Yellow Fever (YF) is a vaccine preventable, zoonotic, arbovirus endemic in tropical regions of Africa and Latin America. It is responsible for approximately 78,000 deaths per year, although under reporting is high and since YF has a non-specific symptom set, misdiagnosis is an issue (Garske et al., 2014). YF has three transmission ‘cycles’ in Africa: urban, zoonotic and intermediate. The urban cycle, mediated by Aedes Aegypti mosquitoes, is responsible for explosive outbreaks such as the one seen in Angola in 2016 (Ingelbeen et al., 2018; Wilder-Smith and Monath, 2017).

While the urban cycle can rapidly amplify transmission, the majority of YF infections are thought to occur as a result of zoonotic spillover from the sylvatic reservoir in non-human primates (NHP). This zoonotic cycle is mediated by a variety of mosquito vectors including Aedes africanus and, as the NHP hosts are mostly unaffected by the infection in Africa, the force of infection due to spillover is fairly constant, although land use change has been shown to affect this (Monath and Vasconcelos, 2015). The intermediate cycle is sometimes called the savannah cycle and is mediated by mosquitoes such as Ae. luteocephalus, who feed opportunistically on humans and NHP, although human-human transmission is limited (Barrett and Higgs, 2007).

The Intergovernmental Panel on Climate Change (IPCC) states that global mean temperatures are likely to rise by 1.5°C, compared with pre-industrial levels, by between 2030 and 2052 if current trends continue (Masson-Delmotte et al., 2018). Increases are projected not only in mean temperature but also in the extremes of temperature, extremes of precipitation and the probability of drought (Kharin et al., 2013; Dunning et al., 2018).

With multiple mosquito vectors and a zoonotic cycle depending on NHP hosts, the impact of climate change on YF is likely to be complex. Focusing on the main urban vector, A. aegypti, there is strong evidence that projected climate change will alter its global distribution and thus, the risk of diseases it carries (Ryan et al., 2019; World Health Organisation, 2018; World Health Organisation, 2018). Climate change has been predicted to increase the regions at risk from dengue and Zika transmission, although seasonal variation in temperature may mitigate the likelihood of outbreaks in areas at the edges of the endemic zone (Mordecai et al., 2017; Huber et al., 2018).

Long-term projections of the future disease burden of YF are needed to inform vaccination planning (VIMC, 2019). Furthermore, differences due to climate change may increase the risk of epidemics, a key consideration for the Eliminate YF Epidemics (EYE) strategy (World Health Organization, 2017).

In this manuscript, we extend an existing model of YF occurrence and disease burden to incorporate a nonlinear temperature suitability metric (Garske et al., 2014). We estimate temperature suitability for YF based on the thermal response of the urban vector, Ae. aegypti, and the YF virus. We combine this with YF occurrence data in a Bayesian hierarchical model in order to account for uncertainty at each stage of the modelling process. This, along with established estimates of transmission intensity informed by serological survey data, allow us to predict current and future transmission intensity. Finally, we use ensemble climate model predictions of future temperature and precipitation to project transmission and thus, burden in 2050 and 2070. Our results are the first examination of YF burden under the potential future effect of climate change.

As we estimate a static force of infection, we focus on transmission as a result of sylvatic spillover rather than including the urban transmission cycle explicitly. As such, the results can be considered the estimated effect of climate change on sylvatic transmission and resulting burden.

Model predictions for baseline scenario

Figure 1 (left) shows occurrence of YF across Africa from 1984 to 2018. Incidence is focused in the West of Africa and, more recently, Angola and the Democratic Republic of the Congo. The model predicts a high probability of YF report in these areas and reflects the general patterns of YF occurrence, see Figure 1 for comparison. Model fit can be characterised by the Area Under the Curve (AUC) statistic (Huang and Ling, 2005), which was 0.9004, similar to the original model formulation of Garske et al., 2014.

Figure 1 with 4 supplements see all

Download asset Open asset

The predicted probability of a YF report is positively informed by temperature suitability with the median posterior predicted distribution shown in Figure 2 (left). This highlights the high suitability of countries such as Nigeria and South Sudan for YF transmission. In contrast, Rwanda, Burundi and areas of Mali and Mauritania have low average temperature suitability. The fit of the thermal response models is shown in Figure 2—figure supplements 1–4.

Figure 2 with 4 supplements see all

Download asset Open asset

Projected transmission intensity

Figure 2 (right) shows the median posterior predicted estimates of the force of infection for the baseline/current scenario, a comparison of the force of infection estimated only from serological studies, and those estimated from the GLM is provided in Figure 2—figure supplement 1. When we incorporate the ensemble projections of temperature and precipitation change we see heterogeneous impacts on force of infection. Figure 3 shows the percentage change in median force of infection for the year 2070. Projections for 2050 are shown in Figure 3—figure supplement 1.

Figure 3 with 1 supplement see all

Download asset Open asset

The posterior distributions of predicted changes in force of infection in different African regions are shown in Figure 4 (region definitions shown in Figure 4—figure supplement 1). Projections for individual countries are given in the Appedix. In West Africa, the predicted change is clustered around zero in the majority of scenarios; this is particularly the case for year 2050. However, due to wider uncertainty in 2070 and for RCP scenario 8.5 in general, there is a more discernible increase. In the East and Central regions, a predicted increase in force of infection is more apparent. Whilst the differences between 2050 and 2070 are difficult to see for RCP scenario 2.6, both peak above zero. In RCP scenarios 4.5, 6.0 and 8.5, the distinction between years is clear, particularly in 8.5, with the greatest increases seen in 2070 as temperatures are expected to continue to rise.

Figure 4 with 1 supplement see all

Download asset Open asset

When we examine the changes at country level, shown in the appendix, the changes are more heterogeneous. For RCP 2.6 Guinea Bissau, the change in force of infection in 2070 is potentially broad, with a credible interval spanning zero: 10.3% (95%CrI [−33.2% , 96.3%]). Whereas in Central African Republic, there is a notable increase by 87.1% (95%CrI [12.4% , 390.2%]).

Projected burden

The projected percentage change in the annual number of deaths caused by YF across Africa is given in Table 1; the projected annual deaths per capita for endemic countries are shown in Figure 5 and in Figure 5—figure supplement 1. These projections assume vaccination is static from 2019 onwards that is that only routine vaccination continues at 2018 levels. Similarly, we assume case management is unvarying. Aggregated numbers of deaths per country and region are shown in the appendix.

Table 1

Year	Scenario	95% CrI low	50% CrI low	Median	50% CrI high	95% CrI high
2050	RCP 2.6	−2.36	4.49	10.84	18.58	37.91
2050	RCP 4.5	−2.40	7.32	16.71	28.16	57.43
2050	RCP 6.0	−2.78	6.79	15.49	25.86	51.85
2050	RCP 8.5	−2.17	11.03	24.92	41.84	88.33
2070	RCP 2.6	−0.74	4.11	9.99	17.03	34.10
2070	RCP 4.5	−2.76	7.77	19.28	33.56	71.08
2070	RCP 6.0	−4.56	8.63	21.35	36.70	77.70
2070	RCP 8.5	−2.90	16.08	39.57	72.43	178.63

Figure 5 with 3 supplements see all

Download asset Open asset

While lower 95% credible intervals in Table 1 are negative, the overall posterior probabilities that climate change will increase YF mortality are very high for each climate scenario. The probability that deaths will increase is 95.5% (95% CrI [95.3%, 95.7%]) for RCP 2.6 in year 2070, rising to 95.9% (95% CrI [95.7%, 96.1%]) for RCP 8.5 in year 2070, values for all scenarios and years are shown in appendix 1.

As with the force of infection projections, the most severe increases are seen for RCP scenario 8.5, especially in year 2070. The distinction between current projected deaths per capita and those under each RCP scenario are most clearly seen for countries in Central Africa, such as Central African Republic, and East Africa, such as Ethiopia. The four countries with the least distinct change, Liberia, Guinea, Sierra Leone and the Gambia, are all in West Africa, commonly thought to see the most intense YF transmission. As such, it appears that the most marked increases in burden are found in East and central Africa.

