1. Introduction
By analyzing Australian surface temperature variations in observations and model simulations, this study is designed to explore whether soil storage of precipitation water can lead to some slow-varying land surface processes that could in turn affect the mean state and the variability of the atmosphere at monthly or even longer scales and modulate the atmospheric responses to SST forcing (e.g., Chen and Kumar 2002). It comprises two components. First, from observational analysis, it investigates if there are observational signals suggesting that land surface conditions can or cannot affect climate variability and predictability in the Australian region. Second, from the analysis of a suite of current AGCM simulations, it assesses whether land surface modeling affects the model-simulated climate variability and predictability over this region.
There are a large number of comprehensive observational studies of the Australian surface temperature variations, and only two recent studies are briefly summarized here. Jones (1999) documented in detail the characteristics of observed land surface temperature variations in the Australian region. Highly significant correlations between rainfall variations and surface temperature variations were noted, and the changes of latent heat cooling directly associated with rainfall variations were attributed to such connections. Following that, Jones and Trewin (2000) reported a detailed observational study of the impacts of the El Niño–Southern Oscillation (ENSO) on Australian surface temperature variation, and they further concluded that the variations of surface energy balance are important in explaining such influences. Results from their analyses, similar to the earlier GCM modeling studies reported by Power et al. (1998) and Simmonds and Hope (1998), demonstrated that the response of land surface processes to the ENSO forcing is important in explaining how ENSO influences Australian surface temperature variations.
The remaining question is, however, whether some slow-varying land surface processes themselves, by responding to anomalous atmospheric forcing, affect surface temperature variations besides the influence from large-scale SST forcing. Despite the fact that previous studies indeed discussed the influence of soil moisture on surface temperature variations, there is still a lack of thorough analysis on trying to separate the impacts of ENSO and the impacts of land surface anomalies on the observed surface temperature variations and predictability. Therefore, the key objective of the observational analysis in this study is to try to go one step further by assessing whether land surface processes can lead to extra predictability of Australian surface temperature variations.
In the bucket-type parameterization, there are two features: (i) The parameterization presents a direct and rapid feedback between evaporation and soil moisture, with no canopy regulation of surface evaporation. If surface resistance is considered in the potential evaporation calculation to account for the physiological resistance exerted by the canopy process, then one can expect a decreased dumping term in such a system due to the decrease in potential evaporation. This means that the soil moisture decay time scale in this case becomes longer. (ii) It has no representation of hydraulic diffusion processes controlling the water movement inside soil and the whole volumetric water is instantaneously available for evaporation. In both accounts, such a hydrological model tends to simulate a rapid response of soil water storage to changes in atmospheric forcing. This leads to a short “memory” of soil moisture in the climate system, which could result in less predictability in the overall model.
An intermediate scheme in soil moisture simulation is the so-called force–restore model in which there is a quick-responding thin top layer and a slowly restoring deep soil layer. The opposite extreme from the “bucket” is a multilayer soil model. This type of scheme has been developed to fully couple soil hydraulic diffusion processes, such as the canopy and root zone processes of water flow in the soil that are affected by root density and distribution, with some schemes even considering horizontal runoff. In this type of scheme, two different time scales are involved in representing the feedbacks between land and the atmosphere: there is a rapid response to the atmospheric forcing in the top thin layer and a slow restore process in the deep layers by soil moisture moving from deep soil to the upper layer for surface evaporation. Surface runoff occurs when the upper layer is saturated even though the deep layer may still be unsaturated. This is contrary to the simulation of runoff in most of the bucket-type models in which runoff only occurs when the whole volumetric soil column is saturated, with only a very few models allowing some water “leaking” in unsaturated soil. In a force– restore or multilayer scheme, surface evaporation is controlled by surface layer soil moisture. Therefore, a dry surface layer can limit surface evaporation while deep soil layers remain wet, leading to a slow release of volumetric soil water. Indeed, when comparing a suite of second-generation land surface schemes in an offline model intercomparison project, Desborough et al. (1996) found no significant dependence of total evaporation on averaged soil moisture in those models, implying weaker feedback between soil water storage and surface evaporation. In addition, in nonbucket schemes, deep soil water was not directly available for surface evaporation. Rather, it depletes by first diffusing through the upper layer(s) and then evaporation from the surface. This leads to slower responses within the subsurface soil layers to atmospheric forcing, particularly outside the root active zone. In both accounts, such schemes can lead to slow-varying land surface processes and introduce a longer memory in the climate system.
However, it should be noted that when the complexity in modeling soil hydrological processes increases, factors controling model-simulated soil moisture variations become complicated. For instance, runoff modeling in bucket-type schemes could counterbalance some effects of the rapid soil moisture depletion due to surface evaporation as a result of the fact that most such models have a large bucket depth, and there is no precipitation water loss to runoff until the whole volumetric soil column is saturated. On the same account, soil water loss due to surface/subsurface runoff/drainage processes and canopy transpiration in multilayer soil hydrological schemes could offset some effects of imposing extra resistance in the pathway of surface evaporation and introducing multilayer soil water restoration processes in such models. Therefore, it is possible that a complex surface scheme may have a shorter soil moisture memory than a model with a bucket-type scheme. Recently, Koster and Suarez (2001) have developed an analytical approach to study the processes affecting soil moisture memory in GCMs. Certainly, applying such an approach in model analysis can shed more light in understanding the different behaviors among the climate models. Furthermore, the translation of soil moisture memory/predictability into the predictability of atmospheric variables is affected by the land–air coupling represented in climate models. Koster et al. (2002) show that the coupling strength varies significantly among the AGCMs, which would affect the influence of soil moisture memory on the atmospheric predictability. Schlosser and Milly (2002) report a model-based study of soil moisture predictability and associated climate predictability and find high associated surface temperature predictability over regions (including a large part of the Australian continent) with strong variability of soil moisture stress on evapotranspiration and abundant surface net radiation.
Considering the differences in soil hydrological modeling and land–air coupling represented in current climate models, the second part of this study is devoted to a suite of AGCM model intercomparison and analysis and serves as a counterpart to the observational analysis in the first part of the study. The second part of the study has benefited from the participation of the diagnostic subproject 12 of the Atmospheric Model Intercomparison Project Phase 2 (AMIP2; Phillips et al. 2000). Zhang et al. (2002) reported a preliminary study on how different the AMIP2 models are in simulating surface climate and fluxes in the Australian region. Here, the modeling analysis is designed to focus on intercomparing a large number of the model 17-yr simulations to explore if soil moisture memory is different in such models and if the differences affect model-simulated climate variability and predictability in the Australian region. Studying the causes of the differences in the model-simulated soil moisture memory is not the primary focus of this study, and it is only briefly discussed. More thorough analyses to explore the mechanisms determining soil moisture variations in these models (e.g., Koster and Suarez 2001) and the translation of soil moisture memory into climate variability (e.g., Koster et al. 2002; Schlosser and Milly 2002), however, are beyond the scope of the current study and will be pursued in the future.
This paper is structured as follows: Section 2 describes the observational data and analysis approaches used in the first part of the study. It also briefly summarizes the 16 AMIP2 models, particularly the land surface components used in these models. Observational assessment of the potential influence of land surface processes on the Australian surface temperature variability is described in section 3. Section 4 presents the analysis of AMIP2 model results demonstrating if land surface parameterizations in GCMs can affect model predictability. Finally, discussion and conclusions from this study are presented in section 5.
2. Data description and analytical approach
The Australian Commonwealth Bureau of Meteorology (BoM) observed monthly mean rainfall, monthly mean daily maximum surface air temperature (Tmax), and daily minimum surface air temperature (Tmin) datasets over the 50-yr period of 1950–99 are used in the observational analysis of this study. The original dataset is described and used in Jones (1999) and Jones and Trewin (2000). In this study, the data are transformed to the T62 Gaussian grids to which all the AMIP2 models used in the second part of the study have been regridded (Zhang et al. 2002).
Figure 1 is the schematic diagram showing the key feedback loops relevant to this study. It is well known that tropical SST forcing plays a dominant role in climate variations in the Australian region and that the chief contributor to the Australian climate variability and predictability is the ENSO phenomenon (e.g., McBride and Nicholls 1983; Nicholls 1989; Jones 1999; Jones and Trewin 2000; Drosdowsky and Chambers 2001). The Southern Oscillation index (SOI) has been used in the operational statistical seasonal outlooks over this region (Hammer et al. 2000). Thus, to keep the analysis simple, in this study we take the correlations between SOI and Australian rainfall and surface temperature as the first-order approximation of the influence of large-scale tropical SST forcing on the Australian climate system, while bearing in mind that there may be other SST forcing effects that cannot be fully represented by the SOI. For instance, in recent years, the impact of the Indian Ocean on the Australian climate has been investigated by a large number of studies (e.g., Drosdowsky 1993; Frederiksen and Balgovind 1994; Saji et al. 1999; IOCIP 2000), although there is still intense scientific debate on whether the Indian Ocean signal is part of the ENSO process or if it operates independently (e.g., Chambers et al. 1999; Nicholls and Drosdowsky 2001).
The following approach is therefore used to try, for the most part, to exclude impacts of large-scale SST forcing from the observed climate variations and then explore the potential contributions from land surface processes. For each of the 12 months, we
remove the seasonal cycle and any linear trend in the BoM observational precipitation (P), Tmax, and Tmin data;
- establish linear regressions of P, Tmax, and Tmin to the observed SOI:
- subtract the effect of SOI components from the observed time series of P, Tmax, and Tmin:
and, finally, calculate the autocorrelations of Tmax′ and Tmin′ and their lag correlations with P′. We will refer to P′, Tmax′, and Tmin′ as “residuals.”
Note that the lack of coherent and long-period observations of land surface properties such as soil moisture and deep soil temperature, as well as surface latent and sensible heat fluxes, has impeded any direct observational studies of the impacts of land surface processes on regional and global climate at different time scales. It is noted that in a recent climate model–based study (Schlosser and Milly 2002), the soil moisture predictability time scale was shown to be about 1 month in the Australian region. Preliminary results from recent efforts in directly measuring soil water content over a number of locations over the Australian region do reveal a slow-varying process of deep soil water content with a time scale larger than a month or two (Richter et al. 2004). Therefore, the land surface can possibly affect the climate variations in the Australian region. In this study, we try to use the available observational data to assess the underlying relationships between large-scale SST forcing and slow-varying land surface processes in affecting Australian climate.
It should be pointed out that even though the term “predictability,” which has been widely used in the climate community and generally refers to what can skillfully be predicted for a given time scale, there is not yet a definition that is generally accepted and clearly defined. Therefore, the measures of the so-called predictability can be different. In this study, persistence from auto- and lag correlations is used as a simple statistical measure of climate variability and predictability.
In the observational analysis of this study, we analyze the lagged correlations between P′ and Tmax′ and Tmin′ to assess if proceeding precipitation could provide extra predictability of surface temperature variations. Such lagged correlations, by and large, reflect the impacts due to the soil moisture memory of precipitation forcing. This is because in the snow-free region, the only mechanism through which the signal of precipitation forcing can be retained in the climate system at a monthly or longer time scale is by the soil storage of precipitation water.
Following the observational analysis, in the second part of the study, 17-yr (1979–95) AMIP2 simulations from 16 AGCMs are examined to assess whether land surface parameterizations can affect model climate variability. As the purpose of this study is not to identify the best model(s) but rather to explore the potential impacts of land surface modeling, the models will remain anonymous in the rest of the analysis, and Table 1 summarizes some key aspects of their configurations that are directly related to the focus of this study. There is a great variety of land surface complexities in these models, ranging from simple Manabe-type bucket models with no canopy-related processes (models O and P), to intermediate bucket plus simple canopy stomatal resistance schemes (model A), to schemes with fully parameterized canopy processes and incorporated carbon cycles. The soil hydrology component varies from a simple one-layer bucket scheme to two-layer force–restore approaches to multilayer soil schemes. Such diversity provides a good opportunity to assess the influence of land surface parameterization on climate simulations.
Since soil moisture is directly available as part of the model outputs, correlations between model-simulated total soil moisture and model-simulated surface climate variations are calculated in the second part of the study. The model-reported total soil moisture (named the total soil water content mrso in the AMIP2 model standard output name list) is used in the analysis. Bearing in mind that soil moisture is a highly conceptual variable in the models, for the purpose of this analysis we are only interested in analyzing the time scale of its variations, not the magnitudes.
3. Observational analysis
Figure 2 shows the 1-month-lag autocorrelations of Tmax from the BoM observations after removing seasonal cycle and linear trend (if any) embedded in the data. For instance, the Tmax autocorrelation in July is the correlation between the June and July Tmax anomalies. In addition, the approach of Livezey and Chen (1983) is used to vigorously test the statistical significance of the correlations calculated from the observational datasets. For each month and at each grid, 500 Monte Carlo random time series are generated, and the Tmax correlation coefficients with the 500 random series are calculated. Following that, the local statistical significance level at each point is derived from the probability distribution of the 500 Monte Carlo random correlation coefficients. Furthermore, the statistical field significance of the correlation spatial pattern calculated from observational data is derived from 500 Monte Carlo correlation calculations following the approach of Livezey and Chen (1983). Based on the results (not shown) for each month, if the total number of grid points passing the local significant test is greater than 28, then the statistical significance of the calculated correlation field from observations is above the 95% confidence level.
Clearly, results show that the persistence of the Tmax variation is higher in the northern and eastern parts of the continent in most of the seasons. Except for the austral summer months, the persistence of Tmax anomalies in the southwestern region is also evident during the April–October period. The spatial pattern of the Tmax autocorrelation is consistent with the five principal temperature variation components identified by Jones (1999).
To what extent is such surface temperature persistence related to slow-varying large-scale tropical SST forcing? This subject has been intensively examined by a large number of studies (e.g., Jones 1999; Jones and Trewin 2000) and is only briefly presented here as 1-month-lag correlations between Tmax and SOI (Fig. 3). In the austral late spring and summer months (November–February), significant correlations occur in the northern and eastern Australian region, which is consistent with the results from Jones (1999). In the austral winter season, the impacts of SOI are mainly in the eastern continent, with significant lag correlations seen in June and August. In addition, it shows large impacts of SOI on Tmax variations in March and April. Nevertheless, the lag correlations between SOI and Tmax shown in Fig. 3 are much weaker and cover much smaller regions than the autocorrelations seen in Fig. 2.
After removing the SOI component from the observed Tmax time series, Fig. 4 exhibits the partial autocorrelation of Tmax′. It offers broadly similar spatial patterns, as seen in Fig. 2. This means that besides the SOI effects, there are other factors that contribute to the persistence of Tmax variations. The observed persistence of the Tmax variations in Fig. 2 is likely the product of (i) the large-scale slow-varying SST forcing as represented by SOI, (ii) some slow-varying SST forcing that cannot be fully represented by SOI (e.g., Indian Ocean), or (iii) other slow-varying atmospheric boundary conditions, such as the land surface. Indeed, as revealed in the studies of Power et al. (1998), Jones (1999), and Jones and Trewin (2000), changes of surface temperature are closely linked to the changes of surface energy partition, including surface radiative forcing and evaporation (e.g., Zhang et al. 1996). It is well known that soil moisture variation is the key factor in determining the partition of surface radiative energy into surface sensible and latent heat fluxes (e.g., Shao and Henderson-Sellers 1996; Zhang et al. 2001). Therefore, in theory, the soil moisture process can affect the variability and predictability of surface temperature.
After removing the SOI component in the observed time series of Tmax, Tmin, and precipitation, Fig. 5 shows the 1-month-lag correlation between P′ and Tmax′. For instance, results in July are the lag correlations of P′ in June to Tmax′ in July. As precipitation itself is a highly stochastic and instantaneous variable, such 1-month-lag correlations in Fig. 5 can largely be attributed to the soil storage of stochastic precipitation forcing and to the slow release of such accumulated effects, which, in turn, affects the Tmax variations in the forthcoming month(s). The correspondence between Figs. 5 and 4 is closely examined next to investigate whether the lag correlation between Tmax′ and P′ can explain the 1-month autocorrelation of Tmax′.
First, Fig. 5 shows remarkably similar patterns to those in Fig. 4 in most months. From the austral late spring to early autumn (November–April), high autocorrelations of Tmax′ in the northern and eastern part of the continent (Fig. 4) correspond well to the locations of high partial correlations of Tmax′ and P′ in Fig. 5. When the autocorrelations of Tmax′ are higher in February and April (Fig. 4), the partial correlations of Tmax′ and P′ are also more evident in Fig. 5. Nevertheless, the coherence of the correlation patterns between Figs. 4 and 5 is relatively weaker in the austral winter months. A chief feature is that the Tmax′ variations in the southwest Australian region cannot be adequately explained by the correlations between Tmax′ and P′ in Fig. 5. The failure of using the lag correlations of P′ and Tmax′ in explaining the Tmax′ autocorrelations over the southwest part of the continent in the winter season can be largely explained by the results from recent studies (e.g., Drosdowsky 1993; Jones 1998). They show that during this period, the Indian Ocean plays an important role in affecting the climate variations in the region. The successful separation of different processes in explaining Tmax′ variations over this region in the current analysis further consolidates the robustness of such a simple analytical approach used here. In addition, in the austral winter months, the partial correlations between Tmax′ and P′ are weaker over the northern part of the continent in Fig. 5 than in Fig. 4. This is because in the Australian winter season, heavy rainfall is shifted to the southern part of the continent, leading to a weak contribution of soil moisture memory from rainfall anomalies to the Tmax variation in the eastern coastal regions.
Similarly, the variations of monthly averaged daily Tmin are also analyzed. Figure 6 shows the 1-month partial autocorrelation of Tmin′ (i.e., after removing the SOI component in Tmin variations). Overall, results clearly show some regional characteristics of the persistence of Tmin anomalies. Such persistence is higher in the northern and eastern coastal regions in most months, while only significant autocorrelations are seen in the summer and autumn months (December–May) in the southern part of the continent. There are also notable signals in the western and southwestern regions, particularly in January, May, and September.
The proportion of Tmin persistence, which can be linked to the impacts of soil moisture variations related to previous precipitation forcing, is much weaker than that seen in the analysis of Tmax. Figure 7 illustrates the lag correlations between Tmin′ and P′ for each of the 12 months. Both the magnitudes and the spatial scales of the correlations between these two variables are smaller than the results shown in Fig. 5. Indeed, most numerical studies show that the strongest signal of the sensitivity of surface energy partition to soil moisture conditions occurs in the daytime (e.g., Bonan 2001; Zhang et al. 2001). Hence, anomalous land surface conditions due to rainfall variations have weaker impacts on Tmin, which occurs during night, than on Tmax, which occurs during the daytime. In addition, some other factors such as cloud coverage become important in affecting the Tmin′ variations (e.g., Power et al. 1998).
Although one cannot draw a firm conclusion on the impacts of soil moisture on surface temperature variations without using observed soil moisture datasets, the analysis here does suggest that the land surface, which presents another important component of slow-varying atmospheric boundary conditions, is likely to play an important role in affecting surface temperature variations and predictability. It needs to be pointed out that even though the observational study here is only focused on presenting 1-month-lag or autocorrelations, such correlation signals can still be significant on seasonal time scales over some regions (results not shown).
4. Results from AMIP2 model intercomparison
In this section, the role of land surface modeling in affecting current AGCM simulations of climate variability will be explored, complementing the observational analysis in section 3. GCM modeling studies of soil moisture impacts on the Australian surface climate have been conducted in a number of studies (e.g., Simmonds and Hope 1998; Power et al. 1998; Timbal et al. 2002; Zhang and Frederiksen 2003). Here, the key objective is to assess how different the soil moisture memories are in current AGCMs and whether such differences affect the climate variability and predictability simulated in the models themselves. This is achieved by an analysis of a suite of AMIP2 model results, described in Table 1.
A series of lag-correlation analysis of 16 AMIP2 model results is conducted in this part of the study. The seasonal cycle and any linear trends are removed before calculating the lag correlations between two variables. With 204 samples from the 17-yr model integrations used in the calculation, correlation coefficients exceeding about 0.14 are statistically significant at the 95% confidence level.
Prior to the analysis of the impacts of soil moisture variations on model simulations, the different behavior of soil moisture variations is shown in Fig. 8. As an example, the monthly soil moisture variations from three models (D, J, and O) at the model grid point near 25°S and 135°E (central Australia) are presented here. These three models have been selected to encompass the range of simulated characteristics of interest in this study. After removing the downward trend from its original data (Zhang et al. 2002), model D exhibits a slow variation of soil moisture anomalies with a time scale of several months for soil moisture depletion. Model J shows moderate variations of soil moisture conditions at this location, while model O that has a bucket-type scheme exhibits rapid responses to the meteorological forcing, with soil moisture anomalies responding rapidly to rainfall anomalies and evaporative demand (not shown). Thus, for this model, an anomalous wet condition decays rapidly over a short time scale.
With such remarkably different characteristics in model-simulated soil moisture variations, it is not surprising to see large different features in the 3-month autocorrelations of soil moisture anomalies across all the AMIP2 models (Fig. 9). Clearly, models O and P, which use simple bucket-type land surface schemes, have the lowest autocorrelation at this time scale. Introducing a consistent canopy resistance in its bucket-type soil hydrological scheme, model A shows an increased persistence of soil moisture anomalies. This agrees well with the results from land surface modeling complexity studies (e.g., Zhang et al. 2001). Adding an extra resistance in the pathway of soil water release decreases the surface potential evaporation rate for a given set of meteorological forcings and therefore reduces the decay rate of soil moisture anomalies in the model simulations (Delworth and Manabe 1988). Such impacts from surface evaporation on soil moisture memory are more evident in the dry regions such as Australia (Koster and Suarez 2001). Some models, such as models D, G, H, and M, show quite large soil moisture autocorrelations even at a 3-month time scale. This behavior can be explained by the slow-varying processes seen in Fig. 8.
Note that model D, which has a nonbucket soil hydrological model but uses a constant volumetric available water-holding capacity of 152 mm, has a similar value as used in the bucket schemes of models O and P (150 mm). Nevertheless, the behavior of its soil moisture variations is remarkably different from models O and P. Similarly, models E and M, using a very similar surface scheme, have the same water-holding capacity but behave differently in their soil moisture variations. Even though the values of field capacity used in all 16 models are not available for a direct intercomparison, results here tend to suggest that field capacity does not chiefly determine the differences in the model-simulated soil moisture memory. In addition, further examination of the relationship between the model performance and the number of soil layers used in the models tends to suggest that when the models use nonbucket schemes, the persistence of soil moisture variations does not depend on the number of soil layers used in the model parameterization. A number of studies, such as Koster and Milly (1997) and Gedney et al. (2000), have proposed approaches in characterizing such differences in a wide range of land surface processes in a linearized framework. Performing such studies (e.g., Koster and Suarez 2001) can be valuable in explaining the differences seen here. Nevertheless, for the context of this study, we only focus on exploring whether such differences affect the model variability and predictability.
In addition to the description of soil moisture retention in Fig. 9, the model-simulated soil moisture memory of meteorological forcing is shown in Fig. 10. It illustrates the lag correlations between precipitation and soil moisture, with soil moisture lagging precipitation by 3 months. Again, models O and P with a simple bucket soil hydrological scheme reveal a very weak signature of precipitation forcing from the previous 3-month period. This is due to the fact that the soil storage of rainfall is rapidly evaporated in the bucket-type scheme as discussed before. By adding a consistent canopy resistance component, the rapid loss of soil water from surface evaporation can be prevented, and model A tends to agree better with models using nonbucket schemes than with models O and P. Of the 16 models, some of them (e.g., C, D, G, and M) show a significant signature of previous precipitation forcing even at a seasonal time scale.
With such differences seen in the model simulations of soil moisture variation and soil moisture memory of meteorological forcing, the key question now is whether the model-simulated climate variability and predictability are affected by the soil moisture variations. Koster et al. (2002) showed that the strength of coupling between the land surface and the atmosphere varied significantly among a number of models. Such coupling strength largely determines how the predictability of soil moisture can be translated into the predictability of the atmosphere and climate. Schlosser and Milly (2002) also reported their model-based study of climate predictability associated with soil moisture predictability. The available data from AMIP2 experiments do not allow us to conduct studies similar to those of Koster et al. (2002) and Schlosser and Milly (2002) to quantify the coupling strengths of the 16 models and the translation from soil moisture predictability to climate predictability. However, intercomparison of the characteristics in terms of model-simulated predictability and their simulated soil moisture variations from a large number of 17-yr model integrations can still be of value in studying the impacts of soil moisture on climate variation and predictability.
Figure 11 shows the 3-month-lag correlations between soil moisture and surface evaporation over the Australian region, with surface evaporation lagging soil moisture. Again, there are remarkable differences among AMIP2 models in terms of the time scale over which soil moisture anomalies affect surface energy partitions. Lag correlations between these two variables are, overall, positive over the continent, except for model C, which has negative correlations over part of the region. Among the 16 models, it is noted that models O and P, together with models C, D, M, and N, exhibit the lowest overall lag correlations. As shown in Fig. 9, such results from models O and P can be directly attributed to the short retention period of soil moisture anomalies in the two models with a bucket-type soil hydrological model. Results from models C, D, M, and N, which show a large soil moisture memory with high soil moisture autocorrelations but weak impacts on surface evaporation calculation, suggest a weak control of soil moisture on surface evaporation and therefore a weak coupling between land surface and the atmosphere in these models.
Due to the dominant role of surface evaporation in determining surface temperature variations (Power et al. 1998; Jones and Trewin 2000; Zhang et al. 2001), it is expected that surface temperature variations will also be influenced. Figure 12 exhibits different features in the 3-month-lag correlations between soil moisture anomalies and monthly Tmax anomalies 3 months ahead. It demonstrates that with a 3-month lag, anomalous soil moisture conditions are useful in forecasting Tmax anomalies in a number of models, particularly over the eastern part of the continent. More significant correlations are seen over larger regions in the results from 1-month- and 2-month-lag correlations (not shown). Among the 16 models, models A, B, F, G, J, and K have the largest area of significant correlations and, again, models O and P with a bucket-type land surface scheme, as well as C, D, H, M, and N, show the lowest lag correlations. Consistent with the results seen in Fig. 11, models C, D, M, and N do not exhibit significant impacts of its soil moisture variations on its surface temperature variations, suggesting that other factors such as the strength of coupling between the land surface and the atmosphere (e.g., Koster et al. 2002; Schlosser and Milly 2002) are also important in determining the translation of soil moisture memory into climate predictability. Similar but slightly weaker correlations are seen in the model Tmin and in the model daily averaged surface temperature results.
Figure 13 further shows areally averaged autocorrelations of soil moisture and the lag correlations between soil moisture, surface evaporation, and monthly averaged daily surface temperature over the Australian region. Figure 13a demonstrates remarkably different characteristics of soil moisture variations simulated in these models, with models O and P showing different features as compared to other AMIP2 models. These two models have the fastest decay rates of soil moisture autocorrelations, underlining the short retention time of soil moisture anomalies and subsequently other surface climate anomalies. Figure 13b shows large differences in model-lag correlations between soil moisture and evaporation. Some AMIP2 models show averaged correlations exceeding 0.15 up to a 5- and 6-month lag, while others exhibit low correlations within a month. These differences are even more pronounced in the correlations between soil moisture and surface temperature (Fig. 13c).
To further explore some possible linkages between soil moisture memory and the model-simulated surface evaporation and temperature variations, Fig. 14 is focused on examining the decay rates of the autocorrelations shown in Fig. 13. The soil moisture autocorrelation decay rate for the first month is calculated as its zero-lag autocorrelations minus the 1-month-lag autocorrelations (r(t+0) − r(t+1)) in Fig. 13a. Figure 14a suggests that such a decay rate is highly correlated to the instantaneous correlations of surface evaporation and soil moisture (r(t+0)). That means, if surface evaporation in one model is heavily determined by soil moisture, then this model has a larger declining rate of soil moisture persistence. This is not only true for the two models (O and P) with a bucket scheme. Such a relationship also holds for other models with different complexity in land surface modeling. Results are in agreement with the analyses of Delworth and Manabe (1988) and recent results from Koster and Suarez (2001), which underlined that soil moisture's control on surface evaporation significantly affects soil moisture memory in climate models, with the evaporation effect more dominant in the dry regions such as Australia (Koster and Suarez 2001).
In addition, when models show a larger decay rate of soil moisture persistence, they also show a more rapid decline of the influence of soil moisture on surface evaporation variations. This is seen in Fig. 14b, which shows the soil moisture 2-month autocorrelations decay rate (r(t+0) − r(t+2)) and the 2-month decay rate of surface evaporation lag correlations with soil moisture. The highly significant correlations between these two decay rates reiterate the impacts of soil moisture memory on surface energy partitions at longer time scales. Such a relationship is also evident for the results of surface temperature persistence (Fig. 14c), with models having a slower decaying of soil moisture anomalies also showing a stronger influence of soil moisture on surface temperature variations and predictability at a longer time scale.
Combining results from the AMIP2 model intercomparison, it is possible to draw two complementary conclusions: (i) in models with a simple one-layer bucket land surface scheme, the retention of soil moisture anomalies is much shorter than others in which compounding contributions from detailed physical processes such as imposing extra resistance in the pathway surface evaporation (Zhang et al. 2001), including soil water vertical distribution and root extraction processes, and effectively imposing different time scales of soil water variations at different soil layers (e.g., Viterbo and Beljaars 1995; Noilhan and Mahfouf 1996), result in a longer soil moisture memory; and (ii) besides the contribution from soil moisture, predictability in surface temperature variations in GCM models is also affected by other factors, such as the strength of the coupling between the land surface and the atmosphere in climate models (Koster et al. 2002) and the contribution of soil moisture predictability to climate predictability (Schlosser and Milly 2002). For instance, models C, D, M, and N show high autocorrelation of soil moisture anomalies in Fig. 9, but they have consistently shown lower lag correlations in Figs. 11 and 12. