Ocean Observing and Climate pages
Other NOCS research groups

Oceanic Precipitation: A comparison of seasonal and interannual variations in different datasets

1. Background

The hydrological cycle of evaporation, precipitation (rain, snow, hail), amassing of ground water in aquifers and its return to the sea is important to us for its effect on the climate where we live, and the ready provision (or not) of drinking water and irrigation of crops. However, the rain that falls at sea is also important for our climate, as the transport of fresh water via the atmosphere affects the salinity of surface waters, with potential impact on density-driven circulation. The broad pattern of rainfall is well known (see Fig. 1), as the atmospheric circulation leads to the development of the ITCZ between converging moist air masses, and tracks of storms in the North Atlantic and North Pacific lead to increased rainfall there. However, what is less clear is the actual quantities of rain, and how they vary seasonally and interannually.

Figure 1 : Mean distribution of precipitation (from GPCP) showing ITCZ (Inter-Tropical Convergence Zone), SPCZ (South Pacific Convergence Zone) and storm tracks in the N. Atlantic and N. Pacific. Note rainfall rates at the east of the ocean basins are considerably less than in the west. Béranger et al. (2006) provide a good comparison of the sp[atial distributions in 10 different datasets.

2. Datasets

To study precipitation globally requires data from satellites, weather forecast models or dense networks of land gauges and ships. Here we consider GPCP (Global Precipitation Climatology Project) as an example of the satellite product, and use NCEP (National Centers for Environmental Prediction) and ECMWF (European Centre for Medium-range Weather Forecasting) reanalyses (atmospheric weather forecast models run continuously with the same code). A priori we cannot state that one dataset is correct and the reference for comparing the others. Close agreement of all gives us confidence that satellite measurements and model forecasts are converging on reality; the nature of any disagreements (when they occur) helps suggest which datasets are likely to be most in error.

3. Long-term consistency

Satellite rainfall measurements have been available since Jan. 1979, but the technology has undergone improvements in that time, the most marked of which was the introduction of passive microwave sensors (SSM/I) in July 1987. Although careful calibration has fixed the mean seasonal cycle to be the same before and after, there is a marked change in interannual variability when the new sensors are included (Fig. 2). Although the reanalyses involve assimilations using unchanging computer code, there are marked changes in the quality of their forecasts, with each introduction of new sources of data e.g. radiosondes and satellite observations.

Fig. 2 : Rainfall for small region of N. Atlantic in GPCP climatology. a) Monthly values, b) Anomalies about monthly mean. Note the variability increases after the introduction of SSM/I in July 1987.


4. Seasonal cycles

Fig. 3 : Mean seasonal cycles of rainfall in the three Atlantic regions.

On a region by region basis we determined the mean rainfall for reach calendar month. The seasonal variation for three parts of the Atlantic is displayed in Fig. 3. All the datasets show good consistency in the N. Atlantic (Fig.3a), with a minimum around May-July and a peak with roughly twice as much rain in Oct-Dec. However, there are some differences in magnitude and timing.

Not surprisingly the S. Atlantic (Fig. 3c) shows a different seasonal cycle, with peaks around April-June and minima in Aug-Sept or Dec-Jan according to dataset. However, the seasonal changes are much weaker than in the north and the differences between datasets more marked, with GPCP having the largest signal, and NCEP and NCEP2 having a seasonal cycle 2 months behind those in GPCP and ECMWF.

The defined 'Tropical Atlantic' (Fig. 3b) encompasses the ITCZ migration, but still shows a seasonal variation, this time with two peaks. In all tropical regions ECMWF gives the largest values, because of a problem with the assimilation of humidity information (detailed in Hagemann et al., 2005).

The greatest disparity between the datasets in terms of seasonal cycle is for the North and Central Indian Ocean and the Tropical Pacific (Quartly et al., 2007). In general, most of the correlation between precipitation records in satellite and reanalysis datasets is due to the seasonal cycle; when this is removed the datasets are much less correlated.


5. North Atlantic Oscillation (NAO)

Fig. 4 : Sensitivity of precipitation to changes in normalized NAO index (based on pressure difference between Azores and Iceland). Example shown is for NCEP; click here for intercomparison of all 4 datasets.

Surprisingly there is good agreement in the changes in precipitation associated with NAO (a large-scale climatic phenomenon affecting the direction of storm tracks across the Atlantic, see LDEO webpage). We characterise the precipitation effect by how much it changes with the NAO index (difference in atmospheric pressure between Azores and Iceland). All the datasets show a tripole pattern, with increased peaks over northern Europe during a positive phase of NAO (e.g. 1986??-1995), with reduction in rainfall over southern Europe and over Greenland and the Labrador Sea.

It is found that the total rainfall over the N. Atlantic and environs does NOT change with NAO (see Kyte et al., 2006); rather it is a redistribution of the freshwater flux (which may be important for where surface waters become dense enough to sink in the convolution cycle.


6. ENSO (El Niño Southern Oscillation)

A similar analysis can be done for the tropical region, showing sensitivity to the Southern Oscillation index, the pressure difference between Tahiti and Darwin that is a measure of the state of development of El Ni&241;o. During the peak of El Ni&241;o (typically Dec-Feb) the ITCZ in the Pacific moves further south and the 'rain pool' east of Indonesia and the whole SPCZ move east. Such a change can be seen in all the datasets -- however, as the NCEP model gets the orientation of the SPCZ wrong (it portrays it mainly zonally), it also gets the change in SPCZ incorrect. However, there is good agreement further afield -- reductions in precipitation in the tropical Atlantic and east Indian Ocean, with increases over the west Indian Ocean and over Africa. The fact that the models and satellites are agreeing in these details gives us confidence that the models are correctly represening the dynamics associated with these interannual changes.

The datasets analysed here only have monthly values, but earlier analysis by Quartly et al. (2000) showed that for the 1997/98 El Ni&241;o the increase in rain in the central Pacific was due to both more intense rain and more frequent events. However, looking at the wider picture we again find the increases and decreases in rainfall to average out, so that ENSO just redistributes the rainfall rather than increasing the total (see Kyte et al., 2006).

Fig. 5 : Sensitivity of precipitation to Southern Oscillation Index Example shown is for ECMWF; click here for intercomparison of all 4 datasets.

7. Long-term trend?

Even with 25+ years data, it is hard to detect a long-term trend given that there are interannual changes with long periodicities.

The changes in satellite sensors in 1987 could have led to a discontinuity in the GPCP records; in fact the algorithms are adjusted so that infra-red and passive microwave sensors lead to the same magnitude records. Whilst there are significant interannual variations in regions, GPCP shows the global mean to be fairly constant (see Fig. 6), implying that total evaporation has not changed significantly in that period either. However, the various adjustments of sensor algorithms may mask a gradual change.

A similar analysis of NCEP does show a trend of significant increase (~3% per decade); NCEP2 and ECMWF show more rapid increases. However, model reanalyses are sensitive to the many changes in the observing systems providing them with data (see Hagemann et al., 2005). Even in the "satellite era'' (1979 onwards) new instruments (such as scatterometers) provide information not previously available.

Fig. 6 : Global (65°S - 65°N) mean precipitation for GPCP and NCEP datasets. Marine precipitation in solid line; land+sea in dashed. Values are smoothed with 11-month moving average. The beige bands indicate the main El Niño events during this period.

References
  • Béranger, K., Barnier, B., Gulev, S., Crépon, M, 2006, 'Comparing 20 years of precipitation estimates from different sources over the world ocean', Ocean Dynamics 56 (2) 104-138, Link to paper
  • Hagemann, S., K. Arpe amd L. Bengtsson, 2005, 'Validation of the hydrological cycle of ERA-40', ERA-40 Project Report No. 24
  • Kyte, E.A., G.D. Quartly, M.A. Srokosz and M.N. Tsimplis, 2006, 'Interannual variations in precipitation: The effect of the North Atlantic and Southern Oscillations as seen in a satellite precipitation data set and in models', J. Geophys. Res., 111, art. no. D24113. doi: 101029/2006JD007138
  • Quartly, G.D., E.A. Kyte, M.A. Srokosz and M.N. Tsimplis, 2007, 'An intercomparison of global oceanic precipitation climatologies', accepted by J. Geophys. Res. preprint
  • Quartly G.D., M.A. Srokosz and T.H. Guymer, 2000, 'Changes in oceanic precipitation during the 1997-98 El Niño', Geophys. Res. Lett., 27, 2293-2296.