Spatiotemporal dynamics of dengue in Colombia in relation to the combined effects of local climate and ENSO

Dengue virus (DENV) is an endemic disease in the hot and humid low-lands of Colombia. We characterize diverse temporal and spatial patterns of monthly series of dengue incidence in diverse regions of Colombia during the period 2007-2017 at different spatial scales, and their association with indices of El Nino/Southern Oscillation (ENSO) at the tropical Pacific and local climatic variables. For estimation purposes, we use linear analysis tools including lagged cross-correlations (Pearson test), cross wavelet analysis (wavelet cross spectrum, and wavelet coherence), as well as a novel nonlinear causality method, PCMCI, that allows identifying common causal drivers and links among high dimensional simultaneous and time-lagged variables. Our results evidence the strong association of DENV cases in Colombia with ENSO indices and with local temperature and rainfall. El Nino (La Nina) phenomenon is related to an increase (decrease) of dengue cases nationally and in most regions and departments, with maximum correlations occurring at shorter time lags in the Pacific and Andes regions, closer to the Pacific Ocean. This association is mainly explained by the ENSO-driven increase in temperature and decrease in rainfall, especially in the Andes and Pacific regions. The influence of ENSO is not stationary (there is a reduction of DENV cases since 2005) and local climate variables vary in space and time, which prevents to extrapolate results from one site to another. The association between DENV and ENSO varies at national and regional scales when data are disaggregated by seasons, being stronger in DJF and weaker in SON. Specific regions (Pacific and Andes) control the overall relationship between dengue dynamics and ENSO at national scale, and the departments of Antioquia and Valle del Cauca determine those of the Andes and Pacific regions, respectively. Cross wavelet analysis indicates that the ENSO-DENV relation in Colombia exhibits a strong coherence in the 12 to 16-months frequency band, which implies the frequency locking between the annual cycle and the interannual (ENSO) timescales. Results of nonlinear causality metrics reveal the complex concomitant effects of ENSO and local climate variables, while offering new insights to develop early warning systems for DENV in Colombia.

For estimation purposes, we use linear analysis tools including lagged 102 cross-correlations (Pearson test), and wavelet analysis, and diverse time-delayed 103 nonlinear causal discovery metrics. Cross-correlations (ρ) between dengue cases and 104 ONI, dengue cases and climatic series, and ONI and climatic series are estimated over a 105 range of lags between 0 and 12 months. Lagged cross-correlograms are estimated by 106 aggregating the information at national, regional, departmental and municipal scales. 107 Furthermore, the correlations between raw dengue cases (without standardization) and 108 ONI are performed by splitting the data into quarters (DJF, MAM, JJA, SON). This 109 kind of analysis quantifies the degree of linear association and the time delay between 110 the time-series. 111 We also perform wavelet analysis to study the time-frequency behavior of monthly 112 series of dengue and ONI, and their conjoint dynamics at the national scale. This 113 technique has the advantage to process scale-dependent non-stationary time series, thus 114 allowing the description of the variability of the spectral properties over time. The 115 coupled dynamics between ONI and dengue cases was evaluated using the wavelet 116 coherence and the cross-wavelet transform. These analyses allow identifying time 117 intervals and period bands in which two time-series are related [50,51]. This technique 118 has already been used to analyze the dynamics of dengue in different places [11,25,33]. 119 We implemented the continuous wavelet transform (CWT) based on Torrence and 120 Compo [50] and the Cross Wavelet Analysis (CWA) based on Maraun and Kurths [52], 121 using the waipy toolkit developed in Python, which was implemented by Mabel Calim 122 Costa and available at https://github.com/mabelcalim/waipy. 123 Furthermore, we use diverse time-lagged causal inference methods aimed at 124 discovering and quantifying the causal interdependencies between time series of weather 125 variables, ENSO indices and dengue incidence at national and regional scales. To this 126 end, we employ the PCMCI method which allows to identify common drivers and links 127 among high dimensional time-lagged variables, by combining a PC Markovian 128 condition-selection step, named after its creators Peter and Clark [53,54] and the 129 Momentary Conditional Independence (MCI) test. PCMCI has been applied recently to 130 a large suite of biogeophysical phenomena [55][56][57][58][59][60]. The PCMCI method uses diverse 131 statistical tests to infer non-linear causalities including linear partial correlations 132 (ParCorr) and three types of nonlinear independence tests: GPDC, CMI, and 133 PCMCIplus. GPDC is based on Gaussian process regressions and a distance correlation 134 test on the residuals, suitable for a large class of nonlinear dependencies with additive 135 noise. CMI is a nonparametric test based on a k-nearest neighbor estimator of 136 conditional mutual information that can accommodate almost any type of 137 dependency [58][59][60]. PCMCIplus can identify the full, lagged and contemporaneous 138 causal graph (up to the Markov equivalence class for contemporaneous causality) under 139 the standard assumptions of Causal Sufficiency, and the Markov condition (J. Runge,140 pers. comm.). For implementation purposes, we use the Tigramite 4.2 python package, 141 which allows to reconstruct graphical models (conditional independence graphs) from increasing biting rates of female mosquitoes and their longevity [28,64,65]. This may 152 result in increases in population size and enhanced egg production. Temperature also 153 influences the geographical range of vector survival [66] and humidity state. High 154 humidity enhances adult mosquito survival, but although low humidity decreases the 155 survival rate of arthropod vectors because of dehydration, and it also may cause an 156 increase in the mosquito blood-feeding rate [67]. Precipitation provides breeding 157 places [13,68], but the impacts depend on its interaction with evaporation [67], soil type, 158 topography and the proximity of water bodies. Besides, heavy or prolonged rainfall 159 events may disrupt vector breeding sites, and indeed, kill the mosquitoes directly [66]. 160 Both the seasonality and interannual variability of dengue cases have shown 161 connections with climate, existing considerable evidence of the role of El Niño-Southern 162 Oscillation (ENSO) on endemic infections [39,69,70]. ENSO is a macroclimatic Pacific Ocean [73].

