a}}^{{{ - 1}}}},{\text{then}}\,{\text{ES}} \times {\text{Norm}}\left[ {{\text{log}}\left( D \right)} \right]} \end{array}\right.\end{align*}$$

where P(I) is the probability of infection, ES is the environmental suitability value and Norm[log(D)] is the log of deposition values, normalised to between 0 and 1 with the upper limit of the deposition values calculated in a pre-processing step. The integrated model allows the user to specify the threshold for the combined suitability/deposition value above which a potential receptor becomes infected. Any value above 0 is assumed to imply infection, i.e. 1 spore ha⁻¹, or more, landing on a susceptible site with a non-zero measure of environmental suitability leads to infection (see also supplementary information for a sensitivity analysis).

Simulations were performed for 2018–2019 under the following scenarios:

• (a)  
  Scenario 1. 'CDLs': simulations restricted to sites with confirmed identification of TTKTT;
• (b)  
  Scenario 2. 'CDLs + Ethiopia RDLs': simulations included sites with confirmed identification of TTKTT plus additional key wheat-growing locations in Ethiopia where Pgt is known to occur but TTKTT has not been confirmed;
• (c)  
  Scenario 3. 'CDLs + Ethiopia RDLs + Stepping-stone RDLs': simulations include sites with confirmed identification of TTKTT plus additional key wheat-growing locations in Ethiopia and locations in theoretical stepping-stone countries of either Eritrea, Yemen, Saudi Arabia and Iran where TTKTT has not been confirmed.

The CDLs are shown in figure 2 for the years 2014–2019. However, with low frequency races like TTKTT it cannot be excluded that the race was present in additional sites, years or countries and went undetected in the field sampling (note: no sampling was undertaken in Yemen or Saudi Arabia).

Zoom In Zoom Out Reset image size

Under Scenario 1, based on the timeline of dates in 2018 where the TTKTT race was identified, the initial donors are assumed to be CDLs in Kenya, Tanzania or Rwanda; note that due to the large number of confirmed sites in Kenya, all of the Kenyan sites are grouped together into the four CDLs shown in figure 3.

Zoom In Zoom Out Reset image size

Under Scenario 2, additional wheat sites in Ethiopia are included where race TTKTT is not confirmed (but its presence in 2018 was possible because TTKTT was confirmed in later years):

• (a)  
  A representative site in Bale zone (figure 3: site #19)—the core wheat belt of Ethiopia and an area of double cropping. Wheat is grown both in the Belg (short rain) season (March/April–July/August) and the Meher (long rain) season (August–January), providing an important green bridge.
• (b)  
  Two representative sites in the newly furrow-irrigated regions along the Awash river (Gebul 2021)—Werer and Metahara (figure 3: sites #14, #15, respectively).

Under Scenario 3, potential stepping-stone sites in key wheat-growing regions of Eritrea, Yemen and Iran are included (figure 3 sites #1–3 and #7–10). The RDLs include locations where races other than TTKTT have been identified and RDLs used in Meyer et al (2017b); most RDLs are assumed to be rainfed but the East Yemen RDL is assumed to be flood-irrigated (Frenken 2009). Due to a change in government policy to re-support domestic wheat production in Saudi Arabia (USDA Foreign Agricultural Service 2019), we also include three additional sprinkler-irrigated locations (Frenken 2009) in Saudi Arabia: Wady Al Dwasser, Al Kharj and Eastern Province (figure 3: sites #4–6).

Although complex models simulating the spread of rust epidemics have been developed (e.g. Eversmeyer et al 1973, Rossi et al 1997, Bregaglio and Donatelli 2015, Savary et al 2015, Naseri and Sabeti 2021, Salotti et al 2022), given the spatial extent of our study area (which covers a wide range of climatic regimes) and the lack of survey data available to calibrate a more complex model, we adopt a simpler approach in our study. The crop calendar is defined in terms of a monthly timestep and planting and harvest date assumptions for each site (figure 3) are based on those used in Meyer et al (2017b), the National Institute of Statistics of Rwanda (2019) and expert knowledge. The main assumptions are that new wheat plants only become receptive to stem rust one month after emergence (to allow for initial growth) and cease to be receptive one month before harvest (when leaf senescence prevents further spore production). Given that these are considered conservative estimates, sensitivity testing to these timings is also conducted.

The following initial donor sites and start dates for spore dispersal were used (see figure 3):

• Rwanda (Muzanse): 1st January 2018 and 1st May 2018,
• Kenya (North Rift, Central Rift, South Rift and Mount Kenya): 1st February 2018 and 1st August 2018
• Tanzania (Kilimanjaro and Hanang): 1st May 2018

The results show there is no meteorology-driven transmission pathway between CDLs in East Africa and the Middle East in the 2018/19 season. Spores released from Muzanse cannot infect any other CDLs. Although spores are deposited from Muzanse over the viable CDLs in Kenya for the January start date (figure 4), environment conditions are not suitable for infection before the crops in Kenya are harvested. For the May start date, the crop calendars do not overlap between the infected donor and susceptible receptor crops.

Zoom In Zoom Out Reset image size

We find that Tanzanian donors can only infect Musanze in Rwanda, which causes no onward transmission of wheat stem rust, as previously described.

With the February start date, Kenyan donors do not infect any other CDLs. With the August start date, Kenyan donors infect all Ethiopian CDLs, but no onward transmission to Iraq is simulated.

Inspection of the crop calendar (figure 3) reveals that the only CDLs that can be infectious at the same time as the CDL in Iraq is receptive are Muzanse, the two Tanzanian CDLs and the four Kenyan CDLs. However, deposition (figure 4) indicates spores released from all these CDLs cannot be transported directly to Halabja, Iraq. This means other potential locations for wheat stem rust where TTKTT has not been identified must be considered as possible stepping-stones.