We build on an established model of YF occurrence and transmission to accommodate temperature and precipitation projections for four climate emissions scenarios. Non-linear dependence on temperature was incorporated by utilising a function of temperature suitability, informed by thermal response data for A. aegypti. We jointly estimated parameters for the temperature suitability and occurrence models in a Bayesian framework, allowing us to quantify the uncertainty in our projections. We found that model fit remained good with a median AUC of 0.9004 despite necessary changes to the covariates used in the occurrence model compared with past work Garske et al., 2014; where changes were required in order to include covariates for which climate change projections were available. This gave us some confidence in the suitability of the model for projecting the impact of climate change on YF transmission through to 2070, the last year for which climate emission scenario projections are available for temperature and precipitation.

The force of infection is projected to increase for the majority of countries in each scenario. Consistently, the Central African Republic is one of the countries most likely to see an increase in transmission, while Liberia and Guinea Bissau have more uncertain projections. This highlights that the most severe proportional increases in force of infection are seen outside West Africa. However, as transmission is currently highest in West Africa, even a small future relative increase of 3% (seen for Liberia in scenario RCP 2.6 in year 2050, see appendix) could equate to a substantial increase in the projected absolute number of annual YF deaths.

In all scenarios, there is a high probability that the number of deaths and deaths per capita will increase in the African endemic region. The most marked changes are seen for RCP 8.5, the most severe emission scenario; however, changes are heterogeneous geographically with large proportional increases occurring in Central and East Africa. We expect the number of deaths per year to increase by 10.0% (95% CrI [−0.7, 34.1]) under RCP scenario 2.6 or 40.0% (95% CrI [−2.9, 178.6]) under RCP scenario 8.5 by 2070 (see Table 1 for other values).

We assume that the force of infection changes linearly between 2018 and 2050, and between 2050 and 2070. Video 1 illustrates this by showing posterior samples of the change in deaths by region for all years between 2018 and 2070. For RCP scenario 2.6, deaths largely cease increasing after year 2050, in line with the assumption that RCP 2.6 represents the situation where contributing carbon activities peak by 2030; however, this scenario has been suggested to be ‘unfeasible’ (Mora et al., 2013; van Vliet et al., 2009). In RCP scenario 8.5, carbon contribution activities are assumed to continue increasing throughout the century. A potential impact of this is seen in the number of YF deaths predicted by our model in East and Central Africa, which accelerate after 2050.

Video 1

Download asset

Climate change may affect not only the magnitude of YF disease burden but also its distribution. We find that, through the projected changes in both temperature and rainfall, transmission may change heterogeneously across the region. This is emphasised by their individual contribution; in the appendix, we explore the effects of changes in only temperature or rainfall. This illustrates that whilst temperature change will drive the variation in transmission intensity with rainfall often acting to mitigate, in some countries there can be a ‘perfect storm’ of altering rainfall and temperature leading to increases in transmission that would not occur if only temperature was varying. This may lead to changing priorities with respect to vaccination. However, it is unclear whether the comparatively low proportional increase in burden seen for West Africa is due to more intensive vaccination or due to the limited increase in force of infection. Our results suggest that there could be drastic proportional increases in burden in East and Central Africa that may lead to greater vaccine demand in areas which have previously been of lower risk. Thus, whilst the countries experiencing the highest numbers of deaths will remain high risk, see Figure 1—figure supplement 3 and Figure 1—figure supplement 4 for the median distribution of deaths per year, countries such as Ethiopia and Somalia may become higher priority targets for vaccination.

Our analysis has a number of limitations. In order to utilise emission scenario projections, we were limited to covariates with projections in 2050 and 2070, namely temperature and precipitation. This meant that we adapted our previous best-fit model (Garske et al., 2014) to include temperature range, temperature suitability and precipitation rather than enhanced vegetation and landcover. This change slightly reduced fit quality, giving an AUC of 0.9004 as opposed to to 0.9157 (Gaythorpe et al., 2019). Vegetation is a key factor determining habitat of non-human primates, an element that may not be captured by the temperature suitability index which focuses on the vector A. aegypti. This omission may lead to an overestimation of the future burden as elements such as desertification and the impact of increasing frequencies of forest fires are not considered (Overpeck et al., 1990; Huang et al., 2016; James et al., 2013).

Similarly, whilst the RCP scenarios model socio-economic and land-use changes, we do not explicitly include these aspects here (van Vuuren et al., 2011). As such, we omit the human choices that may affect population distributions and behaviour, for example urbanisation which has been shown to both reduce disease burden (Wood et al., 2017) and increase emergence of arboviruses (Gubler, 2011; Hotez, 2017). In the same way, while our model accounts for migration through use of the UN WPP population data, climate scenario-specific migration is not included in the model. This may mean that we under estimate the potential increases in burden due to increased infringing of human environments on the sylvatic cycle. Projecting these non-linear relationships between human behaviour and transmission would be highly uncertain and is a source of ongoing research.

Vaccination is the main control method for yellow fever and whilst we account for vaccine coverage and efficacy in this mansucript, we do not explicitly propagate uncertainty in vaccination coverage. This will be uncertain not only through data scarcity on vaccination campaign doses, wastage and clustering of doses, but also through the uncertainty in demography. We have presented a comparison of scenarios where, in all cases, vaccination coverage distribution, is held to be the same. As such, whilst we focus on the effect of changing transmission, we will underestimate the uncertainty in our estimates of burden in the future.

Data availability constrains aspects of our modelling approach. We use A. aegypti and YF-specific datasets to inform the thermal response relationships and thus, temperature suitability index. However, some data, such as information on the extrinsic incubation period are severely limited; we use a dataset of experimental results from 1930s (Davis, 1932). These data may be outdated due to current mosquito species potentially adapting to different climates as well as improved experimental procedures. This is a key data gap for YF and new experimental results concerning the extrinsic incubation period could provide valuable insight into the dynamics of the virus in mosquitoes today.

As further experimental data on thermal responses for A. aegypti and other vectors of YF become available, the temperature suitability index developed here will be able to be enhanced. YF is known to have multiple vectors, each contributing to transmission cycles differently (Monath and Vasconcelos, 2015), which are likely to have different thermal responses. Focusing only on the urban vector of YF, as we have in this manuscript, means that we will likely under-estimated the uncertainty in the thermal response of the vectors of YF and thus future projections of burden. Additionally, whilst we have included a relatively detailed relationship between transmission and temperature, we have only assumed a simple relationship with rainfall. Currently models of thermal response for vectors of diseases such as YF are well parametrised with experimental results; however, this is not yet the case for the influence of rainfall on transmission although there are clear links with aspects such as vector breeding. As these relationships are better characterised, we can further refine the relationships in the current work to reflect the more nuanced relationships between temperature, rainfall and transmission.

We focus only on a constant force of infection model which is similar to assuming the majority of transmission occurs as a result of zoonotic spillover. This assumption is supported by recent studies Gaythorpe et al., 2019; however, the urban transmission cycle, driven by A. aegypti plays a crucial role in YF risk and was responsible for recent severe outbreaks such as that in Angola in 2016. Incorporating climate projections into models that examine multiple transmission routes and thermal responses for multiple vectors, would produce a more realistic picture of how the dynamics of this disease may change with climate.

Climate change is projected to have major global impacts on disease distribution and burden (Mordecai et al., 2017; Huber et al., 2018; Kraemer et al., 2015). Here, we examined the specific effects on YF and find that disease burden and deaths are likely to increase heterogeneously across Africa. This emphasises the need to implement and prepare for new vaccination activities, and consolidate existing control strategies in order to mitigate the rising risk from YF. Intervention through vaccination is the gold standard for YF, and new approaches are being implemented with respect to fractional dosing which is a useful resort to respond to urban outbreaks in case of vaccine shortage (Vannice et al., 2018). Yet, vaccination is not the only potentially effective control for YF, with novel vector control measures such as the use of Wolbachia showing promise, and perspectives to improve clinical management or urban resilience (Rocha et al., 2019; World Health Organization, 2017). Finally, in order to monitor and respond to changing transmission patterns, effective and sensitive surveillance will be essential.

A schematic of data sources and models is shown in Figure 1—figure supplement 1.

