Urban air pollution and industrial air pollution have been shown qualitatively to suppress rain and snow. Here, precipitation losses over topographical barriers downwind of major coastal urban areas in California and in the land of Israel that amount to 15%–25% of the annual precipitation are quantified. The suppression occurs mainly in the relatively shallow orographic clouds within the cold air mass of cyclones. The suppression that occurs over the upslope side is coupled with similar percentage enhancement on the much drier downslope side of the hills. The evidence includes significant decreasing trends of the ratio of hill to coast precipitation during the twentieth century in polluted areas in line with the increasing emissions during the same period, whereas no trends are observed in similar nearby pristine areas. The evidence suggests that air-pollution aerosols that are incorporated in orographic clouds slow down cloud-drop coalescence and riming on ice precipitation and hence delay the conversion of cloud water into precipitation. This effect explains the pattern of greatest loss of precipitation at the midlevel of the upwind slopes, smaller losses at the crest, and enhancement at the downslope side of the hills.
The objective of this study is the first-ever quantification of the microphysical effect of air-pollution aerosols on precipitation on a regional scale (tens to several hundreds of kilometers). Previous studies have shown qualitatively that urban and industrial air pollution suppresses precipitation-forming processes in convective clouds (Rosenfeld 1999, 2000). The pollution aerosols serve as small cloud condensation nuclei (CCN) that form large concentrations of small cloud droplets. This process in turn suppresses the drop coalescence and the warm-rain processes, as well as the ice precipitation (Rosenfeld 2000; Borys et al. 2003), and so prolongs the time required to convert the cloud water that exists in small drops into large hydrometeors that can precipitate. Borys et al. (2003) show that the addition of as little as 1 μg m−3 of anthropogenic sulfate aerosols to a clean background can reduce the orographic snowfall rate in the Colorado Rocky Mountains by up to 50%. The suppression is stronger in shallower clouds with warmer top temperatures. Satellite observations showed that pollution can completely shut off precipitation from clouds that have temperatures at their tops >−10°C (Rosenfeld and Woodley 2003). Therefore, one can expect to find the greatest rain suppression in regions that are dominated by relatively short-lived clouds with relatively warm tops downwind of major urban areas. Because of their short life, such clouds are more sensitive to the slowing down of the conversion of cloud water to precipitation, whereas long-lived clouds would eventually convert their water into precipitation regardless of the conversion rate. The urban heat island has been documented previously as the main cause for precipitation enhancement in the warm season downwind of major urban areas (Changnon 1979; Changnon et al. 1991; Shepherd et al. 2002). Therefore, we had to select regions in which the precipitation is dominated by clouds that are not thermally driven, preferably formed by orographic lifting over mountain ranges downwind of pollution sources during the cold season. We used annual rainfall data because almost all of the precipitation in the areas of interest occurs in the winter or in winterlike storms during the spring and autumn.
The study areas and data collection
In the ideal, the effect would be most pronounced downwind of coastal cities with hills inland that receive precipitation mainly during the winter in maritime onshore flow from shallow convective clouds. The main effect would be, therefore, the suppression of the orographic component of the precipitation, which would be manifested as a reduction in the orographic enhancement factor Ro, where Ro is defined as the ratio between the precipitation amounts at the hills and at the upwind lowland. Such conditions are abundant, especially on the west coast of continents in the subtropics and midlatitudes, where the precipitation over hills is a major source for the scarce water there. This study analyzes historical records of precipitation from California and from the land of Israel as representative of these conditions (see maps in Fig. 1).
The main analysis tool is the time series of Ro based on annual precipitation from rain gauges downwind and sidewind of major urban areas. The underlying assumption is that small-particulate air-pollution emissions have increased with the growth of urban areas, resulting in a decrease in Ro with time. The sidewind Ro time series are not expected to show any trend with time and so serve as controls.
Results and discussion
Trend analyses of rain gauges
Figures 2–5 display the longest annual time series that could be found in California and Israel for urban downwind and sidewind pairs of rain gauges. According to Fig. 2d, Ro decreased by 28% in the mountains to the east of San Diego, California, during the twentieth century, most of it after 1940 when San Diego started to grow, and more recently with the explosive growth of Tijuana just across the Mexican border. No such decrease could be detected farther north, midway between San Diego and Los Angeles, California, but a decrease has occurred again downwind of the Los Angeles (Fig. 6a), Fresno, California, and San Francisco, California, areas. Farther north, in the sparsely populated northern California, again no decreasing precipitation trends were observed (Fig. 3c). Furthermore, there were some statistically insignificant increases. A similar situation is evident in Israel (Figs. 4d, 5d). These trends occurred consistently in all of the pairs of rain gauges that were tested and are summarized in the tables in the appendix. The decreasing trend of Ro does not necessarily mean an absolute decrease of precipitation, because the precipitation decreases may have occurred in the context of long-term increases in the overall precipitation (see absolute trends in Figs. 2a, 2b, 3a, 3b, 4a, 4b, 5a, and 5b).
