We mapped disease location data on corresponding NDVI and climate data anomalies in order to understand associations between specific anomaly patterns in ecological and climate variables and disease outbreak patterns through space and time. We further explored the associations by plotting and comparing disease data against cumulative rainfall and vegetation index anomalies to illustrate the lag time between the driving climate conditions and the timing of first case disease occurrence for Rift Valley fever. For chikungunya we further investigated the relationships among surface air temperature, precipitation anomalies, and chikungunya outbreaks through correlation analysis. For Rift Valley fever we further investigated relationships between precipitation and Rift Valley fever outbreaks through logistic regression. Results were interpreted in terms of vector biology and population dynamics. We suggested caveats for non-outbreak years when climate and ecological conditions would indicate an imminent outbreak.

The map in Fig. 1 shows the correlation between monthly NINO3.4 SST anomalies and monthly GPCP rainfall anomalies for 1979–2010. It illustrates how variations in sea surface temperature in the equatorial eastern Pacific (ENSO) influence rainfall variability at various locations around the world. It is also a depiction of the concept of teleconnections. The central to eastern Pacific Ocean region, Ecuador, Peru, Eastern equatorial Africa, and the southern tier of the United States indicate the tendency for wetter-than-normal conditions in these regions during the El Niño (warm) phase of ENSO. Negative correlations over Southern Africa and the western Pacific region, including Australia, the greater Indonesian Basin, and northern South America, indicate the tendency for drier-than-normal conditions during El Niño [10]. These anomaly patterns are largely reversed during the La Niña (cold) phase of ENSO as shown in previous studies [10]. Extremes in rainfall anomalies resulting from phase shifts in ENSO affect regional ecological patterns [11], [12] differentially, and influence the emergence of different disease vectors and consequently patterns of vector-borne disease outbreaks [13], [14]. In particular, there is a tendency for outbreaks of Rift Valley fever to occur in Eastern Africa during the El Niño phase and for outbreaks of Rift Valley fever to occur in Southern Africa during the La Niña phase, as will be shown later in this study.

The 2004–2009 period of analysis is an illustration of the above teleconnection patterns and demonstrates how different climate and ecological anomaly extremes resulting from these teleconnections influence vector-borne disease outbreaks through variations in temperature, rainfall, and ecology. The clusters of outbreaks and epidemics/epizootics of chikungunya and Rift Valley fever across Africa (Fig. 2) [8], [9], [23], [24] occurred during periods of severe drought for chikungunya (2004–2006; Fig. 3) [19], [26], and during periods of above-normal rainfall for Rift Valley fever (late 2006 to 2009, Fig. 4) [8], [27]. During 2004–2006 there was a regional chikungunya epidemic covering coastal East Africa that expanded to cover western Indian Ocean islands, and later parts of India and Southeast Asia through 2010 (Fig. S4). The bulk of the human cases were reported during the most severe drought period of 2005–2006 in East Africa and the western Indian Ocean islands. This chikungunya outbreak was consistent with drought resulting from circulation anomalies forced by the ENSO cold phase (La Niña) and a regional circulation pattern characterized by anomalously high pressure in the western equatorial Indian Ocean and low pressure in the east. These pressure patterns led to enhanced surface westerlies, causing enhanced descending motion over East Africa and the western Indian Ocean and ascending motion over Indonesia [21], [26], [27]. This pattern of regional circulation hampered convective activity and precipitation and resulted in one of the most severe droughts ever observed in the region (Fig. 2).