Datasets

We use a number of data sets to inform both the generalised linear model (GLM) of YF occurrence and the temperature suitability model. Additionally, we rely on estimates of transmission intensity informed by serological studies which are detailed in Gaythorpe et al., 2019 and described below.

YF occurrence

Request a detailed protocol

Details of YF outbreaks occurring from 1984 to present day were collated into a database of occurrence, extended from Garske et al., 2014. These data were collected from the World Health Organisation (WHO) weekly epidemiological record (WER), disease outbreak news (DON), published literature and internal WHO reports (World Health Organization, 2009; World Health Organization, 1996). The database includes all outbreaks recorded for yellow fever and is resolved at province level, any reports that could not be resolved at province level were excluded. Additionally, reports of suspected YF cases were collected in the WHO African Regional Office YF surveillance database (YFSD); this included data from 21 countries in West and Central Africa. The database was based on the broad case definition of fever and jaundice leading to a large proportion of cases attributed to non-YF causes and cross-reactivity with other flaviviruses was not considered. However, the incidence of suspected cases can be used as a measure of surveillance effort and is included as a covariate in the generalised linear model. We assume this to be constant over time due to scarcity of data on the subject.

YF serological status

Request a detailed protocol

Surveys of seroprevalence were conducted in Central and East Africa. We use these to assess transmission intensity in specific regions of the African endemic zone. The current study includes surveys from published sources (Diallo et al., 2014; Kuniholm et al., 2006; Merlin et al., 1986; Omilabu et al., 1990; Tsai et al., 1987; Werner and Huber, 1984) and unpublished surveys from East African countries conducted between 2012 and 2015 as part of the YF risk assessment process (Mengesha Tsegaye et al., 2018). The surveys were included only if they represent the population at steady state, as such outbreak investigations were omitted (Garske et al., 2014). Additionally, in the majority of surveys, vaccinated individuals were not included; however, in South Cameroon, vaccination status is unclear and so we fit an additional vaccine factor for this survey. Summary details of the seroprevalence studies are included in the apendix.

Past vaccination coverage and demography

Request a detailed protocol

Vaccination coverage is estimated using data on historic large-scale mass vaccination activities taking place between 1940 and 1960 (Durieux, 1956; Moreau et al., 1999), routine infant immunisation reported by the WHO and UNICEF estimates of National Immunization Coverage (WUENIC) (World Health Organization/ UNICEF, 2015), outbreak response campaigns from 1970 onwards which are detailed in the WHO WER and DON (World Health Organization, 2009; World Health Organization, 1996) and recent preventive mass-vaccination campaigns carried out as part of the yellow fever initiative (World Health Organisation, 2016). The coverage is estimated with the methodology of Garske et al. and Hamlet et al. and is visualised in the polici shiny application (Garske et al., 2014; Hamlet et al., 2018a). The application provides vaccination coverage estimates at province level for 34 endemic countries in Africa which can be downloaded for years between 1940 and 2050. We assume all targeted age groups have an equal chance of vaccination irrespective of vaccination staus.

Demography is obtained from the UN World Population Prospect (UN WPP) (DoE United Nations, 2017). We dis-aggregate this to province level by combining it with estimates of spatial population distributions from LandScan 2014 (Dobson et al., 2000). This allows us to estimate population sizes at province level for each year of interest assuming that the age structure is relatively similar across all provinces in each country.

Environmental and climate projections

Request a detailed protocol

We use three main environmental covariates within the generalised linear model of YF occurrence: mean annual rainfall, average temperature and temperature range, shown in Figure 6 and listed in Table 2. These are gridded data at various resolutions, ranging from approximately 1 km to 10 km, which we average at the first administrative unit level (Nasa LPD, 2001; Xie and Arkin, 1996; Hijmans et al., 2004).

Table 2

Covariate	Interpretation
log(survey quality)	Log of the survey quality for countries in YFSD.
adm05	Country factors for countries not in YFSD.
longitude	Longitude of province centroid
temperature suitability	Temperature suitability at average suitability of province.
temperature range	Temperature range in province.
rainfall	Mean Precipitation in province.
log(pop)	Log of the human population size of the province

Figure 6

Download asset Open asset

Projected temperature and rainfall changes under climate change scenarios were obtained from worldclim version 1.4 (Hijmans et al., 2005; Fick and Hijmans, 2017). These data provided the 5th Intergovernmental panel on climate change (IPPC5) climate projections for four Representative Concentration Pathways (RCPs): 2.6, 4.5, 6.0 and 8.5 (van Vuuren et al., 2011). The different RCPs indicate different possible emission scenarios and represent the resulting radiative forcing in 2100, measured in W/m² or watts per square metre, see Table 3 for further information (Stocker, 2013). Each scenario is assumed to peak at a different times, with emissions peaking between 2010 and 2020 for RCP 2.6, but rising throughout the century for RCP 8.5. Projections of the mean global temperature rise by 2046–2065 are 1 or 2 °C for RCPs 2.6 or 8.5, respectively, compared to pre-industrial levels of the 1880s. By the end of the century, these projections suggest a rise of 1 [0.3 to 1.7] or 3.9 [2.6 to 4.8]°C for RCPs 2.6 or 8.5 (Stocker, 2013; Rogelj et al., 2012). Current warming is estimated to be 0.85 °C since pre-industrial levels (Stocker, 2013). Based on current commitments through aspects such as the Paris agreement, scenarios where temperatures are expected to rise by more than 3 °C have been suggested to be most likely (Sanford et al., 2014). As such, a recent study omitted the RCP 2.6 scenario as it is unlikely now to occur (Mora et al., 2013; van Vliet et al., 2009).

Table 3

Scenario	Temperature rise (°C) [range]	CO2 concentrations (ppm)
RCP 2.6	1 [0.3 to 1.7]	421
RCP 4.5	1.8 [1.1 to 2.6]	538
RCP 6.0	2.2 [1.4 to 3.1]	670
RCP 8.5	3.7 [2.6 to 4.8]	936

Projected mean rainfall, maximum temperature and minimum temperature are available for each RCP scenario in years 2050 and 2070. We take the midpoint and range of the temperature as inputs for the model of YF occurrence, where the midpoint temperature is used to calculate the temperature suitability index.

We do not model changes in climate prior to 2018, instead using Worldclim baseline estimates described as representative of conditions from 1960 to 1990 (Hijmans et al., 2005).

Temperature suitability

Request a detailed protocol

We estimate the components of the temperature suitability index from YF-specific sources of information on extrinsic incubation period, vector mortality and bite rate for A. aegypti, the urban vector of YF (Davis, 1932; Tesla et al., 2018; Hamlet et al., 2018b; Mordecai et al., 2017). The extrinsic incubation period was estimated from the experimental results of Davis which were calculated specifically for YF in A. aegypti (Davis, 1932). We included bite rate data from both Mordecai et al., 2017 and Martens, 1998 which both describe A. aegypti. Finally, vector mortality was estimated from the experimental data of Tesla et al., 2018. Where data was provided in figure form, plots were digitised to extract the information. All data used for fitting the temperature suitability model are made available in the GitHub repo (https://github.com/mrc-ide/YF_climateChange; Gaythorpe, 2020; copy archived at https://github.com/elifesciences-publications/YF_climateChange). Whilst we focus only on thermal response of the urban vector of YF due to data availability, we estimate the thermal response models within a Bayesian hierarchical framework in order to capture some of the uncertainty that we miss from examining one vector species.

Public repository data: Vaccination coverage: coverage is available to download from the PoLiCi shiny app : https://shiny.dide.imperial.ac.uk/polici/. Serology surveys: There are seven published surveys used, available at DOI:10.1016/0147-9571(90)90521-T, DOI:10.1093/trstmh/tru086, DOI: 10.1186/s12889-018-5726-9, DOI: 10.4269/ajtmh.2006.74.1078, PMID: 3501739, PMID: 4004378, PMID: 3731366 Demographic data: Population level data was obtained from UN WPP https://population.un.org/wpp/, this was disaggregated using Landscan 2014 data https://landscan.ornl.gov/landscan-data-availability. Environmental data: This was obtained from LP DAAC: https://lpdaac.usgs.gov/ and worldclim http://www.worldclim.org/ Yellow fever outbreaks: These were compiled from the WHO weekly epidemiologic record and disease outbreak news https://www.who.int/wer/en/ and https://www.who.int/csr/don/en/. Data elsewhere: The data from the WHO YF surveillance database and from recent serological surveys from WHO member states in Africa underlying the results presented in the study are available from World Health Organization (contact: William Perea, pereaw@who.int or Laurence Cibrelus, cibrelusl@who.int or Jennifer Horton, jhorton@who.int).

