• Adler, R. F., and Coauthors, 2003: The Version-2 Global Precipitation Climatology Project (GPCP) Monthly Precipitation Analysis (1979–present). J. Hydrometeor., 4 , 11471167.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Arkin, P. A., , and Meisner B. N. , 1987: The relationship between large-scale convective rainfall and cold cloud over the Western Hemisphere during 1982-1984. Mon. Wea. Rev., 115 , 5174.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Chelliah, M., , and Arkin P. , 1992: Large-scale interannual variability of monthly outgoing longwave radiation anomalies over the global tropics. J. Climate, 5 , 371389.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Ellingson, R. G., , Yanuk D. J. , , Lee H-T. , , and Gruber A. , 1989: A technique for estimating outgoing longwave radiation from HIRS radiance observations. J. Atmos. Oceanic Technol., 6 , 706711.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Gu, G., , Adler R. F. , , Huffman G. J. , , and Curtis S. , 2007: Tropical rainfall variability on interannual-to-interdecadal and longer time scales derived from the GPCP monthly product. J. Climate, 20 , 40334046.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Hilburn, K. A., , and Wentz F. J. , 2008: Intercalibrated passive microwave rain products from the Unified Microwave Ocean Retrieval Algorithm (UMORA). J. Appl. Meteor. Climatol., 47 , 778794.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Huffman, G. J., and Coauthors, 1997: The Global Precipitation Climatology Project (GPCP) combined precipitation dataset. Bull. Amer. Meteor. Soc., 78 , 520.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Lee, H. T., , Gruber A. , , Ellingson R. G. , , and Laszlo I. , 2007: Development of the HIRS outgoing longwave radiation climate dataset. J. Atmos. Oceanic Technol., 24 , 20292047.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Liebmann, B., , and Smith C. A. , 1996: Description of a complete (interpolated) outgoing longwave radiation dataset. Bull. Amer. Meteor. Soc., 77 , 12751277.

    • Search Google Scholar
    • Export Citation
  • Lucas, L. E., , Waliser D. E. , , Xie P. , , Janowiak J. E. , , and Liebmann B. , 2001: Estimating the satellite equatorial crossing time biases in the daily, global outgoing longwave radiation dataset. J. Climate, 14 , 25832605.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Ohring, G., , Gruber A. , , and Ellingson R. , 1984: Satellite determinations of the relationship between total longwave radiation flux and infrared window radiance. J. Climate Appl. Meteor., 23 , 416425.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Sapiano, M. R. P., , Smith T. M. , , and Arkin P. A. , 2008: A new merged analysis of precipitation utilizing satellite and reanalysis data. J. Geophys. Res., 113 , D22103. doi:10.1029/2008JD010310.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Schneider, U., , Fuchs T. , , Meyer-Christoffer A. , , and Rudolf B. , 2008: Global precipitation analysis products of the GPCC. Global Precipitation Climatology Centre (GPCC) Rep., 12 pp.

    • Search Google Scholar
    • Export Citation
  • Waliser, D. E., , and Zhou W. , 1997: Removing satellite equatorial crossing time biases from the OLR and HRC datasets. J. Climate, 10 , 21252146.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Wentz, F. J., , Ricciardulli L. , , Hilburn K. , , and Mears C. , 2007: How much more rain will global warming bring? Science, 317 , 233235. doi:10.1126/science.1140746.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Xie, P., , and Arkin P. A. , 1997: Global precipitation: A 17-year monthly analysis based on gauge observations, satellite estimates, and numerical model outputs. Bull. Amer. Meteor. Soc., 78 , 25392558.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Xie, P., , and Arkin P. A. , 1998: Global monthly precipitation estimates from satellite-observed outgoing longwave radiation. J. Climate, 11 , 137164.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • View in gallery

    Time series plots of (a) uncorrected tropical mean OLR from AVHRR (W m−2) and (b) ECTs used of the AVHRR OLR estimates.

  • View in gallery

    Example of the correction (W m−2) for a single grid box, 16.25°S, 8.75°E, where the gray line denotes the uncorrected time series and the black line denotes (a) the fitted value from the regression model (which is the correction) and (b) the corrected AVHRR OLR for the same grid box.

  • View in gallery

    Map of correlations between fitted regression lines and ECTS at each grid box.

  • View in gallery

    Mean corrected and uncorrected AVHRR OLR (W m−2) for (a) all lats and (b) low to midlats (40°S–40°N) and (c) power spectrum for corrected and uncorrected AVHRR OLR.

  • View in gallery

    Maps of the correlations between corrected and uncorrected precipitation for (a) December–February, (b) March–May, (c) June–August, and (d) September–November.

  • View in gallery

    Time series of spatial correlations between the corrected and uncorrected AVHRR OLR for (a) the raw data and (b) anomalies.

  • View in gallery

    The third EOF loadings for the corrected and uncorrected AVHRR OLR.

  • View in gallery

    Maps (mm day−1) of the eight combined EOF for (a) the uncorrected AVHRR OLR, (b) Lee HIRS, (c) LW01 OLR, and (d) corrected AVHRR OLR, as well as (e) the time series associated with the eighth combined EOF.

  • View in gallery

    Maps (mm day−1) of (a) the mean precipitation from the OPI based on the uncorrected AVHRR OLR and difference between this field and the OPI based on (b) Lee HIRS, (c) LW01 OLR, and (d) corrected AVHRR OLR.

  • View in gallery

    Time series of global mean OPI precipitation (mm day−1) anomalies for (a) land and (b) ocean from the OPI based on each correction technique and the uncorrected AVHRR OLR.

  • View in gallery

    Correlation of each EOF time series with the ECTs for the OPI based on each correction technique and the uncorrected AVHRR OLR.

  • View in gallery

    Mean low to midlatitude (40°S–40°N) precipitation (mm day−1) for (a) land and (b) ocean from the four OPIs, GPCC Full V4 (land), and UMORA V06 (ocean).

