1. Introduction
Nordeste, Brazil, extending from 2° to 12°S, 36° to 43°W is a semiarid region with recurrent drought conditions. The climatological evolution of precipitation in Nordeste is directly affected by the migratory nature of the Atlantic intertropical convergence zone (ITCZ). The ITCZ starts to shift southward in late boreal summer, reaches the north coast of Nordeste in February, continues to move southward in March, and reaches its southernmost position in April, then starts to retreat northward in May, and moves out of Nordeste in June. Therefore, most of Nordeste experiences one rain season during the year, that is, from February to May. Previous investigations have firmly established that sea surface temperature (SST) anomaly forcing is the primary factor responsible for the interannual climate variability in Nordeste (Harzallah et al. 1996; Hastenrath and Heller 1977; Mechoso et al. 1990; Moron et al. 1996; Moura and Shukla 1981; Pezzi and Cavalcanti 2001; Ropelewski and Halpert 1996; Roucou et al. 1996; Uvo et al. 1998; Ward and Folland 1991). The current atmospheric general circulation models (AGCMs) forced by observed SSTs can capture the large-scale circulation in northeast South America (Moron et al. 1998; Sperber and Palmer 1996). However, they are unable to resolve the local precipitation patterns in Nordeste due to the coarse resolution (Nobre et al. 2001).
Dynamical downscaling is one of the approaches used to investigate the precipitation variability at local scales. A high-resolution limited-area model is run for the region of interest, forced by the large-scale circulation prescribed from a lower-resolution AGCM. The strategy underlying this technique is that AGCMs can provide the response of global circulation to large-scale forcings, and nested regional models can account for the effects of local, sub-AGCM grid-scale forcings (Giorgi and Mearns 1999). Regional models have been tested for climate downscaling over North America, South America, Africa, Asia, and Europe (Fennessy and Shukla 2000; Giorgi and Marinucci 1991; Hong et al. 1999; Kanamtisu and Juang 1994; Nobre et al. 2001; Roads 2000; Seth and Rojas 2003; Sun et al. 1999a,b; Takle et al. 1999; Ward and Sun 2002). The simulation of both finer spatial-scale details and overall monthly or seasonal mean precipitation were improved in the regional models. Examination of other atmospheric variables indicated that the regional model simulations were generally as good or better than those from AGCMs alone. Finer spatial-scale features developed in regional models are attributed to four types of sources: (i) the surface forcing such as detailed topography, (ii) the nonlinearities presented in the atmospheric dynamical equations, (iii) hydrodynamic instabilities (Denis et al. 2002), and (iv) noises generated at the lateral boundaries.
Studies that concern analysis of daily precipitation in climate models are relatively few. Mearns et al. (1990) examined several versions of the AGCMs and found errors of too high and too low daily variability of precipitation depending on the version investigated. A study by Gershunov and Barnett (1998) indicated that the AGCM missed important aspects of the ENSO signal in seasonal statistics of daily precipitation, although it is capable of capturing the ENSO signal in seasonal-averaged precipitation. The analysis of daily precipitation in AGCMs probably is of limited value, given the crude horizontal resolution (i.e., the model can neither resolve important topographic influences on precipitation, nor synoptic-scale precipitation processes) and the crude parameterizations of precipitation (Mearns et al. 1995). However, the model parameterization schemes are steadily improving, and regional models have relatively fine horizontal resolution. This mitigates some of the limitations of AGCMs, and examination of daily precipitation may prove more fruitful. Giorgi and Marinucci (1996) investigated the sensitivity of simulated precipitation to model resolution and concluded that the effect of model resolution on daily precipitation statistics is evident. Several studies indicate that high-resolution regional models appear to be capable of improving the statistics of high-frequency precipitation toward more realism on finer spatial and time scales (Chen et al. 1999; Hong and Leetmaa 1999; Mo et al. 2000; Sun et al. 1999a).
Although dynamical downscaling provides enhanced details of climate simulations, there is a need for more research to examine the regional model ability in simulating observed local climate conditions and evaluate the statistical structures of climate signals at various spatial and temporal scales to answer the question of whether predictability of climate systems is improved with regional, over global, modeling. A logical approach is to use multiple AGCMs with multiple ensembles and force multiple regional models (Leung et al. 2003). This task greatly exceeds our current computational limits. In this study, we use one AGCM and one regional model with 30-yr multiple ensembles to study the climate dynamical downscaling of precipitation in Nordeste. Our primary objectives are to find out
(i) is the regional model capable of producing large-scale information?
(ii) is probabilistic information improved in the regional model compared to that in the AGCM?
(iii) is the finer spatial-scale information produced in the regional model skillful, or is it just “noise” on the top of the large-scale signal?
(iv) is temporal character of variability skillful in the regional model?
This paper is organized as follows. Section 2 describes the experimental design and observed data available for this study. Results from dynamical downscaling are discussed in section 3, and the summary and concluding remarks are given in section 4.
2. Experimental design and data
a. Experimental design
The AGCM used in this study is the European Community–Hamburg (ECHAM) AGCM version 4.5, developed at the Max Planck Institute for Meteorology (MPI) in Germany. The model was configured at triangular 42 (T42) spectral truncation, giving a spatial resolution of about 2.8° lat × lon, with 19 vertical layers (Roeckner et al. 1996). The ECHAM4.5 AGCM is one of the forecast AGCMs used operationally at the International Research Institute for Climate Prediction (IRI). It produces well the large-scale climate variability over northeast South America (e.g., see http://iri.columbia.edu/forecast/skill/SkillMap.html).
The regional model used in this study is the Regional Spectral Model (RSM) version 97, developed at the National Centers for Environmental Prediction (NCEP) (Juang and Kanamitsu 1994). It is based on the primitive equations in terrain-following sigma coordinates. The NCEP RSM is one of the regional models involved in the IRI/Applied Research Centers (ARCs) Regional Model Intercomparison over Brazil, and the overall performance of the RSM is good among all the regional models (Roads et al. 2003). A case study with the RSM on downscaling of ECHAM AGCM seasonal prediction over Nordeste also produced encouraging results (Nobre et al. 2001).
