Over 150 years of investigations into global terrestrial precipitation are revisited to reveal how researchers estimated annual means from in situ observations before the age of digitization. After introducing early regional efforts to measure precipitation, the pioneering estimates of terrestrial mean precipitation from the late nineteenth and early twentieth centuries are compared to successive estimates, including those using the latest gridded precipitation datasets available. The investigation reveals that the range of the early estimates is comparable to the interannual variation in terrestrial mean precipitation derived from the latest Climatic Research Unit (CRU) dataset. In-depth revisions of the estimates were infrequent up to the 1970s, due in part to difficulty obtaining and maintaining up-to-date datasets with global coverage. This point is illustrated in a “family tree” that identifies the key publications that subsequent authors referenced, sometimes decades after the original publication. Significant efforts to collate global observations facilitated new investigations and improved data exchange, for example, in the International Hydrological Decade (1965–74) and following the establishment of the Global Telecommunication System under the World Weather Watch Programme of the World Meteorological Organization. Also in the 1970s were the first attempts to adjust in situ observations on a global scale to account for gauge undercatch, and this had a noticeable impact on mean annual estimates. There remains no single satisfactory approach to gauge bias adjustment. Echoing the repeated message of past researchers, today’s authors cite poor spatial coverage, temporal inhomogeneity, and inadequate sharing of in situ observations as the key obstacles to obtaining more accurate estimates of terrestrial mean precipitation.
Beginning with regional observations in preindustrial times, we focus on methods used before the age of digitization for determining global precipitation amount over land.
The desire to quantify precipitation1 over land has a long history. In the fourth century BCE, some of the earliest known measurements were made in India, where a network of rain gauges was used to collect observations for land tax calculations and to inform agricultural practices (Biswas 1970). Subsequent short-lived and isolated periods of observation activity arose in various other parts of the world. In Palestine, for example, documents note the use of rain gauges between the second century BCE and second century CE, along with descriptions of rainfall seasonality and the associated stages of crop development (Biswas 1970).
In China, precipitation gauges were used from at least the midthirteenth century (Biswas 1970), and during the Qing Dynasty (1644–1912) an official observation network existed for agricultural purposes, although this used traditional measurement methods—depth of rain infiltration and snow depth above ground—instead of gauges (Wang and Zhang 1988). In Korea in the fifteenth century, a network of rain gauges was established across the nation’s districts, marking the beginning of what would become a long history of measuring precipitation on the peninsula (Chun and Jeon 2005). A key development here was that the gauges, designed by Prince Munjong, were of a standardized size (Biswas 1970; Chun and Jeon 2005).
Meanwhile in Europe, key inventions such as Christopher Wren’s tipping-bucket rain gauge in the 1660s (Bentley 1905; Biswas 1970) heralded a surge of interest in meteorology that extended into the eighteenth century. This period saw organized efforts to collect precipitation data and new motivations for measuring precipitation. France’s Royal Society of Medicine, for example, was interested in the link between climate and disease, and consequently physicians were enlisted to collect data and publish them in “medical topologies” featuring regional information of climate and disease, alongside other geographical information (Feldman 1990). The combined efforts of the society and meteorologist Louis Cotte achieved a network of 150 meteorological observers across France (Feldman 1990). Organized regional efforts were therefore gaining momentum at this time, yet at what point did these quantification attempts turn their attention to the areas beyond national borders?
EARLY INTERNATIONAL NETWORKS AND WORLD PRECIPITATION MAPS.
In Britain in 1723, Royal Society secretary James Jurin made an early endeavor to establish an international observation network, publishing a notice (Jurin 1723) in Philosophical Transactions that called for observers who could record meteorological data, including daily precipitation (Feldman 1990). His request was met by 15 individuals in several countries including the United States, Sweden, and Russia, and their observations were later published in the same periodical (Feldman 1990). Later that century in 1780, the German Palatinate–Bavaria’s state meteorological society established a 37-station network that utilized standardized meteorological instruments and trained observers across continental Europe and in one U.S. city (Cassidy 1985). In total, the society published almost 15 years of observations (Council of the American Academy of Arts and Sciences 1879). Missionaries from France and Russia went farther afield, carrying their instruments to China in the mideighteenth century and recording precipitation observations intermittently in Beijing and Shanghai, until at least the early twentieth century (Wang and Zhang 1988). Such efforts were hardly global nor long-lasting, however, in many cases hampered by difficulties in communication, unsatisfactory performance of instruments, and political unrest (Cassidy 1985; Feldman 1990).
Estimating mean annual precipitation over land requires not only a global distribution of gauge stations but also long-term records of continuous observations. Biswas (1970) suggested that the interest in assessing annual precipitation arose in the mideighteenth century, but it was hindered by insufficient long-term observational data. A century later, however, the continued efforts of organizations and individuals to collect regional observations resulted in some of the earliest maps of global precipitation, which would prove pivotal for the first estimates of terrestrial mean precipitation.