1. 1. Andrieu C
   2. Thoms J

   (2008) A tutorial on adaptive MCMC

   Statistics and Computing 18:343–373.

   https://doi.org/10.1007/s11222-008-9110-y

   • Google Scholar
2. 1. Barrett AD
   2. Higgs S

   (2007) Yellow fever: a disease that has yet to be conquered

   Annual Review of Entomology 52:209–229.

   https://doi.org/10.1146/annurev.ento.52.110405.091454

   • PubMed
   • Google Scholar
3. 1. Brady OJ
   2. Johansson MA
   3. Guerra CA
   4. Bhatt S
   5. Golding N
   6. Pigott DM
   7. Delatte H
   8. Grech MG
   9. Leisnham PT
   10. Maciel-de-Freitas R
   11. Styer LM
   12. Smith DL
   13. Scott TW
   14. Gething PW
   15. Hay SI

   (2013) Modelling adult aedes aegypti and aedes albopictus survival at different temperatures in laboratory and field settings

   Parasites & Vectors 6:351.

   https://doi.org/10.1186/1756-3305-6-351

   • PubMed
   • Google Scholar
4. 1. Brady OJ
   2. Golding N
   3. Pigott DM
   4. Kraemer MU
   5. Messina JP
   6. Reiner RC
   7. Scott TW
   8. Smith DL
   9. Gething PW
   10. Hay SI

   (2014) Global temperature constraints on aedes aegypti and ae albopictus persistence and competence for dengue virus transmission

   Parasites & Vectors 7:338.

   https://doi.org/10.1186/1756-3305-7-338

   • PubMed
   • Google Scholar
5. 1. Cowlishaw G
   2. Hacker JE

   (1997) Distribution, diversity, and latitude in african primates

   The American Naturalist 150:505–512.

   https://doi.org/10.1086/286078

   • PubMed
   • Google Scholar
6. 1. Davis NC

   (1932) The effect of various temperatures in modifying the extrinsic incubation period of the yellow fever virus in aëdes aegypti*

   American Journal of Epidemiology 16:163–176.

   https://doi.org/10.1093/oxfordjournals.aje.a117853

   • Google Scholar
7. 1. Diallo M
   2. Janusz K
   3. Lewis RF
   4. Manengu C
   5. Sall A
   6. Staples JE

   (2014) Rapid assessment of yellow fever viral activity in the central african republic

   Transactions of the Royal Society of Tropical Medicine and Hygiene 108:608–615.

   https://doi.org/10.1093/trstmh/tru086

   • Google Scholar
8. 1. Dobson JE
   2. Bright EA
   3. Coleman PR

   (2000)

   LandScan: a global population database for estimating populations at risk

   Photogrammetric Engineering and Remote Sensing 66:849–857.

   • Google Scholar
9. Report

   1. DoE United Nations

   (2017) Projections World Population Prospects

   United Nations.

   https://www.un.org/development/desa/publications/world-population-prospects-the-2017-revision.html

   • Google Scholar
10. 1. Dunning CM
    2. Black E
    3. Allan RP

    (2018) Later wet seasons with more intense rainfall over africa under future climate change

    Journal of Climate 31:9719–9738.

    https://doi.org/10.1175/JCLI-D-18-0102.1

    • Google Scholar
11. 1. Durieux C

    (1956)

    Mass yellow fever vaccination in french africa south of the sahara

    Yellow Fever Vaccination, Monograph Series 30:115–121.

    • Google Scholar
12. 1. Fick SE
    2. Hijmans RJ

    (2017) WorldClim 2: new 1‐km spatial resolution climate surfaces for global land Areas

    International Journal of Climatology 37:4302–4315.

    https://doi.org/10.1002/joc.5086

    • Google Scholar
13. Software

    1. Garnier S

    (2018) Viridis: Default Color Maps from ’Matplotlib’, version 0.5.1

    CRAN.

    https://CRAN.R-project.org/package=viridis
14. 1. Garske T
    2. Ferguson NM
    3. Ghani AC

    (2013) Estimating air temperature and its influence on malaria transmission across africa

    PLOS ONE 8:e56487.

    https://doi.org/10.1371/journal.pone.0056487

    • PubMed
    • Google Scholar
15. 1. Garske T
    2. Van Kerkhove MD
    3. Yactayo S
    4. Ronveaux O
    5. Lewis RF
    6. Staples JE
    7. Perea W
    8. Ferguson NM
    9. Yellow Fever Expert Committee

    (2014) Yellow fever in africa: estimating the burden of disease and impact of mass vaccination from outbreak and serological data

    PLOS Medicine 11:e1001638.

    https://doi.org/10.1371/journal.pmed.1001638

    • PubMed
    • Google Scholar
16. 1. Gaythorpe KAM
    2. Jean K
    3. Cibrelus L
    4. Garske T

    (2019) Quantifying model evidence for yellow fever transmission routes in africa

    PLOS Computational Biology 15:e1007355.

    https://doi.org/10.1371/journal.pcbi.1007355

    • PubMed
    • Google Scholar
17. Software

    1. Gaythorpe K

    (2020) YF_climateChange, version 22c19c2

    GitHub.

    https://github.com/mrc-ide/YF_climateChange
18. 1. Gubler DJ

    (2011) Dengue, urbanization and globalization: the unholy trinity of the 21st century

    Tropical Medicine and Health 39:S3–S11.

    https://doi.org/10.2149/tmh.2011-S05

    • Google Scholar
19. Report

    1. Hamlet A
    2. Jean K
    3. Garske T

    (2018a) POLICI: POpulation Level Immunization Coverage Imperial

    Imperial college.

    https://polici.shinyapps.io/yellow_fever_africa/

    • Google Scholar
20. 1. Hamlet A
    2. Jean K
    3. Perea W
    4. Yactayo S
    5. Biey J
    6. Van Kerkhove M
    7. Ferguson N
    8. Garske T

    (2018b) The seasonal influence of climate and environment on yellow fever transmission across africa

    PLOS Neglected Tropical Diseases 12:e0006284.

    https://doi.org/10.1371/journal.pntd.0006284

    • PubMed
    • Google Scholar
21. Software

    1. Hijmans RJ
    2. Cameron SE
    3. Parra JL
    4. Jones PG
    5. Jarvis A

    (2004) The Worldclim Interpolated Global Terrestrial Climate Surfaces, version 1.3

    Worldclim.

    https://worldclim.org/data/v1.4/methods.html
22. 1. Hijmans RJ
    2. Cameron SE
    3. Parra JL
    4. Jones PG
    5. Jarvis A

    (2005) Very high resolution interpolated climate surfaces for global land Areas

    International Journal of Climatology 25:1965–1978.

    https://doi.org/10.1002/joc.1276

    • Google Scholar
23. 1. Hotez PJ

    (2017) Global urbanization and the neglected tropical diseases

    PLOS Neglected Tropical Diseases 11:e0005308.

    https://doi.org/10.1371/journal.pntd.0005308

    • PubMed
    • Google Scholar
24. 1. Huang J
    2. Yu H
    3. Guan X
    4. Wang G
    5. Guo R

    (2016) Accelerated dryland expansion under climate change

    Nature Climate Change 6:166–171.

    https://doi.org/10.1038/nclimate2837

    • Google Scholar
25. 1. Huang J
    2. Ling CX

    (2005) Using AUC and accuracy in evaluating learning algorithms

    IEEE Transactions on Knowledge and Data Engineering 17:299–310.

    https://doi.org/10.1109/TKDE.2005.50

    • Google Scholar
26. 1. Huber JH
    2. Childs ML
    3. Caldwell JM
    4. Mordecai EA

    (2018) Seasonal temperature variation influences climate suitability for dengue, Chikungunya, and zika transmission

    PLOS Neglected Tropical Diseases 12:e0006451.