The NCEP RSM97 is used to downscale ECHAM4.5 AGCM simulations. The one-way nesting of the NCEP RSM97 into the ECHAM4.5 AGCM is done in a way that is different from convectional methods, which use global model results along the lateral boundary zone only. The perturbation nesting method used here allows the global model outputs to be used over the entire regional domain, not just in the lateral boundary zone. The dependent variables in the regional domain are defined as a summation of perturbation and base. The base is a time-dependent prediction from the AGCM. All other variability that cannot be predicted by the AGCM but can be resolved and predicted by the RSM in the regional domain is defined as perturbation. Nesting is done in such a way that the perturbation is nonzero inside the domain but zero outside the domain. Five prognostic variables only from the global model outputs are used as the base fields in the RSM. They are zonal and meridional winds, temperature, humidity, and surface pressure. Perturbations are often concentrated in the smaller scales. In some cases, the perturbations can be strong at larger scales (e.g., Ward and Sun 2002). All diagnostic variables (e.g., precipitation) are generated by the regional model itself.
We focus on the downscaling of precipitation in this study. The cumulus convection scheme developed by Tiedtke (1989) is used in the ECHAM4.5 AGCM, and the simplified Arakawa–Schubert scheme is used in the NCEP RSM (Hong and Pan 1996). Precipitation difference between RSM simulations and the driving AGCM outputs is attributed not only to the different resolution forcing but also to the difference in the convection schemes used. It is necessary to verify if the RSM can reproduce the large-scale information of precipitation in the driving AGCM.
The choice of regional model domain and resolution is very important when setting up the downscaling experiment (Giorgi and Mearns 1999). Several test simulations were performed to determine the optimum horizontal resolution and the size of the computational domain. The resolution varied between 30 and 100 km, and the lateral boundaries were placed as far west as 100°W, as far east as 10°E, as far north as 20°N, and as far south as 30°S. The simulation period is January–June. We chose the Nordeste wet year of 1985 and dry year of 1993 for the test simulations. Compared with observations, the simulated seasonal precipitation bias over Nordeste is the smallest with model horizontal resolution of 60 km and the domain defined in Fig. 1 (109 × 72 grid points). Thus, this configuration is chosen for our downscaling study. Note that Nordeste is far from the lateral boundaries. This prevents possible noise generated at the lateral boundaries from excessively contaminating the solution over Nordeste (Seth and Giorgi 1998). The test simulations show that the Atlantic ITCZ is very critical for the precipitation simulation in Nordeste. The entire tropical Atlantic Ocean has to be included in the domain. The displacement of the ITCZ occurs when only a portion of tropical Atlantic Ocean is included in the domain. With the resolution of 60 km, the main topographical features are resolved by the RSM. The São Francisco Valley runs roughly in a south–north direction, bounded on the west by the range of Serra Geral de Goiás and on the east by the ranges of Serra do Espinhaço and Chapada Diamantina. The smaller ranges of Serra Ibiapaba, Chapada do Araripe, and Borborema in northeast Nordeste can also be identified in Fig. 1. These topographic features cannot be resolved in the ECHAM4.5 AGCM at T42 resolution. Instead, the land rises smoothly from the coast toward inland with the peak at 18°S, 46°W (not shown).
Our downscaling study will focus on the rain season in Nordeste. An ensemble of 10 ECHAM4.5 AGCM integrations forced with observed time-evolving SSTs was done at the IRI from the late 1940s to present. For each of the ECHAM4.5 AGCM integrations a nested integration with the NCEP RSM was done for the period January–June 1971–2000. Six-hour wind, temperature, humidity, and surface pressure data from ECHAM4.5 AGCM outputs were linearly interpolated in time and in space onto RSM grids as base fields. Soil wetness and soil temperature in RSM were predicted after initialization. They were initialized by ECHAM AGCM outputs on 1 January every year. Verification was done for all 6 months (i.e., January–June), and in this paper we focus on the February–April period, which comprises a large proportion of the rain season in Nordeste. Furthermore, February–April is the critical period for crop management in Nordeste (Sun et al. 2004, manuscript submitted to J. Appl. Meteor., hereafter SLW).
b. Observational datasets
Precipitation is the primary variable for performing the quantitative assessment of the model performance. The observed monthly precipitation data used for verification are a blend of several observational datasets. The idea is to get the best possible quality, while maximizing coverage. Global precipitation data with resolution of 0.5° from the Climate Research Unit at University of East Anglia (CRU05 dataset) (New et al. 2000) served as the foundation. This dataset covers the period 1901–98. The data for the period 1971–98 are used here in the model simulation period. Over 19 800 stations were used to construct this global dataset. Nordeste is one of the regions of the world with high station density. These station data have undergone extensive quality control. Note that the station density in 1990s is dramatically reduced compared to that in 1970s. The CRU05 dataset was linearly interpolated onto the RSM grid. Then monthly precipitation data with resolution of 0.25° over Brazil for the period 1994–2000 from the Instituto Nacional de Meteorologia was linearly interpolated onto the RSM grid (hereafter referred to as Brazil dataset). The Brazil dataset was constructed using more than 1000 stations in Brazil and about 300 stations in Nordeste. However, the Brazil dataset was not as rigorously quality controlled. The Brazil dataset and the CRU05 dataset were consolidated on the RSM grid. We gave an equal weight when both of them were available on a RSM grid square. The third dataset is the Ceará dataset. It includes rain gauge data with 95 stations over the state of Ceará for the period 1971–2000. Figure 2 shows the station locations as well as the station altitudes. All of the 95 station data have undergone extensive quality control. The station data was also linearly interpolated onto the RSM grid. The so-called IRI-blended precipitation data refers to the combined CRU05 dataset and Brazil dataset everywhere except for the state of Ceará, where only the Ceará dataset was used.
Observed daily precipitation data are available only in the state of Ceará for the period 1974–2000. There are at least 300 rainfall reporting stations every year, and there are 95 stations with a completed record for all years. These daily data were interpolated onto the RSM grid and will be referred to as the Ceará daily dataset.