  • View in gallery

    Trends (mm day−1 decade−1) from the OPIs and GPCC Full V4 over land and GPCP V2 over land and ocean. All trends were calculated over the common period 1979–99.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 8 8 1
PDF Downloads 2 2 0

Corrections for Temporal Discontinuities in the OPI

View More View Less
  • 1 Department of Atmospheric Science, Colorado State University, Fort Collins, Colorado, and Earth System Science Interdisciplinary Center, University of Maryland, College Park, College Park, Maryland
  • 2 Earth System Science Interdisciplinary Center, University of Maryland, College Park, College Park, Maryland
  • 3 NOAA/NESDIS/STAR/SCSB, College Park, Maryland
  • 4 NOAA/CPC, Camp Springs, Maryland
© Get Permissions
Full access

Abstract

The longest record of precipitation estimated from satellites is the outgoing longwave radiation (OLR) precipitation index (OPI), which is based on polar-orbiting infrared observations from the Advanced Very High Resolution Radiometer (AVHRR) instrument that has flown onboard successive NOAA satellites. A significant barrier to the use of these data in studies of the climate of tropical precipitation (among other things) is the large bias caused by orbital drift that is present in the OLR data. Because the AVHRR instruments are deployed on the polar-orbiting spacecraft, OLR observations are recorded at specific times for each earth location for each day. Discontinuities are caused by the use of multiple satellites with different observing times as well as the orbital drift that occurs throughout the lifetime of each satellite. A regression-based correction is proposed based solely on the equator crossing time (ECT). The correction allows for separate means for each satellite as well as separate coefficients for each satellite ECT. The correction is calculated separately for each grid box but is applied only at locations where the correction is correlated with the OLR estimate. Thus, the correction is applied only where deemed necessary.

The OPI is used to estimate precipitation from the OLR estimates based on the new corrected version of the OLR, the uncorrected OLR, and two earlier published corrected versions. One of the earlier corrections is derived by removing variations from AVHRR based on EOFs that are identified as containing spurious variations related to the ECT bias, whereas the other is based on OLR estimates from the High Resolution Infrared Radiation Sounder (HIRS) that have been corrected using diurnal models for each grid box. The new corrected version is shown to be free of nearly all of the ECT bias and has the lowest root mean square difference when compared to gauges and passive microwave estimates of precipitation. The EOF-based correction fails to remove all of the variations related to the ECT bias, whereas the correction based on HIRS removes much of the bias but appears to introduce erroneous trends caused by the water vapor signal to which these data are sensitive. The new correction for AVHRR OLR works well in the tropics where the OPI has the most skill, but users should be careful when interpreting trends outside this region.

Corresponding author address: M. R. P. Sapiano, Department of Atmospheric Science, Colorado State University, 1371 Campus Delivery, Fort Collins, CO 80523-1371. Email: msapiano@atmos.colostate.edu

Abstract

The longest record of precipitation estimated from satellites is the outgoing longwave radiation (OLR) precipitation index (OPI), which is based on polar-orbiting infrared observations from the Advanced Very High Resolution Radiometer (AVHRR) instrument that has flown onboard successive NOAA satellites. A significant barrier to the use of these data in studies of the climate of tropical precipitation (among other things) is the large bias caused by orbital drift that is present in the OLR data. Because the AVHRR instruments are deployed on the polar-orbiting spacecraft, OLR observations are recorded at specific times for each earth location for each day. Discontinuities are caused by the use of multiple satellites with different observing times as well as the orbital drift that occurs throughout the lifetime of each satellite. A regression-based correction is proposed based solely on the equator crossing time (ECT). The correction allows for separate means for each satellite as well as separate coefficients for each satellite ECT. The correction is calculated separately for each grid box but is applied only at locations where the correction is correlated with the OLR estimate. Thus, the correction is applied only where deemed necessary.

The OPI is used to estimate precipitation from the OLR estimates based on the new corrected version of the OLR, the uncorrected OLR, and two earlier published corrected versions. One of the earlier corrections is derived by removing variations from AVHRR based on EOFs that are identified as containing spurious variations related to the ECT bias, whereas the other is based on OLR estimates from the High Resolution Infrared Radiation Sounder (HIRS) that have been corrected using diurnal models for each grid box. The new corrected version is shown to be free of nearly all of the ECT bias and has the lowest root mean square difference when compared to gauges and passive microwave estimates of precipitation. The EOF-based correction fails to remove all of the variations related to the ECT bias, whereas the correction based on HIRS removes much of the bias but appears to introduce erroneous trends caused by the water vapor signal to which these data are sensitive. The new correction for AVHRR OLR works well in the tropics where the OPI has the most skill, but users should be careful when interpreting trends outside this region.

Corresponding author address: M. R. P. Sapiano, Department of Atmospheric Science, Colorado State University, 1371 Campus Delivery, Fort Collins, CO 80523-1371. Email: msapiano@atmos.colostate.edu

1. Introduction

Perhaps the most significant barriers to the use of satellite data for climate studies are the relatively short lengths of such datasets and the discontinuities introduced by the need to concatenate multiple sensors, which encompasses satellite intercalibration, differing observing times, and satellite drift. In the case of precipitation, satellite estimates are usually made from either passive microwave (PMW) or infrared (IR) information, with the former being more direct and reliable and the latter having a longer record. Geostationary IR-based precipitation estimates are currently available from 1986, and polar-orbiting measurements of outgoing longwave radiation (OLR) are available from 1975 with continuous coverage from 1979 to present; thus, the OLR record is the longest single-source satellite record available for precipitation.

The two most well-established precipitation datasets for climate, the Global Precipitation Climatology Project (GPCP; Huffman et al. 1997; Adler et al. 2003) and the Climate Prediction Center (CPC) Merged Analysis of Precipitation (CMAP; Xie and Arkin 1997), use IR-based estimates for the period before PMW estimates were available. Both of these datasets use the OLR precipitation index (OPI; Xie and Arkin 1998) based on NOAA interpolated OLR (Liebmann and Smith 1996). Most importantly, the OPI is used in both datasets to fill the gap from 1979 to 1986. The OLR data used for the OPI are the average of the ascending and descending passes from the Advanced Very High Resolution Radiometer (AVHRR; Ohring et al. 1984) instruments flown aboard the National Oceanic and Atmospheric Administration (NOAA) satellites. This monthly regression-based technique works by relating rain rates to the cloud-top temperature, with the assumption that lower monthly OLR is associated with deeper clouds, colder cloud-top temperatures, and thus higher rain rates (Arkin and Meisner 1987). In general, this assumption holds for convective situations, so the OPI technique tends to work best in the tropics where convective precipitation dominates.

The main drawback of the current OPI record is that the OLR derived from AVHRR contains temporal discontinuities caused by the use of multiple satellites with varying observing times. Figure 1 shows the time series of tropical mean anomaly NOAA interpolated OLR (Liebmann and Smith 1996) from 1979 to the present, which is composed of the satellites listed in Table 1. Figure 1 also shows the equator crossing times (ECTs) for each of the satellites (note that NOAA-10 is excluded but had an ECT around 0730 LT). Not only do the ECTs vary among the satellites, but each satellite exhibits orbital drift that causes the ECT to vary with time within the record of a single satellite. These lead to artificial variability in the OLR and ultimately in OPI precipitation estimates. The sudden switches between satellites and the effect of the drift can be seen in the time series of tropical OLR anomalies in Fig. 1, which could also be affected by differences in satellite calibration. These discontinuities are particularly obvious in the results of an EOF analysis of uncorrected OLR. Chelliah and Arkin (1992) performed a rotated EOF analysis and found that one EOF in particular was associated with changes in the ECT that was characterized by large loadings over Africa and slightly smaller loadings of opposite sign over the Indian Ocean.

The existence of these discontinuities makes these data unsuitable for use in climate studies without some form of correction. Waliser and Zhou (1997) suggested a correction based on the monthly rotated EOF that corresponded to the ECT bias. The time series of this EOF was modified to represent only the spurious variability, and this was removed from the original data. Lucas et al. (2001, hereafter LW01) used a similar approach but estimated the ECT bias separately from morning and afternoon observations of the daily OLR. They used synthetic corrections based on rotated EOFs from the morning and afternoon OLR to correct for the ECT bias. This general approach relies the ability to partition the erroneous variability associated with the ECT bias into a single (or dual in the case of LW01) EOF. This technique has the advantage of not removing too much variability, with the disadvantage that erroneous variability not included in the single (or dual) EOF is not removed.

An alternative to the AVHRR OLR is to use the OLR estimates from the High Resolution Infrared Radiation Sounder (HIRS) developed by Ellingson et al. (1989). Lee et al. (2007) described a climate version of the HIRS OLR dataset with improved radiance calibration, intersatellite calibration, and corrections for the ECT biases based on fitted diurnal models. The application of these corrections means that the HIRS OLR estimates are more consistent than the AVHRR OLR estimates. As a radiation flux, the HIRS OLR is more accurate than the AVHRR OLR retrieval; besides the extra efforts in the postprocessing, the HIRS OLR is more sensitive to water vapor, which is desirable for an OLR flux. However, this water vapor sensitivity is a major drawback for the estimation of precipitation that can to lead to a confounding signal, particularly in the tropics.

Both of these corrected OLR datasets can be used to produce precipitation estimates using the OPI technique, and it is likely that they would reduce the erroneous signal of the uncorrected NOAA interpolated AVHRR. However, both techniques have drawbacks that could lead to erroneous signals remaining in the OPI estimates, and there appears to be room for improvement. The aim of this manuscript is to present a simple monthly correction for OLR based on the NOAA interpolated AVHRR that is intended for use with the OPI for the estimation of rainfall. Rainfall estimates from the OPI based on the new correction will be compared with new OPI estimates based on the technique of LW01 and Lee et al. (2007), as well as those from the uncorrected AVHRR OLR to assess the relative merits of each correction. The first use of this corrected OPI dataset will be for the construction of a new merged precipitation estimate based on the technique of Sapiano et al. (2008) that will use optimum interpolation to blend OPI and reanalysis estimates, with the former used in the tropics and the latter elsewhere.

2. Data and methods

a. Data

The NOAA interpolated AVHRR OLR data are available at daily and monthly resolution on a 2.5° grid. The daily data are the average of the ascending and descending passes of the satellite, which are interpolated to remove any gaps before averaging. The daily record is then composed of data from a single satellite that changes in time, as indicated in Table 1. The NOAA interpolated OLR monthly product is the monthly average of all days within a month. However, there are a number of instances where a single day of data from NOAA-11 was missing and the daily record was filled with data from NOAA-10 and NOAA-12 (see caption of Table 1). Also, in months when the transition between satellites falls in the middle of the month, the monthly average is composed of days from the satellite that has the most available days. For most months, this means the exclusion of a few days, apart from September 1994, which is composed of 16 days from NOAA-11 and 15 days from NOAA-12. The NOAA-12 data for September 1994 had an anomalous signal over India that was far larger than the climatology for this satellite (not shown) so the monthly average was calculated with September 1994 based only on NOAA-11. This new monthly average of the NOAA interpolated OLR is shown in Fig. 1 and is referred to here as the “uncorrected AVHRR OLR.” Note that the monthly NOAA interpolated AVHRR OLR product is not used here, becuase that dataset uses the average of NOAA-11 and NOAA-12 data for September 1994. ECTs (shown in Fig. 1) were obtained from the NOAA Satellite Services Branch, which maintains a historical archive of ECTs for the NOAA series of satellites.

