1. Introduction
Outgoing longwave radiation (OLR) estimates have been made since June 1974 from the window channel measurements of the operational National Oceanographic and Atmospheric Administration (NOAA) polar-orbiting satellites (Gruber and Winston 1978; Gruber and Krueger 1984). These estimates have been used extensively in meteorological studies, both as one component of the radiation balance of the atmosphere (Ohring and Gruber 1982) and to infer changes in the amount and height of clouds. In particular, the dominant role of clouds in determining OLR over the Tropics has led to extensive use as an indicator of convective activity, andits broad spatial coverage and long period of record have made it invaluable in investigating atmospheric phenomena with a variety of temporal and spatial scales (e.g., Winston and Krueger 1977; Murakami 1980a–c; Heddinghaus and Krueger 1981; Liebmann and Hartmann 1982; Lau and Chan 1983a,b, 1985, 1986; Weickmann 1983; Weickmann et al. 1985; Wang 1994).
In the Tropics, where surface temperatures vary modestly through the annual cycle, the strongest variations in OLR result from changes in the amount and depth of clouds. This direct physical connection with clouds led to the use of OLR in quantitative precipitation estimation beginning in the early 1980s. The first published work of this sort was by Lau and Chan (1983a), in which the number of days in a month with OLR less than 240 W m−2 was used as a predictor of monthly precipitation and calibrated against estimates based on microwave observations over tropical oceans (Rao et al. 1976). Morrissey (1986) analyzed the relationships among the OLR, precipitation, and moisture budget over the tropical Pacific and found that daily OLR correlates negatively and significantly with precipitation over tropical areas where deep convection dominates. Motell and Weare (1987) related the total OLR flux to precipitation as observed by gauges at carefully selected islands over the tropical Pacific and developed a linear statistical model to estimate monthly precipitation from total OLR flux. Weare (1987) applied this model to produce gridded fields of monthly precipitation over the tropical Pacific for the 10-yr period from June 1974 to February 1984 and used them to investigate the relationship between precipitation and SST. Yoo and Carton (1988) estimated total precipitation over the tropical Atlantic Ocean from OLR for the period 1983–84. Arkin (1984)investigated the relationship between seasonal mean OLR and gauge-observed precipitation for 2.5° lat–long grid areas over the tropical Pacific and found that 57% of the variance in the total seasonal precipitation could be explained by a linear function of the mean OLR. Arkin et al. (1989) compared the Geostationary Operational Environmental Satellite (GOES) Precipitation Index (GPI) (Arkin 1979; Arkin and Meisner 1987) withOLR as derived from the Indian National Satellite System and found that the spatial distribution patterns of the OLR are very similar to those of the precipitation as estimated by the GPI, with low values of OLR corresponding quite closely to large amounts of precipitation. Janowiak and Arkin (1991) used a regression analysis to develop a linear equation to estimate precipitation from OLR over the global Tropics. These estimates have been used to fill in the gaps of the GPI estimates over areas where geostationary satellites do not provide coverage (Huffman et al. 1995; Xie and Arkin 1996).
All of these techniques use the total OLR flux, in one way or the other, as the predictor and have been developed to estimate precipitation for tropical areas where the total OLR flux is strongly modulated by deep convective clouds and where most of the precipitation is associated with deep convection. None of these techniques is applicable to extratropical areas where variations in surface temperatures are as important as or more important than cloudiness in determining the magnitude of the total OLR flux. However, Arkin (1984) showed that the seasonal anomaly of precipitation in the Tropics is also strongly negatively correlated with that of OLR. He found that 45% of the variance in the seasonal precipitation anomaly over the tropical Pacific can be explained by the OLR anomaly, compared to 57% of the total precipitation amount by the total OLR flux. Whether this relationship between OLR anomaly and precipitation anomaly holds over extratropical areas and for other temporal scales (e.g., monthly) has not been determined, largely because of the lack of high quality global precipitation datasets. However, Susskind et al. (1997) have shown that monthly anomalies in OLR are correlated positively with surface temperature over land and negatively with cloudiness over the ocean in mid- and high latitudes. Since continental temperature anomalies are often correlated with precipitation, and since total cloud cover might be expected to exhibit a positive correlation with precipitation, it seems worth investigating the relationship between OLR and precipitation globally. This has been made a practical possibility by the recent success of global monthly precipitation analyses based on the merging of various kinds of satellite and in situ observations (Huffman et al. 1995; Xie and Arkin 1996; Huffman et al. 1997).
The objective of this paper is to investigate the relationship between precipitation and OLR globally, and to develop a new technique to estimate monthly precipitation anomalies from the OLR data. The primary motivation for this study comes from the increasing requirements from both the meteorological and hydrological communities for a global precipitation dataset with high quality and extended temporal coverage. Although recent efforts to merge the various existing individual data sources have succeeded in quantitatively describing global monthly precipitation with complete spatial coverage and improved quality (Huffman et al. 1995; Xie and Arkin 1996; Huffman et al. 1997), these merged analyses rely heavily on the GPI (Arkin and Meisner 1987) and on products derived from microwave observations made from the Special Sensor Microwave/Imager (SSM/I). Unfortunately, these are not available for the period before January 1986 and July 1987, respectively. To make possible the extension of the merged analysis to periods before 1986, other satellite-based precipitation estimates must be found. One such estimate is that based on observations of the Microwave Sounding Unit produced by Spencer (1993). However, those estimates are limited to oceanic regions. The extended recording period and the near-complete global coverage of the OLR observations made by the NOAA polar-orbiting satellites, and the use of such data, both implicitly and explicitly, as precipitation estimates in many regions encouraged us to investigate the possibility of retrieving precipitation information from the OLR data over the globe.
Section 2 describes the data used in this study, while section 3 depicts the relationship between precipitation and OLR over various areas. Section 4 describes the development and verification of the OLR-based precipitation estimation technique, section 5 illustrates some applications of a 22-yr time series of precipitation estimates constructed by the new technique, and conclusions are given in section 6.
2. Data
We begin by comparing monthly OLR to analyses (gridded fields) of precipitation over land based on gauge observations alone, and for the entire globe derived from a combination of gauge observations, estimates based on satellite imagery, and numerical model predictions. We will refer to the gridded fields as the gauge-based and merged analyses, respectively. All three datasets are defined on a global grid of 2.5° latitude × 2.5° longitude, although the gauge-based analysis is available only over land areas.