3. Results
As noted in section 2, our analyses focus on the season of February–April (FMA). Most figures show Nordeste and the adjoining ocean areas, which is the main area of interest, or the state of Ceará only due to the paucity of observations, rather than whole domain, which is shown in Fig. 1.
a. Precipitation variability at seasonal time scales
To first evaluate how the nested NCEP RSM97 simulations compare to those from the ECHAM4.5 AGCM alone, Fig. 3 shows the climatology of observed and simulated FMA precipitation by the AGCM and the nested RSM. The climatology used here is the 1971–2000 mean of FMA precipitation, except for the observed precipitation over the ocean. The IRI blended precipitation dataset covers land area only. Climatology over the ocean in Fig. 3a is obtained from the 1979–2000 mean of the Global Precipitation Climatology Project (GPCP) dataset (Xie and Arkin 1996). The observed precipitation maximum is located in northwest Nordeste. Relatively high precipitation amounts are also found in the central Nordeste near 7°S, 39°W and along the Nordeste coast. Low precipitation amounts occur over most of southeastern Nordeste. The Atlantic ITCZ is over the equatorial region (Fig. 3a). The AGCM captured the general patterns of precipitation as well as the gradient across Nordeste (Fig. 3b). There are two major differences between the observed and AGCM climatology:
the AGCM is unable to resolve the regional or local scale precipitation patterns such as the relatively high precipitation amounts in the central Nordeste and along Nordeste east coast, undoubtedly due to its coarse resolution, and
the simulated Atlantic ITCZ is southward shifted compared with the observation.
As shown in Fig. 4, the maximum precipitation along longitude 35°W is near 2°N (4°S) in the observation (AGCM simulation). The RSM tends to correct the AGCM’s ITCZ bias toward observation by placing the maximum precipitation near 1°S.
As shown in Fig. 3c, the RSM produced the precipitation maximum in northwest Nordeste. Observed high precipitation amounts in central Nordeste near 7°S, 39°W is also resolved by the RSM. Relatively high precipitation amounts along the coast and the low-precipitation band next to the coastal zone were also captured in the RSM simulations. The main weakness in the nested RSM simulation is that the precipitation maximum in northwest Nordeste is shifted southward compared with observations. We also note that the RSM produced dry biases in the area of 6.5°–3°S, 38.5°–37°W relative to observations.
Easterly flow prevails over Nordeste during rain seasons. In response to the topographic forcing, windward high pressure patterns and leeward lows are generated in RSM simulations. Thus, windward divergence and leeward convergence are produced (Fig. 5b). Windward divergence leads to lower precipitation amounts, and leeward convergence leads to high precipitation amounts. This explains the local precipitation patterns over inland regions in Nordeste in RSM simulations, and maybe the observed precipitation patterns as well. As expected from its high resolution, the RSM can resolve the land–sea breeze and its diurnal cycle. A narrow convergence zone along the coast is shown in Fig. 5b. This explains the high precipitation amounts along the coast in Fig. 3c. The difference of strong convergence patterns in the Atlantic equatorial region is evident between the AGCM and RSM simulations (Figs. 5a,b). The RSM produced a narrow strong convergence zone with convergence stronger than 0.4 × 10−5 s−1. This may be attributed to the high resolution of the model (Nobre et al. 2001). This strong convergence zone is located further north than the AGCM simulations. This may explain the northward shift of the ITCZ in RSM simulations.
In Fig. 6, we have plotted the FMA precipitation probability distribution function/curve (PDF) over Ceará from observations, AGCM, and RSM simulations, respectively. Note that both model PDFs are based on a sample of 300 measurements (i.e., 10 ensembles for the period 1971–2000). The observed precipitation PDF is based on a sample of 30 measurements only. They all suggest non-Gaussian distributions. The RSM-simulated PDF is similar to that of the observations. A positive skew is shown, and the median of the seasonal precipitation total is substantially lower than the mean. This indicates that there are more dry years than wet years during the period 1971–2000. By contrast, there is a negative skew in the ECHAM AGCM simulated precipitation—that is, the median is substantially higher than the mean, and there are more wet years than dry years in the ECHAM AGCM simulation for the period 1971–2000. While the limited sample size prohibits rigorous analysis of the differences between AGCM outputs and observations, it appears that the nested RSM is better than the AGCM in simulating the observed precipitation distribution.
The ability of the nested RSM in simulating precipitation interannual variability is illustrated in Fig. 7. Anomaly correlations between observed and simulated FMA precipitation exceed 0.5 over most of Nordeste. The correlation drops rapidly beyond the 11°S latitude. This result confirmed that the SST anomaly forcing is the primary factor responsible for the interannual variability of precipitation in Nordeste since the only external forcing for nested RSM simulations is observed SSTs.
Having shown that the RSM is able to simulate FMA precipitation variability in Nordeste, we applied a spatial-scale separation technique to analyze the added value of RSM compared with the AGCM. We upscaled observed and RSM-simulated precipitation to the resolution of the ECHAM4.5 AGCM (i.e., about 2.8°). This is done by using running average over 5 × 5 grid points. The upscaled precipitation is treated as the large-scale component, and the precipitation difference between the total field and the large-scale component is treated as the local component. There is no local-scale component in ECHAM AGCM simulations. The precipitation in the AGCM simulation is treated as the large-scale component. Comparison of the climatology of the large-scale component of the observations, RSM-simulated precipitation (not shown), and AGCM-simulated precipitation (Fig. 3b) shows the large-scale features in agreement, with higher precipitation amounts in northwest Nordeste and along the northern coast, and low precipitation amounts in southeast Nordeste. Figure 8 shows the anomaly correlations for the large-scale component between observations and model simulations. The high correlation values given by the ECHAM AGCM are reproduced by the nested RSM in Nordeste. These results suggest that the nested RSM can capture the large-scale information as well as its interannual variability. The climatology of the local-scale component of precipitation (1971–2000) is shown in Fig. 9. The observed local-scale component of precipitation is presented in the state of Ceará and along the Nordeste coast, and displays near-zero values elsewhere (Fig. 9a). For the areas with near-zero values of local-scale component climatology, we found that the local component is also very weak during all the years (i.e., 1971–2000). However, the RSM does simulate at least a small local-scale component everywhere (Fig. 9b). The RSM local-scale component is masked out over the ocean since there are no observations for verification. There are two possible explanations for the regions with near-zero values of local-scale component in the observation: 1) the large-scale component is dominant and the local-scale component is negligible or 2) the local-scale component does exist but the observed dataset used here fails to detect it. Considering that both the CRU05 and Brazil datasets were generated from a limited number of observational stations, it is possible that the local-scale component is missed for some regions. On the other hand, the Ceará dataset was derived from 95 stations (Fig. 2). This makes the Ceará dataset capable of having a local-scale component, and Fig. 9a confirmed this. We tend to accept the second explanation. This result may also suggest that more station data are required for regional model evaluation.