The corrected OLR datasets of LW01 and Lee et al. (2007) were also used for this intercomparison. The LW01 monthly OLR product was obtained from the National Center for Atmospheric Research (NCAR) Computational and Information Systems Laboratory Research Data Archive on the same 2.5° as the uncorrected AVHRR OLR. The LW01 OLR record is the shortest of the datasets in the intercomparison and currently ends in 1999. The Lee et al. (2007) monthly HIRS OLR 2.0 product (referred to as the Lee HIRS product) was obtained from the University of Maryland Cooperative Institute for Climate Studies Web archive on the same 2.5° as the other data. This product currently ends in September 2003.

b. Monthly regression correction to AVHRR OLR

A new correction for ECT bias was developed using a regression approach based on the ECTs alone. First, ordinary least squares linear regression was used to estimate the relationship between the ECTs and the AVHRR OLR. Separate intercepts and slopes were estimated at each grid box, and fitted values of AVHRR OLR were calculated as
i1520-0426-27-3-457-e1
where xi represents the ECT, β̂0 and β̂1 are the estimated intercept and slope, and the i index denotes month from 1 to n. Separate regression models were used for each satellite period. The model is unable to fit monthly or even annual variations, thus these are left intact by the correction. However, the model can remove real low-frequency variability at interannual and longer scales. To avoid removing genuine interannual and lower-frequency signals from the OLR record, the correction was applied only in boxes where the correlation of the fitted model with the ECTs was greater than or equal to 0.2. Locations with correlations between 0.1 and 0.2 received a weighted correction, where the weight was linearly scaled between zero for correlations of 0.1 and one for correlations of 0.2 to avoid spatial discontinuities. This threshold was chosen to be above the approximately statistically significant threshold of 0.10 (p = 0.05 and N = 360 months). The weighted fitted values were then subtracted from the uncorrected OLR anomalies to get the corrected AVHRR OLR anomalies. The corrected AVHRR OLR anomalies were then added to the uncorrected AVHRR OLR monthly climatology to obtain the final corrected AVHRR OLR product.

Figure 2 shows an example of the fitted model in a single grid box. The model is capable of fitting any mean within each satellite period, but the shape of the line is fixed by the ECTs and modulated only by a scaling factor. In this way, the model is able to adapt to most shapes, which are controlled by the shape of the diurnal cycle and the local sampling of the diurnal cycle in each box. Note that NOAA-10 was used for only one month (November 1988), which is insufficient to estimate a correction for this period, so November 1988 was treated as though it came from NOAA-9.

As stated earlier, the correction is weighted based on its correlation with the ECT biases. The intent of this step is to avoid using the correction in areas where there is no discernable ECT bias in the AVHRR OLR and thus to avoid removing real signal. For example, at a location where the ECT bias is weak (or nonexistent), a genuine linear trend could be fitted by the model as a series of steps. This would be undesirable and so the preliminary test is performed for each model at each grid box. The fitted values from the model are compared with the original ECTs at each grid box using the Pearson product-moment correlation coefficient, and these values are used to obtain the weights for the fitted values. These correlations are shown in Fig. 3. The highest correlations are found in the areas highlighted by Chelliah and Arkin (1992) and are similar to the rotated EOF used by LW01. However, there are relatively large correlations in the midlatitudes and high latitudes as well as over many tropical areas. Note that high correlations do not necessarily equate with larger corrections.

3. Validation of new technique

Figure 4 shows time series of mean global and tropical OLR from the corrected and uncorrected products. The major low-frequency variability is preserved in the corrected version with the same general pattern of lower anomalies in the early 1990s and around the turn of the millennium. Both the global and tropical anomalies show that the correction is larger during certain satellite periods. The periods where observations are from NOAA-6 or NOAA-12 have particularly large corrections that reflect the large shift in ECT during these periods (Fig. 1). The difference between the corrected and uncorrected AVHRR OLR is also larger during the NOAA-16 and NOAA-18. A slightly different behavior is evident during the NOAA-11 and NOAA-14 periods that saw large satellite drift. The corrected OLR is slightly lower than the uncorrected OLR at the start and larger at the end of each period, because the drift was slightly larger at the end of each satellite’s lifetime. Figure 4c shows the power spectrum of the corrected and uncorrected datasets. Both datasets show nearly identical peaks at 6 and 12 months, although the uncorrected data also have peaks for 30, 15, 10, 7.5, and 6 yr, which are unlikely to be physical and are most likely related to the ECT bias. These peaks are far smaller in the corrected dataset.

The correction appears to have removed many of the larger and longer-term discontinuities, but there are still several “jumps” in the OLR record that coincide with changes between satellites. These jumps are most pronounced around the transition from NOAA-11 to NOAA-12 to NOAA-14 and are more severe in the global mean than in the tropical mean. It cannot be ruled out that these jumps are genuine variability, but it is fortunate that they are less severe in the tropics, because this is where the OPI technique (which will be applied in the next section) works best and is most likely to be used. Note that the apparent jump around 2003 occurred in the middle of the NOAA-16 record and cannot be explained by a change of satellites or satellite drift.

One of the most undesirable properties of a correction such as the one we describe here would be the removal of genuine signal, particularly at high and middle frequencies. Figure 5 shows maps of the anomaly correlation between the corrected and uncorrected AVHRR OLR for each season. Correlations are generally high everywhere, with most locations being greater than 0.9. The lowest correlations are seen in areas where the effect of the diurnal cycle is largest, particularly over the African continent. Also, correlations are generally lower globally during December–February and June–August, suggesting that the impact of the correction on these seasons is larger.

Figure 6 shows the time series of global correlations between the corrected and uncorrected AVHRR OLR records for land and ocean. The top panel shows the correlations between the total values (i.e., not anomalies), and the bottom panel shows the correlations between the anomalies. The two panels have largely the same pattern, but the correlation map for the totals shows much higher correlations, which reflect the fact that the correction (which is ∼1–2 W m−2) is relatively small when viewed against the total OLR. Both the total and anomaly correlations show lower agreement between the corrected and uncorrected OLR at the beginning and end of each satellite period, which reflects the correction for satellite drift. The lowest correlations occur during the period of the NOAA-12 satellite, which has a different observing time to most of the other satellites and is therefore expected to be least consistent with data from other satellites. The anomaly correlations (bottom panel) also show that the difference between corrected and uncorrected OLR is slightly larger over the land as would be expected, particularly during the NOAA-6 period.