One of the earliest global precipitation maps was by German geographer Heinrich Berghaus (Berghaus 1841). His map identified zones of rainfall over the Earth, and, in a step toward estimating global precipitation, he included quantitative estimates of annual terrestrial rainfall for the tropics and temperate regions. Berghaus’s map was translated and published in English several years later by fellow geographer Alexander Johnston of Scotland (Johnston 1848), amended with his own updates to the data (Freeman 1971). A slight improvement in the spatial resolution of such maps came two decades later, when Russian geographer Alexander Wojeikof published his own map, accompanied by tables of annual mean rainfall over various regions of the Earth (Wojeikof 1874). Like Berghaus’s (1841) map, the zonal categories were mostly descriptive, denoting, for example, “subtropical rain” and “Asiatic monsoon” zones, although there was one semiquantitative category included for annual rainfall totaling over 1,200 mm.
The expansion of gauge networks and the revision of regional precipitation maps facilitated further improvements to global precipitation maps. In the United States, German-American scientist Charles Schott published compilations (e.g., Schott 1881) containing decades of observations from across the United States, where various institutions had maintained networks during the nineteenth century (Schott 1881). These included New York University from 1825 (Blodget 1853), the U.S. Army Medical Department, Pennsylvania’s Franklin Institute, the Smithsonian Institution, and from 1870, the U.S. Signal Corps (Blodget 1853; Schott 1881). By the mid-1860s Schott’s compilation contained observations from some 1,200 stations (Schott 1881).
Growing collections of regional precipitation data such as Schott’s became indispensable to those investigating precipitation on a global scale. One of these investigators, Yale Professor Elias Loomis, included Schott’s data in his own world map of terrestrial mean annual rainfall, published in 1882 (Fig. 1, top). For the Indonesian Archipelago, Loomis sourced data from the world’s first tropical meteorology center, the Royal Magnetic and Meteorological Observatory, established in Jakarta in 1873 (Schereschewsk 1977). The observatory was established by the Dutch in the 1860s to receive records from the archipelago’s meteorological network (Boomgaard 2006), and Loomis (1882) includes data from 124 of these stations. Berghaus’s Physical Atlas (edition unknown) provided Loomis with data for the Guianas, Penang in Malaysia, Trinidad, and Abyssinia (Loomis 1882). Loomis sourced observations for parts of Brazil and various islands of the Lesser Antilles and the Azores from the 1857 collection of German professor and director of the Prussian Meteorological Institute, Heinrich Wilhelm Dove. Dove’s work in itself was the result of two decades spent collating scattered records (Dove 1851).
As a step forward from the maps by Berghaus (1841) and Johnston (1848), which had featured mostly qualitative terrestrial rainfall zones, Loomis’s (1882) map showed the distribution of rainfall as quantitative zones. This attempt did not immediately satisfy Wojeikof, however, who was quick to point out the shortcomings of the map, such as the rainfall depths Loomis assigned some parts of central Asia and tropical South America (Wojeikof 1882). The data insufficiencies in these areas at this time are apparent on the map. For example, the entire area enclosing the Amazon basin, east to Villa Rica and south to Buenos Aires, is assigned just one broad category: 25–50 in. (approximately 650–1,270 mm) of mean annual precipitation (Fig. 1, top).
To compare Loomis’s (1882) map with one of the latest estimates of the climatological mean map of annual precipitation, Fig. 1 (bottom) was provided by the Global Precipitation Climatology Project (GPCP; Adler et al. 2003). Surprisingly, the overall distributions are similar between these two maps estimated more than a century apart. However, close inspection reveals underestimation on Loomis’s map for the middle of the Amazon River basin, the midwest of Canada, central Asia, eastern Siberia, across the western part of Russia, and through to Poland and the Ukraine. This is partially due to the gauge undercatch adjustment applied to the GPCP climatology (see sidebar “How to adjust?” for a discussion of undercatch adjustment).
The problem of gauge undercatch caused by wetting on the interior walls, evaporation, and strong winds, for solid precipitation in particular, has been recognized for as long as terrestrial mean precipitation has been examined (Legates and Willmott 1990). BH30, for example, recognized that their only two observations for the Antarctic were “probably subject to considerable error” (BH30, p. 141) due to the combined effects of strong winds and solid precipitation. However, it was not until the UNESCO (1977) ,Atlas of World Water Balance and its accompanying monograph (UN78), that an attempt to address this bias was made on the global precipitation dataset as a whole (UNESCO 1977; UN78). Further, the exact methods used to adjust the data were not consistent and not all authors chose to adjust their raw data.