Over Ceará, the RSM captured the main features of the observed climatology: two local maxima along the coast, one local minimum in central Ceará, and relatively high values in southern Ceará. We note that the local-scale component accounts for a small part of total precipitation climatology (i.e., about 15% averaged over Ceará). However, it significantly contributes to the total precipitation variability. Figure 10 shows the observed and RSM-simulated precipitation standard deviation. The standard deviation of the observed local-scale component is roughly one-half of that of the observed total precipitation. The standard deviation of the observed large-scale component shows near uniformity over Ceará with a weak north–south gradient (not shown). Thus, the spatial patterns of standard deviation of total precipitation are significantly attributed to variability of the local-scale component (Figs. 10a,b). The RSM has some ability in producing variability of the local component (Figs. 10c,d). It can generate about one-half and one-third of observed variations of the local-scale component of precipitation in central Ceará and the Ceará coastal region, respectively. The RSM’s weakness in producing smaller precipitation variability at smaller spatial scales may be attributed to the lack of various small-scale forcings in the model. For example, SSTs with resolution of 2° lat × lon were used in the model, and the vegetation cover was constant during the 30-yr simulations. Use of high-resolution SSTs (e.g., 0.5°) and varying vegetation cover may improve the model simulation.
Having shown the importance of the local-scale component to the total precipitation variability and the RSM’s ability in generating local-scale information, we now examine the predictability of the local-scale information produced by the RSM. The physical climate anomaly signal in both RSM simulations and observed precipitation data tend to be significantly contaminated by noise, particularly for the local-scale component. In order to separate the signal from the noise, we filtered both the observed and RSM-simulated local-scale component of precipitation by retaining only the leading empirical orthogonal function (EOF). We used standardized data for EOF analysis. Figures 11 and 12 display the observed and simulated leading EOF patterns, as well as the corresponding principal components. The RSM generally captured the observed EOF1 spatial patterns over Ceará. It failed to produce the observed positive amplitude at the area centered at 5.2°S, 40.5°W. The observed leading EOF explains about 17% of the total variance, while the explained variance for the leading EOF of RSM is much higher (47%). This may be associated with the absence of several important forcings in the model, such as small-scale SST forcing and local vegetation changes from year to year.
As shown in Fig. 12, the time series of the leading EOF between observations and RSM simulations are in good agreement, with a correlation coefficient of 0.44. Further analysis indicated that the observed time series of EOF1 is correlated to the tropical Atlantic dipole, with a correlation coefficient of −0.47. The correlation exceeds the 99% confidence level. The index for the tropical Atlantic dipole is defined as the SST difference between area 5°–28°N and area 20°S–5°N (Servain 1991). The correlation between the observed time series for leading EOF and Niño-3.4 is −0.29, which is below the 90% confidence level. For RSM simulations, the time series of EOF1 is well correlated to both Niño-3.4 and the tropical Atlantic dipole, with correlations of −0.80 and −0.46, respectively. The model intensified the link between the local-scale component of observed precipitation and the tropical Atlantic dipole. The model also generated a false “link” between the local-scale component of precipitation and Niño-3.4.
We also performed an EOF analysis for the large-scale component of precipitation and found that the observed leading EOF accounts for more than 90% of total variance and is highly correlated with both the Niño-3.4 and the tropical Atlantic dipole, with correlation −0.56 and −0.60, respectively. The RSM reproduced these observed links. The results may suggest that both Niño-3.4 and the tropical Atlantic dipole affect the large-scale component of precipitation in Ceará, while the tropical Atlantic dipole is more important than the Niño-3.4 for the local component of precipitation.
b. Precipitation predictability at seasonal time scales
Here M is the number of ensembles, pij represents the seasonal precipitation at a given grid cell during year i of ensemble member j, 
Note that, if every ensemble member produced identical time series of precipitation, then σ2p̂ would attain its maximum value of σ2p. At the other extreme, if all values of pij for each year i were completely independent, then σ2p̂ would attend its lowest value of σ2p/M. The value of Ωp ranges from 0 to 1. For large enough M, Ωp reduces to the simple “signal to total” variance ratio σ2p̂/σ2p. Higher values of Ωp imply a greater control of SST variations over the timing of precipitation anomalies. Therefore, Ωp quantifies the robustness of the precipitation response to the prescribed time-varying SSTs. In other words, Ωp can be interpreted as a field of “maximum potential prediction skill” in a model system that relies solely on SST forecasts.
The coherence index for both the AGCM and nested RSM simulations for the precipitation over Ceará is listed in Table 1. The relatively high values of coherence index produced by the AGCM and the RSM suggest good prediction skills. Note that the coherence index produced by the AGCM is larger than that produced by the RSM. Both large-scale and local-scale information are included in the RSM simulation, and the larger intraensemble noise (i.e., larger spread of ensemble members) in the RSM is mainly attributed to simulation of the local-scale component (Fig. 13). A similar value of coherence index was obtained when we used the large-scale component of precipitation only in the RSM. We evenly divided the 30-yr period 1971–2000 into three categories: above-normal, near-normal, and below-normal precipitation years. Further analysis indicated that both models show good prediction skills for the below-normal and above-normal years, particularly the below-normal years, and no skill for the near-normal years. This is consistent with most AGCMs’ performance (Van den Dool and Toth 1991).
Can model prediction skills be improved if we increase the ensemble size? Although the ensemble sizes used in this study are moderate, an attempt is made to shed light on this question by comparing the coherence index with different ensemble sizes. Table 2 lists the number of combinations for the ensemble size ranging from 2 to 10, and Fig. 14 shows the distribution of coherence indices. The spread of coherence indices is consistently reduced as the ensemble size increases. This is because the estimation of the coherence becomes more stable with larger ensemble sizes. This implies that the reliability of prediction is consistently increased as the ensemble size increases. However, the coherence index averaged across all combinations with the same ensemble size is almost constant for all the ensemble sizes. This implies that the accuracy of the prediction is not necessarily improved with an increase in the ensemble size. Similar results are obtained for the ECHAM AGCM simulation (not shown).
c. Daily precipitation variability
An aspect of precipitation variability that is important for climate impact assessments is the frequency and intensity distribution of daily precipitation events, which can be as important, or even more important, than the seasonal-average precipitation (e.g., SLW). Observed daily precipitation is available only over the state of Ceará. Since Ceará is not a homogeneous region for the local-scale component of precipitation, we target a relatively small area (i.e., a nearly homogeneous region) in this section. The shaded area in Fig. 2 shows the rain-fed agriculture region in Ceará. It covers 13 RSM grid cells, about half the size of the AGCM grid cell. We use this region, containing data for 64 stations, for detailed investigation of daily precipitation variability. Analysis of daily precipitation shows high covariance among these stations. Therefore, observed daily precipitation averaged over all the stations is compared with the RSM daily precipitation averaged over the 13 corresponding grid points. Daily precipitation events within the intensity intervals 0–1, 1–5, 5–10, and >10 mm were tallied, and the frequency values normalized by the total number of days in the FMA season are used for the analysis of precipitation intensity. Figure 15 shows the climatology of precipitation intensity distribution (PID) and the correlation between observed and RSM-simulated seasonal precipitation intensity. The events with daily precipitation in the range of 0–1, 1–5, 5–10, and >10 mm are categorized as no-rain events, light rain events, medium rain events, and heavy rain events, respectively. It is clear that the RSM overestimated the frequency of no-rain events and underestimated the frequencies of light, medium, and heavy rain events. This has been attributed in part to the model cumulus parameterizations in the RSM. The RSM does capture the interannual variability of no-rain events, medium rain events, and heavy rain events well since the correlations are well above the 99% confidence level. The RSM also shows some ability in producing the interannual variability of light rain events. The RSM skill in simulating the frequency variability of precipitation events is partly related to the magnitude of observed variability. The model is usually capable of capturing large observed variability but has difficulty capturing small observed variability. The observed standard deviation–to-mean ratio is 49%, 25%, 33%, and 57% for no rain events, light rain events, medium rain events, and heavy rain events, respectively.
Dry spells and wet spells are rarely discussed in climate modeling papers, although their importance to applications is demonstrated (SLW). An attempt is made in this study to investigate the RSMs’ skill in generating dry and wet spells since it is more important for the semiarid rain-fed agricultural region, where there is great vulnerability of the local population to recurrent drought conditions (Chimeli and Filho 2004, unpublished manuscript). In this study, dry spells are defined as precipitation below 2 mm day−1 for a certain period of time. Figure 16 shows the climatology (1974–2000) of observed and RSM-simulated dry spell numbers with different dry spell thresholds and the RSM skill in simulating the observed dry spells. We note that the RSM systematically overestimated the dry spell numbers with the dry spell threshold ranging from 3 to 18 days. However, the RSM does show skill in capturing the interannual variability of dry spell frequency for dry spell thresholds ranging from 2 to 15 days. Within this range, correlations between the observed and RSM- simulated dry spells exceed 0.3 with the correlation decreasing quickly for dry spells lasting more than 15 days. The model’s inability to simulate extremely long dry spells is expected since dry spells longer than 15 days occurred only four times during the FMA period from 1974 to 2000.
Α strong weight is given to dry spells lasting at least 11 days because they cause severe damage to crop yield in this region. SLW found that crop yield is highly correlated to the observed drought index, with correlation −0.75, and the link between crop yield and observed seasonal precipitation total is weak, with correlation of 0.24. Figure 17a shows the observed and RSM-simulated drought index for the period 1974–2000. The RSM captured the interannual variability of the observed drought index well, with a correlation between them of 0.67. We found, as would be expected, that large positive (negative) anomalies of seasonal precipitation total are usually associated with smaller (larger) drought indices. We also found larger interannual variability for the drought index than for the seasonal precipitation total. The standard deviation-to-mean ratio is 40% (43%) for observed (RSM simulated) seasonal precipitation total and 77% (66%) for observed (RSM simulated) drought index. This higher variability provides further evidence of the meaningfulness of the drought index as compared to the seasonal total precipitation.
Figure 17 shows the observed and RSM-simulated flooding index for the period 1974–2000. The RSM captures the interannual variability of observed flooding index well. The correlation between them is 0.75. As was the case for the drought index, we found larger interannual variability for the flooding index than the seasonal precipitation total. The standard deviation–to-mean ratio is 124% (81%) for observed (RSM simulated) flooding index.
Drought index and flooding index are highly correlated to the seasonal precipitation anomaly for both observations and RSM simulations. The RSM is also not able to capture observed indices variability during the years with near-normal seasonal precipitation. However, the two indices represent precipitation information at intraseasonal scale, and RSM-simulated indices, not the seasonal precipitation, are highly correlated to crop yield in this region (SLW).
4. Summary and concluding remarks
We have employed the NCEP Regional Spectral Model version 97 with horizontal resolution of 60 km to downscale the climate information from the ECHAM4.5 AGCM simulations forced with observed time evolving SSTs in Nordeste, Brazil, for the rain season of 1971–2000. The ECHAM4.5 AGCM is able to produce the observed large-scale information in general. The main deficiency of the AGCM is that it systematically places the tropical Atlantic ITCZ to the south of its observed position. At seasonal time scales, the RSM captures the spatial characteristics of the observed precipitation and the interannual variability as well. The RSM places the Atlantic tropical ITCZ to the north of the AGCM’s ITCZ position, which partly corrects the AGCM’s bias with respect to observations. The precipitation PDF analysis over Ceará indicated that the RSM distribution is consistent with observations—that is, a positive skew with more dry years than wet years for the period 1971–2000—while the AGCM produced more wet years than dry years for the same period.