One of the most effective methods for highlighting the ECT biases is the EOF analysis such as the ones by Chelliah and Arkin (1992) and LW01 (who used rotated EOFs). Figure 7 shows the EOF loadings for the third EOF from the corrected and uncorrected OLR. EOFs were fitted after removing the annual cycle. The third uncorrected EOF shows similar patterns to those obtained by Chelliah and Arkin (1992), although there are some differences resulting from the longer period now available. This third EOF shows the characteristic pattern associated with the ECT bias. The corrected OLR is very different from that for the uncorrected OLR and is not dominated by land-only features but by a strong center over the Indian Ocean that is very similar to the fourth uncorrected EOF (not shown). Thus, the EOF analysis suggests that the ECT bias has been effectively removed from the OLR record without the loss of at least the major interannual variability.

To isolate modes associated with discontinuities, a combined EOF analysis was conducted. Monthly anomalies of each dataset were concatenated for the common period 1979–99, and an EOF analysis was performed on the combined data. Correlations between the ECTs and the EOF time series were highest for the eighth mode, which is shown in Fig. 8. Each of the maps shows the loading associated with each of the OLR datasets, which have been separated and plotted individually. Figure 8e shows the time series associated with the eighth mode, which shows similarities to the ECT pattern of Fig. 1, with drift effects evident and large offsets during the NOAA-6 and NOAA-12 periods. The combined EOF shows a physical pattern that is global but dominated by a large center over North Africa and Eurasia in the uncorrected NOAA AVHRR and the LW01 datasets. This strong land signal is associated with the genuine physical mode but is amplified by the ECT bias and dominates this particular combined EOF. The strong land signal is far smaller in both the Lee HIRS and the corrected AVHRR OLR, and it is consistent with the erroneous signal associated with the ECT bias. Thus, the ECT bias as described by this particular mode appears to be removed from the corrected AVHRR OLR and the Lee HIRS datasets and partially removed from the LW01 OLR. The HIRS OLR also shows larger signals over much of the globe that are stronger than those shown in the AVHRR OLRs, which suggests amplification of this mode rather than extra (erroneous) signal.

4. Comparison of corrected OPIs

The main use of the new corrected AVHRR OLR is in the estimation of rainfall from the OPI. To assess the usefulness of this new correction, the OPI has been estimated from the uncorrected AVHRR OLR, the corrected AVHRR OLR, the LW01 OLR, and the Lee HIRS OLR. The OPI is run operationally at the CPC for use in the GPCP version 2 (V2) dataset. The algorithm requires the specification of a calibration dataset that needs to be a global gridded time series of precipitation. The operational code was ported to run at the University of Maryland, and a modified version of the GPCP V2 estimate was constructed from the interim products that usually combine to form the full GPCP dataset. The GPCP V2 product is formed from a multisatellite product and a gauge product; the multisatellite product is based on an interim product that includes merged land–ocean PMW and calibrated precipitation estimates from the Television and Infrared Observation Satellite (TIROS) Operational Vertical Sounder (TOVS). This PMW–TOVS product is then blended with precipitation estimates from the adjusted Geostationary Operational Environmental Satellite (GOES) precipitation index (AGPI) with OPI used before 1987. Both the GPI and OPI use IR data, so they were excluded as calibration data. Thus, the PMW–TOVS estimate was merged with the gauge data used in GPCP with the gauges used over land and the PMW–TOVS estimates used over the oceans. These same calibration data were used for each of the OPIs for the period 1979–99, which is the common period of all the input OLR data.

a. Intercomparison

In this section, we explore the basic characteristics of the OPI estimates of precipitation based on the four input products. Figure 9a shows the annual mean precipitation from the OPI based on the uncorrected AVHRR OLR. This mean is highly dependent on the calibration data used, and interpretation of these values is not the goal here, although this figure is given for reference. The OPI is calculated for the anomalies from the monthly climatology of the calibration data, with this monthly climatology added back to the estimated anomalies to obtain the final values. In this case, the calibration data were for the period from 1987 to 1999, so the monthly climatology of each estimate is identical over this shorter period. Differences between the annual means are thus expected to be small in magnitude. Figures 9b–d show the difference between the OPI from the uncorrected AVHRR OLR and the OPI from each of the correction techniques, and all the differences are small at the mean annual scale (O 0.1 mm day−1). The Lee HIRS data are the most different from the uncorrected AVHRR OPI, which is expected, because these data are from a different instrument. As might be expected, differences are smaller in the dryer, cloud-free areas. However, biases are large throughout the global oceans. The main drawback of the Lee HIRS OLR estimates is that they are expected to be associated with water vapor that would contaminate the precipitation signal, although this will be explored more by looking at time series of precipitation. The OPI based on the LW01 OLR is only slightly different than the uncorrected OPI that indicates that LW01 made a fairly small adjustment in terms of the global mean. The difference between the OPIs from the corrected and uncorrected AVHRR OLR datasets is larger than that for LW01 because of the correction technique allowing a separate bias adjustment for each period.

Figure 10 shows the global mean time series of precipitation anomalies for land and ocean from the OPI. These precipitation variations are generally similar to those seen in the OLR record in Fig. 4, although they are of opposite sign (as would be expected). Over land, the LW01 OPI generally follows the uncorrected OPI, although they deviate a little more over the ocean. The LW01 correction appears to perform badly in the first part of the dataset during the TIROS-N and NOAA-6 periods and seems to overcompensate over land. The LW01 OPI is also very similar to the uncorrected data around 1995 during the NOAA-12 period, and these two periods show steps in both the uncorrected and the LW01 OPIs. In contrast, the OPI based on the corrected AVHRR OLR does not suffer from these jump points but is similar to the uncorrected OPI. Over land, the OPI based on the Lee HIRS OLR is similar to the other OPIs that are based on AVHRR; however, it shows some striking differences over the ocean. It is not clear whether these differences are erroneous, but it does suggest that they could be caused by additional variations in water vapor and might suggest that this dataset is unsuitable for the estimation of precipitation.