172
Relationships between climate variables and factors affecting DENV transmission are 173 complex, non-linear, and non-stationary [11,20,31,33,67]. Climate variables may impact 174 the mosquito's populations in different ways depending on local conditions since the 175 fundamental processes of heat and water exchange are determined by the microclimate 176 states [74]. Furthermore, ENSO's impacts vary markedly, affected by the ENSO 177 diversity and the modes of variability within and outside the tropical Pacific [73]. 178 Climate of Colombia 179 Colombia's climate varies remarkably in time and space owing to its equatorial location 180 and the temperature and rainfall gradients associated with the Andes topography. The 181 temporal distribution of rainfall highly depends on the meridional oscillation of the 182 Inter-Tropical Convergence Zone (ITCZ), the dynamics of three low-level jets: Chocó, 183 Caribbean and Orinoco [75][76][77][78], changes in topography (0 to 6,000 m), the 184 ocean-atmosphere dynamics of the Pacific, Atlantic and Caribbean Sea, the Amazon 185 and Orinoco River basins, and strong land surface-atmosphere feedbacks [79,80]. Mean 186 annual temperature and rainfall depends on altitude, and geographic location. Mean 187 annual rainfall ranges from 50 mm in deserts to 10,000-13,000 mm over the Pacific coast 188 in one of the rainiest spots on Earth [81]. Besides, uni-modal and bi-modal annual 189 cycles of precipitation vary throughout the different regions [22]. A bi-modal annual   Cross-correlations between ONI and climate variables on the regional scale. Maximum cross-correlation (p ≤0.05) and lag between ONI and precipitation (a,b), maximum temperature (c,d), and minimum temperature (e,f).
Also, correlations between ONI and temperatures are positive (ρ max ≈0.75 for T max and 203 ρ max ≈0.65 for T min ), denoting the increase in both temperatures during El Niño. But 204 there is variability among regions. Maximum linear correlations between ONI and P are 205 simultaneous (lag 0) in all regions, except in the Amazon, which occurs at 2-months lag. 206 Between ONI and T max , maximum correlation occur at 1-month lag for the Amazon, 207 Orinoco, and Caribbean regions, and at 2-months lag for the Andes and Pacific regions, 208 and between ONI and T min , at 1-month lag for the Caribbean, 3-months for the Pacific 209 and Amazon, 4-months for the Andes, and 5-months for the Orinoco region.