To examine the RSM’s added value compared with the AGCM, a spatial-scale separation technique is applied to the observations and the RSM outputs. The observed and RSM-simulated precipitation are upscaled to the AGCM’s resolution. The upscaled precipitation is referred to as the large-scale component of precipitation, and the precipitation difference between the total field and the large-scale component is treated as the local-scale component. The AGCM-simulated precipitation and its interannual variability are well reproduced by the RSM as the large-scale component of precipitation, which is also consistent with the observed large-scale component of precipitation. We argued that the local-scale component of precipitation is not included in the observed dataset except for the state of Ceará due to the paucity of observations elsewhere. Dense networks of monitoring stations are required to reflect the local-scale component of precipitation. For the state of Ceará, the RSM captured the observed spatial patterns of local component climatology—that is, two local maxima along the Ceará coast, one local minimum in the central Ceará, and relatively high values in southern Ceará. EOF filtering of the observed and the RSM-simulated local-scale component of precipitation was used to suppress noise. The time series for the leading principal component between observation and the RSM simulation are in good agreement. We found that the local-scale component of variation is closely related to the tropical Atlantic dipole, while its link with Niño-3.4 is weak. However, the large-scale component is highly related to both the Atlantic dipole and Niño-3.4. We noted that the RSM generated a false link between the local-scale component and Niño-3.4. While the local-scale component accounts for a relatively small portion of the total precipitation climatology, it significantly contributes to the total precipitation variability. One RSM deficiency is that it produced weaker than the observed variability for the local-scale component of precipitation.
We examined the significance and reliability of the model predictions. Good potential prediction skills are obtained from the ECHAM4.5 AGCM in predicting large-scale information and from the NCEP RSM in predicting both large-scale and local-scale information. The skill is higher in the AGCM than the RSM. This is mainly due to the noise in the simulation of the local-scale component of precipitation. The reliability of prediction is consistently increased as the ensemble size increases. However, the expected accuracy of the mean of the prediction is not affected by the increase in the ensemble size.
Seasonal statistics of daily precipitation in the RSM is also examined over the Ceará rain-fed agricultural region. The RSM shows reasonable skill in producing the interannual variability of precipitation intensity distribution, although it systematically overestimated the frequency of no-rain events and underestimated the frequency of light, medium, and heavy rain events. The RSM has measurable skill in capturing the variability of the dry spells, with a threshold ranging from 2 to 15 days for the FMA period from 1974 to 2000. A drought index and a flooding index are used in this paper to indicate the severity of drought and flooding conditions. The RSM captured the observed variability of these indices fairly well.
This study confirmed that the SST anomaly forcing is the primary factor responsible for the interannual variability of precipitation in Nordeste. These results provide the ground for climate dynamical downscaling forecasts as they show that the AGCM forced by observed SSTs is able to capture the large-scale waves and the RSM is able to produce regional- or local-scale precipitation at both seasonal and intraseasonal time scales. SST anomalies, particularly in the tropical oceans, can be predicted with measurable skill (Mason et al 1999; Repelli and Nobre 2001; Zebiak and Cane 1987). A dynamical downscaling prediction system based on the ECHAM4.5 AGCM and NCEP RSM97 has been developed for Nordeste, and the forecasts have been issued since December 2001 and are updated monthly during rain seasons. A summary and validation of the downscaling forecasts will be presented in another paper.
Acknowledgments
Thanks are due to David DeWitt for running the AGCM and to Wagner Luis Barbosa Melciades, Emerson Mariano da Silva, and Vicente Silva Filho for performing the RSM integrations. The authors also acknowledge the enlightening discussion with Steve Zebiak, Anthony Barnston, M. Neil Ward, Matayo Indeje, and Anji Seth.
REFERENCES
Chen, S-C., J. O. Roads, H-M. H. Juang, and M. Kanamitsu, 1999: Global to regional simulation of California’s wintertime precipitation. J. Geophys. Res., 104 , 31517–31532.
Denis, B., R. Laprise, D. Caya, and J. Cote, 2002: Downscaling ability of one-way nested regional climate models: The big-brother experiment. Climate Dyn., 18 , 627–646.
Fennessy, M. J., and J. Shukla, 2000: Seasonal prediction over North America with a regional model nested in a global model. J. Climate, 13 , 2605–2627.
Gershunov, A., and T. P. Barnett, 1998: ENSO influence on intraseasonal extreme rainfall and temperature frequencies in the contiguous United States: Observations and model results. J. Climate, 11 , 1575–1586.
Giorgi, F., and M. R. Marinucci, 1991: Validation of a regional atmospheric model over Europe: Sensitivity of wintertime and summertime simulations to selected physics parameterizations and lower boundary conditions. Quart. J. Roy. Meteor. Soc., 117 , 1171–1207.
Giorgi, F., and M. R. Marinucci, 1996: An investigation of the sensitivity of simulated precipitation to model resolution and its implications for climate studies. Mon. Wea. Rev., 124 , 148–166.
Giorgi, F., and L. O. Mearns, 1999: Introduction to special section: Regional climate modeling revisited. J. Geophys. Res., 104 , 6335–6352.
Harzallah, A., J. O. R. Aragao, and R. Sadourny, 1996: Interannual rainfall variability in north-east Brazil: Observation and model simulation. Int. J. Climatol., 16 , 861–878.
Hastenrath, S., and L. Heller, 1977: Dynamics of climatic hazards in northeast Brazil. Quart. J. Roy. Meteor. Soc., 103 , 77–92.
Hong, S-Y., and H-L. Pan, 1996: Nonlocal boundary layer vertical diffusion in a medium-range forecast model. Mon. Wea. Rev., 124 , 2322–2339.
Hong, S-Y., and A. Leetmaa, 1999: An evaluation of the NCEP RSM for regional climate modeling. J. Climate, 12 , 592–609.
Hong, S-Y., H-M. H. Juang, and D-K. Lee, 1999: Evaluation of a regional spectral model for the East Asian monsoon case studies for July 1987 and 1988. J. Meteor. Soc. Japan, 77 , 553–572.