The ECT bias issue is clearly seen in the results of an EOF analysis with the third mode containing the majority of the erroneous signal. However, some apparently erroneous signals were also evident in the other EOFs shown in Fig. 7. To isolate the ECT bias signal, an EOF analysis was performed on the anomalies of each of the OPIs, and the corresponding EOF time series for each mode was compared with the actual ECT biases. Figure 11 shows the correlation between the EOF time series and the ECTs for each of the 252 modes; lower mode numbers correspond to EOFs that explain greater variability. The OPI based on the uncorrected AVHRR OLR shows that many of the first 50 modes have correlations larger than the reference line that is the approximate level at which the correlations are statistically significantly different from zero at the 5% level (±0.12 for a two-sided test with sample size of 252 months). The Lee HIRS OPI has some correlations above and below the lines of significance, which signifies some weak remaining ECT bias, but these are very low and it is expected that 1 in 20 correlations would be falsely significant. The LW01 OPI shows several statistically significant correlations in the higher mode numbers, although still far less than the uncorrected data. This implies that the ECT bias is not confined to a single EOF (rotated or not) and so corrections based on that approach are likely to leave some erroneous variability uncorrected. Finally, the OPI based on the corrected AVHRR OLR shows extremely low correlations with the ECT bias in the lower modes, although there are some relatively large correlations in the lower variance modes. The correction has removed the bulk of the erroneous signal but has introduced some additional signal into the series. However, this additional signal appears only in the lower variance modes that are comprised of mostly noise and thus does not contribute much to the overall dataset.

b. Comparison with other datasets

Finally, the OPIs based on the various correction methods are compared with precipitation estimates from other sources. Figure 12 shows time series of tropical monthly mean rainfall anomalies separately for ocean and land. The Global Precipitation Climatology Center (GPCC) Full Data Reanalysis V4 (Full V4) gauge estimate (Schneider et al. 2008) is included for reference over land, and the Unified Microwave Ocean Retrieval Algorithm (UMORA; Hilburn and Wentz 2008) V6 is included over ocean. The GPCC Full V4 gauge analysis is based on 10 000–25 000 gauges over the period of use and is higher quality than the near-real time monitoring product used by GPCP. Note that this is different from the gauge analysis used in GPCP V2 that uses the monitoring product; the monitoring product was also used to calibrate the OPI. The UMORA V6 precipitation product is an ocean-only algorithm based on a physical retrieval from passive microwave information.

The agreement over land is fairly good and the OPI captures many of the same monthly variations present in GPCC Full V4. The effect of the ECT bias is clear in the uncorrected OPI during the periods based on NOAA-6 and NOAA-11 during which times it is higher than the gauges. The LW01 OPI has similar issues at the same times that support the conclusions from Fig. 10. The OPI based on the corrected AVHRR OLR follows the gauges quite closely, as does the OPI based on the Lee HIRS OLR. There is greater disagreement between the PMW estimates and the IR-based OPI over the ocean. The interannual variations are similar between the UMORA and the OPIs, although the monthly variations do not always coincide. The Lee HIRS dataset deviates slightly from UMORA in two periods during 1989 and 1998 but has a similar tendency to UMORA during 1999 when the others do not.

Table 2 shows the root-mean-square difference (RMSD) of each of the OPIs compared to GPCC Full V4 over land and UMORA V6 over ocean over the globe and the tropics. The results are qualitatively the same over the tropics and the whole globe with higher RMSDs naturally occurring over the tropics where the variance is higher. The OPIs based on the Lee HIRS OLR and the corrected AVHRR OLR have the lowest RMSD over land. The LW01 OPI has a slightly lower RMSD than the uncorrected OPI over land, but it is higher than the other two corrected OPIs. The Lee HIRS OPI has a higher RMSD than the uncorrected OPI over the ocean. This higher RMSD is most likely caused by the water vapor signal that the HIRS OLR is more sensitive to and that would be expected to be stronger over the ocean. The water vapor signal appears to lead to an extra source of variability that is undesirable for the OPI. The corrected AVHRR OPI has lower RMSD than the LW01 OPI over the global oceans, but they are very similar over the tropical oceans. The LW01 correction removes more of the ECT bias in the tropics but does not remove much bias outside the tropics, because the rotated EOF method favors tropical variations. Thus, the new correction more adequately addresses the global problem of ECT biases.

Time trends represent a particular problem when correcting for ECT bias. The ECT bias could reasonably be expected to contribute to any linear time trend calculated from the uncorrected data (from either the OLR or the OPI estimates); thus, these trends are likely to contain erroneous signals and are untrustworthy. On the other hand, linear time trends in short time series are often sensitive to changes in the series, so it is important that any correction remove erroneous signals and not genuine ones. Figure 13 shows the linear time trends from each of the OPIs compared with those obtained from the GPCC Full V4 and GPCP V2 for the period 1979–99. Note that these trends are used for comparison only and should not be interpreted in the context of climate change. The UMORA estimates are not used, because the common period is too short. It is important to note that the GPCP V2 uses the OPI based on the uncorrected OLR over the oceans before 1987 (and calibrated to the GPCP V2 satellite gauge estimates over the time period 1988–2007) and is therefore not entirely independent. GPCP V2 is included here because it is a well-known dataset that has been used for climate in many studies (e.g., Gu et al. 2007), although this does not imply that the trend estimates are free from errors at all locations on the globe. The reader should also note that the GPCP is mainly composed of information from PMW sensors and cannot be expected to match those from the OPI datasets. Figures 13e,f show that the trends are generally largest over the tropics. There is some disagreement between GPCC Full V4 and GPCP V2, with the latter having more areas with increasing or decreasing trends, particularly over Africa, North America, and Eurasia. There is good qualitative agreement between the AVHRR-based OPIs, but the Lee HIRS OPI has negative trends over all ocean areas that are inconsistent with other published results (Wentz et al. 2007). The LW01 OPI is a very close match to the uncorrected OPI, but both of these have differences with GPCP V2, particularly at high latitudes where they have some slight positive trends and over the Pacific and Southern Oceans. The corrected OPI has slightly weaker trends than the uncorrected OPI, with fewer trends outside the tropics. The corrected OPI does not show the negative trends over the southern parts of South America and Africa that exist in both GPCC Full V4 and GPCP V2 and that appear in the uncorrected OPI and LW01 OPI.

5. Discussion

A new correction for ECT bias in the AVHRR OLR has been suggested, applied, and tested. The correction is based on separate regressions for each grid box that are a function of the ECTs alone and that are applied only to grid boxes where they are required. The correction was developed for use in the OPI that is a regression-based technique for calculating global precipitation estimates, although the OPI works best in the tropics. The corrected AVHRR OLR reduces the signs of ECT bias as judged by both EOF analysis and visual inspection of global and tropical anomalies, but it maintains the variability within the record and is well correlated with the uncorrected OLR.

Two other corrected OLR time series are also available, including one based on the AVHRR and one based on the HIRS instrument. The OPI was used to obtain precipitation estimates from the corrected and uncorrected AVHRR OLR as well as from these other two products. Out of these, the OPI based on the new correction performed the best overall, with virtually no trace of the ECT bias remaining. The OPI based on the new correction had the lowest RMS when compared to PMW data. The Lee HIRS OPI had the lowest RMS when compared to gauges, although the RMS for the new correction was only slightly higher and was far lower than the RMS for the other techniques. A linear time trend analysis of the corrected OPI showed relatively weak trends outside the tropics when compared to GPCP V2, which suggests that the new correction might erroneously remove some of these extratropical trends. The tropical trends are more consistent between the AVHRR-based OPIs and GPCP, which is important because the OPI technique works mainly in convective regimes. In contrast, the LW01 correction gives almost identical time trends to the uncorrected OLR but fails to remove some of the erroneous variability. The LW01 technique involves removing variability from only a very limited set of EOFs, which means that ECT bias is not removed if it cannot be represented by this limited set. The OPI based on the Lee et al. (2007) HIRS OLR appeared to have very little sign of the ECT bias, although it was in generally poor disagreement with PMW estimates of precipitation over the ocean and had completely different time trends than those observed in precipitation. Although the Lee et al. (2007) method appears to work well, the currently available HIRS OLR technique appears to be unsuitable for precipitation estimation. In the future, an OPI estimate based on a stable window radiance estimate using similar diurnal corrections to Lee et al. (2007) would most likely be superior to the corrected AVHRR OLR–based OPI, and the development of such a dataset is an important topic of further work.

The main use of this new corrected AVHRR OLR product is for the production of OPI precipitation estimates to be used in new merged analyses of precipitation, such as that by Sapiano et al. (2008), where satellite estimates are used in the tropics and reanalysis estimates are used in higher latitudes. However, the OPI estimates would also be suitable for use in GPCP and CMAP, which rely on these estimates for the early period, as well as any other applications requiring a long series of tropical precipitation. One of the benefits of this correction over the existing corrections is that it is relatively simple to calculate and can be fully automated. Thus, this correction could easily be used for the routine production of corrected AVHRR OLR in an operational setting, and it is hoped that these corrected data, as well as the correction technique, will be useful for the wider scientific community as well as for scientists studying global rainfall.

Acknowledgments

The authors are grateful to David Bolvin and another anonymous reviewer for their helpful comments. Interpolated OLR data were provided by the NOAA–CIRES Climate Diagnostics Center, Boulder, Colorado, from their Web site (available online at http://www.cdc.noaa.gov/). The LW01 corrected OLR data were provided by the Data Support Section of the Computational and Information Systems Laboratory at the National Center for Atmospheric Research. NCAR is supported by grants from the National Science Foundation. UMORA data are produced by Remote Sensing Systems (data available online at http://www.remss.com) and sponsored by the NASA Earth Science MEaSUREs DISCOVER Project. The GPCP combined precipitation data were developed and computed by the NASA Goddard Space Flight Center Laboratory for Atmospheres as a contribution to the GEWEX Global Precipitation Climatology Project.

REFERENCES

  • Adler, R. F., and Coauthors, 2003: The Version-2 Global Precipitation Climatology Project (GPCP) Monthly Precipitation Analysis (1979–present). J. Hydrometeor., 4 , 11471167.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Arkin, P. A., , and Meisner B. N. , 1987: The relationship between large-scale convective rainfall and cold cloud over the Western Hemisphere during 1982-1984. Mon. Wea. Rev., 115 , 5174.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Chelliah, M., , and Arkin P. , 1992: Large-scale interannual variability of monthly outgoing longwave radiation anomalies over the global tropics. J. Climate, 5 , 371389.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Ellingson, R. G., , Yanuk D. J. , , Lee H-T. , , and Gruber A. , 1989: A technique for estimating outgoing longwave radiation from HIRS radiance observations. J. Atmos. Oceanic Technol., 6 , 706711.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Gu, G., , Adler R. F. , , Huffman G. J. , , and Curtis S. , 2007: Tropical rainfall variability on interannual-to-interdecadal and longer time scales derived from the GPCP monthly product. J. Climate, 20 , 40334046.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Hilburn, K. A., , and Wentz F. J. , 2008: Intercalibrated passive microwave rain products from the Unified Microwave Ocean Retrieval Algorithm (UMORA). J. Appl. Meteor. Climatol., 47 , 778794.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Huffman, G. J., and Coauthors, 1997: The Global Precipitation Climatology Project (GPCP) combined precipitation dataset. Bull. Amer. Meteor. Soc., 78 , 520.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Lee, H. T., , Gruber A. , , Ellingson R. G. , , and Laszlo I. , 2007: Development of the HIRS outgoing longwave radiation climate dataset. J. Atmos. Oceanic Technol., 24 , 20292047.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Liebmann, B., , and Smith C. A. , 1996: Description of a complete (interpolated) outgoing longwave radiation dataset. Bull. Amer. Meteor. Soc., 77 , 12751277.

    • Search Google Scholar
    • Export Citation
  • Lucas, L. E., , Waliser D. E. , , Xie P. , , Janowiak J. E. , , and Liebmann B. , 2001: Estimating the satellite equatorial crossing time biases in the daily, global outgoing longwave radiation dataset. J. Climate, 14 , 25832605.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Ohring, G., , Gruber A. , , and Ellingson R. , 1984: Satellite determinations of the relationship between total longwave radiation flux and infrared window radiance. J. Climate Appl. Meteor., 23 , 416425.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Sapiano, M. R. P., , Smith T. M. , , and Arkin P. A. , 2008: A new merged analysis of precipitation utilizing satellite and reanalysis data. J. Geophys. Res., 113 , D22103. doi:10.1029/2008JD010310.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Schneider, U., , Fuchs T. , , Meyer-Christoffer A. , , and Rudolf B. , 2008: Global precipitation analysis products of the GPCC. Global Precipitation Climatology Centre (GPCC) Rep., 12 pp.

    • Search Google Scholar
    • Export Citation
  • Waliser, D. E., , and Zhou W. , 1997: Removing satellite equatorial crossing time biases from the OLR and HRC datasets. J. Climate, 10 , 21252146.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Wentz, F. J., , Ricciardulli L. , , Hilburn K. , , and Mears C. , 2007: How much more rain will global warming bring? Science, 317 , 233235. doi:10.1126/science.1140746.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Xie, P., , and Arkin P. A. , 1997: Global precipitation: A 17-year monthly analysis based on gauge observations, satellite estimates, and numerical model outputs. Bull. Amer. Meteor. Soc., 78 , 25392558.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Xie, P., , and Arkin P. A. , 1998: Global monthly precipitation estimates from satellite-observed outgoing longwave radiation. J. Climate, 11 , 137164.

    • Crossref
    • Search Google Scholar
    • Export Citation
Fig. 1.
Fig. 1.

Time series plots of (a) uncorrected tropical mean OLR from AVHRR (W m−2) and (b) ECTs used of the AVHRR OLR estimates.

Citation: Journal of Atmospheric and Oceanic Technology 27, 3; 10.1175/2009JTECHA1366.1

Fig. 2.
Fig. 2.

Example of the correction (W m−2) for a single grid box, 16.25°S, 8.75°E, where the gray line denotes the uncorrected time series and the black line denotes (a) the fitted value from the regression model (which is the correction) and (b) the corrected AVHRR OLR for the same grid box.

Citation: Journal of Atmospheric and Oceanic Technology 27, 3; 10.1175/2009JTECHA1366.1

Fig. 3.
Fig. 3.

Map of correlations between fitted regression lines and ECTS at each grid box.

Citation: Journal of Atmospheric and Oceanic Technology 27, 3; 10.1175/2009JTECHA1366.1

Fig. 4.
Fig. 4.

Mean corrected and uncorrected AVHRR OLR (W m−2) for (a) all lats and (b) low to midlats (40°S–40°N) and (c) power spectrum for corrected and uncorrected AVHRR OLR.

Citation: Journal of Atmospheric and Oceanic Technology 27, 3; 10.1175/2009JTECHA1366.1

Fig. 5.
Fig. 5.

Maps of the correlations between corrected and uncorrected precipitation for (a) December–February, (b) March–May, (c) June–August, and (d) September–November.

Citation: Journal of Atmospheric and Oceanic Technology 27, 3; 10.1175/2009JTECHA1366.1

Fig. 6.
Fig. 6.

Time series of spatial correlations between the corrected and uncorrected AVHRR OLR for (a) the raw data and (b) anomalies.

Citation: Journal of Atmospheric and Oceanic Technology 27, 3; 10.1175/2009JTECHA1366.1

Fig. 7.
Fig. 7.

The third EOF loadings for the corrected and uncorrected AVHRR OLR.

Citation: Journal of Atmospheric and Oceanic Technology 27, 3; 10.1175/2009JTECHA1366.1

Fig. 8.
Fig. 8.

Maps (mm day−1) of the eight combined EOF for (a) the uncorrected AVHRR OLR, (b) Lee HIRS, (c) LW01 OLR, and (d) corrected AVHRR OLR, as well as (e) the time series associated with the eighth combined EOF.

Citation: Journal of Atmospheric and Oceanic Technology 27, 3; 10.1175/2009JTECHA1366.1

Fig. 9.
Fig. 9.

Maps (mm day−1) of (a) the mean precipitation from the OPI based on the uncorrected AVHRR OLR and difference between this field and the OPI based on (b) Lee HIRS, (c) LW01 OLR, and (d) corrected AVHRR OLR.

Citation: Journal of Atmospheric and Oceanic Technology 27, 3; 10.1175/2009JTECHA1366.1

Fig. 10.
Fig. 10.

Time series of global mean OPI precipitation (mm day−1) anomalies for (a) land and (b) ocean from the OPI based on each correction technique and the uncorrected AVHRR OLR.

Citation: Journal of Atmospheric and Oceanic Technology 27, 3; 10.1175/2009JTECHA1366.1

Fig. 11.
Fig. 11.

Correlation of each EOF time series with the ECTs for the OPI based on each correction technique and the uncorrected AVHRR OLR.

Citation: Journal of Atmospheric and Oceanic Technology 27, 3; 10.1175/2009JTECHA1366.1

Fig. 12.
Fig. 12.

Mean low to midlatitude (40°S–40°N) precipitation (mm day−1) for (a) land and (b) ocean from the four OPIs, GPCC Full V4 (land), and UMORA V06 (ocean).

Citation: Journal of Atmospheric and Oceanic Technology 27, 3; 10.1175/2009JTECHA1366.1

Fig. 13.
Fig. 13.

Trends (mm day−1 decade−1) from the OPIs and GPCC Full V4 over land and GPCP V2 over land and ocean. All trends were calculated over the common period 1979–99.

Citation: Journal of Atmospheric and Oceanic Technology 27, 3; 10.1175/2009JTECHA1366.1

Table 1.

Satellites used for the NOAA interpolated OLR. For the NOAA interpolated OLR, missing data from NOAA-11 were filled with data from NOAA-10 between 1 and 4 Jul 1990, on 5 and 13 Mar 1991, and on 14 Aug 1991; NOAA-12 was used on 15 Oct 1992.

Table 1.
Table 2.

RMSD (mm day−1) between each of the OPIs, GPCC FULL V4 gauges (1979–99) over land, and UMORA V6 (1987–99) over ocean for the tropics and the whole globe.

Table 2.
Save