Many more rain gauges have been available for such trend analysis for the last 60 yr, making it possible to compare clusters of rain gauges in the hill and upwind plain areas. Their analysis shows a clear signal of decreasing Ro in the San Diego, Los Angeles, and San Francisco areas and in central Israel. Additional analyses of other areas in California and Israel (see the appendix) show that the decreasing trends occur at the western slopes of the hills that are located downwind of pollution sources. Again, such a trend was not found in hills downwind of pristine areas. The continuous decrease of Ro is consistent with the increasing trend of air pollution in Israel throughout the period. However, the values of standard air-quality measurements in California have improved since the late 1970s (Malm et al. 2002) while Ro continues to decrease at a somewhat lesser rate, as shown in the appendix. This pattern raises a question as to whether it is really the particulate air pollution that is responsible for the observed trends in Ro.
The radiosonde model
The most likely alternative explanation to the reduction in Ro is a decreasing trend in the cross-mountain component of the low-tropospheric wind velocity and moisture flux during rain events. Orographic precipitation has already been related quantitatively to the low-level winds and moisture in both California (Pandey et al. 1999; Neiman et al. 2002) and Israel (Alpert and Shafir 1991; Rosenfeld and Farbstein 1992). We applied the radiosonde regression model after Rosenfeld and Farbstein (1992) to predict the daily rain amounts in the hilly stations, where the upwind coastal precipitation and the cross-mountain 850-hPa wind velocity and the absolute humidity at that level are RMM = RCM(WS × W), where RMM is the predicted precipitation in the mountains (mm day−1), RCM is the gauged precipitation at the coast (mm day−1), WS is the wind speed component toward the mountain (m s−1), and W is the mixing ratio (g kg−1).
Radiosonde data have been available since the early 1950s for both California and Israel. The model results show that the ratio between the measured rain and the predicted rain (which is unaffected by pollution) for polluted mountain areas decreased in both California and Israel in a manner very similar to the observed trends of Ro. The model results for clean areas show no difference between the measured and predicted rain. In essence, this result reflects that the relevant meteorological conditions during rain days did not change systematically along the years, and the observed trends in Ro are likely caused by nonmeteorological reasons, such as anthropogenic air pollution.
In line with these considerations, despite the reported decreases in the pollutant emissions in California during the last two decades, the total amount of soluble pollution ions in precipitation particles aloft has not shown any decreasing trend and even showed a slight increasing trend in Sequoia National Park (Table 1). Therefore, the expectation for a recent recovery of Ro with the improving standards cannot be supported by these observations of the recent trend of steady to increasing concentrations of pollution in the precipitation. The contaminants in the precipitation were monitored in the Sierra Nevada at elevated locations that eliminated the possibility of scavenging the pollutants below cloud base, because the monitoring stations were mostly above cloud-base level during precipitation events. Local sources for pollutant species such as SO2, NOx, and NH3 were found to be relatively unimportant in the Yosemite and Sequoia National Parks and were attributed to transport from the urban areas and from the oil industry in the southern San Joaquin Valley (Collett 1989; Collett et al. 1989, 1990; Takemoto et al. 1995; Carroll and Dixon 2002). Indeed, the combined concentrations of non–sea salt ions in the precipitation (Ca, Mg, K, NH4, NO3, and SO4) at the mountain stations of the Sierra Nevada, showing a significant decreasing trend of precipitation (Fig. 7), were 2 times the concentration found in the northern Sierra Nevada, where no trend in Ro is indicated (see Table 1). These ions, especially sulfates and nitrates, are major constituents of pollution-produced CCN particles.
Trend of very small CCN
A prime suspect for the lack of recovery of Ro, despite the decreasing levels of standard measures of air pollution in California, is the constancy of the very small aerosols that account for the bulk of the CCN concentrations. A major source of very small (<0.1 μm) aerosols, which are as efficient as CCN as are biomass smoke particles, is diesel engines (Lammel and Novakov 1995). A diesel car produces several orders of magnitude more such particles per mile than a gasoline car with the same fuel consumption (Pierson and Brachaczek 1983; Williams et al. 1989; Lowenthal et al. 1994; Weingartner et al. 1997; Maricq et al. 1999). The consumption of diesel fuel in transportation has been increasing in California at 2 times the pace of gasoline consumption since 1980, as has, also, the production of these small CCN aerosols, which are not reflected in any of the standard air-quality measures. Furthermore, these small and numerous particles have the greatest potential for precipitation suppression. In contrast, the larger (>1 μm) pollution particles could actually induce large drops and enhance precipitation, but these particles are the ones that have been most effectively eliminated from the emissions.
