Irrigation increases and stabilizes mosquito populations and increases West Nile virus incidence

Humans have greatly altered earth’s terrestrial water cycle with the majority of fresh water being used for agriculture. Irrigation changes spatial and temporal water availability and alters mosquito abundance and phenology. Previous studies evaluating the effect of irrigation on mosquito abundance and mosquito-borne disease have shown inconsistent results and little is known about the effect of irrigation on variability in mosquito abundance. We examined the effect of irrigation, climate and land cover on mosquito abundance and human West Nile virus (WNV) disease cases across California. Irrigation made up nearly a third of total water inputs, and exceeded precipitation in some regions. Abundance of two key vectors of several arboviruses, including WNV, Culex tarsalis and the Culex pipiens complex, increased 17–21-fold with irrigation. Irrigation reduced seasonal variability in C. tarsalis abundance by 36.1%. Human WNV incidence increased with irrigation, which explained more than a third (34.2%) of the variation in WNV incidence among California counties. These results suggest that irrigation can increase and decouple mosquito populations from natural precipitation variability, resulting in sustained and increased disease burdens. Shifts in precipitation due to climate change are likely to result in increased irrigation in many arid regions which could increase mosquito populations and disease.

USA have been mixed.Some studies have found positive associations between increased rainfall and C. tarsalis mosquito populations 41,43 while others have found no effect 44 .WNV incidence has been associated with both increased and decreased rainfall 40,45 , decreased rainfall and drought [46][47][48] , and moderate rainfall amounts 49 .These inconsistent relationships with rainfall may result from unmeasured effects of irrigation or land cover that affect mosquito larval habitat.
Our goal was to examine the effect of irrigation on the magnitude and variability in mosquito populations and WNV human disease incidence across California.First, we quantified water amounts coming from irrigation and precipitation across space and time.We then examined irrigation, climate, and land cover data as predictors of mosquito abundance, seasonal variability in mosquito populations, and WNV human disease incidence.We focused on three key WNV vectors, C. tarsalis, C. erythrothorax, and C. pipiens complex (C.pipiens and C. quinquefasciatus).We hypothesized that irrigation would increase mosquito abundance and WNV incidence and reduce seasonal variability in mosquito populations, likely by stabilizing water availability for larval mosquito development 50 .

Results
We examined the influence of irrigation, climate, and land use on spatial and temporal variability in mosquito abundance and across hydrological detailed analysis units (DAU) in California.Culex tarsalis abundance varied over four orders of magnitude across California (0.03-555 mosquitoes per trap/night), and seasonal variability of C. tarsalis populations (measured using percent coefficient of variation) varied five-fold (76-412%) (Fig. 1).Irrigation was also spatially variable (0-2.97Acre-feet per acre; Fig. 1), with irrigation making up 18.7% of total annual water (irrigation + precipitation) across the state and 30.9% of total water inputs in the 78 DAUs where there were mosquito data (Fig. 2a).Water from irrigation exceeded that from precipitation during the seasonally dry mosquito trapping season for all but one of 13 years of the study period (Fig. 2b).
In univariate analyses, C. tarsalis abundance increased with irrigation (Fig. 3A,B; Table S2), temperature, wetland cover, and decreased with precipitation (Table S2).In the multiple regression model C. tarsalis abundance increased with irrigation, average temperatures, wetland landcover, and decreased with developed land cover, but was not significantly correlated with precipitation or open water cover (Table S3).In the univariate model, C. tarsalis mosquito abundance more than doubled (2.74-fold higher) with each tenfold increase in irrigation, and increased 20.6-fold across the 1000-fold range of irrigation amounts (Fig. 3A,B, Table S2).
In univariate analyses, C. pipiens complex mosquito abundance also increased with irrigation (increasing 17.0-fold across the range of irrigation), temperature developed land cover, and decreased with precipitation and open water (Fig. 3C, D; Table S4).In the multiple regression model, C. pipiens complex mosquito abundance increased with irrigation and developed land cover, and decreased with open water, but was uncorrelated with other climate and land use predictors (Table S5).Finally, C. erythrothorax abundance decreased with irrigation and precipitation, and increased with wetland land cover (Table S6).
Seasonal variability in C. tarsalis abundance across DAUs decreased by 36.1% across the range of irrigation observed (Fig. 4) but was not significantly correlated with climate or land cover (Table S7).The decrease in variability was due primarily to sustained periods of higher abundance in high-irrigation areas, whereas in lowirrigation areas there was usually a single sharp peak in abundance (Fig. 4A).Seasonal variation in in C. pipiens complex mosquito abundance was marginally negatively correlated with irrigation but was not correlated with any other predictors (Table S8).Seasonal variation in C. erythrothorax abundance decreased with greater wetland land cover but was uncorrelated with irrigation or other variables (Table S9).
Finally, we examined the effect of irrigation, climate and land use on human WNV disease incidence among California counties.WNV incidence between 2004 and 2010 increased with irrigation, precipitation, temperature, developed land cover, wetland land cover and open water cover in univariate relationships (Table S10).However, the strongest relationship was with irrigation and incidence was not significantly correlated with any climate (precipitation or temperature) or land cover variable once irrigation was accounted for (Fig. 5; Table S11).Irrigation alone explained 34.2% of the variation in human WNV disease incidence among California counties (Fig. 5).