UN78 placed considerable emphasis on making undercatch bias adjustments to their data, which goes some way to explaining why their final estimate for global average precipitation is higher than many preceding estimates (Fig. 2) and is a key point of difference from Baumgartner and Reichel (1975, hereinafter BR75) and Baumgartner (1981). The UN78 authors created multiple precipitation maps (published in UNESCO 1977), starting with a global map based on the raw, unadjusted observational data or existing unadjusted regional maps. Next, an adjustment factor for each region of Earth was determined and represented on an intermediate map. The adjustment factor tended to increase with latitude and elevation due to the increasing influence of solid precipitation and strong winds. For Greenland and the Tibetan Plateau, for example, these maps (UNESCO 1977) show an adjustment factor of 60%–70%. Similarly, a large adjustment factor was applied in areas prone to extreme evaporation rates. In some of the desert regions west of the Andes in South America, for example, observed precipitation was adjusted upward as much as 300% (UNESCO 1977; UN78). From these intermediate maps, the final maps (UNESCO 1977) of adjusted average annual precipitation for Earth were produced.
In contrast, BR75 were reluctant to make adjustments for gauge error over such an extensive geographical scale and focused instead on adjusting observational data sourced from the northern polar regions (BR75; Baumgartner 1981). Even in these areas BR75 exercised caution, adjusting upward by 10%–12% in latitudes around 60°N and 14%–33% for 70°N, compared to the authors of the original precipitation maps for those regions who had generally adjusted data upward by 20%–30% (BR75). BR75 did not further adjust terrestrial precipitation in the Canadian Arctic and subarctic regions, Greenland, and mountainous regions because the undercatch problem in these areas had already been accounted for by the original data providers (BR75).
In the later twentieth century, the exact method that should be applied for adjustment remained under constant review and discussion (Legates and Willmott 1990; Adam and Lettenmaier 2003; Legates and McCabe 2005; Schneider et al. 2014). Legates and Willmott (1990) also placed a strong emphasis on adjustment and provided the first climatology adjusted for gauge-induced bias that was available digitally (Legates and McCabe 2005). In contrast, Hulme (1992) disputed the value of making such adjustments at all, with the view that the approach potentially produces even greater errors than those of the original data.
Figure 3 [cf. Legates (1987) bias adjusted and unadjusted] demonstrates the potential difference between adjusted and unadjusted estimates. Certainly, the design of the precipitation gauge itself and its variation over time has also affected the severity of this bias and overall consistency in the long-term record, although discussion of this is beyond the scope of this article. A further difficulty is the absence for many gauge stations of the meteorological observations and metadata detailing specifications of the actual gauge and site setup, which are required to make the adjustments (Schneider et al. 2014). We can conclude that the best method to account for this gauge undercatch bias will most likely continue to be a point of disagreement among future investigations, but this summary shows the important station metadata that are required.
While Loomis’s (1882) map is still rather crude by today’s standards, and while he did not go as far as actually computing terrestrial mean precipitation for the Earth as a whole, his work, which had built upon the work of those before him, is notable for facilitating further investigations by subsequent researchers. This leads us to the focus of the present article, which is to understand how researchers extended these early efforts in order to estimate terrestrial mean precipitation and to discuss the limitations they faced in achieving this task.
THE PIONEERS OF TERRESTRIAL MEAN PRECIPITATION ESTIMATES.
John Murray was one researcher who was quick to make use of Loomis’s (1882) map. As an oceanographer, Murray was interested in quantifying the proportion of precipitation over land that would eventually be discharged back into the sea (Murray 1887). Loomis’s data were a significant help for this purpose, and Murray used them to generate his own rainfall map (Murray 1887), supplemented by the most recently available observations. Using a planimeter, Murray measured the area within each of the five rainfall depth categories on his map, and from this he estimated the total precipitation falling annually on the terrestrial regions of Earth. In January 1887, Murray presented his findings to the Royal Society of Edinburgh (Murray 1887), including an estimate for mean terrestrial precipitation of 840 mm yr–1 (Table 1).
One of the more comprehensive investigations after Murray was by Alexander Supan (Supan 1898). According to one review (Herbertson 1899), Supan’s (1898) investigation was the first to include world maps of seasonal rainfall distribution. After Supan, however, it appears that no entirely new revisions of global terrestrial precipitation estimates were made for over three decades (Fig. 2; Table 1). As stated in Brooks and Hunt (1930, hereinafter BH30) and Jaeger (1983), many authors continued to use Supan’s (1898) map as a basis for their own calculations (Fig. 2). According to BH30, it was used by Bezdek, Fritsche, and Kerner (cited in BH30) in each of their investigations (Fig. 2), for example, and they differed only in the actual methods they used to arrive at their final estimates. German oceanographer Georg Wüst later made a revision of mean terrestrial precipitation, although he used Fritsche’s values supplemented with his own values for the polar regions (Wüst 1922; Fig. 2).