The OLR data used here are derived from window channel measurements of the Advanced Very High Resolution Radiometer (AVHRR) on the NOAA operational sun-synchronous polar-orbiting satellites. The flux is estimated for day and night separately from IR window channel brightness temperatures using a nonlinear regression based on comparisons to broadband earth radiation budget observations by Ohring et al. (1984). The total OLR flux for a month is obtained by averaging all nighttime and daytime estimates within the period. The monthly set of total OLR flux is available from the Climate Prediction Center (CPC) of NOAA for the 22- yr period from June 1974 to December 1995, with a 10- month gap from March to December 1978. Technical details concerning the definition of the dataset can befound in Gruber and Winston (1978) and Gruber and Krueger (1984).
The gauge-based analysis used in this study was constructed by Xie et al. (1996) and consists of 2.5° latitude × 2.5° longitude gridded fields of global monthly precipitation created by statistically interpolating station observations of monthly precipitation collected from the Climate Anomaly Monitoring System of NOAA/CPC and from the Global Historical Climatology Network of the Carbon Dioxide Information Analysis Center of the Department of Energy’s Oak Ridge National Laboratory. The gauge-based analysis covers the 25-yr period from 1971 to 1995 and is available for global grid areas with 50% or more land coverage. Xie et al. (1996) showed that the gauge-based analysis is capable of adequately representing the spatial and temporal variability in large-scale precipitation, although bias remains over gauge-sparse areas.
As the gauge-based analysis does not cover oceanic areas, the atoll gauge precipitation dataset of Morrissey and Greene (1991) is compared to the OLR data over the tropical Pacific. The atoll gauge dataset comprises station observations of monthly precipitation from over 100 gauges located on atolls and small islands without high terrain. Monthly mean precipitation for all 2.5° lat–long grid areas with at least one gauge is defined as thearithmetic average of the monthly values for all gauges within the grid box. The atoll gauges are mainly located over the western Pacific along a northwest to southeast axis extending from 10°N, 140°E across the equator to 20°S, 140°W (Xie and Arkin 1995, their Fig. 1). The number of gauges in each grid area varies from one to eight, with an average of two to three.
The merged analysis of global monthly precipitation (Xie and Arkin 1996) was constructed by combining five kinds of individual data sources with different characteristics: a gauge-based analysis (Xie et al. 1996), three kinds of estimates based on satellite observations of IR (GPI; Arkin and Meisner 1987), microwave scattering (Grody estimates; Grody 1991; Ferraro et al. 1994) and microwave emission (Chang estimates; Wilheit et al. 1991), and numerical model predictions from the model used at the European Centre for Medium- Range Weather Forecasts (ECMWF) (Arpe 1991). The merged analysis is defined by a two-step algorithm (Xie and Arkin 1996), in which the three kinds of satellite estimates and the model predictions are first combined linearly through the maximum likelihood estimation method to reduce the random error inherent in the individual data sources. The output of this first step is then blended with the gauge-based analysis using the method of Reynolds (1988) to remove the bias. Tests showed that the quality of the merged analysis was improved substantially compared to the individual data sources, with random error reduced significantly and bias removed almost completely. The merged analysis of global monthly precipitation used in this study has complete spatial coverage over global land as well as oceanic grid areas and extends for the 8-yr period from July 1987 to June 1995. Note that the OLR data are not completely independent from the merged analysis, since both the OLR and the GPI estimates are obtained from satellite observation in the infrared window near 11 μm. Thus, the results of our comparison must be used cautiously over tropical and subtropical areas where the GPI estimates (and in some instances, estimates based directly on OLR) are used as one of the five input data sources. The impact of this interdependence should be limited by the fact that the merged analysis is determined principally by gauge observations over land, while over ocean it is defined by a combination of the IR-based estimates with three other data sources (Xie and Arkin 1996). Furthermore, IR-based estimates are not used at all in the merged analysis poleward of 40°.
3. OLR–precipitation relationship
Generally speaking, the OLR at the top of the earth’s atmosphere is modulated by two factors: clouds and surface temperature, both of which are correlated to some degree with precipitation. Frequent occurrence and broad coverage of deep clouds with cold tops are highly characteristic of convective precipitation (Richards and Arkin 1981; Arkin and Xie 1994) and depress OLR, resulting in a negative correlation between total precipitation and OLR in regions where convective precipitation dominates. In mid- and high latitudes, warm surface temperatures are associated with elevated OLR and increased available moisture, and to some extent with heavier precipitation than found with cold surface temperatures. Of course, this correlation is confounded by the fact that, for a given surface temperature, increased cloud cover of any type, and hence decreased OLR, exhibits some correlation with increased precipitation. We will attempt to disentangle these relationships by examining the annual and interannual variability of OLR and precipitation separately.
We begin by comparing the merged analysis of precipitation of Xie and Arkin (1996) and the monthly OLR flux obtained from the NOAA/CPC for the 8-yr period from July 1987 to June 1995. The total precipitation amount and the total OLR flux are each partitioned into a mean annual cycle and an anomaly, which are defined for each 2.5° × 2.5° grid area and for each calendar month as the mean value for the 8-yr period and the difference between the total and the 8-yr mean for that month, respectively. Correlation coefficients between the total OLR and total precipitation, and between their mean annual cycles and anomalies, are computed over various areas and for various seasons. Figures 1 and 2 show time series of the pattern correlation for the total value (thin solid line), mean annual cycle (thick solid line), and anomaly (dotted line) over seven latitudinal zones over land and oceanic areas, respectively. Figure 3 presents the global distribution of the temporal correlations for the totals and their two components. These correlations are statistically significant at the 1% level for values greater than 0.2 in Figs. 1 and 2 and greater than 0.26 in Fig. 3.
Over tropical land areas (Fig. 1: 20°N–20°S), both the total flux and the mean annual cycle of OLR exhibit large stable negative correlations with their corresponding precipitation fields, with the mean annual cycles showing better pattern agreements than the totals. This result agrees well with earlier findings (e.g., Arkin 1984;Motell and Weare 1987; Arkin et al. 1989; Janowiak and Arkin 1991) that variations in OLR over the Tropics are dominated by variations in deep cloudiness and canbe used to estimate total precipitation through a simple linear model. The correlation between the anomalies is negative as well, although the absolute value of the correlation is smaller than those for the totals and the mean annual cycles, with values consistently around −0.4. This indicates that the anomaly of OLR is modulated by the anomaly of coverage and/or height of clouds that precipitate.