Classification by synoptic conditions
The next step focused on the cloud types that are most vulnerable to the suppression effects. We expect that most strongly affected would be the clouds forming in the cold sector of the cyclone, because their roots are near the surface so that they can readily ingest the air pollution, as was found by Dayan and Lamb (2003). In addition, the cloud-top temperatures in the cold sector are typically much higher than for the frontal and warm-sector clouds. The roots of the latter clouds are usually not connected directly to the surface, especially in the cold season, but rather feed on long-range transport of moisture. These frontal and warm-sector clouds are forced typically by synoptic lifting, so that their formation is much less dependent on orographic lifting than the clouds in the cold sector, which experience little synoptic forcing. Because all that we had to work with is the radiosonde data, we divided the rain days into “warm” and “cold” classes, delimited at a temperature of −3°C at 700 hPa. This threshold was selected because it is the highest temperature that would still allow convection through that level, starting from a typical winter sea surface temperature of 15°C off the coast of southern California. According to Fig. 7, the orographic enhancement Ro was much greater [by a factor of 5 at the start (S) of the period and a factor of 3.28 at the end (E)] for the clouds in the colder situation. Respectively, Ro for the clouds in the cold conditions showed a strong decrease (E/S = 3.28/5.00) in accordance with our conceptual expectations, although it is not statistically significant because of the large scatter. The scatter of the points for the warmer conditions is much smaller, in agreement with the large degree of organization of the frontal and synoptically forced cloud systems versus the poorly organized postfrontal clouds, which is responsible for the large scatter and the lack of statistical significance in the colder conditions. Similar analysis could not be done for Israel because almost all rainfall there occurs from the cold air masses. Nevertheless, this process might explain the large effect in Israel despite the modest height (up to 1 km) of the hills.
Trend analyses of mountain snowpack water content
All of the analyses presented up to this point suggest that anthropogenic aerosols suppress the orographic precipitation in stations located at elevations lower than 2 km on the upwind slopes of topographic barriers. The next natural question is, What happens when the clouds are forced across very high barriers and reach very cold temperatures, where aerosols suppress precipitation to a lesser extent because ice precipitation processes become more efficient? The answer was obtained from end-of-winter measurements of the water value of the snowpack near the divide of the Sierra Nevada downwind of Sacramento and Fresno. These data showed only a small (7%–8%) and statistically insignificant decreasing trend of Ro. To make sure that air pollution was still delaying the formation of precipitation in these clouds, the trend of Ro farther upwind and at lower elevations was tested and was found to decrease strongly (22%–24%) and highly significantly (Fig. 8). Similar analysis in pristine areas in northern California showed near-zero to slightly positive trends of Ro at all elevations (see the appendix).
Downslope compensating effects
Next, we explore the question of what happens downwind of the ridgeline to the excess cloud water that was not converted into precipitation because of the suppression of precipitation on the upslope side. It is apparent that the added cloud water and newly formed precipitation particles that pass over the divide cause enhancement of the precipitation on the downslope side. This effect occurs in all areas with available data where suppression was observed on the upslope side—both in California and Israel (see, e.g., Figs. 8, 9). No trends in Ro were found at the upslope or downslope of hills downwind of pristine areas. The opposite trends on the western and eastern slopes downwind of pollution sources are of comparable magnitude in percent, but because the absolute amount of precipitation over the western slope is about 4 times that over the eastern slope (in the case shown in Fig. 9), the net result is dominated by the decrease of Ro over the western (upwind) slopes.
Summary of the results
Figures 10, 11, and 12 summarize the results both in California and Israel. Figure 10 displays the trends in Ro in polluted areas downwind of Sacramento, Fresno, and Los Angeles in California and in the Samaria and Judea Hills in Israel. In Fig. 11 it can be seen that no such trend accrued in pristine areas that are located north of San Francisco and in the Hebron Mountains of Israel. Figure 12 shows in a schematic cross section from west to east the full process as it occurs in the polluted mountains of the Sierra Nevada in California (downwind of Fresno). It can be seen that the western slopes are the most sensitive areas to the air-pollution effect. The decrease in the orographic ratio between the plains and the western slopes because of this effect is around 20% (statistically significant). This trend becomes weaker as elevation increases, and the opposite trend occurs in the eastern slopes of the Sierra Nevada. The proportion of the postfrontal precipitation from the colder air masses, which is more orographically controlled (see Fig. 7), has not decreased with time at all, especially at the coastal stations. For example, the fraction of rainfall that occurs with 700-hPa T < −3°C, which is associated with the more orographic rainfall, has increased between 1952 and 2000 from 52% to 57% in San Diego while decreasing from 73% to 72% in Cuyamaca (see details in the appendix). This result does not leave any conceivable alternative mechanism of which the authors are aware, except for the aerosol effects that can explain the observed patterns of suppression of the orographic component of the precipitation at the upslope side of the western slopes while increasing it on the eastern slopes.