Thus by the time BH30 undertook their investigation, a complete revision of global precipitation was long overdue. BH30 sourced the most up-to-date regional precipitation maps available from around the world. They then divided each map into smaller units of 1° × 1°, and from these they estimated the area within each rainfall depth category. They noted the difficulty of working with such a wide array of regional maps. For example, maps on Mercator projection could be easily divided into the desired square-degree units by ruling straight lines across them, although the authors seemed to overlook the need to adjust for the spatial bias inherent in such maps of nonequal projection. To divide maps with curved coordinates, they required a little more creative visualization, and they estimated the area between rainfall contours at the smallest division by eye (BH30).
BH30 summarized the 1° × 1° values into mean terrestrial precipitation for 5° latitudinal “strips” across Earth, providing both annual and monthly means for these zones. Their resulting estimate for global mean annual terrestrial precipitation was 659 mm, much less than the estimates of those earlier authors such as Fritsche and Wüst, and 200 mm lower than Murray’s in 1887 (Table 1). BH30 concluded that this was mostly due to differences in their own evaluations for the polar and equatorial zones, compared to the prior authors. Murray (1887), for example, estimated mean annual precipitation for the Antarctic continent to be 760 mm—massive compared to today’s estimate of 166 mm (Vaughan et al. 1999, cited in Schneider et al. 2014). BH30 fared better than Murray in their estimate for the frozen continent. Applying an assumption that precipitation decreases toward the center of Antarctica, they used two observations—116 and 265 mm—measured near the Antarctic Peninsula to infer 50 mm of mean annual precipitation for the South Pole (BH30).
These discrepancies between authors’ estimates, arising from calculations based on few actual observations, were symptomatic of the state of the global precipitation dataset as a whole at this time and for the years that would follow. BH30 had criticized the authors preceding them for depending on outdated datasets, in some cases for decades after original publication. However, remaining inadequacies in the global dataset impeded further revisions following BH30, and so this publication itself became the subject of continued reuse in subsequent investigations, for at least another 20 years (Fig. 2; Table 1). Meinardus’s (1934) map (cited in Jaeger 1983), for example, which was at the time regarded as one of the most important in German literature, was based on BH30’s figures (Jaeger 1983; Fig. 2). Not surprisingly, Meinardus’s estimate for terrestrial precipitation was almost identical to BH30, at 665 mm (Fig. 2). In fact, between the late 1930s and 1960s, numerous other authors gave estimates of around 665–670 mm (Table 1). Even as late as 1951, Möller considered BH30’s monthly terrestrial precipitation figures to be the most reliable available at that time, and consequently he used them in his own tables of seasonal variation in precipitation, while referring to Meinardus’s estimate for terrestrial mean precipitation (Möller 1951; see also Fig. 2).
It therefore seems that this pioneering period from the late nineteenth to midtwentieth centuries was characterized by sporadic reevaluations of mean global precipitation, followed by periods of recycling data collated by previous authors, until a sufficient number of new observations were collated to enable the next revision. Confirming this situation, Möller reported in 1951 that the use of outdated data was a persistent problem in the field. Therefore, almost 70 years after Loomis’s (1882) original investigation, midtwentieth-century researchers were apparently struggling with a lack of new observational data, which restricted their ability to regularly revise their estimates. The data problem consisted of not only spatial gaps in the global precipitation gauge network (New et al. 2001) but also the lack of a satisfactory coordinated international effort toward data exchange (Nace 1964). It would require a monumental effort to stimulate worldwide interest in precipitation and the larger hydrological field, before quantification of terrestrial mean precipitation could be improved.
WATER BALANCE IN THE AGE OF THE INTERNATIONAL HYDROLOGICAL DECADE (1965–74).
This lack of available data in the midtwentieth century presented a problem not only for researchers interested in precipitation but also for those investigating the state of global water resources as a whole. From around this time, population growth, industrial development, and agricultural intensification were just some of the factors generating intense pressure on water resources (Keller 1976). Yet it was estimated that in 1950 the amount of runoff, for example, was known for only half the world (Nace 1964). To avoid “a foreseeable crisis of human affairs” (Nace 1964, p. 414), figures such as Raymond L. Nace of the U.S. Geological Survey called for international cooperation, “because neither water nor science recognizes geographic boundaries” (Nace 1964, p. 414). It was hoped that such cooperation would facilitate the international cooperation in data collection, sharing, and interpretation necessary for constructing water inventories and world water balances (Nace 1964).
It was in this context that some of the first international scientific collaborations were organized, regenerating interest in the question of global precipitation and drawing attention to the still-developing field of hydrology. At the forefront of this era of global scientific cooperation was the International Geophysical Year (IGY; 1957–58). With the backdrop of the Cold War, it attracted the participation of scientists from almost 70 countries (Nicolet 1984). The IGY facilitated the establishment of a large number of new meteorological observation stations worldwide, stimulated global interest in precipitation measurement, and, under the recommendations of the World Meteorological Organization (WMO), promoted international standardization of observations (Bleasdale 1959; Nicolet 1984). Overall it was considered to be widely successful and set a high benchmark for global scientific collaboration (Nace 1967; Nicolet 1984; Aronova et al. 2010).
The impetus of the IGY continued into the next decade, and, as meteorological satellites became more common, the demand for observational data to calibrate and verify weather forecasting models also increased (Ashford 1970). Realizing this demand and further recognizing the need for international cooperation, WMO’s Fourth Congress approved the establishment of the World Weather Watch (WWW) Programme in 1963, with multiple objectives including improved availability of global meteorological data for its member states (Ashford 1970; WMO 2005). This objective called for improved observation, data transfer, and processing systems, which would be facilitated by three of its components in particular: the Global Observation System (GOS), Global Telecommunication System (GTS), and Global Data-Processing and Forecasting System (GDPFS; Ashford 1970; WMO 2005).
Shortly following the establishment of WWW, UNESCO convened the International Hydrological Decade (IHD; 1965–74), with the specific objectives of drawing international attention to the field of hydrology and making information on water globally available (Nace 1980). It was long overdue, according to Nace, because until this point “only a few hardy souls had lifted their eyes from the ground at their feet to the hydrological horizon” (Nace 1980, p. 1241). The hydrological cycle—including the precipitation component—was still considered to be poorly understood and its study characterized by a lack of data (Nace 1964). This suggests that, even in light of the achievements of the IGY, further efforts were required to improve the data availability situation described by Möller (1951).
To a large degree, the IHD rose to this challenge, bringing more developments in hydrology in one decade than had been seen in the several that preceded it (Keller 1976). Nace (1980) concluded, for example, that the IHD was especially successful in drawing awareness to the need for international data sharing. The momentum generated by the IHD spilled over to the activities of other programs and agencies, including the International Hydrological Program and the WMO (Nace 1980), the latter of which was the leader of several central activities within the IHD (Nace 1967) and had already contributed hugely to the improved worldwide sharing of regional data by establishing the GTS under WWW (Kohnke et al. 1976; Menne et al. 2012). This action, along with the achievements of the IHD, presumably facilitated the new wave of original investigations into global terrestrial precipitation that were to appear in the years that followed (Fig. 2).
Two IHD-era global precipitation investigations stand out in the literature: the monograph originally published in 1974 (English version 1978) by the U.S.S.R. Committee for the IHD (UNESCO 1978, hereinafter UN78) and the second, a monograph by Baumgartner and Reichel (1975, hereinafter BR75), which was funded in part by the German IHD World Water Balance Research Group. These studies are two of the most detailed after BH30, including revised estimates for the continents within each 5° latitudinal zone and new world maps of precipitation compiled from the most up-to-date regional maps and data, newly available to them thanks to the activities of the IHD (BR75; UN78). Both monographs noted improved understanding of precipitation in the polar, subpolar, and temperate latitudes of the Northern Hemisphere, thus going some way to fill the spatial gaps in the investigations of the previous decades. The end result according to BR75 was a world map of precipitation more comprehensive than any previously published.
It was not only the global precipitation dataset that benefited from the increased hydrological activities of this period. There were newly available data for the other components of the water balance—evaporation E and runoff R—which in turn proved useful for cross-checking estimates of precipitation P where actual observations were limited, such as in mountainous areas (UN78). Similarly, BR75 and UN78 applied the water balance principle to identify zones where the estimated precipitation was too low to satisfy the mass balance equation (i.e., P − E = R). Using this method, BR75 identified underestimation in the subarctic regions, for example, where the combined effects of strong winds and large volumes of solid precipitation exacerbated gauge undercatch. The best method of undercatch bias adjustment was not definitive, however, and is a fundamental point of difference between BR75 and UN78 (see sidebar “How to adjust?”).
Despite relative improvements in the pool of available data brought about by the IHD and the WMO, for example, and even in light of the significant efforts of BR75 and UN78 to conduct comprehensive investigations into global precipitation, spatial and temporal gaps in the global network persisted. For example, while there were precipitation maps for almost every country of the world, data remained scarce for Greenland and Antarctica owing to the impracticalities of data collection in such harsh environments (UN78; BR75). In the several years before BR75, new information became available for Antarctica; however, this was mostly related to the outermost regions of the continent (BR75). Similarly, the meteorological data needed for undercatch error adjustment, such as wind velocity, were lacking, meaning that the adjustment carried out by UN78 was challenging and particularly unreliable for some areas, especially northern Canada and parts of Asia and Africa (UN78; see also sidebar “How to adjust?”).
In addition to the spatial gaps, many areas of the world had only relatively short data records or datasets with intermittent periods of observation, and therefore a long-term global precipitation dataset of uninterrupted temporal coverage was still not a reality. The guiding instructions from UNESCO to the authors of UN78 was to standardize the period of data used so that they only included observations collected between 1931 and 1960 in the investigation (UN78). However, the initial attempt to do this revealed that few regions on Earth had satisfactory datasets available to fit this 30-yr period. The authors of UN78 therefore had to greatly extend the period of observations to 1891–1970 to ensure that they included an adequate number of observations to satisfy the spatial coverage they required (UN78).
THE AGE OF DIGITIZATION.
Toward the end of the twentieth century, a new wave of demand for a comprehensive global gauge-based precipitation dataset appeared. In addition to their continued requirement in water resources investigations, such data became crucial for calibrating satellite-derived estimates and for validating climate models, while the temporal length of the gauge-based record increasingly provided value in global climate change research (Legates 1995; New et al. 2001). Nonetheless, the demand for a homogeneous gauge-based global dataset was not adequately met on the ground in many regions.
Even in recent years, authors continue to reiterate the shortcomings that have plagued the global precipitation dataset since the pioneering investigations of the late nineteenth century: incomplete spatial and intermittent temporal coverage of precipitation observations, the continued question of how to deal with gauge undercatch, and issues with quality control within regional datasets (Legates 1995; New et al. 2001; Schneider et al. 2014; Hegerl et al. 2015; see also sidebar “The question of oceanic precipitation” for discussion of these issues over oceans). Further, the methods for interpolating precipitation depth from gauge-based point observations present additional errors (Legates 1995), and variation in methods employed among the different analyses ultimately produces different precipitation estimates (New et al. 2001). Echoing the authors of the preceding decades, recent authors cite deficiencies in spatial coverage in mountainous, polar, arid, and tropical regions and severe underrepresentation in the global network of regions such as South America and Africa, compared to industrialized nations (Legates 1995; New et al. 2001; Kidd et al. 2017).
Today we are fully aware of the vital interconnection between the meteorological processes on the land and over oceans, and thus the important need to estimate oceanic precipitation. What, however, motivated the first observations of precipitation over the oceans? Sailors recognized that voyage efficiency was at the mercy of the weather; thus merchant ship-based observations dating from the early nineteenth century comprise the earliest sources of oceanic precipitation data (Wilson-Barker 1904; Woodruff et al. 1987), although these were mostly of rainfall frequencies rather than quantitative measurements (Brooks and Hunt 1930).
An early figure to describe oceanic precipitation was Matthew Fontaine Maury, superintendent of the then U.S. Naval Observatory and Hydrographic Office. With the objective of improving shipping routes, he actively gathered vessel logbooks containing routine observations of winds, currents, and other meteorological observations (Maury 1855; Leighly 2003). These were published as charts (e.g., Maury 1855) of seasonal rainfall frequencies over the North and South Atlantic Ocean, based on over 250,000 days of observations. The German Naval Observatory similarly collated the thousands of logbooks distributed to German vessels for recording maritime meteorological observations between the 1830s and 1880s (Kaspar et al. 2015). England, the Netherlands, Austria, France, Spain, and Russia were also active in collecting oceanic rainfall observations during this period (Wilson-Barker 1904). By the early twentieth century, there were still relatively few observations for the Pacific Ocean compared to the Atlantic and south Indian Oceans, however, probably due to the higher concentration of shipping routes in the latter regions and the wider expanse of the Pacific Ocean (Wilson-Barker 1904).
Just as measuring terrestrial precipitation has particular challenges, so too does measurement over oceans. Writing over a century ago, Wilson-Barker (1904, p. 111) exclaimed that, of all types of maritime meteorological data, precipitation is probably “the most unsatisfactory with which to deal.” The physical structure of the ship itself, for example, made it difficult to obtain reliable gauge data (Wilson-Barker 1904). In addition to this, more recent assessments confirm that the instability of the vessel or oceanic platform, and the generally exposed conditions of the open sea, exacerbate the usual gauge undercatch bias (Legates and Willmott 1990; Strangeways 2004). Various alternative methods for estimating global mean oceanic precipitation can be applied that make use of other available data. These include extrapolation of quantitative estimates from the ship-based weather reports, coastal and island gauge observations, or a combination of these sources (e.g., Legates and Willmott 1990).
Precipitation over coasts and islands is not necessarily comparable to precipitation over the open ocean, however, meaning that there is a reluctance to use gauge observations from these areas to directly inform estimates of oceanic precipitation (Reed and Elliott 1973). One-third of the Comprehensive Pacific Rainfall Database (PACRAIN) does comprise atoll stations; however, it is commonly argued that the small landmass of these islands has a negligible effect on precipitation measurement, making such observations useful proxies for the open ocean (Morrissey et al. 1995). The Global Tropical Moored Buoy Array (GTMBA), comprising multinational networks such as the Tropical Atmosphere Ocean/Triangle Trans-Ocean Buoy Network (TAO/TRITON) in the Pacific, measures precipitation from gauges on its ocean surface monitoring buoys in Earth’s tropical oceans (McPhaden et al. 2010), although the number of buoys in the TAO component has been reduced in recent years (Tollefson 2014).
Overall, spatial coverage of gauge-based observations over oceans remains sparse, and accurate measurement is difficult for the reasons described; therefore oceanic precipitation estimates remain unsatisfactory compared to other global water cycle components (Reed and Elliott 1977; New et al. 2001; Trenberth et al. 2007). While the advent of satellite offers to address the spatial shortcomings of surface observations and thus improve estimates over oceans, gauge-based observations remain vital for the calibration of these data (Hegerl et al. 2015; Kidd et al. 2017). Further, satellite data date back only to the late 1970s; therefore surface observations will remain indispensable for the assessments of long-term trends (New et al. 2001).
In fact, UN78, one of the key publications to come out of the IHD, appears to have remained an important reference for those investigating global water resources (e.g., Oki and Kanae 2006), despite it now being decades old. Perhaps this is a tribute to the extensive undertaking that it entailed, but this also suggests that the era in which it was published was a heyday for investigations derived from the global precipitation gauge network.
Supporting this point is that the number of gauge stations in the global network peaked in the 1970s (New et al. 2001), mainly in response to the First Global Atmospheric Research Program (GARP) Global Experiment (FGGE) organized in 1978–79. Further, according to Legates and Willmott (1990), until the time of their own publication, UN78 was the only available climatology adjusted for gauge undercatch on a global scale. Shiklomanov (2003), writing almost 30 years after the close of the IHD, described the state of water resources research as relying on old data, much of it still sourced from UN78. This suggests, again, that new investigations based on gauge observations continued to be hindered by the absence of an adequate global precipitation dataset.
Nevertheless, significant advances were made beyond the 1970s in terms of data digitization (Table 1), which improved the availability of quantitative estimates of precipitation for grid sections, rather than only summaries by latitudinal zones or continents. Notably, Jaeger’s (1976) digitized, gridded global precipitation dataset and maps included mean monthly precipitation, seasonal variation, and mean annual precipitation estimates. To estimate global precipitation within 5° grid sectors, he used climatic maps and additional data. He also attempted to use a standardized period of observations (1931–60), although this was not possible for some regions. Jaeger estimated 759 mm for mean annual terrestrial precipitation over Earth. This was 100 mm higher than BH30 half a century earlier, which may be explained by the higher value he obtained for the equatorial regions (Jaeger 1983).
Another step in the availability of digitized climatologies came in 1990, when the first one adjusted for gauge error became available (Legates and Willmott 1990, cited in Legates and McCabe 2005). Legates and Willmott’s (1990) investigation drew primarily on three digitized datasets [Wernstedt 1972; Willmott et al. 1981; Spangler and Jenne 1984 (cited in Legates and Willmott 1990)], which in themselves had required significant effort to amass. The Pennsylvania State University’s Frederick Wernstedt, for example, spent decades obtaining invaluable datasets, particularly from nonindustrialized countries, where personal contacts had to be exploited and political barriers overcome (Vose et al. 1995). Wernstedt’s dataset, sourced from around 6,000 stations, was one of the largest contributed to version 2.0 of the Global Historical Climatology Network (GHCN), an initiative of the WMO, Carbon Dioxide Information Analysis Center, and National Climatic Data Center, launched in 1990 to improve the quality and exchange of global data (Vose et al. 1995).
Other large contributions to GHCN included Michael Hulme’s Climatic Research Unit (CRU) dataset, sourced from around 8,000 stations, and the National Center for Atmospheric Research’s World Monthly Surface Station Climatology, comprising observations from around 4,000 stations thanks to Roy Jenne’s efforts (Spangler and Jenne 1990, cited in Peterson and Vose 1997; Vose et al. 1995; National Research Council 2007). The GHCN data form a substantial part of the Global Precipitation Climatology Centre’s (GPCC) full database, which itself comprises data the GPCC obtained from the National Meteorological and Hydrological Services (NMHSs) in almost 200 countries, as well as large historical datasets for underrepresented regions such as Africa (Schneider et al. 2014).
Today, global precipitation data centralization and quality control practices have improved for databases such as those maintained by the GPCC and CRU and a number of other publicly available precipitation gauge datasets (the International Precipitation Working Group maintains a list of these, available at www.isac.cnr.it/∼ipwg/data/datasets4.html). The GPCC, for example, has developed a semiautomated quality control system in order to detect errors in the raw data, addressing them before the data are added to the databases (Schneider et al. 2014). Importantly, time series of monthly precipitation are also available through such databases, including the CRU TS datasets (Harris et al. 2014) and GPCC’s full dataset, the latter comprising monthly totals from over 85,000 stations (Schneider et al. 2014).
Even with these improvements, however, the remaining issues of spatial coverage and temporal inconsistencies cannot be resolved unless the global observation network is maintained and improved and data exchange is kept a priority. Not only has the number of precipitation stations in the worldwide network been shrinking since the 1970s (New et al. 2001), but the lack of timely availability of in situ observations from source countries is also a key obstacle preventing the realization of a comprehensive global precipitation dataset (Hegerl et al. 2015; Kidd et al. 2017). The GPCC’s full database, for example, peaks at around 47,000 stations in the late 1980s, followed by a decline, which is attributed to the time it takes for data to reach the GPCC and then pass through its quality-control system (Schneider et al. 2014).
HOW FAR HAVE WE COME IN 130 YEARS OF TERRESTRIAL PRECIPITATION ANALYSES?
Recounting over a century of endeavors to quantify terrestrial mean precipitation from in situ observations reveals that the difficulties faced by the pioneering researchers continue to hamper efforts to make more definite estimates today. Despite the relative success of the International Hydrological Decade, which facilitated two exhaustive studies of world water balance, and even with the improved infrastructure for global data exchange through WWW and its GTS, for example, today’s issues echo those of the early researchers: poor data density in developing regions and in difficult areas such as mountainous and polar regions, long-term datasets characterized by gaps in records, and difficulty addressing gauge undercatch error due to poor availability of the necessary ancillary data and lack of a definitive adjustment method (see sidebar “How to adjust?”).
The CRU, GPCC, and other agencies promote data centralization and sharing; however, it appears to be at ground level where difficulties persist, as indicated, for example, by the global decrease in the number of gauges in recent decades (New et al. 2001; Hegerl et al. 2015). Furthermore, collating regional precipitation observations for global analyses remains challenging owing to the discrepancies in data availability among the local collection agencies (Kidd et al. 2017).
What is remarkable, however, in a direct comparison of selected investigations spanning a 130-yr period (Fig. 3), is the similarity between the terrestrial mean precipitation estimates. Even the earliest estimate by Murray (1887) falls within the interannual range of the most recently available CRU data. It is a testament to pioneering researchers such as Murray (1887) and Brooks and Hunt (1930) that they were able to achieve this despite the daunting task of pulling together far-flung datasets and analyzing them in a predigitized world. Further, the impetus of the IHD, the activities of the WMO, and numerous other individuals are duly recognized for stimulating new investigations and improving data sharing.
Despite this, the final message remains that, because of these issues of data availability, estimates of “global” mean precipitation in reality represent less than one-quarter of Earth’s surface (New et al. 2001). Even the GPCC’s dataset, one of the world’s largest for monthly precipitation totals (Schneider et al. 2014), is estimated to represent just 1% of Earth’s surface area (Kidd et al. 2017). This number is likely to become even smaller if the number of gauges worldwide continues to decline. Satellite-derived observations provide improved coverage in areas poorly represented by gauge stations, and they form the basis of datasets such as GPCP, the Tropical Rainfall Measuring Mission Multisatellite Precipitation Analysis (Huffman et al. 2007), Global Satellite Mapping of Precipitation (Ushio et al. 2009), Tropical Applications of Meteorology Using Satellite Data (Tarnavsky et al. 2014), and others, yet these datasets still depend on in situ observations for calibration. The momentum of past eras in hydrometeorology, such as the IHD, perhaps needs to be reignited, so that the indispensable role of the humble rain gauge in global precipitation research is not forgotten.
This work was conducted under the framework of the “Precise Impact Assessments on Climate Change” supported by the SOUSEI and TOUGOU Program of the Ministry of Education, Culture, Sports, Science, and Technology (MEXT), the Japan Society for the Promotion of Science’s Grants-in-Aid for Scientific Research (KAKENHI) (16H06291), and the Environment Research and Technology Development Fund (S-10 and S-14) of the Ministry of the Environment, Japan. Figure 1b was courteously provided by Drs. Jackson Tan (USRA; NASA/GSFC), George J. Huffman (NASA/GSFC), and Chris Kidd (University of Maryland; NASA/GSFC). We would like to express our gratitude to Petra Döll and Jacob Schewe for kindly providing the German literature. We sincerely thank David R. Legates for his invaluable, detailed review and further help sourcing additional references. We also extend our thanks to two anonymous reviewers for their constructive comments.
This article intends the term “precipitation” to mean rainfall and snowfall. However, the authors of the historical references we cite commonly used the term “rainfall” when in some cases their observations may have included snowfall. For these references we have retained the original terms as they were given by the authors.
L’vovich (1970) cited the 1939 map by Drozdov, while UNESCO (1977) mentioned earlier maps by Drozdov (1937), however neither authors included Drozdov (1939 or 1937, respectively) in their reference lists. A full reference for these Drozdov maps therefore remains a mystery to the present authors. It is possible that Drozdov’s map is the global map of annual precipitation published in the Boleshoi Sovetskii Atlas Mira (Great Soviet World Atlas) (Motylev et al. 1937, map 34), however inspection of this map did not reveal any reference to Drozdov.