210
The effects of ENSO are more noticeable in the Andes, Pacific, and Caribbean 211 regions (see Fig 1a and Fig 1c). This is due to: (1) the weakening of the westerly 212 low-level Chocó Jet [72,75,81,85,90,91]; (2) the reduction of the 700 hPa equatorial 213 easterly jet (over South America and the eastern equatorial Pacific); (3) the anomalous 214 Hadley cell circulation that sets in during El Niño over tropical South America; (4) 215 changes in the flow of atmospheric moisture into the continent; (5) changes in 216 atmospheric pressures resulting in the displacement of the convection centers within the 217 ITCZ to the west and south; (6) feedbacks between precipitation and surface 218 convergence in tropical South America [72,92]; and (8) land-atmosphere interactions 219 due to the regional coupling between soil moisture, precipitation, and 220 vegetation [22,80,83,84,93]. The physical mechanisms explaining the climatic anomalies 221 August 18, 2020 6/34 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 28, 2020. . https://doi.org/10.1101/2020.08.24.20181032 doi: medRxiv preprint related to ENSO in Colombia have been studied by [83-86, 90, 94-99], among others. In 222 the Pacific region the maximum correlation between ONI and precipitation is ≈0. 65,223 while in the Amazon and Orinoco regions, correlations are not statistically significant. 224 These results are in agreement with those found by Salas et al. [100].

225
The largest correlation between ONI and precipitation at national scale is  For our purposes, monthly dengue cases data and climate variables data are aggregated 254 at national, regional, departmental, and municipal scales. At national scale, we use 255 three methods to quantify the association between dengue cases, climate variables, and 256 August 18, 2020 7/34 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 28, 2020. . . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 28, 2020.  smaller political scales, we only estimate cross-correlation due to the few data available. 261 At national and regional scales, we also analyze the association between dengue cases 262 and ONI at seasonal timescales.  correlations for lags between 2 and 7 months, and also suggest that El Niño (La Niña) 279 events are associated with an increase (decrease) of dengue cases in Colombia with a 2 280 to 6-months lag.

281
In spite of the previously shown high cross-correlations between ONI and dengue in 282 Colombia, they are lower than those found by Acosta and co-authors [47] for the period 283 spanning from 2005 to 2013. With the aim to understand this difference, we calculate 284 the dynamic correlations between the ONI and dengue cases for lags from 0 to 5 months. 285 These cross-correlations are estimated by aggregating the data from both series, one 286 month at a time, after December 2009, that is: where t 0 is the December of 2009 and t i the time at i. . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted August 28, 2020.   . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted August 28, 2020. standardized data to investigate seasonal differences of ENSO and climatic variables on 307 dengue incidence. Purple bars correspond to correlations between ONI and dengue cases 308 during the same season. To describe the results, we use brackets with two positions, 309 where the first position indicates the quarter of ONI and the second the quarter of 310 DENV cases. Higher cross-correlation appear for (DJF,DJF) and (SON,DJF), and lower 311 for (MAM,DJF). SON is the season of ONI that results in higher correlations, and DJF 312 is the season of dengue cases that seems to be more affected by ENSO 313 (see [79,84])(except for ONI in MAM). Remarkably, the only negative cross-correlation 314 occurs in (JJA,JJA), but it does not exceed the confidence interval of statistical 315 significance.

316
As mentioned before, ENSO does not affect directly the dynamics of dengue 317 dynamics, but modify weather and climate variables that are directly related to dengue 318 transmission. To identify which variables affected by ENSO influence dengue in 319 Colombia, we performed cross-correlation analysis between the number of DENV cases 320 and precipitation, maximum and minimum temperature, wind velocity, and relative 321 humidity. Table 1 shows the maximum cross-correlations between climate variables and 322 dengue cases and the peak lag associated with them. Unlike other sites 323 (e.g., [8,27,110,111]), precipitation is negatively correlated with the number of dengue 324 cases in Colombia. Maximum temperature is highly and positively correlated with 325 dengue cases as expected, and wind speed is positively related as it can help the spread 326 of mosquitoes. Finally, relative humidity is negatively correlated to DENV cases (even if 327 correlations do not exceed the CI), suggesting that the water stress effect is important 328 in Colombia. Confronting these results with those in Fig 4, it is possible to suspect that 329 the key variable in the relationship between ENSO and dengue cases is temperature, as 330 noted in other sites (e.g., [22,27,67,[112][113][114]  . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted August 28, 2020.  Results from the ParCorr test indicate positive relations between ESPI and T max for 342 lags of 1 and 6 months and 1 month, respectively. PCMCIplus shows no association but 343 indicates that ENSO modifies precipitation and this, in turn, changes temperature (for 344 lags between 4 and 7 months). The ParrCorr method also indicates that the effects of 345 ENSO has on T min , if they occur, are brought about by precipitation. The non-linear 346 causalities between P and both temperatures (T max and T min ) are negatives for the 347 ParrCorr method (for T max the causality is indirect, and for T min it occurs for a lag of 348 2 months). PCMCIplus shows a positive causality of P with T max and a negative with 349 T min (for lags of 2 and 4 months for T max and of 2 months for T min ). The PCMCIplus 350 method shows an indirect causality with T . Negative and positive relationships are . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted August 28, 2020. . https://doi.org/10.1101/2020.08.24.20181032 doi: medRxiv preprint precipitation, and then ET [115]. 360 ParCorr shows negative non-linear causalities between T and P (for lags of 3 and 6 361 months for T max and 6 and 7 months for T min ), and PCMCIplus indicates a positive 362 relation of T max and P for a lag of 7 months and a negative indirect relationship.