Juang, H-M. H., and M. Kanamitsu, 1994: The NMC nested regional spectral model. Mon. Wea. Rev., 122 , 3–26.
Kanamtisu, M., and H-M. H. Juang, 1994: Simulation and analysis of an Indian Monsoon by the NMC Nested Regional Spectral Model. Proc. Int. Conf. on Monsoon Variability and Prediction, Vol. 2, WMO TD-619, Trieste, Italy, ICTP/WMO/ICSU/IOC, 50–59.
Koster, R. D., M. J. Suarze, and M. Heiser, 2000: Variance and predictability of precipitation at seasonal-to-interannual timescales. J. Hydrometeor., 1 , 26–46.
Leung, L. R., L. O. Mearns, F. Giorgi, and R. L. Wilby, 2003: Regional climate research: Need and opportunities. Bull. Amer. Meteor. Soc., 84 , 89–95.
Mason, S. J., L. Goddard, N. E. Graham, E. Yulaeva, L. Sun, and P. A. Arkin, 1999: The IRI seasonal climate prediction system and the 1997/98 El Niño event. Bull. Amer. Meteor. Soc., 80 , 1853–1873.
Mearns, L. O., S. H. Schneider, S. L. Thompson, and L. R. McDaniel, 1990: Analysis of climate variability in general circulation models: Comparison with observations and changes in variability in 2× CO2 experiments. J. Geophys. Res., 95 , 20469–20490.
Mearns, L. O., F. Giorgi, L. Mcdaniel, and C. Shields, 1995: Analysis of daily variability of precipitation in a nested regional climate model: Comparison with observations and double CO2 results. Global Planet. Change, 10 , 55–78.
Mechoso, C. R., S. W. Lyons, and J. A. Spahr, 1990: The impact of sea surface temperature anomalies on the rainfall over northeast Brazil. J. Climate, 3 , 812–816.
Mo, K., M. Kanamitsu, H-M. H. Juang, and S-Y. Hong, 2000: Ensemble regional and global climate prediction for the 1997/1998 winter. J. Geophys. Res., 105 , 29609–29623.
Moron, V., M. N. Ward, and A. Navarra, 1996: Observed and SST-forced seasonal rainfall variability across tropical America. Int. J. Climatol., 21 , 1467–1501.
Moron, V., A. Navarra, M. N. Ward, and E. Roeckner, 1998: Skill and reproducibility of seasonal rainfall patterns in the tropics in ECHAM-4GCM simulations with prescribed SST. Climate Dyn., 14 , 83–100.
Moura, A. D., and J. Shukla, 1981: On the dynamics of droughts in northeast Brazil: Observations, theory and numerical experiments with a general circulation model. J. Atmos. Sci., 38 , 2653–2675.
New, M., M. Hulme, and P. Jones, 2000: Representing twentieth-century space–time climate variability. Part II: Development of 1901–1996 monthly grids of terrestrial surface climate. J. Climate, 13 , 2217–2238.
Nobre, P., A. D. Moura, and L. Sun, 2001: Dynamical downscaling of seasonal climate prediction over Nordeste Brazil with ECHAM3 and NCEP’s Regional Spectral Models at IRI. Bull. Amer. Meteor. Soc., 82 , 2787–2796.
Pezzi, L. P., and I. F. A. Cavalcanti, 2001: The relative importance of ENSO and tropical Atlantic sea surface temperature anomalies for seasonal precipitation over South America: A numerical study. Climate Dyn., 17 , 205–212.
Repelli, C. A., and P. Nobre, 2001: Statistical prediction of sea surface temperature over the tropical Atlantic. Int. J. Climatol.24, 45–55.
Roads, J., 2000: The Second International Regional Spectral Model Workshop: Meeting summary. Bull. Amer. Meteor. Soc., 81 , 2979–2982.
Roads, J., and Coauthors, 2003: International Research Institute/Applied Research Centers (IRI/ARCs) regional model intercomparison over South America. J. Geophys. Res., 108 .4425, doi:10.1029/2002JD003201.
Roeckner, E., , and Coauthors. 1996: The atmospheric general circulation model ECHAM-4: Model description and simulation of present-day climate. Tech. Rep. 218, Max-Planck Institute for Meteorology, Hamburg, Germany, 90 pp.
Ropelewski, C. F., and M. S. Halpert, 1996: Quantifying Southern Oscillation–precipitation relationships. J. Climate, 9 , 1043–1059.
Roucou, P., J. O. R. Aragao, A. Harzallah, B. Fontaine, and S. Janicot, 1996: Vertical motion changes related to north-east Brazil rainfall variability: A GCM simulation. Int. J. Climatol., 16 , 879–891.
Servain, J., 1991: Simple climatic indices for the tropical Atlantic Ocean and some applications. J. Geophys. Res., 96 , 15137–15146.
Seth, A., and F. Giorgi, 1998: The effects of domain choice on summer precipitation simulation and sensitivity in a regional climate model. J. Climate, 11 , 2698–2712.
Seth, A., and M. Rojas, 2003: Simulation and sensitivity in a nested modeling system for South America. Part I: Reanalyses boundary forcing. J. Climate, 16 , 2437–2453.
Sperber, K. R., and T. N. Palmer, 1996: Interannual tropical rainfall variability in general circulation model simulations associated with the Atmospheric Model Intercomparison Project. J. Climate, 9 , 2727–2750.
Sun, L., F. H. M. Semazzi, F. Giorgi, and L. Ogallo, 1999a: Application of the NCAR regional climate model to eastern Africa 1. Simulation of the short rains of 1988. J. Geophys. Res., 104 , 6529–6548.
Sun, L., F. H. M. Semazzi, F. Giorgi, and L. Ogallo, 1999b: Application of the NCAR regional climate model to eastern Africa 2. Simulation of interannual variability of short rains. J. Geophys. Res., 104 , 6549–6562.
Takle, E. S., and Coauthors, 1999: Project to Intercompare Regional Climate Simulations (PIRCS). J. Geophys. Res., 104 , 19443–19449.
Tiedtke, M., 1989: A comprehensive mass flux scheme for cumulus parameterization in large-scale models. Mon. Wea. Rev., 117 , 1779–1800.