Over subtropical land areas (Fig. 1: 20°–40°N–S), the correlations for both the totals and the mean annual cycles exhibit relatively small but clear annual cycles, with maxima (r ∼ −0.2) and minima (r ∼ −0.8) during winter and summer, respectively. The influence of precipitating clouds on OLR is still seen but is substantially less strong during the cold season. Linear models such as those of Motell and Weare (1987) and Janowiak and Arkin (1991) can still be confidently used to estimate precipitation during the warm season, but their use will result in large errors during the cold season. However, the anomaly correlation exhibits no strong seasonality and remains near a value of −0.4, similar to that found over tropical land areas.
Over midlatitude land areas (40°–60°N–S), the correlations, of the total value and the mean annual cycle exhibit seasonal variations similar to, but larger in amplitude than, those observed over subtropical land areas, ranging from 0.2 during winters to −0.8 during summers. Variations associated with precipitating clouds appear to strongly influence the OLR during summer, while variations associated with surface features apparently are more important during the winter. As over tropical and subtropical land areas, the correlation between the anomalies of OLR and precipitation is relatively stable with negative values of about −0.4.
Over high latitudes (Fig. 1: 60°–90°N–S), the positive correlations observed for both the total flux and the mean annual cycle indicate that the annual cycle in OLR there is associated with that of surface temperature. However, the correlation of the anomalies exhibits an interesting pattern, particularly in the Northern Hemisphere. Correlations are strongly negative (<−0.4) during summer and rise to peak values of near or slightly above 0 during the fall and early winter. The relationship between anomalies in OLR and precipitation in high latitudes is clearly complex, with interannual variations in both clouds and snow cover potentially important. The poor quality of precipitation observations available for use in the merged analysis, as well as the difficulty in accurately measuring snowfall, might also contribute to the confusion.
The correlation between precipitation from the merged analysis and OLR over oceanic areas is shownin Fig. 2. The results are similar to those described above for land areas, except that the magnitude of the correlation is larger and the amplitude of the annual cycle in the correlation is smaller than observed over land areas. The simplest explanation for this is the simplicity of the radiating surface, which restricts the possible influences on OLR to changes in surface water and/or ice temperature and changes in cloudiness.
The latitudinal dependence of the relationship between OLR and precipitation inferred from Figs. 1 and 2 is confirmed by the spatial distribution of the serial correlation between OLR and precipitation as obtainedfrom the merged analysis of Xie and Arkin (1996) for the 8-yr period (Fig. 3). The correlations between both the total values and the mean annual cycles exhibit negative values over tropical and subtropical areas with positive values over most mid- and high-latitude areas. Extremely abrupt transition zones are found between the two regimes. The correlation between the anomaly of precipitation and that of OLR, however, is negative over the entire globe except for Antarctica, with higher values observed over the Tropics.
These results show clearly that the relationship between OLR and precipitation differs depending upon whether one examines the mean annual cycle or the interannual variability. The mean annual cycle of OLR is dominated by two competing factors: variation in cloud extent and top temperature, and variation in surface temperatures, both of which are related to precipitation but in opposite ways in terms of correlation. The amplitude of the annual cycle is large compared to the anomalies, making it difficult to establish a globally uniform and seasonally independent simple relationship between precipitation and total OLR flux. The interannual variability of OLR, however, is more closely related to changes in cloudiness, which has larger interannual variability than does surface temperatures. The separation of anomaly from total flux makes it possible to identify the precipitation information in the OLR data in a simple and consistent way.
To further investigate the possibility of making quantitative estimates of precipitation anomaly from OLR, we performed a regression analysis between the anomaly of the merged analysis of precipitation and the anomaly of OLR for each 2.5° lat–long grid area for the 8- yr period from July 1987 to June 1995. Figure 4 shows the distributions of the proportional coefficient (top; in mm day−1 w−1 m2) of the regression equations for the globe. The intercepts are very small everywhere, while the proportional coefficient (note that the sign has been reversed) exhibits coherent spatial inhomogeneity, with variations from over −0.25 in the tropical western Pacific to less than −0.025 in central Russia. These results suggest that, while the anomaly of precipitation is proportional to the anomaly of OLR, adoption of a globally uniform proportional coefficient will result in substantial systematic error in converting the OLR anomaly to the precipitation anomaly. Yoo and Carton (1988) compared OLR flux with the precipitation data available from gauge observations over several islands in the tropical Atlantic and found similar spatial dependence in the OLR–precipitation relationship.
However, the spatial variability of the proportional coefficient is similar to that of the mean precipitation from the merged analysis for the 8-yr period (Fig. 4, bottom), with areas of large (negative) proportional coefficient corresponding roughly to those of heavy precipitation. Figure 5 shows the scatterplot illustrating the relationship between the proportional coefficient (Fig. 4 top) and the mean precipitation (Fig. 4, bottom). It is clear that there is a linear relationship between the two quantities. Least square linear fits were conducted for global land and oceanic areas, respectively, and the results are shown in Table 1. The correlations between the proportional coefficient and the mean precipitation are relatively strong (≤−0.7), while the slopes and the intercepts obtained exhibit little difference among land and ocean areas, suggesting that the proportional coefficient relating the precipitation anomaly and the OLR anomaly can be expressed with usable accuracy as a globally uniform linear function of the mean precipitation.
4. Precipitation estimation from OLR
a. Estimation method
Thus, from (1)–(5), precipitation (P) for a particular month for each grid box can be estimated from δOLR and the long-term MP at that point for that month. The new technique has been applied to estimate the global monthly precipitation for the 22-yr period from June 1974 to December 1995 (March–December 1978 missing). For convenience, hereafter we will call the estimates derived by this technique the OLR-based precipitation index (OPI). An example of the OPI for August1987 is shown in Fig. 6, together with the merged analysis of Xie and Arkin (1996) for comparison. Generally speaking, the OPI exhibits similar but smoother patterns of large-scale precipitation as those found in the merged analysis over both tropical and extratropical areas. It appears to underestimate some areas of heavy precipitation (e.g., over the western Pacific) while overestimating light precipitation (e.g., over Russia).
b. Cross validation
Before the OPI can be utilized in any application, it is necessary to investigate its performance in estimating global monthly precipitation and to examine the stability of the coefficients employed in the technique. For that purpose, cross validations are conducted. First, 2 yr are selected randomly from the 8-yr period and the data for the remaining 6 yr are used to develop the technique, which is then applied to estimate the global monthly precipitation for the withheld 2 yr. These procedures are repeated four times so that the entire 8-yr period is covered by the estimates based on the 6-yr datasets.