363
Non-linear causalities between precipitation and DENV cases are negative for all 364 methods for lags between 3 and 6 months. Both negative and positive causalities are 365 found worldwide as discussed in Section Discussion.

366
All methods show positive causalities between T max and DENV cases for lags 367 between 1 and 7 months. This association has been identified in many sites, however, it 368 is not direct and depends on the temperatures range. Furthermore, ParCorr indicates 369 positive causalities between T min and DENV cases for lags between 2 and 3 months, but 370 the PCMCIplus method shows no causality, suggesting that minimum temperature only 371 affects through precipitation, with which it has a positive indirect association.

372
Note that, as expected, all results obtained with the non-linear methods do not show 373 any causality between local climatic variables on ESPI, DENV cases on ESPI, and 374 DENV cases on the climatic variables.    Table 2 shows results of maximum correlations, ρ max s, between climatic variables 431 and dengue cases and the corresponding lag at regional scales. The linear association Non-linear causalities between DENV cases, ESPI, P , and T max at the regional scale at regional scale. Non-linear causalities given by ParCorr (left panel), and PCMCIplus (rigth panel) for standardized data.
August 18, 2020 15/34 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted August 28, 2020. . https://doi.org/10.1101/2020.08.24.20181032 doi: medRxiv preprint  . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted August 28, 2020. . exceed the confidence interval in the Orinoco region, the negative correlations between 437 DENV cases and T min can explain the negative association between ONI and DENV in 438 this region. Similarly, the negative associations between P and DENV and the positive 439 associations between temperature and DENV would explain the high positive 440 correlations between ONI and DENV in Andes and Pacific regions. The mechanistic 441 processes involved in these associations are discussed in Section Discussion.   . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted August 28, 2020.    Table 3 shows the cross-correlation at lag 0 between the three climatic variables and 484 the number of DENV cases. In Medellín, no correlation exceeds the confidence interval, 485 but the signs remain the same as for department of Antioquia. In Cali, the correlation 486 with precipitation does not exceed the CI either, but the association with minimum and 487 maximum temperatures seems to play an important role, especially with the former one. 488 Table 3. Cross-correlation between climatic variables and dengue cases in Medellín

489
Our results confirm the strong relationship between ENSO and the incidence of 490 arboviruses like dengue in many parts of the world [11,22,43,47,70,72]. Analyzes . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted August 28, 2020. . spatial and political scales, unlike the non-statistically significant relationship found 494 by [116] and in agreement with other previous studies [22,37,47]. El Niño (La Niña) 495 events are associated with the increase (decrease) in dengue cases in the country.

496
Detailed analysis at finer spatial and temporal resolutions allowed us to identify 497 regional, departmental and municipal differences, as well as within the year (see Figs 4b, 498 8, 9, 10a, 11, and 13a). Furthermore, we have identified that the association is not 499 stationary, pointing to a reduction in dengue cases in Colombia since 2015 (see Fig 5), 500 as in the case of the study by Cazelles et al. [11] in Thailand.

501
The highest (and positive) cross-correlations between ONI and DENV cases appear 502 in the Pacific and Andes regions, while negative in the Orinoco region although below 503 the CI (see Fig 10a). Note that as the Ae. Aegypti is tipically an urban mosquito [117] 504 and the majority of Colombia's population is concentrated in the Pacific, Andes, and  13a).

512
Regarding results between ENSO and dengue cases at regional scales, the Pacific 513 region has shorter lags associated with higher correlations than the Andes region (see 514 Fig 10b), confirming the traveling wave found by Acosta [47], although spanning a 515 broader geographical setting. This traveling wave is not so clear in the non-linear 516 causality methods since several relationships are indirect and have no lag associated.

517
The ParCorr method indicates that when the maximum temperature is taken into 518 account, the effects of ENSO have a longer delay in the Andes than in the Pacific region 519 (see Fig 11). The presence of such delayed effect of ENSO on the occurrence of dengue 520 at departmental level can be used as a pro-time to take preventive measures before 521 potential outbreaks. In Orinoquia the correlation between ENSO and DENV cases is 522 negative correlations (El Niño (La Niña) is related to less (more) cases of dengue) and 523 associated with the highest lag (12 months), but it is not statistically significant  Results at the national scale are consistent with those at regional scales (see Figs 7 540 and 12). Again, it is worth noticing the strong influence of the Andes and Pacific 541 regions at the national scale, since the combination of seasons showing higher 542 cross-correlations in these regions dominate the aggregated behavior. The above is also 543 suggested by the results obtained with the non-linear causality metrics, especially the 544 influence of the Pacific region at national scale (see Fig 11).

545
August 18, 2020 20/34 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted August 28, 2020. . https://doi.org/10.1101/2020.08. 24.20181032 doi: medRxiv preprint The association between El Niño occurrence and outbreaks of DENV cases can be 546 explained in terms of an increase in temperatures (minimum and maximum) and a 547 decrease in rainfall, which may favor the ecological, biological, and entomological 548 components related to this disease [22]. Other variables such as wind velocity and 549 relative humidity may play an important role in dengue dynamics, but the available 550 data is not enough to perform analyzes below the national scale. Wind velocity shows a 551 positive correlation with dengue cases, while relative humidity is negatively correlated, 552 but it does no exceed the CI (see Table 1).

553
Precipitation is negatively correlated with dengue cases in most of the country, 554 except in the Amazon region (although it does not exceed the CI), and in the 555 departments of Chocó, Caquetá, Amazonas, and Putumayo (see Tables 1 and 2, and 556 Fig 14). Linear correlations between P and DENV cases result at national and regional 557 scales are in agreement with those found with the non-linear causality methods (see and distribution of the mosquito [67]. If so, the amount of mosquitoes and dengue cases 562 should increase (decrease) during La Niña as in Hawaii [110]. In some places, 563 precipitation is only important when breeding sites are not handmade [27] or in wetter 564 areas, where rainfall is not a limiting-transmission factor [13,20,79]. However, in 565 Colombia, the decrease in rainfall brought about by El Niño is mostly associated with 566 increases in the number of dengue cases, so decreased rainfall can lead to the formation 567 of stagnant ponds [22,72]. A similar result was found by Poveda et al. [79] for malaria 568 in Colombia and Juliano et al. [118] for dengue in Florida, USA.  [67,72,113,114]. High temperatures result in larger virus replication rates, less 586 time of mosquito incubation and gonotrophic cycles, more frequent biting rates, and 587 greater survival at all stages within the observed temperature values [18,22,67,112,114]. 588 This means more time available for the mosquito reproduction and transmission of the 589 virus [72], and a greater basic reproductive number [114] ,R 0 , defined as the expected 590 number of cases directly generated by one case in a population where all individuals are 591 susceptible to infection. 592 Focks et al. [114] concluded that under tropical conditions, temperature does not 593 affect adult mosquito abundance, but the quantity of water-holding containers and the 594 amount of available food for larval survival. However, extremely high temperatures can 595 contribute to the reduction of oviposition sites by evaporation [27], lifespan reduction, 596 and the inability of mosquitoes to feed and fly [18,21]. These facts can help to explain 597 August 18, 2020 21/34 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted August 28, 2020. . the low correlations found in the Caribbean region since temperatures are always high 598 owing to low altitude, which are even higher during the El Niño (see Figs 1b,c). On the 599 other hand, when temperature is very low (below 18 • C according to [113]), the 600 oviposition abruptly diminishes. This implies that the effect of temperature on dengue 601 dynamics is non-linear and depends on local conditions. Furthermore, not only the 602 average or instantaneous temperature values can influence this phenomenon, but also 603 their variability and fluctuations, which have not been considered in this study, as it 604 occurs in other sites [40,111]. 605 Almeida et al. [13] and Dibo et al. [19] suggest that air humidity is one of the most 606 significant climatic variables in determining Ae. Aegypti abundance. In general, 607 positive correlations are expected between relative humidity and dengue cases, i.e., that 608 low values of relative humidity result in fewer dengue cases, given that the mosquito 609 survival and egg development decline [67], and mosquitoes use their available cellular 610 resources for their maintenance and not the virus [18]. At the same time, low humidity 611 can place mosquitoes under stress, impeding them from fighting off a viral infection [18]. 612 Besides, dehydration may cause an increase in blood feeding (biting rates), as female 613 mosquitoes seek to ensure their reproduction [119]. Atmospheric humidity also 614 influences evapotranspiration rates [114] and, as mentioned before, it may or may not 615 favor the creation of breeding sites. Although humidity data is limited and the 616 estimated correlations with dengue cases are not statistically significant, the value 617 surpasses -0.41 (see Table 1). This suggests that the increase in blood feeding due to 618 dehydration plays an important role in Colombia. Strong winds can extend the 619 mosquito's flight distance, but it can also reduce the biting frequency [61]. In Colombia, 620 the first mechanism seems to have a greater influence, since dengue cases show a 621 statistically significant and positive correlation with wind velocity (see Table 1).

622
Notably, temperature, relative humidity, wind speed, and precipitation are not 623 independent variables, with multiple nonlinear feedbacks among them.

624
Dengue transmission is a multi-factorial phenomenon [37], however, the impact of 625 macroclimatic and local variables is evident in most of Colombia and many other 626 regions in the world, as reported by WHO and the IPCC. Advances in understanding 627 causality between climate, weather and dengue incidence can lead to an appropriate 628 approach, both in time and in space, to design and implement preventive measures 629 against potential dengue outbreaks and early warning systems based on ENSO and local 630 climate variables. Also, our results contribute to understand and anticipate the possible 631 consequences of climate change on this type of diseases. 632 We note that even if we considered the "natural regions" of Colombia to study the 633 possible association between dengue cases, ENSO and climate variables, these regions 634 were defined based on many variables, such as relief, climate, vegetation and soil type, 635 that may not be determinant in the studied phenomenon. Besides, other relevant The relationship between climate, weather, human behavior, and arbovirus diseases is is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted August 28, 2020. . https://doi.org/10.1101/2020.08.24.20181032 doi: medRxiv preprint contribute to the development of climate-based surveillance, prevention and control 648 programs for dengue fever in endemic areas of Colombia, shedding light on 649 decision-makers about the adequate timing to implement prevention and control 650 measures and to anticipate the effects of climate variability and climate change.

651
The spatial and temporal distribution of dengue fever cases in Colombia are 652 associated with macroclimatic and local conditions. El Niño (La Niña) phenomenon is 653 related to the intensification (weakening) of dengue cases at the national level and in 654 most regions. Our results suggest that this association is mainly explained by 655 ENSO-driven increase in temperature and decrease in rainfall. However, these 656 associations are not simultaneous, and the temporal delay varies regionally.

657
The influence of ENSO and different climatic variables affected by this phenomenon 658 vary in space and time, and is not stationary, so it is not easy to extrapolate results 659 from one site to another. The association between dengue cases and ENSO varies when . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 28, 2020. August 18, 2020 24/34 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 28, 2020. . https://doi.org/10.1101/2020.08.24.20181032 doi: medRxiv preprint  We are grateful to Mabel Calim Costa and Jakob Runge for making available the Waipy 702 and Tigramite toolkits, respectively. We also thank the Instituto de Hidrología,

703
Meteorología y Estudios Ambientales -IDEAM, and the Instituto Nacional de Salud de 704 Colombia -INS, by the availability of the climatic and dengue data, respectively. The 705 work of G. Poveda was supported by Universidad Nacional de Colombia at Medellin, 706 Colombia, as a contribution to the project "Detección temprana de transmisión de 707 dengue basada en diagnóstico molecular e información ambiental y climática en