Given that spores from Muzanse in Rwanda and the two Tanzanian CDLs cannot infect any other CDLs, it is highly unlikely these were the origin of the confirmed TTKTT infections in Ethiopia in 2018 or the confirmed TTKTT infection in Iraq in 2019. For this reason, we confine all further analyses to Kenyan initial donors, and we assume an August start date only because the February start date assumption did not lead to any infections outside of Kenya.

The potential Ethiopian RDLs considered in Scenario 2 are Bale (two annual rainfed crops), Werer, and Metahara (both with single irrigated crops). The simulations assume Kenyan initial donors starting in August, as described in section 3.1.

Inclusion of the Bale RDL or the two irrigated Ethiopian RDLs (Metahara and Werer) result in infection at those sites (in September 2018, December 2018, and January 2019, respectively), but the infection does not reach Halabja. Spores from the Ethiopian sites could, however, reinfect the CDLs in Kenya and Rwanda before the crops are harvested in June 2019.

Inclusion of the additional Ethiopian RDLs cannot facilitate infection of Halabja because spore deposition from Bale (figure 5(A)) or Metahara/Werer (figure 5(B)) does not extend as far as Iraq. This means that other potential RDLs in key wheat-growing regions, in other countries, where sampling has not been undertaken and TTKTT has not been identified, must be considered as possible 'stepping-stones'.

Zoom In Zoom Out Reset image size

The potential Ethiopian RDLs considered in Scenario 3 are the irrigated locations of Werer and Metahara; and Bale with two annual rainfed crops in both the Belg and Meher seasons. Other possible stepping-stone RDLs are located in Eritrea, Yemen, Saudi Arabia and Iran. Each potential stepping-stone RDL is considered individually, and, in each case, we consider the results both with and without the irrigated Ethiopian RDLs. The simulations assume Kenyan initial donors starting in August, as described in section 3.1.

In all cases, when the two irrigated Ethiopian RDLs are excluded from the analysis, the crop is harvested before the infection reaches Halabja. When the irrigated RDLs in Ethiopia are included, stepping-stone RDLs in either Saudi Arabia or Yemen are necessary to facilitate infection at Halabja. We now describe these two simulations in more detail.

3.3.1. Yemen

The West Yemen RDL is infected from the Hararghe CDL in August 2018 (figure 6(A)). Spores from West Yemen in November 2018 are deposited over Halabja when there is no susceptible wheat. The Guji CDL and/or the Bale RDL (infected in September 2018 from Meher Ethiopian CDLs) infect the Metahara RDL, which then infects the neighbouring Werer RDL (figure 6(B)). The irrigated Ethiopian RDLs infect the irrigated East Yemen RDL (figure 6(C)) and then East Yemen infects Halabja in March 2019 (figure 6(D)). East Yemen also reinfects West Yemen in March 2019 and the infection spreads back into Bale in April 2019, and the cycle repeats.

Zoom In Zoom Out Reset image size

3.3.2. Saudi Arabia

The simulation is identical to that described for Yemen until the two irrigated Ethiopian RDLs have been infected. In February 2018, those irrigated Ethiopian RDLs then infect the central irrigated Saudi Arabian RDL of Al Kharj (figure 7(A)), which then infects Halabja (and the other two Saudi Arabian RDLs) in March 2019 (figure 7(B)). Unlike the East Yemen route, though, no reinfection of the Ethiopian sites occurs in this scenario and the infection transmission ceases in July 2019.

Zoom In Zoom Out Reset image size

We have shown that both Yemen and Saudi Arabia can act as stepping-stone locations between East Africa and the Middle East, but only when we include the irrigated wheat RDLs in Ethiopia, as summarised in figure 8. We find that infection of Halabja from spores released from East Yemen, though, can only happen on three days in March and a single day in April 2019. Conversely, infection of Halabja from spores released from the Saudi Arabian (irrigated) RDLs can happen on 29 d between February and April 2019 and therefore we consider the route through Saudi Arabia more likely. Infection of West Yemen by spores released from East Yemen, however, is necessary for the infection to return to Ethiopia.

Zoom In Zoom Out Reset image size

Several sensitivity tests evaluate the robustness of the results for these two Scenario 3 simulations (see supplementary information). The 'Control scenario' consists of CDLs + irrigated Ethiopian RDLs + Bale RDL + Stepping-stone RDLs in Saudi Arabia and Yemen, initial donors in Kenya and an August start date.

The results are sensitive to the latency period (refer to section 2.1): with a 30 day latency period, only a pathway through Saudi Arabia provides a viable transmission route from East Africa to the Middle East. The Saudi Arabian pathway is also the only viable route found when the probability of infection threshold is increased above 0 but ceases to be a viable pathway when the threshold is raised above 0.1. The results are also sensitive to the timing of receptiveness to infection, but only to the time at which crops become susceptible after sowing. A one-month delay in susceptibility at receptor locations results in no infection of the irrigated RDLs. Extending the period of susceptibility closer to harvest does not change the results from the Control scenario.

The sensitivity tests confirm that it is not just the overlap in the crop calendar that is important in creating a green bridge between the Guji CDL and/or the Bale RDL and the irrigated Ethiopian RDLs, but also the fact that those RDLs are irrigated: spore deposition does not coincide with environmental suitability for infection when those RDLs are not irrigated. Future work should consider sensitivity of the environmental suitability model to the type of irrigation system used (Swett 2020).