Since the estimates for each year are independent ofthe data used to develop the technique, comparison of the estimates with the merged analysis of precipitation should be able to help us get insight into the performance of the new technique, while examination of the differences among the coefficients (A and B) obtained in the four tests, which are based on slightly different datasets, might help us understand the stability of the technique. Figure 7 shows the time series of the pattern correlation between the estimates defined in the cross validations and the merged analysis for several latitude bands over land (solid line) and ocean (dotted line) areas; Fig. 8 presents the distributions of the serial correlation (top), bias (middle), and rms difference; Table2 gives the coefficients (A and B) obtained from the four tests; and Table 3 summarizes the comparison results for the combined time–space domains over various areas.
The correlation (Fig. 7; Fig. 8, top; and Table 3) is near 0.8 for all seasons and for all latitudes over both land and oceanic areas, except over Antarctica, where the merged analysis is less reliable. The bias (Fig. 8, middle) is less than ±1% of the merged analysis of precipitation over most global areas, while under- and overestimates are observed over areas with heavy and light precipitation, respectively. The relative rms differences (Fig. 8, bottom) are less than 50% over mostareas with substantial precipitation and larger than 100% only for some areas with little precipitation. The correlation between the proportional coefficient C and the MP (Table 2) exhibits relatively high values, about 0.7, for all four tests. Only minor differences are observed in the coefficients (A and B, Table 2) among the four tests, and all of them are very close to those calculated by using data for the entire 8-yr period (Table 1) suggesting that coefficients based on data for 6 yr or longer are relatively stable.
c. Comparison with gauge observations
To further examine the ability of the new technique to estimate global monthly precipitation, the OPI estimates are compared with the gauge-based precipitation analysis of Xie et al. (1996) over global land areas and with the atoll gauge data of Morrissey and Greene (1991) over the tropical Pacific Ocean for the 22-yr period from June 1974 to June 1995. To get insight into the performance of the OPI relative to other satellite estimates, the same comparisons are conducted for the IR-based GPI (Arkin and Meisner 1987) and the scattering-based Grody (Ferraro et al. 1994) estimates. Since the gauge-based analysis exhibits significant value-dependent bias over gauge-sparse areas (Xie et al. 1996), the comparisons are limited to grid areas with at least one gauge.
Figure 9 shows the distributions of mean precipitation from the OPI, GPI, Grody estimates, and the gauge-based analysis of Xie et al. (1996) for the 8-yr period from July 1987 to June 1995. Since the OPI is based on the 8-yr merged analysis of Xie and Arkin (1996), which uses the gauge data as one of its input data sources, the OPI is not independent of the gauge data for that period. All three kinds of satellite estimates display similar large-scale patterns of mean precipitation. The OPI estimates exhibit better agreement in amplitude when compared to the gauge-based analysis than do the other two satellite estimates, which overestimate precipitation over land areas (e.g., over tropical Africa and South America). To ensure that the success of the OPI in estimating precipitation over global land areas is not due entirely to its dependence on the gauge observations for that period, we compare the mean distributions of precipitation from the OPI estimates and the gauge-based analysis for the period from June 1974to June 1987 (Fig. 10), during which the OPI estimates are completely independent of the gauge data. The GPI and the Grody estimates are not included in this comparison because the GPI estimates are only available for a short time (from January 1986), while the Grody estimates are not available at all. Once again, the patterns of the OPI estimates are similar and the amplitude is very close to the gauge-based analysis over tropical as well as extratropical land areas.
To further evaluate the performance of the three satellite estimates in estimating precipitation over various areas and for various seasons, quantitative comparisons are conducted with the gauge-based analysis over global land areas and with the atoll gauge data over the tropical Pacific Ocean. Tables 4 and 5 present the comparison results for the subperiod from June 1974 to June 1987 when the OPI is independent of the gauge data and forthe subperiod from July 1987 to June 1995 when the OPI is partially dependent on the gauge data, while Figs. 11–14 show the time series of the calculated statistics for the tropical Pacific and for tropical (20°S–20°N), subtropical (20°–40°N), and midlatitude (40°–60°N) land areas, respectively. Generally speaking, the OPI estimates exhibit high skill in estimating precipitation for all seasons and for almost all global areas except Antarctica, with correlations of ∼0.8, bias of ±10%, and random error of 50%–100% with little difference in the results for the two subperiods. Over tropical areas, the OPI estimates exhibit similar correlations and random error as those for the GPI and the Grody estimates but with smaller bias. Over extratropical areas, the OPI estimates perform better than both the GPI and the Grody estimates and display no seasonal variation in the correlation, bias, and the random error as observed forthe other two satellite estimates. Noticeable is the continuously increasing trend of the bias of the OPI estimates from 1974 to 1985. While we did not investigate this carefully, it is known that substantial changes in OLR have been caused by changes in the instruments and equator-crossing times in different satellites (Chelliah and Arkin 1992), and recent work by Zhou and Waliser (1997) indicates that a correction is possible.
To explore the reasons for the excellent performance of the OPI in estimating global precipitation, time series (Fig. 15) of the mean annual cycle of precipitation (thick solid line), the anomalies as defined in the merged analysis (thin solid line), and in the OPI estimates (dotted line) are plotted for 2.5° lat–long grid areas centered at 178.75°E, 1.25°S and 118.75°E, 28.75°N, which were selected to represent tropical and extratropical areas, respectively. Over the tropical grid area, the magnitudeof the mean annual cycle and the anomaly are comparable, and the estimated anomaly from the OPI correlates well (r = 0.921) with that in the merged analysis. Over the extratropical grid area, the anomaly is reproduced by the OPI with less accuracy (r = 0.643 for the grid) compared to that over the tropical grid area, but the mean annual cycle there dominates the total precipitation, limiting the impact of the estimation error for the anomaly on the total value. It is the combination of the successful definition of the mean annual cycle by the merged analysis and the skillful estimation of the anomaly from the OLR data that makes it possible for the OPI to estimate global monthly precipitation with spatially uniform and seasonally independent high accuracy.