Uvo, C. B., C. A. Repelli, S. E. Zebiak, and Y. Kushnir, 1998: The relationships between tropical Pacific and Atlantic SST and Northeast Brazil monthly precipitation. J. Climate, 11 , 551–562.
Van den Dool, H. M., and Z. Toth, 1991: Why do forecasts for “near normal” often fail? Wea. Forecasting, 6 , 76–85.
Ward, M. N., and C. K. Folland, 1991: Prediction of seasonal rainfall in the north Nordeste of Brazil using eigenvectors of sea surface temperature. Int. J. Climatol., 11 , 711–743.
Ward, M. N., and L. Sun, 2002: Intensive course on dynamical downscaling of seasonal to interannual climate predictions. IRI Tech. Rep. 014, Palisades, NY, 215 pp.
Xie, P., and P. A. Arkin, 1996: Analysis of global monthly precipitation using gauge observations, satellite estimates, and numerical model predictions. J. Climate, 9 , 840–858.
Zebiak, S. E., and M. A. Cane, 1987: A model El Niño–Southern Oscillation. Mon. Wea. Rev., 115 , 2262–2278.
Topographic field used by the NCEP RSM97 with horizontal resolution of 60 km. Units are meters above sea level and the contour interval is 50 m. The São Francisco Valley, and the ranges of Serra Ibiapaba, Chapada do Araripe, and Borborem are marked with SA, SI, CA, and BB, respectively. The location of Nordeste is indicated by the rectangle.
Citation: Journal of Climate 18, 4; 10.1175/JCLI-3266.1
Observational stations in Ceará used to create monthly precipitation data on RSM grid. Station locations are marked by small squares, and the colors indicate the station altitudes. Units are meters above sea level. The shaded area is the rain-fed agriculture region.
Citation: Journal of Climate 18, 4; 10.1175/JCLI-3266.1
Precipitation climatology (mm day−1) for Feb–Apr season. (a) Observation; (b) ECHAM4.5 AGCM simulation; (c) NCEP RSM97 simulation.
Citation: Journal of Climate 18, 4; 10.1175/JCLI-3266.1
Precipitation climatology (mm day−1) for the FMA season along longitude 35°W. (a) Observation; (b) ECHAM4.5 AGCM simulation; (c) NCEP RSM97 simulation.
Citation: Journal of Climate 18, 4; 10.1175/JCLI-3266.1
Convergence field (×10−5 s−1) for the FMA season averaged over the period 1979–98. (a) ECHAM4.5 AGCM simulation; (b) NCEP RSM97 simulation.
Citation: Journal of Climate 18, 4; 10.1175/JCLI-3266.1
Climatological precipitation PDF for the period FMA 1971–2000 in Ceará. The unit for zonal axis is one standard deviation, and the unit for the vertical axis is percentage of occurrences.
Citation: Journal of Climate 18, 4; 10.1175/JCLI-3266.1
Temporal correlations between observed and RSM- simulated precipitation anomalies for FMA 1971–2000.
Citation: Journal of Climate 18, 4; 10.1175/JCLI-3266.1
Temporal correlations between the observed large-scale component of precipitation and (a) ECHAM4.5 AGCM-simulated precipitation and (b) NCEP RSM97-simulated large-scale component of precipitation for FMA 1971–2000.
Citation: Journal of Climate 18, 4; 10.1175/JCLI-3266.1
Climatology of the local-scale component of precipitation (mm day−1) for the FMA season averaged over the period 1971–2000. (a) Observation; (b) NCEP RSM97 simulation. The contour interval is 0.5 mm day−1. The solid black line indicates the location of Ceará.
Citation: Journal of Climate 18, 4; 10.1175/JCLI-3266.1
Standard deviation of precipitation (mm day−1) for the period FMA 1971–2000 in Ceará. (a) Total field of observation; (b) local-scale component of observation; (c) total field of RSM simulation; (d) local-scale component of RSM simulation. The contour interval is 0.2 mm day−1.
Citation: Journal of Climate 18, 4; 10.1175/JCLI-3266.1
The leading EOF patterns for the local-scale component of precipitation for the period FMA 1971–2000 in Ceará. (a) Observation; (b) NCEP RSM97 simulation. The contours show the actual eigenvector values multiplied by 10. Contour interval is 0.5.
Citation: Journal of Climate 18, 4; 10.1175/JCLI-3266.1
Time series of the leading EOF mode for the local-scale component of precipitation.
Citation: Journal of Climate 18, 4; 10.1175/JCLI-3266.1
Precipitation anomaly distribution for the FMA season over Ceará. Unit is mm day−1.
Citation: Journal of Climate 18, 4; 10.1175/JCLI-3266.1
Coherence index distribution with different ensemble sizes for the NCEP RSM97 based on the FMA period from 1971 to 2000 in Ceará. The coherence index averaged across all the combinations with the same ensemble size is shown by the red dashed line.
Citation: Journal of Climate 18, 4; 10.1175/JCLI-3266.1
(a) Climatology of precipitation intensity distribution. (b) Correlations between observation and NCEP RSM97 simulation for no-rain events, light rain events, medium rain events, and heavy rain events. Both are for the FMA period from 1974 to 2000 in the rain-fed agriculture region of Ceará.
Citation: Journal of Climate 18, 4; 10.1175/JCLI-3266.1
Climatology of the occurrence of dry spells with different thresholds (frequency on left-hand axis). Also plotted is the correlation between observed and NCEP RSM97-simulated dry spells with different thresholds (right-hand axis). Both are for the FMA period from 1974 to 2000 in the rain-fed agriculture region of Ceará.
Citation: Journal of Climate 18, 4; 10.1175/JCLI-3266.1
(a) Observed and NCEP RSM97-simulated drought index. (b) Observed and NCEP RSM97-simulated flooding index. Both are for the FMA season in the rain-fed agriculture region of Ceará.
Citation: Journal of Climate 18, 4; 10.1175/JCLI-3266.1
Coherence index for both the NCEP RSM and the ECHAM AGCM for the precipitation over Ceará for the FMA period from 1971 to 2000.
The number of combinations for different ensemble size.