An important question is the net gain or loss of precipitation along the whole cross section. This is calculated in Table 2 for a 1-km-wide strip across the average width of the Sierra Nevada stages with their respective rainfall changes as defined in Fig. 12. The reduction in the western slopes dominates the overall hydrological budget, incurring a large net loss of about 20 million m3 for each 1-km segment of the mountain ranges. This means an overall loss of 4 × 109 m3 yr−1 of precipitation water just for the 200-km-long section of the mountains that are located to the east of the line between Sacramento and Fresno. This calculation is highly oversimplified and provides us merely with the magnitude of the problem. More exact calculations and their hydrological meaning are required. Further, these precipitation losses may not even be evident to water managers because of masking by the long-term increases in the overall upwind base level of precipitation.
In this study, we analyzed only two geographical areas for which we had the physical basis to expect that such unfavorable redistribution of precipitation occurs and for which quality precipitation data at the hilly areas were available to us. Similar effects can be expected in other areas, such as the Snowy Mountains and Victorian Alps in Australia (Rosenfeld 2000), the Atlas Mountains, the Mediterranean coastal hill ranges, Chile, Puerto Rico, and many more locations.
In this study, we avoided addressing the possible confounding effects of the glaciogenic cloud seeding of the orographic clouds that has been taking place in both Israel and California. If seeding did enhance precipitation, the effects in the absence of seeding may have been larger than indicated in this study. Additional research is needed to separate the seeding and pollution effects. In addition, measurements of CCN are needed to strengthen the case for pollution as causal of the apparent losses in precipitation.
In summary, strong circumstantial evidence quantitatively links anthropogenic air pollution and the suppression of orographic precipitation downwind of the pollution sources by 15%–25%, in the following ways:
The decreasing trend is linked to the period of urbanization and industrialization upwind, whereas similar analysis of the orographic rainfall in nearby pristine areas showed no trends.
The suppression occurs mainly in the relatively shallow clouds within the cold air masses of cyclones, which ingest the pollution from the boundary layer while ascending over the mountains.
The suppression that occurs over the upslope side is coupled with a similar percentage of enhancement on the much drier downslope side of the hills, probably because more cloud water passes over the divide.
The main hydrological recharge zones of the water resources in the study areas overlap with the areas for which large and statistically significant suppression of precipitation was measured, with water losses ranging between 15% and 25% of the annual precipitation. The downwind areas in which compensatory enhancement occurs have a much lower absolute amount of precipitation (about 25%) than the upslope side of the hills, and, therefore, the compensatory relative increases on the downwind side are manifested as much smaller amounts of added water when compared with the losses in the upslope areas. For example, a net loss of precipitation water volume of about 4 × 109 is estimated over the mountains to the east of the line connecting Fresno and Sacramento. These are startling results for regions that already now experience severe water shortages and have to resort to seawater desalination (in Israel) to meet their water needs. In addition to the obvious ramifications for water resources, climate impacts are also important and must be considered.
This study was funded by the Israeli Ministry of Science and by the Eshcol Foundation. The authors thank Dr. W. L. Woodley for valuable discussions and for reviewing the manuscript.
Description of the Additional Tables
This appendix contains additional information about the tables that are presented in this paper. Many more cases and results from California and Israel are discussed here to show that the results that are presented in the article are examples of the general situation. Each of Tables A1–A13 describes different aspects of the discussion, which together help to show the full picture of the time series that we present. The sources for the rain guage data are the National Climatic Data Center, the California Department of Water Resources, and the Israeli Meteorological Service. The names of the stations appear as in the datasets.
The tables contain any of the following information: station name—the rain gauges or snowpack name; lat, lon—the geographical latitude and longitude (in decimal degrees); years—the analyzed period in starting and ending years; elev—the elevation of the rain gauges or snowpacks (m MSL); avg yearly precipitation—the average annual amount of precipitation at each station for the tested years (mm yr−1); correlation—the correlation coefficient between the annual rainfall of pairs of stations; E/S—the ratio (between the precipitation amounts at the hills and at the upwind lowlands) in the end of the time series divided by the ratio in the beginning gives the change, as calculated using the regression line, which is the trend along the years; P value—statistical significance of the trend corresponding to the t-test statistic, which is the probability that the indicated trend occurs because of random variability and is not a real trend (values ≤ 0.05 are boldfaced to highlight their significances); E/S precipitation—same as E/S, but for individual rain gauge time series, not pairs of station ratios; cluster—groups of rain gauges or snowpacks that represent the average annual precipitation in a selected geographical area.
Corresponding author address: Daniel Rosenfeld, Institute of Earth Sciences, The Hebrew University of Jerusalem, Jerusalem 91904, Israel. firstname.lastname@example.org