    https://doi.org/10.1371/journal.pntd.0006451

    • PubMed
    • Google Scholar
27. 1. Ingelbeen B
    2. Weregemere NA
    3. Noel H
    4. Tshapenda GP
    5. Mossoko M
    6. Nsio J
    7. Ronsse A
    8. Ahuka-Mundeke S
    9. Cohuet S
    10. Kebela BI

    (2018) Urban yellow fever outbreak-Democratic republic of the Congo, 2016: towards more rapid case detection

    PLOS Neglected Tropical Diseases 12:e0007029.

    https://doi.org/10.1371/journal.pntd.0007029

    • PubMed
    • Google Scholar
28. 1. James R
    2. Washington R
    3. Rowell DP

    (2013) Implications of global warming for the climate of african rainforests

    Philosophical Transactions of the Royal Society B: Biological Sciences 368:20120298.

    https://doi.org/10.1098/rstb.2012.0298

    • Google Scholar
29. 1. Jean K
    2. Hamlet A
    3. Benzler J
    4. Cibrelus L
    5. Gaythorpe KAM
    6. Sall A
    7. Ferguson NM
    8. Garske T

    (2020) Eliminating yellow fever epidemics in africa: vaccine demand forecast and impact modelling

    PLOS Neglected Tropical Diseases 14:e0008304.

    https://doi.org/10.1371/journal.pntd.0008304

    • PubMed
    • Google Scholar
30. 1. Johansson MA
    2. Arana-Vizcarrondo N
    3. Biggerstaff BJ
    4. Staples JE

    (2010) Incubation periods of yellow fever virus

    The American Journal of Tropical Medicine and Hygiene 83:183–188.

    https://doi.org/10.4269/ajtmh.2010.09-0782

    • PubMed
    • Google Scholar
31. 1. Johansson MA
    2. Vasconcelos PF
    3. Staples JE

    (2014) The whole iceberg: estimating the incidence of yellow fever virus infection from the number of severe cases

    Transactions of the Royal Society of Tropical Medicine and Hygiene 108:482–487.

    https://doi.org/10.1093/trstmh/tru092

    • PubMed
    • Google Scholar
32. 1. Kharin VV
    2. Zwiers FW
    3. Zhang X
    4. Wehner M

    (2013) Changes in temperature and precipitation extremes in the CMIP5 ensemble

    Climatic Change 119:345–357.

    https://doi.org/10.1007/s10584-013-0705-8

    • Google Scholar
33. 1. Kraemer MU
    2. Sinka ME
    3. Duda KA
    4. Mylne AQ
    5. Shearer FM
    6. Barker CM
    7. Moore CG
    8. Carvalho RG
    9. Coelho GE
    10. Van Bortel W
    11. Hendrickx G
    12. Schaffner F
    13. Elyazar IR
    14. Teng HJ
    15. Brady OJ
    16. Messina JP
    17. Pigott DM
    18. Scott TW
    19. Smith DL
    20. Wint GR
    21. Golding N
    22. Hay SI

    (2015) The global distribution of the arbovirus vectors aedes aegypti and ae albopictus

    eLife 4:e08347.

    https://doi.org/10.7554/eLife.08347

    • PubMed
    • Google Scholar
34. 1. Kuniholm MH
    2. Wolfe ND
    3. Huang CY
    4. Mpoudi-Ngole E
    5. Tamoufe U
    6. LeBreton M
    7. Burke DS
    8. Gubler DJ

    (2006) Seroprevalence and distribution of Flaviviridae, togaviridae, and Bunyaviridae arboviral infections in rural cameroonian adults

    The American Journal of Tropical Medicine and Hygiene 74:1078–1083.

    https://doi.org/10.4269/ajtmh.2006.74.1078

    • PubMed
    • Google Scholar
35. Book

    1. Macdonald G

    (1957)

    The Epidemiology and Control of Malaria and Others

    CABI.

    • Google Scholar
36. 1. Martens WJ

    (1998) Health impacts of climate change and ozone depletion: an ecoepidemiologic modeling approach

    Environmental Health Perspectives 106:241–251.

    https://doi.org/10.1289/ehp.98106s1241

    • PubMed
    • Google Scholar
37. Report

    1. Masson-Delmotte V
    2. Zhai P
    3. Pörtner HO
    4. Roberts D
    5. Skea J
    6. Shukla PR
    7. Pirani A

    (2018) IPCC, 2018: Summary for Policymakers, Global Warming of 1

    IPCC.

    https://www.ipcc.ch/sr15/chapter/spm/

    • Google Scholar
38. 1. McKinley TJ
    2. Ross JV
    3. Deardon R
    4. Cook AR

    (2014) Simulation-based bayesian inference for epidemic models

    Computational Statistics & Data Analysis 71:434–447.

    https://doi.org/10.1016/j.csda.2012.12.012

    • Google Scholar
39. 1. Mengesha Tsegaye M
    2. Beyene B
    3. Ayele W
    4. Abebe A
    5. Tareke I
    6. Sall A
    7. Yactayo S
    8. Shibeshi ME
    9. Staples E
    10. Belay D
    11. Lilay A
    12. Alemu A
    13. Alemu E
    14. Kume A
    15. H/Mariam A
    16. Ronveaux O
    17. Tefera M
    18. Kassa W
    19. Bekele Weyessa A
    20. Jima D
    21. Kebede A
    22. Tayachew A

    (2018) Sero-prevalence of yellow fever and related flavi viruses in Ethiopia: a public health perspective

    BMC Public Health 18:1011.

    https://doi.org/10.1186/s12889-018-5726-9

    • PubMed
    • Google Scholar
40. 1. Merlin M
    2. Josse R
    3. Kouka-Bemba D
    4. Meunier D
    5. Senga J
    6. Simonkovich E
    7. Malonga JR
    8. Manoukou F
    9. Georges AJ

    (1986)

    Evaluation of immunological and entomotological indices of yellow fever in Pointe-Noire, people’s Republic of Congo

    Bulletin De La Societe De Pathologie Exotique Et De Ses Filiales 79:199–206.

    • PubMed
    • Google Scholar
41. 1. Monath TP
    2. Vasconcelos PF

    (2015) Yellow fever

    Journal of Clinical Virology 64:160–173.

    https://doi.org/10.1016/j.jcv.2014.08.030

    • PubMed
    • Google Scholar
42. 1. Mora C
    2. Frazier AG
    3. Longman RJ
    4. Dacks RS
    5. Walton MM
    6. Tong EJ
    7. Sanchez JJ
    8. Kaiser LR
    9. Stender YO
    10. Anderson JM
    11. Ambrosino CM
    12. Fernandez-Silva I
    13. Giuseffi LM
    14. Giambelluca TW

    (2013) The projected timing of climate departure from recent variability

    Nature 502:183–187.

    https://doi.org/10.1038/nature12540

    • PubMed
    • Google Scholar
43. 1. Mordecai EA
    2. Cohen JM
    3. Evans MV
    4. Gudapati P
    5. Johnson LR
    6. Lippi CA
    7. Miazgowicz K
    8. Murdock CC
    9. Rohr JR
    10. Ryan SJ
    11. Savage V
    12. Shocket MS
    13. Stewart Ibarra A
    14. Thomas MB
    15. Weikel DP

    (2017) Detecting the impact of temperature on transmission of zika, dengue, and Chikungunya using mechanistic models

    PLOS Neglected Tropical Diseases 11:e0005568.

    https://doi.org/10.1371/journal.pntd.0005568

    • PubMed
    • Google Scholar
44. 1. Moreau JP
    2. Girault G
    3. Dramé I
    4. Perraut R

    (1999)

    Reemergence of yellow fever in west africa: lessons from the past, advocacy for a control program

    Bulletin De La Societe De Pathologie Exotique 92:333–336.