Discussion
We found that irrigation was correlated with increased abundance and reduced seasonal variability of the two most important mosquito vectors in California, C. tarsalis and C. pipiens complex, and with increased human WNV incidence.The reduced variability in C. tarsalis populations resulted from irrigation increasing periods of elevated abundance, which is consistent with previously observed multi-modal seasonal peaks of C. tarsalis abundance in regions of California with heavy irrigation inputs 35 .These relationships suggest that irrigation can partially decouple mosquito abundance, for some species, from natural precipitation patterns 29 and result in elevated and sustained transmission of mosquito-borne disease.The effect of irrigation on both C. tarsalis and C. pipiens abundance likely results from irrigation increasing larval habitat for these species 50 .
Irrigation has also been found to alter seasonal variation in mosquito abundance and transmission of some other mosquito-borne diseases, with the best studied disease being malaria in Africa.There, as we found in California, irrigation both increased Anopheles mosquito abundance and dampened seasonal variability, with the strongest effects occurring in dry regions 16,52,53 .Irrigation in drier months enabled year-round malaria transmission 29,52,54 .As a result, as we found for mosquitoes in California, human malarial transmission in heavily irrigated regions was no longer tightly coupled with natural precipitation patterns 29 .
We found correlations between mosquito abundance and land use that match previous associations, but climate relationships were somewhat unexpected.Both C. tarsalis and C. erythrothorax abundance increased with wetland area and C. pipiens complex mosquitoes increased with developed land, as expected from other studies 15,23,[55][56][57][58][59][60] .However, while C. tarsalis and C. pipiens complex mosquito abundance increased with temperature in univariate analyses, the significant correlation was not present in multiple regression analyses, suggesting the correlation with temperature for this group of mosquitoes may not have been causal.In addition, there was no evidence that the relationship between C. tarsalis abundance and temperature, was non-linear, as might be expected based on the increased developmental rate and decreased survival rate with temperature [61][62][63] .Finally, only C. erythrothorax abundance was correlated with precipitation, and abundance decreased with increasing precipitation and irrigation, which is somewhat counterintuitive.None of the climate predicators were correlated with seasonal variability in abundance of any of the three species.Overall, the data from California suggest that spatial variation in mosquito abundance and seasonal variability at the scale we examined is more influenced by irrigation and land use than climate.
Three notable limitations affected this study.First, irrigation data was only available at the DAU spatial scale and annual temporal scale.Finer scale analyses may have uncovered even stronger or more detailed relationships, especially for the short-term effects of irrigation on mosquito abundance within a year, which are critical for reducing mosquito larval habitat 64 .The effects of climate variables may have been more apparent at other spatial or temporal scales.Second, we were unable to obtain data after 2010, so we weren't able to examine the impact of multiple severe droughts and floods on mosquitoes and WNV that occurred after 2010.Similarly, it's possible the effects of irrigation on WNV might differ between the initial years after introduction (2004-2010) and the most recent decade.Third, we were not able to incorporate vector control efforts which may have reduced mosquito abundance.Despite these limitations, the strength of the relationships we found between irrigation and mosquito abundance, variability, and WNV suggest that irrigation likely plays an important role in transmission of mosquito borne pathogens in this region.
More broadly, the patterns we observed suggest that manipulations to the water cycle increase the magnitude and reduce the variability of multiple mosquito species.Improvements in irrigation water delivery systems (e.g.fixing leaky irrigation canals), irrigation methods (e.g.sub-surface drip irrigation), irrigation pricing schemes (farmers paying for actual amounts of water used rather than supply), and potentially, intermittent irrigation 64 , might reduce the impacts of irrigation on mosquito populations by limiting potential breeding sites 53,65,66 .A key challenge in the coming decades will be finding ways to use irrigation to maintain food security, without increasing mosquito populations and disease risk.

Irrigation data
We obtained annual irrigation estimates ("Applied water", Acre-feet per acre) for each "detailed analysis unit" (DAU) in California from 1998-2010 from the California department of water resources 51 (2010 was the last year we were able to obtain data).DAUs are sub-watershed scale areas and are the smallest hydrologic study area used for the analysis of water supply and use by the California Water Plan 67 .These annual irrigation estimates for each DAU are based on the growing areas of 20 agricultural crop categories, each with specific crop coefficients corresponding to estimated annual water use 51 .In each DAU, we multiplied the growing area of each crop by estimates of the irrigation water for growing each crop to calculate annual estimates of total irrigated water for each DAU.