Historically, chikungunya is known to be enzootic in west and central Africa in a sylvatic cycle involving wild non-human primates and forest species of Aedes mosquitoes [28]. The first and subsequent documented outbreaks of chikungunya occurred during periods of heavy rainfall, filling natural and artificial containers that serve as immature mosquito habitats [29], [30]. However, in 2004–2006 the severe drought described above [26] led to widespread storage of water in containers around households. Unprotected stored water allowed Aedes aegypti on the East African coast [9], [31] to reproduce in large numbers and permit establishment of a Central/East African genotype of chikungunya virus in highly populated areas of the East African coast and the western Indian Ocean islands [28]. In addition and possibly most importantly, elevated temperatures during the drought facilitated the amplification of the virus and increased vectorial capacity in the mosquitoes [9], [31]. Exposure of Ae. aegypti larvae to similarly elevated temperatures in the laboratory has been shown to select for strains adapted to survive longer at these temperatures and significantly increase susceptibility of adult mosquitoes to chikungunya virus, enhancing vectorial capacity and vector competence, respectively [32]. This suggests Ae. aegypti was adaptable to actual environmental conditions of severe drought and elevated temperatures presented here. The initial chikungunya cases were identified at health care facilities in Lamu (June, 2004) in the initial stages of the drought, and Mombasa (November, 2004) as the drought peaked in Kenya (Fig. 3).

In time the chikungunya outbreak spread and impacted the western Indian Ocean islands including Seychelles, Comoros, Mayotte, Mauritius, and La Reunion, all in 2005, infecting between 30–75% of the populations in affected areas [2], [21] (Figs. 2 and 3). The entire western equatorial Indian Ocean region was impacted by a large scale drought in 2005 (Fig. 3) [26]. In December, 2005 a mutation was observed in some of the chikungunya isolates from La Reunion that caused a single amino acid substitution in the E1 envelope glycoprotein (E1-A226V) [33], [34]. Although Aedes albopictus was considered a competent but secondary vector of chikungunya [35], the E1-A226V mutation increased virus replication and dissemination in this mosquito, enhancing its vector competence and increasing the potential for the virus to permanently extend its geographic range [33], [34]. As many as 3 other independent events led to the E1-A226V mutation in India, Cameroon, Gabon, and Sri Lanka, and geographical expansion of the adaptation of the virus to Ae. albopictus [36], [37], [38]. Subsequently the outbreak continued eastward affecting many Indian Ocean countries and in conjunction with human travel and tourism affected some European countries ([9], [24], [33], Text S1).

Cessation of the drought in eastern Africa was marked by the development of contrasting patterns of SST anomalies in the equatorial Indian Ocean, with positive SST anomalies in the equatorial western sector and negative SST anomalies in the eastern sector, and a warm ENSO event in the eastern Pacific Ocean. These patterns of SST anomalies (Fig. S1) and associated precursor pressure anomalies led to enhanced easterlies, fast southern trade winds and ascending motion over East Africa, and descending motions over Southeast Asia [19], [26], [27]. The combined effect of these conditions resulted in persistent, above-normal, and widespread rainfall in excess of 200 mm per month during September–December, 2006 in endemic regions of Eastern Africa (Fig. 4A) and severe drought conditions in Southeast Asia. The above-normal rainfall flooded low lying wetlands known as dambos [39], the primary habitats of Aedes mcintoshi mosquitoes pre-infected with Rift Valley fever virus from prior epizootics. The abnormally high and sustained precipitation led to not only rapid emergence and development of the mosquitoes, but also enhanced and sustained survival because of rapid emergence of protective vegetation habitat, as indicated by large positive NDVI anomalies (Fig. 4B) [10], [23]. Large increases in numbers of virus-carrying mosquitoes in northeastern Kenya and southern Somalia quickly followed, and the first human Rift Valley fever cases were identified from serological surveys in mid-December, 2006 in Kenya and Somalia.

The enhancement of El Niño and WIO positive SST conditions and then the shift in the main rainfall belt southwards into Tanzania resulted in a shift of the area at risk from Rift Valley fever activity southwards into Tanzania (Fig. 4A). The first cases of Rift Valley fever in humans were reported in Tanzania in February 2007 [23], [40]. While the Pacific NINO3.4 region shifted to a cold phase with the emergence of La Niña conditions in May, 2007, the western equatorial Indian Ocean (Fig. S2) continued to warm with SST anomalies as high as +1°C between May–July, 2007. These conditions led to enhanced convective activity across the Sahel region with persistent and above-normal rainfall centred over central Sudan between June–October, 2007 (Fig. 4C). As in East Africa, these anomalous rains flooded dambo habitats and areas within the Gezira irrigation scheme as shown by the positive NDVI anomalies (Fig. 4D) resulting in the emergence of large numbers of Rift Valley fever-carrying Aedes and Culex mosquitoes. The first cases of Rift Valley fever in humans and livestock in Sudan were reported in mid-October, 2007. This was the first reported outbreak of Rift Valley fever in Sudan since 1976, although serologic evidence suggests an outbreak also occurred in 1981 [41], [42].