    • PubMed
    • Google Scholar
45. Report

    1. Nasa LPD

    (2001) NASA Land Processes Distributed Active Archive Center (LP DAAC) USGS/Earth Resources Observation and Science (EROS) Center

    NCRIS.

    https://portal.tern.org.au/nasa-land-processes-lp-daac/20991

    • Google Scholar
46. 1. Omilabu SA
    2. Adejumo JO
    3. Olaleye OD
    4. Fagbami AH
    5. Baba SS

    (1990) Yellow fever haemagglutination-inhibiting, neutralising and IgM antibodies in vaccinated and unvaccinated residents of Ibadan, Nigeria

    Comparative Immunology, Microbiology and Infectious Diseases 13:95–100.

    https://doi.org/10.1016/0147-9571(90)90521-T

    • PubMed
    • Google Scholar
47. 1. Overpeck JT
    2. Rind D
    3. Goldberg R

    (1990) Climate-induced changes in forest disturbance and vegetation

    Nature 343:51–53.

    https://doi.org/10.1038/343051a0

    • Google Scholar
48. Software

    1. R Development Core Team

    (2014) R: A Language and Environment for Statistical Computing

    R Foundation for Statistical Computing, Vienna, Austria.

    http://www.R-project.org/
49. 1. Reinhold J
    2. Lazzari C
    3. Lahondère C

    (2018) Effects of the environmental temperature on aedes aegypti and aedes albopictus mosquitoes: a review

    Insects 9:158.

    https://doi.org/10.3390/insects9040158

    • Google Scholar
50. 1. Roberts GO
    2. Rosenthal JS

    (2009) Examples of adaptive MCMC

    Journal of Computational and Graphical Statistics 18:349–367.

    https://doi.org/10.1198/jcgs.2009.06134

    • Google Scholar
51. 1. Rocha MN
    2. Duarte MM
    3. Mansur SB
    4. Silva B
    5. Pereira TN
    6. Adelino TÉR
    7. Giovanetti M
    8. Alcantara LCJ
    9. Santos FM
    10. Costa VRM
    11. Teixeira MM
    12. Iani FCM
    13. Costa VV
    14. Moreira LA

    (2019) Pluripotency of Wolbachia against arboviruses: the case of yellow fever

    Gates Open Research 3:161.

    https://doi.org/10.12688/gatesopenres.12903.2

    • PubMed
    • Google Scholar
52. 1. Rogelj J
    2. Meinshausen M
    3. Knutti R

    (2012) Global warming under old and new scenarios using IPCC climate sensitivity range estimates

    Nature Climate Change 2:248–253.

    https://doi.org/10.1038/nclimate1385

    • Google Scholar
53. 1. Ryan SJ
    2. Carlson CJ
    3. Mordecai EA
    4. Johnson LR

    (2019) Global expansion and redistribution of Aedes-borne virus transmission risk with climate change

    PLOS Neglected Tropical Diseases 13:e0007213.

    https://doi.org/10.1371/journal.pntd.0007213

    • PubMed
    • Google Scholar
54. 1. Sanford T
    2. Frumhoff PC
    3. Luers A
    4. Gulledge J

    (2014) The climate policy narrative for a dangerously warming world

    Nature Climate Change 4:164–166.

    https://doi.org/10.1038/nclimate2148

    • Google Scholar
55. 1. Sherlock C
    2. Thiery AH
    3. Roberts GO
    4. Rosenthal JS

    (2015) On the efficiency of pseudo-marginal random walk metropolis algorithms

    The Annals of Statistics 43:238–275.

    https://doi.org/10.1214/14-AOS1278

    • Google Scholar
56. Report

    1. Stocker TF

    (2013) The Physical Science Basis. Contribution of Working Group I to the Fifth Assessment IPCC, 2013: Summary for Policymakers Climate Change

    Report of the Intergovernmental Panel on Climate Change.

    https://www.ipcc.ch/report/ar5/wg1/

    • Google Scholar
57. Preprint

    1. Tennant W
    2. McKinley T

    (2019) Inferring the ecological drivers of arboviral outbreaks

    bioRxiv.

    https://doi.org/10.1101/632133

    • Google Scholar
58. 1. Tesla B
    2. Demakovsky LR
    3. Mordecai EA
    4. Ryan SJ
    5. Bonds MH
    6. Ngonghala CN
    7. Brindley MA
    8. Murdock CC

    (2018) Temperature drives zika virus transmission: evidence from empirical and mathematical models

    Proceedings of the Royal Society B: Biological Sciences 285:20180795.

    https://doi.org/10.1098/rspb.2018.0795

    • Google Scholar
59. 1. Tjaden NB
    2. Caminade C
    3. Beierkuhnlein C
    4. Thomas SM

    (2018) Mosquito-Borne diseases: advances in modelling Climate-Change impacts

    Trends in Parasitology 34:227–245.

    https://doi.org/10.1016/j.pt.2017.11.006

    • PubMed
    • Google Scholar
60. 1. Tsai TF
    2. Lazuick JS
    3. Ngah RW
    4. Mafiamba PC
    5. Quincke G
    6. Monath TP

    (1987)

    Investigation of a possible yellow fever epidemic and serosurvey for Flavivirus infections in northern Cameroon, 1984

    Bulletin of the World Health Organization 65:855.

    • PubMed
    • Google Scholar
61. 1. van Vliet J
    2. den Elzen MGJ
    3. van Vuuren DP

    (2009) Meeting radiative forcing targets under delayed participation

    Energy Economics 31:S152–S162.

    https://doi.org/10.1016/j.eneco.2009.06.010

    • Google Scholar
62. 1. van Vuuren DP
    2. Edmonds J
    3. Kainuma M
    4. Riahi K
    5. Thomson A
    6. Hibbard K
    7. Hurtt GC
    8. Kram T
    9. Krey V
    10. Lamarque J-F
    11. Masui T
    12. Meinshausen M
    13. Nakicenovic N
    14. Smith SJ
    15. Rose SK

    (2011) The representative concentration pathways: an overview

    Climatic Change 109:5–31.

    https://doi.org/10.1007/s10584-011-0148-z

    • Google Scholar
63. 1. Vannice K
    2. Wilder-Smith A
    3. Hombach J

    (2018) Fractional-Dose yellow fever vaccination - Advancing the evidence base

    New England Journal of Medicine 379:603–605.

    https://doi.org/10.1056/NEJMp1803433

    • PubMed
    • Google Scholar
64. Website

    1. VIMC

    (2019) Vaccine impact modelling consortium

    Accessed February 2, 2020.

    https://www.vaccineimpact.org
65. 1. Werner GT
    2. Huber HC

    (1984)

    Prevalence of yellow fever antibodies in north Zaire

    In Annales De La Societe Belge De Medecine Tropicale 65:91–93.

    • Google Scholar
66. Book

    1. Wickham H

    (2016)

    Ggplot2: Elegant Graphics for Data Analysis

    New York: Springer-Verlag.

    • Google Scholar
67. 1. Wilder-Smith A
    2. Monath TP

    (2017) Responding to the threat of urban yellow fever outbreaks

    The Lancet Infectious Diseases 17:248–250.

    https://doi.org/10.1016/S1473-3099(16)30588-6

    • PubMed
    • Google Scholar
68. Software

    1. Wilke C

    (2018) ggridges: Ridgeline Plots in ’ggplot2’, version 0.5.2

    GitHub.

    https://CRAN.R-project.org/package=ggridges
69. 1. Wood CL
    2. McInturff A
    3. Young HS
    4. Kim D
    5. Lafferty KD

    (2017) Human infectious disease burdens decrease with urbanization but not with biodiversity

    Philosophical Transactions of the Royal Society B: Biological Sciences 372:20160122.

    https://doi.org/10.1098/rstb.2016.0122

    • Google Scholar
70. Report

    1. World Health Organisation

    (2016) The Yellow Fever Initiative: An Introduction

    WHO.

    https://www.who.int/csr/disease/yellowfev/introduction/en/

    • Google Scholar
71. Website

    1. World Health Organisation

    (2018) Climate change and health

    Accessed February 2, 2020.

    https://www.who.int/en/news-room/fact-sheets/detail/climate-change-and-health
72. Website

    1. World Health Organization

    (1996) Disease outbreak news since

    Accessed February 2, 2020.

    http://www.who.int/csr/don/en/
73. Report

    1. World Health Organization

    (2009) Weekly Epidemiological Record (WER) Issues

    WHO.

    https://www.who.int/wer/en/

    • Google Scholar
74. Conference

    1. World Health Organization

    (2017)

    Eliminate yellow fever epidemics (Eye): A global strategy, 2017–2026–Éliminer Les Épidémies De Fièvre Jaune(EYE): Une Stratégie Mondiale, 2017-2026

    Weekly Epidemiological Record= Relevé Épidémiologique Hebdomadaire. pp. 193–204.