Mosquito abundance data
We obtained mosquito trapping data collected by vector control districts across California via the CALSURV vector-borne disease surveillance system 68 .These public health agencies focus mosquito trapping efforts near human population centers, in regions with urban and agricultural land use.We used CO 2 trapping data from the months April through November over a 13-year period during which irrigation data was also available (1998-2010).We included trap sites with at least 10 visits per year (n = 1336 sites), for a total of ~ 80,000 unique site visits in 78 DAU.We calculated average C. tarsalis abundance estimates (mosquitoes per CO 2 trap-night) for each DAU by first by averaging across each year for each site (site-year average), then averaging across sites within an DAU (DAU-year), and then averaging all years.We quantified seasonal variability in C. tarsalis abundance

Temperature and precipitation data
We obtained spatial estimates of average monthly temperature and precipitation data across California at 4 km resolution 36 .We used ArcGIS to calculate average annual estimates of temperature and precipitation for each DAU across California (n = 278), including the 78 detailed analysis regions with mosquito trapping data.We calculated the average annual precipitation (Dec-Nov, in mm) 41 and average temperature (°C) during each mosquito trapping season (Apr-Nov) in each DAU for the years 1998-2010.To make precipitation and irrigation amounts comparable, we converted precipitation (mm) into volumetric amounts of water per acre (Acre-feet per acre; 304 mm/foot).

Land cover data
Within each DAU, we summarized percent land cover of several land cover classes that past analyses indicated were important in mosquito abundance (developed areas, wetlands, open water; 23,31,69 ) using datasets available from 2010 USDA National Agricultural Statistics Service (30 m resolution; 70 ).Developed areas included a combination of low, medium, and high intensity, as well as open space developed.Wetland areas included a combination of woody and herbaceous wetlands.We did not include fraction agricultural land as an independent variable because the irrigation predictor was created using agricultural land use fractions of 20 different crops and was therefore highly correlated with irrigation (r = 0.94, N = 78, P < 0.001).

WNV human disease cases
We compiled reported human cases of WNV for each California county from the Centers for Disease Control and Prevention (CDC) ArboNET program for the years 2004-2010 71 (2010 was the last year we were able to   S3, S4.The three larger points with light green, dark green and black outlines in (B) and (D) show example DAUs shown in panels (A) and (C), respectively.obtain data).Average human WNV disease incidence was calculated as the mean number of all cases (fever and neuro-invasive cases combined) per year divided by the county's population, multiplied by 100,000.We converted DAU estimates of irrigation to county estimates based on weighted averages of county area overlapping with DAU areas.Counties with < 75% overlap with DAU areas were discarded.

Statistical analyses
All statistical analyses were done using the program R (v. 3.5.0)with spatial summaries and maps made using ArcGIS (v.10.5).We used generalized least squares models in the R package gls, with exponential spatial autocorrelation using the DAU centroid, to examine the effects of climate, land cover, and irrigation on the abundance and seasonal variation in mosquito populations, and human WNV disease incidence.We log 10 transformed predictor and response variables to linearize relationships and increase normality of residuals (Table S1).We examined non-linearity in relationships between response variables and temperature by comparing models with a quadratic temperature term in each analysis to models without it.For all response variables, models with quadratic terms had higher Akaike Information Criteria values so linear models were used.We examined normality with Shapiro-Wilke tests and used R package PiecewiseSEM to calculate R 2 values for gls models.S11).WNV human disease incidence is average number of WNV cases per year/100,000 people for 2004-2010.

Fig. 1 .
Fig. 1.Mosquito trap locations, temperature, precipitation and irrigation shown across California for 1998-2010.(A) mosquito CO 2 trap locations (n = 1336) and detailed analysis units (DAU) with mosquito traps (n = 78).(B) Average annual irrigation (typically from April-Oct) for each DAU obtained from California department of water resources 51 .(C) Average C. tarsalis per CO 2 trap-night in each detailed analysis unit {average across sites [average across years (average per trap site-year)]}.(D) Average coefficient of variation in C. tarsalis abundance in each DAU (average across sites (average across years (CV per trap site-year))).

Fig. 2 .
Fig. 2. Yearly and warm seasonal precipitation and irrigation.(A) Average yearly precipitation (Dec.-Nov., solid black line) and average yearly irrigation (dashed blue line) within study region (n = 78 detailed analysis units).(B) Average precipitation during mosquito trapping season (April-November) compared to irrigation within study region.Dashed lines display average across all years.

Fig. 4 .
Fig. 4. Irrigation effects on seasonal variation in mosquito abundance.(A) Example time series for 2010 showing C. tarsalis abundance per CO 2 trap-night at 2 trap sites located in adjacent DAU's which differed in irrigation amounts.(B) Average coefficient of variation in C. tarsalis abundance on a log 10 scale in each DAU {average across sites [average across years (CV per trap site-year)]} plotted against average irrigation amount (Acre-feet/acre) on a log 10 scale in each DAU.Univariate general least squares model including spatial autocorrelation: Log 10 CV C. tarsalis abundance = 2.20-0.065(± SE = 0.016) * log 10 (Irrigation + 0.001); R 2 = 24%; N = 78; P < 0.0001).The full model with land use and climate predictors is shown in TableS5.The two colored points show the same DAUs as the time series in panel A.