The shift from El Niño conditions to La Niña (Fig. S2) conditions in the summer and fall of 2007 moved the main area of enhanced rainfall from Eastern to Southern Africa (Fig. 4E). The above-normal and widespread rainfall between October, 2006–February, 2007 resulted in flooding and anomalous green-up of vegetation in Rift Valley fever-endemic areas of Southern Africa (Fig. 4F) and Madagascar. Cases of Rift Valley fever in livestock and humans were identified in mid-February, 2008 and subsequently in some farm workers and veterinary students in South Africa. Human cases of Rift Valley fever were identified in Madagascar over a large part of the country from March–May, 2008 [23]. The persistence of La Niña conditions and enhanced westerlies in the equatorial Indian Ocean through mid-2009 [20] resulted in recent and continuing severe drought conditions in East Africa and continued above-normal rainfall conditions in Southern Africa [43]. As a result there was a recurrence of Rift Valley fever activity in Southern Africa during the southern hemisphere summer period through 2011. Figure 5 summarizes the temporal and spatial anomaly patterns in vegetation (i.e., ecology) as they relate to the temporal patterns in SST anomalies and the resulting temporal distribution patterns of Rift Valley fever outbreaks for East Africa, Sudan, and Southern Africa during this period.

An analysis of the January, 1979–February, 2010 relationships between chikungunya activity and surface air temperature anomalies and precipitation anomalies is shown in Figs. 6 and 7, respectively. The figures show the frequency distribution of the number of reported chikungunya outbreak events against rainfall and temperature anomalies. Persistent temperature anomalies over a four month period were classified as hot if anomalies were >0 or cool if <0. Persistent precipitation anomalies over a 4 month period were classified as drought if <0 and wet if >0. In East Africa, Central Africa, and South Asia, 94%, 68%, and 80% of the outbreaks occurred during warmer-than-normal temperatures, and these differences were significant at p<0.05 (Figs. 6A–C). However, in Southeast Asia 52% of the outbreaks occurred during cooler-than-normal temperatures (Fig. 6D). In East Africa outbreaks were also significantly positively correlated with drought conditions at p<0.05 (Fig. 7A), and not significantly correlated in South Asia (Fig. 7C). In Central Africa (Fig. 7B) and in Southeast Asia (Fig. 7D) outbreaks were significantly negatively correlated with drought at p<0.05. The positive correlation between chikungunya outbreaks and warmer-than-normal temperatures in Africa and South Asia was consistent with non-sylvatic transmission by Ae. aegypti and Ae. albopictus in highly populated domestic settings where domestic and peri-domestic stored water supplies were the likely source of the mosquitoes [9], [32], [33], [44], [45].

The relationship between precipitation and Rift Valley fever was determined through a logistic regression of Rift Valley fever presence/absence on cumulative precipitation anomalies for the 4 months immediately preceding each Rift Valley fever outbreak [7], [8] (Fig. S3; Text S1). Results by region (East Africa, Sudan, South Africa, and Madagascar) are presented in Table 2. For all 4 regions a significant relationship was found between cumulative rainfall anomalies and Rift Valley fever presence with at least 99.9% confidence. For East Africa, Sudan, and South Africa this relationship was strongly positive (β₁>0), with the highest cumulative rainfall anomalies yielding the highest odds of Rift Valley fever presence, equivalent to an outbreak, as indicated by the positive rainfall anomaly terms estimated in Table 2. This relationship confirmed experimental findings by Linthicum et al. [46] that persistent, widespread, and above-normal rainfall is required to flood dambo habitats in order to create ideal conditions to spawn abundant mosquito populations on a large scale that would result in a Rift Valley fever epizootic. As shown in Fig. S3, each of the selected outbreak locations for each region was preceded by above-normal rainfall for 3–4 months before the first case of Rift Valley fever, which would likely create ideal ecological conditions for an increase in Rift Valley fever mosquito vector emergence and survival.

In contrast, for Madagascar a negative relationship was found (β₁<0), with the model predicting higher odds of Rift Valley fever outbreaks when rainfall was less than normal. Although the 2008 Madagascar Rift Valley fever outbreak was initially triggered by rainfall [23], [47], the subsequent spread of the outbreak appears to have been related to the introduction of infected livestock. In Madagascar, livestock located in heavy rainfall areas in the south had become infected with Rift Valley fever and were then transported to other parts of the country to the north. Even though rainfall was only slightly above normal in these northern areas, precipitation was sufficient to produce abundant Culex populations originating from flooded domestic and semi-domestic immature mosquito breeding habitats. The Culex mosquitoes, efficient vectors of Rift Valley fever, then transferred the virus from the newly arrived infected livestock to surrounding human populations.

Our knowledge of teleconnection events and the quasi-cyclical nature of climate variability may allow parts of Africa, the Indian Ocean basin islands, and elsewhere within the greater tropics to have more than a year warning prior to Rift Valley fever outbreaks [23] and other humanitarian climate-related conditions, permitting more precise targeting of vaccine strategies, mosquito control, animal quarantine, and public education strategies. Additionally, identifying the potential for Rift Valley fever outbreaks to occur in Africa and controlling these outbreaks will be of interest to regions of the world that are not currently endemic for Rift Valley fever. The documented expansion of the range of Rift Valley fever beyond sub-Saharan Africa into Egypt in 1977 [56] and into the Arabian Peninsula in 2000 [57], [58], [59] makes Rift Valley fever a likely candidate for further expansion and a significant threat to most of the world where immunologically naïve animal species and humans exist, and competent Culex or Aedes vector mosquito species are present. Furthermore, there is a significant economic threat from Rift Valley fever in non-endemic countries. For example, the U.S. had beef related exports in 2003 of $5.7 billion and should Rift Valley fever occur in the U.S. the OIE will impose a 4 year trade ban until free of the disease for 6 months [60]. The climate-disease teleconnections examined here can also be used to better understand the temporal dynamics of the burden of other diseases such as dengue, malaria, or cholera which can be influenced by major climate events [11], [12].

Outbreaks of mosquito-borne diseases on epidemic scales, such as those experienced during 2005–2009 in Africa and the western Indian Ocean islands, place a huge burden on public healthcare systems and the economy. Outbreaks of chikungunya are also an impediment to tourism, a major contributor to the gross national product of countries and island nation states in the region. The costs to the economies of East Africa in lost trade in livestock due to Rift Valley fever outbreaks were estimated to be $65 million [61] during the 2006–2007 outbreak. Furthermore, there is extended disruption in trade due to livestock movement bans following an outbreak. Given the greater risk of spread and recurrence of outbreaks of these diseases, it is critical that countries in the region have the capacity to anticipate the changing and variable nature of the climate in the region to prevent or minimize the emergence and re-emergence of such diseases. Therefore there is need for public health authorities to take advantage of climate observations and analyses in times of extreme climate variability to aid response and mitigation planning including vector surveillance and control, vaccination, and public education in areas that may be impacted by disease outbreaks. In addition, climate-based predictions offer opportunities for virologists, epidemiologists, entomologists, physicians, and veterinarians to understand the biological and cyclic nature of the disease and how its episodic occurrence relates to livestock immunity in recently infected areas, and the potential for re-emergence of the disease in livestock and human populations.

It is apparent from our analyses that in changing and variable climate, arboviruses and their mosquito vectors are going to adapt to the existing climatic and ecological conditions in a new region, and the resultant disease transmission will vary accordingly and may not be the same manifestation as observed in the original endemic regions. Combining satellite-derived measurements and analyses of climate and ecology with an understanding of mosquito vector biology and human and animal population immunity status can contribute substantially towards reducing the global burden of vector-borne diseases.

All data analyzed were anonymized. We only used GPS latitude-longitude coordinates for cases, and we did not handle or deal with any human or animal specimens.