5. Application
The spatial distribution and temporal variation of global precipitation as observed in the OPI estimates are investigated for the 22-yr period from 1974 to 1995 and compared to the long-term means of Jaeger (1976, referred to hereafter as Jaeger) and Legates and Willmott (1990, referred to hereafter as L–W). We will describe the spatial distributions of OPI, together with the Jaeger and L–W climatologies for the full year (Fig. 16), and for the December–January–February (DJF—Fig. 17) and June–July–August (JJA—Fig. 18) periods.
In general, all three datasets exhibit similar distribution patterns of large-scale precipitation characterized by rainbands associated with the intertropical convergence zone (ITCZ), the South Pacific convergence zone(SPCZ), the midlatitude oceanic storm tracks, and the continental monsoon maxima. A distinct split in the ITCZ is observed over the central Pacific in the annual mean of the L–W climatology (Fig. 16), while Jaeger exhibits smoother distributions for the rainbands. A large area of heavy precipitation appears in the eastern Pacific in L–W, but there is no corresponding features in either Jaeger or the OPI estimates. The rainbands associated with the storm tracks over the southern oceans are strongest in Jaeger, while they are relatively weak in the OPI estimates and L–W. During the DJF period (Fig. 17), the SPCZ is strong and the ITCZ is relatively weak over the central and eastern Pacific compared to JJA (Fig. 18). The midlatitude oceanic storm tracks exhibit a connection with the ITCZ in the summer hemisphere, while they are rather separated in the winter hemisphere. Similar behavior is exhibited in Jaeger and L–W, except that the seasonal variations of the ITCZ over the central and eastern Pacific are not as strong as in the OPI estimates. In Jaeger, the ITCZ is generallyweaker than that found in the OPI, while in L–W, the ITCZ is much stronger during both the DJF and JJA periods.
We also used the OPI estimates to examine the interannual variability associated with the El Niño–Southern Oscillation (ENSO) cycle. Shown in Fig. 19 are the time series of SST anomaly over the Niño3 area (5°S–5°N; 150°–90°W) and the Hovmö_er diagram for the precipitation averaged over the Pacific sector (140°E–100°W). The annual cycle of the precipitation is quite significant and is clearly modified by the SST anomaly. The magnitude of the estimates becomes larger and large values extend farther south when the positive SST anomalies are found, while the precipitation is weaker in general when the SST is lower than normal.
Since the Hovmö_er diagram is not able to reveal the east–west variation in precipitation, spatial distributions are composited for warm and cold ENSO episodes, respectively, and the differences are investigated. Although there are many ways to define an ENSO episode,we adopted a rather simple and objective approach in which a cold/warm ENSO episode is declared for a season or a year if the SST anomaly over the Niño3 area is ≤−0.5°C or ≥0.5°C for the period. The classification based on this method results in four cold and six warm event years for the 22-yr period, which are in general agreement with those of Ropelewski and Halpert (1989) based on the Southern Oscillation index (SOI). Figure 20 shows the distributions of the difference observed during the warm and cold events for the DJF and JJA periods, in which areas that are significantly (at the 90% level) wetter/drier during warm ENSO episodes are shaded in heavy/light hatching. During DJF, warm episodes are characterized by more precipitation over the central tropical Pacific, the southern portion of the South America, the extreme northeastern Pacific, and over a belt extending from eastern Pacific, across the Gulf of Mexico well into the Atlantic, and by less precipitation over the western Pacific and the northern portion of the South American continent. During JJA, there is more precipitation all the way across the Pacific and less precipitation over some of the southeastern monsoon area during warm episodes. The storm track over the northwestern Pacific is farther north than during the cold episodes, bringing more precipitation over Japan and its adjacent oceanic areas during warm episodes. Although this comparison is based on a relatively short period, the results agree well with those of Ropelewski and Halpert (1987, 1989), which are based on longer historical records of gauge observations.
6. Conclusions
The relationship between OLR and precipitation has been investigated using the OLR data estimated from the window channel observations of the NOAA polar- orbiting satellites and the merged analysis of global precipitation of Xie and Arkin (1996) for the 8-yr period from July 1987 to June 1995. The total flux and the mean annual cycle of OLR are dominated by two factors: clouds and surface temperatures, with differing degrees of influence in different latitudes. This results in strong negative correlations between total OLR and precipitation over tropical areas, where the effect of clouds is greatest, and weak positive correlations in higher latitudes, where the effects of surface temperatures dominate. The anomaly of OLR, meanwhile, correlates negatively with precipitation for every season and all latitudes except over Antarctica, presumably reflecting the dominant effect of interannual variability in cloudiness on the OLR and the positive correlation between cloudiness and precipitation. Quantitative comparisons using the merged analysis showed that the anomaly of precipitation is proportional to the anomaly of OLR and that the proportional coefficient, which is regionally and seasonally dependent, can be defined with high accuracy by a globally uniform linear function of the mean precipitation.
These findings provided the basis for a new technique to estimate global monthly precipitation on a 2.5° lat–long grid from OLR data. First, the mean annual cycle of precipitation is defined for each grid area and for each month from the 8-yr merged analysis of Xie and Arkin (1996). The precipitation anomaly is then estimated from the anomaly of OLR flux using proportional coefficients that are defined as a linear function of the mean annual cycle of precipitation. Finally, the total precipitation is defined by adding the anomaly to the mean annual cycle. Cross validation and verification against independent observations showed that the estimates, or OPI, based on the new technique are able to estimate global monthly precipitation with spatially uniform and temporally stable high quality that is similar/superior to that of the IR-based GPI and the microwave- scattering-based Grody estimates over tropical/extratropical areas.
OPI estimates of global monthly precipitation on a 2.5° lat–long grid were produced for the 22-yr period from 1974 to 1995 and used to describe the spatial and temporal variations of global precipitation. The mean distribution and seasonal variation, which are determined by the merged analysis of Xie and Arkin (1996), are in good agreement with those of Jaeger (1976) and Legates and Willmott (1990), although some differences are observed. The interannual variability associated with the El Niño–Southern Oscillation is similar to that found by previous studies based on historical gauge observations, but with additional interesting details, especially over the oceans.