    • Google Scholar
75. Website

    1. World Health Organization/ UNICEF

    (2015) Estimates of national immunization coverage (WUENIC) 2015

    Accessed February 4, 2020.

    http://apps.who.int/immunization_monitoring/globalsummary/timeseries/tswucoverageyfv.html
76. 1. Xie P
    2. Arkin PA

    (1996) Analyses of global monthly precipitation using gauge observations, satellite estimates, and numerical model predictions

    Journal of Climate 9:840–858.

    https://doi.org/10.1175/1520-0442(1996)009<0840:AOGMPU>2.0.CO;2

    • Google Scholar

Acceptance summary:

This is an important piece of work, of both clear public health importance and of scientific interest, that will guide policies and programmes against Yellow Fever across sub-Saharan Africa in the years ahead due to the potential impact of climate change on the intensity and distribution of yellow fever transmission and disease burden there.

Decision letter after peer review:

Thank you for submitting your article "The effect of climate change on Yellow Fever disease burden in Africa" for consideration by eLife. Your article has been reviewed by three peer reviewers, one of whom is a member of our Board of Reviewing Editors, and the evaluation has been overseen by Eduardo Franco as the Senior Editor. The reviewers have opted to remain anonymous.

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

Summary:

The reviewers agreed that this is an important piece of work – of both clear public health importance and of scientific interest – that will guide policies and programmes against Yellow Fever across sub-Saharan Africa in the years ahead due to the potential impact of climate change on the intensity and distribution of yellow fever transmission and disease burden there.

Essential revisions:

1) The description of approach would benefit greatly from a relatively simple schematic diagram to make it very clear how the different data (seroprevalence; occurrence etc.) and modelling components (temperature suitability model; seroprevalence/transmission/FOI model; covariate-driven GLM) link together.

2) Framework for inferring contemporary pattern of transmission and burden

- The authors should please provide further details on how the YF occurrence database was compiled and contents of the database. For example, what was the inclusion/exclusion criteria for an occurrence? Was potential cross-reactivity with other flaviviruses considered? For data extracted from WERs, DONs and literature – at the very least a reference list should be provided.

- Can you present a map showing areas where you infer FOI based on serological surveys and where it is inferred as a function of the GLM 'probability of a report' estimate.

- Some more narrative would also be helpful in explaining conceptually how the FOI-to-probability of occurrence component was achieved. It appears that the ingredients you have in each province for which no sero survey exists are (i) the GLM-derived 'probability of > = 1 case in a given province'; (ii) population; (iii) an idea of the probability that a case is detected given that it occurs. Your objective is to infer FOI. It is not clear if you do this within your Bayesian inference framework-thereby somehow borrowing strength from those provinces for which you have both sero surveys and GLM estimates, or, if this is essentially an arithmetic transform that converts the GLM output to an FOI estimate. It appears to be the latter-and if so it is then not clear how robust this transform is, how you infer probability of detection etc. Presumably you could also report on the degree of alignment between your GLM outputs and you FOI calculations in provinces that do have sero-surveys, as an indicator of how well this approach will then work in provinces without surveys.

- On YF burden from zoonotic/intermediate transmission versus urban transmission cycles ("the transmission intensity is assumed to be a static force of infection, akin to the assumption that most YF infections occur as a result of sylvatic spillover." And also in the Discussion.) Instead of only touching on it in the Discussion perhaps it should be highlighted that the estimate of FOI is limited to burden from zoonotic spillover earlier/throughout the manuscript. Consider describing results as estimated effect of climate change on zoonotic/sylvatic YF burden?

- How appropriate is it to use temperature suitability pertaining only to the urban vector rather than vectors involved in the intermediate and sylvatic transmission cycles. Also, whether occurrences from urban outbreaks should be excluded from the occurrence database?

- Can you clarify also what is the implication of focusing on Ae. aegypti (subsection “Temperature suitability”) instead of other vectors?

3) Framework for inferring contemporary relationship with rainfall and temperature

- There is a much greater emphasis/level of sophistication around temperature effects than around rainfall. For temperature, the authors recognise the highly non-linear response via a mechanistic model capturing deterministically the temperature effects on different components of the transmission cycle. For rainfall, however, the presumably similarly complex mechanisms of impact on aedes breeding, vector densities, interactions with humans etc., are represented much more crudely via a single covariate in a GLM, with implicit assumptions around linearity. Some justification on this would be useful.

- While the paper builds on previous work, the very small number of covariates in the GLM is surprising and needs justification in this current paper. Is there no rationale for including, for example, human population density, or metrics of landscape proxies for non-human primate habitats etc?

- Further, could there be a bit more discussion of the reasonableness and motivation for omitting land cover and vegetation index? This would seem to be important given that at least in Brazil there's a very strong effect of proximity to forest?

4) Framework for inferring impact of changing climate

- Given it would be straightforward in your current framework, please consider the utility of running projections with only the effects of temperature change and only the effects of rainfall change? The rationale being that you have differing levels of sophistication in your ability to link these to YF risk; that there are differing levels of certainty in the climate projections for these two aspects; and that you could provide a richer interpretation of their respective contributions to elevated risk.

- You elude to it in the Discussion, but more consideration of what you do not include in your framework is crucial here. Temperature/rainfall changes may exert upward pressure on transmission and burden – but how does the magnitude of this effect compare to, for example, a modest increase in vaccination coverage, or changes in exposure due to urbanisation or modernisation of housing etc. How might global change impact other components of the transmission cycle such as the col-location of humans an NHP zoonotic reservoir species?

5) Impact of vaccination

- It is not clear from this paper or the references precisely what vaccination coverage data was compiled and how it was used in vaccination coverage estimates. The data and methods are not fully described in this paper or the referenced papers. It seems that uncertainty in vaccination coverage estimates was considered in Garske et al., 2014, by exploring alternate vaccination targeting assumptions. For the analysis in this manuscript, it appears that only the scenario where supplemental vaccination campaigns target unvaccinated individuals is considered? Lessler and colleagues (journal.pmed.1001110) have previously described issues with assumptions about vaccination targeting when making coverage estimates. It is currently not clear whether any uncertainties were propagated through the analyses presented in this paper. It is important to explore the possible impact of these (potentially large) uncertainties on the results.

6) Impact of future population changes

- Why were demographic conditions held constant? There ought to be projections for future population growth (and ageing?) for these countries that would give a more accurate representation of future disease burden.

- Possible to forecast total numbers of deaths as well as the country specific ones?

Essential revisions:

1) The description of approach would benefit greatly from a relatively simple schematic diagram to make it very clear how the different data (seroprevalence; occurrence etc.) and modelling components (temperature suitability model; seroprevalence/transmission/FOI model; covariate-driven GLM) link together.

Thank you, we have included a schematic of this information (Figure 1—figure supplement 1).

2) Framework for inferring contemporary pattern of transmission and burden

- The authors should please provide further details on how the YF occurrence database was compiled and contents of the database. For example, what was the inclusion/exclusion criteria for an occurrence? Was potential cross-reactivity with other flaviviruses considered? For data extracted from WERs, DONs and literature – at the very least a reference list should be provided.

We have included further details in the methodology to expand on how this was collated (section 2.1.1). This database was detailed first in Garske et al., 2013, and has been updated since. Cross reactivity with other flaviviruses was not included.

- Can you present a map showing areas where you infer FOI based on serological surveys and where it is inferred as a function of the GLM 'probability of a report' estimate.

Thank you, a map of serological study locations has been included in the Appendix (Appendix—figure 1).

- Some more narrative would also be helpful in explaining conceptually how the FOI-to-probability of occurrence component was achieved. It appears that the ingredients you have in each province for which no sero survey exists are (i) the GLM-derived 'probability of > = 1 case in a given province'; (ii) population; (iii) an idea of the probability that a case is detected given that it occurs. Your objective is to infer FOI. It is not clear if you do this within your Bayesian inference framework-thereby somehow borrowing strength from those provinces for which you have both sero surveys and GLM estimates, or, if this is essentially an arithmetic transform that converts the GLM output to an FOI estimate. It appears to be the latter-and if so it is then not clear how robust this transform is, how you infer probability of detection etc. Presumably you could also report on the degree of alignment between your GLM outputs and you FOI calculations in provinces that do have sero-surveys, as an indicator of how well this approach will then work in provinces without surveys.

We have added further description to clarify the transform (subsection “Mapping probability of occurrence to force of infection”) and added a comparison in the Appendix for the alignment between the FOI from serology and from the GLM.

- On YF burden from zoonotic/intermediate transmission versus urban transmission cycles ("the transmission intensity is assumed to be a static force of infection, akin to the assumption that most YF infections occur as a result of sylvatic spillover." And also in the Discussion.) Instead of only touching on it in the Discussion perhaps it should be highlighted that the estimate of FOI is limited to burden from zoonotic spillover earlier/throughout the manuscript. Consider describing results as estimated effect of climate change on zoonotic/sylvatic YF burden?

Thank you, we have reviewed the text and included a statement at the beginning of the Results section to highlight this.

- How appropriate is it to use temperature suitability pertaining only to the urban vector rather than vectors involved in the intermediate and sylvatic transmission cycles. Also, whether occurrences from urban outbreaks should be excluded from the occurrence database?

Urban outbreaks result first from a spillover event and the majority of the outbreaks are considered to be sylvatic in Africa, as such it would not be necessary to exclude further data form the database.

Using information on the urban vector is an approximation made due to data limitations. However, Aedes aegypti is well examined and there are studies available investigating the thermal response of them in different contexts. We also fit the temperature dependence in a hierarchical framework in order to capture some of the uncertainty resulting from the lack of sylvatic vector data. We have added further information in the description of data sources and Discussion to describe this limitation.

- Can you clarify also what is the implication of focusing on Ae. aegypti (subsection “Temperature suitability”) instead of other vectors?

As we focus on one vector rather than the multitude that are related to YF transmission, it is likely that we have underestimated uncertainty in the response of the vector to changes in climate. We have added to the Discussion to highlight this.

3) Framework for inferring contemporary relationship with rainfall and temperature

- There is a much greater emphasis/level of sophistication around temperature effects than around rainfall. For temperature, the authors recognise the highly non-linear response via a mechanistic model capturing deterministically the temperature effects on different components of the transmission cycle. For rainfall, however, the presumably similarly complex mechanisms of impact on aedes breeding, vector densities, interactions with humans etc., are represented much more crudely via a single covariate in a GLM, with implicit assumptions around linearity. Some justification on this would be useful.

The relationship between the thermal response of vector attributes such as mortality are well characterised in comparison the influence of rainfall. We have added further description on this in the Discussion.

- While the paper builds on previous work, the very small number of covariates in the GLM is surprising and needs justification in this current paper. Is there no rationale for including, for example, human population density, or metrics of landscape proxies for non-human primate habitats etc?

We include human population density – this was accidentally omitted from the table in the Appendix which is now corrected (Appendix —table 2).

Human population is included and the landcover metrics are omitted due to uncertainties oh how they will be affected by climate change. Similarly, whilst we use projections of human population sizes which include measures of immigration, the distribution of humans and non-human primates are uncertain and are likely to vary in the future and as a product of climate change. We have added to the description of this in the Discussion.

- Further, could there be a bit more discussion of the reasonableness and motivation for omitting land cover and vegetation index? This would seem to be important given that at least in Brazil there's a very strong effect of proximity to forest?

Thank you. This limitation is detailed in the Discussion. We required a suite of covariates that both reflected the transmission now but also could be parametrised for the 4 climate scenarios. The vegetation index is thus far not included for this reason.

4) Framework for inferring impact of changing climate

-Given it would be straightforward in your current framework, please consider the utility of running projections with only the effects of temperature change and only the effects of rainfall change? The rationale being that you have differing levels of sophistication in your ability to link these to YF risk; that there are differing levels of certainty in the climate projections for these two aspects; and that you could provide a richer interpretation of their respective contributions to elevated risk.

Thank you, this was an interesting point. We have included a further subsection, “Environmental and climate projections”, where we show the change to the mean force of infection per country in 2050 and 2070 given either rainfall or temperature changes under climate scenario RCP 8.5. We find that the dominant factor is the change in temperature but that rainfall and temperature change can have interesting effects: in some cases, the change in rainfall mitigates the influence of changing temperature, in others, such as Uganda, the change in both creates a ‘perfect storm’ where transmission is projected to be highest if both change under the projected scenario.

We have added a further section to the Discussion for this point.

- You elude to it in the Discussion, but more consideration of what you do not include in your framework is crucial here. Temperature/rainfall changes may exert upward pressure on transmission and burden – but how does the magnitude of this effect compare to, for example, a modest increase in vaccination coverage, or changes in exposure due to urbanisation or modernisation of housing etc. How might global change impact other components of the transmission cycle such as the col-location of humans an NHP zoonotic reservoir species?

Thank you, we have expanded the Discussion to further highlight these limitations and that, as this time, projecting these highly non-linear influences may not be possible with available data.

5) Impact of vaccination

- It is not clear from this paper or the references precisely what vaccination coverage data was compiled and how it was used in vaccination coverage estimates. The data and methods are not fully described in this paper or the referenced papers. It seems that uncertainty in vaccination coverage estimates was considered in Garske et al., 2014, by exploring alternate vaccination targeting assumptions. For the analysis in this manuscript, it appears that only the scenario where supplemental vaccination campaigns target unvaccinated individuals is considered? Lessler and colleagues (journal.pmed.1001110) have previously described issues with assumptions about vaccination targeting when making coverage estimates. It is currently not clear whether any uncertainties were propagated through the analyses presented in this paper. It is important to explore the possible impact of these (potentially large) uncertainties on the results.

We have included further details in how the vaccination coverage was calculated and highlighted the limitations around quantifying uncertainty in vaccination and subsection “Past vaccination coverage and demography”, Discussion). We have also emphasised that this is based on methodology that is available in an R package, as a shiny application and has been detailed in both the POLICI application paper [Hamlet et al., 2019] including how campaigns are targeted, and in [Garske et al., 2014]. In this manuscript, vaccination campaigns are assumed to target people at random rather than only unvaccinated individuals. We also focus on using the same vaccination scenarios in all projections and do not specifically quantify vaccine impact here, although, as noted in the Discussion, it may mitigate some of the effects of changing transmission.

6) Impact of future population changes

- Why were demographic conditions held constant? There ought to be projections for future population growth (and ageing?) for these countries that would give a more accurate representation of future disease burden.

Demographic conditions are assumed to follow the United Nations World Population Prospects (UNWPP) projections of population growth. As such, demographic conditions are not held constant and future projections of burden take these into account. The counterfactual also assumes population growth.

- Possible to forecast total numbers of deaths as well as the country specific ones?

Sure, we have added a table to the Appendix for 2050 and 2070 for all scenarios (Appendix—table 4).

Author details

1. Katy AM Gaythorpe

   Imperial College London, London, United Kingdom

   Contribution

   Conceptualization, Resources, Data curation, Software, Formal analysis, Validation, Investigation, Visualization, Methodology, Writing - original draft, Project administration, Writing - review and editing

   For correspondence

   k.gaythorpe@imperial.ac.uk

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0003-3734-9081

2. Arran Hamlet

   Imperial College London, London, United Kingdom

   Contribution

   Formal analysis, Validation, Investigation, Visualization, Methodology, Writing - review and editing

   Competing interests

   No competing interests declared

3. Laurence Cibrelus

   World Health Organisation, Geneva, Switzerland

   Contribution

   Data curation, Writing - review and editing

   Competing interests

   No competing interests declared

4. Tini Garske

   Imperial College London, London, United Kingdom

   Contribution

   Resources, Supervision, Funding acquisition, Methodology, Project administration

   Competing interests

   No competing interests declared

5. Neil M Ferguson

   Imperial College London, London, United Kingdom

   Contribution

   Supervision, Project administration, Writing - review and editing

   Competing interests

   Senior editor, eLife

• 4,511

  Page views

• 387

  Downloads

• 22

  Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.