While useful for describing interannual variability in precipitation over the globe in an internally consistent manner, the OPI estimates undoubtedly are subject to important errors and more work is needed to improve the estimation technique. In particular, the OPI is, to a significant extent, derived from other precipitation estimates, including the GPI and those based on the SSM/I, through their inclusion in the merged analysis. In areas such as the tropical oceans, accuracy of the OPI in estimating total and anomalous precipitation is less than that of the GPI. It is important to note that the mean distributions of OPI are determined by the source used for the mean annual cycle, in this case the 8-yr merged analysis of Xie and Arkin (1996). Anomalies in OPI, on the other hand, are much less dependent on the specific source used to determine the mean annual cycle. The extension of the merged analysis and improvements to its quality are necessary to ensure the accuracy and the stability of the mean annual cycle and thereby the total estimates of precipitation based on the OPI. The implementation of the Tropical Rainfall Measurement Mission (TRMM) project (Simpson et al. 1988), which will for the first time obtain radar observations of rainfall from space, will help to clarify the uncertainty in the actual amount of precipitation over oceanic areas equatorward of 35°, and will thus improve the quality of the merged analysis and the OPI. Futurespace-based radar observations of precipitation at higher latitudes will be more useful when used in conjunction with the OPI.
In order to ensure spatial completeness, the version of the merged analysis used in this paper included forecasts from the operational model of the ECMWF in areas where other observations were unavailable. Since some applications, such as numerical weather prediction forecast model validation, require a dataset based solely on observations, while others, such as diagnostic studies of the global hydrologic cycle, require a spatially complete and temporally homogenous dataset, two versionsof the OPI have been constructed and made available to the scientific community. One is defined based on a merged analysis using observation-based data sources only, while the other is based on a merged analysis that supplements these sources with the precipitation distributions from the National Centers for Environmental Prediction–National Center for Atmospheric Research reanalysis (Kalnay et al. 1996). Differences between these two products are small and similar to the differences between the model and no-model versions of the merged analysis (Xie and Arkin 1996).
Furthermore, while correlations consistently in thevicinity of −0.6 indicate that the relationship between OLR anomalies and precipitation anomalies is sufficiently strong in the Tropics and midlatitudes to justify this approach, the density and quality of precipitation observations in high latitudes is inadequate to support strong conclusions. Where the best gauge data are available, in Canada and northern Eurasia, correlations between OLR anomalies and precipitation anomalies resemble those in lower latitudes. However, correlations over Antarctica, where several long-term and presumably high quality gauge time series are available, do not support this conclusion. Improvement of the gauge- based precipitation analysis is needed to understand these issues.
Despite these issues, the OPI appears to provide the best coverage, both in space and time, of any set of estimated precipitation based on satellite observations. Since the OLR flux data cover a relatively long period compared to other satellite observations, and since they are available on a quasi-real-time basis, the high quality global precipitation estimates of the OPI are expected to be useful in many-applications such as climate analysis, model verification, and climate monitoring.
Acknowledgments
The authors would like to express their thanks to C. Ropelewski and J. Janowiak of the Climate Prediction Center (CPC) and A. Gruber of the National Environmental Satellite Data and Information Service (NESDIS) of NOAA for invaluable discussions throughout this research. Reviews of an early version of this paper by F.-Z. Weng, R. Ferraro, R. Joyce, and T. Oki were most helpful. Comments by Dr. Joanne Simpson and two anonymous reviewers were extremely helpful.
REFERENCES
Arkin, P. A., 1979: The relationship between fractional coverage of high-cloud and rainfall accumulations during GATE over the B- scale array. Mon. Wea. Rev.,107, 1382–1387.
——, 1984: An examination of the Southern Oscillation in the upper tropospheric tropical wind field. Ph.D. dissertation, University of Maryland, 240 pp. [Available from Dept. of Meteorology, University of Maryland, College Park, MD 20742.].
——, and B. N. Meisner, 1987: The relationship between large-scale convective rainfall and cold cloud over the Western Hemisphere during 1982–84. Mon. Wea. Rev.,115, 51–74.
——, and P. Xie, 1994: The Global Precipitation Climatology Project:First Algorithm Intercomparison Project. Bull. Amer. Meteor. Soc.,75, 401–419.
——, A. V. R. K. Rao, and R. R. Kelkar, 1989: Large-scale precipitation and outgoing longwave radiation from INSAT-1B during the 1986 southwest monsoon season. J. Climate,2, 619–628.
Arpe, K., 1991: The hydrological cycle in the ECMWF short-range forecasts. Dyn. Atmos. Oceans,16, 33–60.
Chelliah, M., and P. A. Arkin, 1992: Rotated principal component analysis of outgoing longwave radiation anomalies over the global tropics. J. Climate,5, 371–389.
Ferraro, R. R., N. C. Grody, and G. F. Marks, 1994: Effects of surface conditions on rain identification using SSM/I. Remote Sens. Environ.,11, 195–209.
Grody, N. C., 1991: Classification of snow cover and precipitationusing the Special Sensor Microwave Imager. J. Geophys. Res.,96, 7423–7435.
Gruber, A., and J. S. Winston, 1978: Earth–atmosphere radiative heating based on NOAA scanning radiometer measurements. Bull. Amer. Meteor. Soc.,59, 1570–1573.
——, and A. F. Krueger, 1984: The status of the NOAA outgoing longwave radiation data set. Bull. Amer. Meteor. Soc.,65, 958–962.
Heddinghaus, T. R., and A. F. Krueger, 1981: Annual and interannual variations in outgoing longwave radiation over the tropics. Mon. Wea. Rev.,109, 1208–1218.
Huffman, G. J., R. F. Adler, B. R. Rudlof, U. Schneider, and P. R. Keehn, 1995: Global precipitation estimates based on a technique for combining satellite-based estimates, rain gauge analysis, and NWP model precipitation information. J. Climate,8, 1284–1295.
——, and Coauthors, 1997: The Global Precipitation Climatology Project (GPCP) combined precipitation dataset. Bull. Amer. Meteor. Soc.,78, 5–20.
Jaeger, L., 1976: Monatskarten des Niederschlags fur die ganze Erde. Berichte des Deutscher Wetterdienstes, Offenbach, 33 pp. and plates.
Janowiak, J. E., and P. A. Arkin, 1991: Rainfall variation in the tropics during 1986–1989, as estimated from observations of cloud-top temperature. J. Geophys. Res.,96, 3359–3373.
Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-year reanalysis project. Bull. Amer. Meteor. Soc.,77, 437–472.
Lau, K. M., and P. H. Chan, 1983a: Short-term climate variability and atmospheric teleconnections from satellite-observed outgoing longwave radiation. I: Simultaneous relationships. J. Atmos. Sci.,40, 2735–2750.
——, and ——, 1983b: Short-term climate variability and atmospheric teleconnections from satellite-observed outgoing longwave radiation. II: Lagged correlations. J. Atmos. Sci.,40, 2751–2767.
——, and ——, 1985: Aspects of the 40–50-day oscillation during northern winter from outgoing longwave radiation. Mon. Wea. Rev.,113, 1889–1909.
——, and ——, 1986: Aspects of the 40–50-day oscillation during northern summer as inferred from outgoing longwave radiation. Mon. Wea. Rev.,114, 1354–1367.
Legates, D. R., and C. J. Willmott, 1990: Mean seasonal and spatial variability in gauge corrected, global precipitation. Int. J. Climatol.,10, 111–127.
Liebmann, B., and D. L. Hartmann, 1982: Interannual variations of the outgoing IR associated with tropical circulation changes during 1974–78. J. Atmos. Sci.,39, 1153–1162.
Morrissey, M. L., 1986: A statistical analysis of the relationships among rainfall, outgoing longwave radiation, and the moisture budget during January–March 1979. Mon. Wea. Rev.,114, 931–942.
——, and J. S. Greene, 1991: The Pacific atoll raingage data set. Tech. Rep. Univ. Hawaii, Manoa, Honolulu, HI, 45 pp. [Available from School of Ocean and Earth Science and Technology, University of Hawaii, Manoa, Honolulu, HI 96822.].
Motell, C. E., and B. C. Weare, 1987: Estimating tropical Pacific rainfall using digital satellite data. J. Appl. Meteor.,26, 1436–1446.
Murakami, T., 1980a: Empirical orthogonal function analysis of satellite-observed outgoing longwave radiation during summer. Mon. Wea. Rev.,108, 205–222.
——, 1980b: Temporal variations of satellite-observed outgoing longwave radiation over the winter monsoon region. Part I: Long- period (15–30 day) oscillations. Mon. Wea. Rev.,108, 408–426.
——, 1980c: Temporal variations of satellite-observed outgoing longwave radiation over the winter monsoon region. Part II: Short- period (4–6 day) oscillations. Mon. Wea. Rev.,108, 427–444.
Ohring, G., and A. Gruber, 1982: Satellite radiation observations and climate theory. Advances in Geophysics, Vol. 25, Academic Press, 237–304.
——, ——, and R. Ellingson, 1984: Satellite determinations of therelationship between total longwave flux and infrared window radiance. J. Appl. Meteor.,23, 416–425.
Rao, M. S. V., V. Abbott III, and J. S. Theon, 1976: Satellite-Observed Global Oceanic Rainfall Atlas. NASA, 31 pp.
Reynolds, R. W., 1988: A real-time global sea surface temperature analysis. J. Climate,1, 75–86.
Richards, C. F., and P. A. Arkin, 1981: On the relationship between satellite-observed cloud cover and precipitation. Mon. Wea. Rev.,109, 1081–1093.
Ropelewski, C. F., and M. S. Halpert, 1987: Global and regional scale precipitation patterns associated with El Niño/Southern Oscillation. Mon. Wea. Rev.,115, 1606–1626.
——, and ——, 1989: Precipitation patterns associated with high index phase of the southern oscillation. J. Climate,2, 268–284.
Simpson, J., R. F. Adler, and G. R. North, 1988: A proposed tropical rainfall measuring misson. Bull. Amer. Meteor. Soc.,69, 278–295.
Spencer, R. W., 1993: Global oceanic precipitation from the MSU during 1979–91 and comparisons to other climatologies. J. Climate,6, 1301–1326.
Susskind, J., P. Piraino, L. Iredell, and A. Mehta, 1997: Characteristics of the TOVS Pathfinder Path A dataset. Bull. Amer. Meteor. Soc.,78, 1449–1472.
Wang, B., 1994: Climatic regimes of tropical convection and rainfall. J. Climate,7, 1109–1118.
Weare, B. C., 1987: Relationship between monthly precipitation and SST variation in the tropical Pacific region. Mon. Wea. Rev.,115, 2687–2698.
Weickmann, K. M., 1983: Intraseasonal circulation and outgoing longwave radiation modes during Northern Hemisphere winter. Mon. Wea. Rev.,111, 1838–1858.
——, G. R. Lussky, and J. E. Kutzbach, 1985: Intraseasonal (30–60- day) fluctuations of outgoing longwave radiation and 250 mb streamfunction during northern winter. Mon. Wea. Rev.,113, 941–961.
Wilheit, T. J., A. T. C. Chang, and L. S. Chiu, 1991: Retrieval of the monthly rainfall indices from microwave radiometric measurements using probability distribution functions. J. Atmos. Oceanic Technol.,8, 118–136.
Winston, J. S., and A. F. Krueger, 1977: Diagnosis of the satellite- observed radiative heating in relation to the summer monsoon. Pure Appl. Geophys.,115, 1131–1144.
Xie, P., and P. A. Arkin, 1995: An intercomparison of gauge observations and satellite estimates of monthly precipitation. J. Appl. Meteor.,24, 1143–1160.
——, and ——, 1996: Analyses of global monthly precipitation using gauge observations, satellite estimates, and numerical model predictions. J. Climate,9, 840–858.
——, B. Rudolf, U. Schneider, and P. A. Arkin, 1996: Gauge-based monthly analysis of global land precipitation from 1971–1994. J. Geophys. Res.,101(D14), 19023–19034.
Yoo, J., and J. A. Carton, 1988: Spatial dependence of the relationship between rainfall and outgoing longwave radiation in the tropical Atlantic. J. Climate,1, 1047–1054.
Zhou, W., and D. E. Waliser, 1997: Removing satellite equatorial crossing time biases from the OLR and HRC datasets. J. Climate,10, 2125–2146.
Time series of the pattern correlation between the OLR flux and precipitation from the merged analysis over several latitudinal zones over land. The results for the total value and its two components, the mean annual cycle and the anomaly, are plotted in thin solid, thick solid, and dotted lines, respectively.
Citation: Journal of Climate 11, 2; 10.1175/1520-0442(1998)011<0137:GMPEFS>2.0.CO;2
As in Fig. 1 except for the results over ocean.
Citation: Journal of Climate 11, 2; 10.1175/1520-0442(1998)011<0137:GMPEFS>2.0.CO;2
Distributions of the serial correlation between OLR flux and precipitation from the merged analysis. The results for the total value and its two components, the mean annual cycle and the anomaly, are plotted in the top, middle, and bottom panels, respectively.
Citation: Journal of Climate 11, 2; 10.1175/1520-0442(1998)011<0137:GMPEFS>2.0.CO;2
Distributions of the negative of the proportional coefficient (top, in mm day−1 W−1 m2) from local regression between the anomaly of precipitation and that of OLR, along with the distribution of the mean precipitation for the 8- yr period from July 1987 to June 1995 as obtained from the merged analysis (bottom).
Citation: Journal of Climate 11, 2; 10.1175/1520-0442(1998)011<0137:GMPEFS>2.0.CO;2
Scatterplot of the proportional coefficient and the mean precipitation as depicted in the top and bottom panels in Fig. 4.
Citation: Journal of Climate 11, 2; 10.1175/1520-0442(1998)011<0137:GMPEFS>2.0.CO;2
Global precipitation distribution (in mm day−1) for August 1987, as obtained by the OPI estimates (top) and by the merged analysis (bottom).
Citation: Journal of Climate 11, 2; 10.1175/1520-0442(1998)011<0137:GMPEFS>2.0.CO;2
Time series of the pattern correlation between precipitation from the merged analysis and that estimated from OLR data using equations based on independent data in the cross validations. The results for land and oceanic areas are plotted in solid and dotted lines, respectively.
Citation: Journal of Climate 11, 2; 10.1175/1520-0442(1998)011<0137:GMPEFS>2.0.CO;2
Correlation (top), bias (middle), and rms error (bottom) of the precipitation estimates based on independent data in the cross-examination tests compared to the merged analysis. Both the bias and the rms error are plotted in percentage relative to the merged analysis.
Citation: Journal of Climate 11, 2; 10.1175/1520-0442(1998)011<0137:GMPEFS>2.0.CO;2
Distributions of the mean precipitation (in mm day−1) for the 8-yr period from July 1987 to June 1995 from the OPI (left, top), GPI (right, top), Grody (left, bottom), and the gauge-based analysis of Xie et al. (1996) (right, bottom).
Citation: Journal of Climate 11, 2; 10.1175/1520-0442(1998)011<0137:GMPEFS>2.0.CO;2
Distributions of the mean precipitation (in mm day) for the period from June 1974 to June 1987, as obtained from the OPI estimates (top) and from the gauge-based analysis of Xie et al. (1996).
Citation: Journal of Climate 11, 2; 10.1175/1520-0442(1998)011<0137:GMPEFS>2.0.CO;2
Time series of the pattern correlation (top), bias (middle), and random error (bottom) between the atoll gauge data and the OPI (solid), GPI (dot), and Grody estimates (dash) over the tropical Pacific. Both the bias and the random error are plotted in percentage relative to the gauge data.
Citation: Journal of Climate 11, 2; 10.1175/1520-0442(1998)011<0137:GMPEFS>2.0.CO;2
Time series of the pattern correlation (top), bias (middle), and random error (bottom) between the gauge-based analysis of precipitation and the OPI (solid), GPI (dot), and Grody (dash) estimates over land areas from 20°S to 20°N. Both the bias and the random error are plotted in percentage relative to the gauge data.
Citation: Journal of Climate 11, 2; 10.1175/1520-0442(1998)011<0137:GMPEFS>2.0.CO;2
As in Fig. 12 except for land areas from 20° to 40°N.
Citation: Journal of Climate 11, 2; 10.1175/1520-0442(1998)011<0137:GMPEFS>2.0.CO;2
As in Fig. 12 except for land areas from 40° to 60°N.
Citation: Journal of Climate 11, 2; 10.1175/1520-0442(1998)011<0137:GMPEFS>2.0.CO;2
Time series of the mean annual cycle defined from the 8-yr merged analysis (thick solid line), anomaly calculated from the merged analysis (thin solid line), and anomaly estimated by the OPI (dotted line) for the 2.5° lat–long grid areas centered at 178.75°E, 1.25°S (top) and at 118.75°E, 28.75°N (bottom).
Citation: Journal of Climate 11, 2; 10.1175/1520-0442(1998)011<0137:GMPEFS>2.0.CO;2
Distributions of precipitation for the entire annual period averaged from the 22-yr OPI estimates (top). Long-term means of Jaeger (1976) and Legates and Willmott (1990) for the same period are plotted as well for comparison.
Citation: Journal of Climate 11, 2; 10.1175/1520-0442(1998)011<0137:GMPEFS>2.0.CO;2
As in Fig. 16 except for the Dec–Feb (DJF) period.
Citation: Journal of Climate 11, 2; 10.1175/1520-0442(1998)011<0137:GMPEFS>2.0.CO;2
As in Fig. 16 except for the Jun–Aug (JJA) period.
Citation: Journal of Climate 11, 2; 10.1175/1520-0442(1998)011<0137:GMPEFS>2.0.CO;2
Time series of the SST anomaly over the Niño3 area (5°S–5°N; 150°–90°W) (top) and the Hovmö_er diagram for the precipitation (in mm day−1) averaged over the Pacific section from 140°E to 100°W (bottom).
Citation: Journal of Climate 11, 2; 10.1175/1520-0442(1998)011<0137:GMPEFS>2.0.CO;2
Differences between warm and cold ENSO episodes during the 22-yr period from 1974 to 1995 for DJF (top) and JJA (bottom) mean precipitation as estimated by the OPI. Areas that are significantly (at the 90% level) wetter/drier during warm ENSO episodes are shaded in heavy/light hatching.
Citation: Journal of Climate 11, 2; 10.1175/1520-0442(1998)011<0137:GMPEFS>2.0.CO;2
Relationship between the proportional coefficient (C, see section 3) and mean precipitation (MP) for the entire 8-yr period from July 1987 to June 1995.
Relationship between the proportional coefficient (C) and mean precipitation (MP) from the cross-validation tests.
Results of the comparison between the OPI and independent precipitation data in the cross-validation tests. (LAN—land, OCN—ocean, GLB—globe)
Results of the comparison of the satellite estimates with gauge data for the period from June 1974 to June 1987.
Results of the comparison of the satellite estimates with gauge data for the period from July 1987 to June 1995.
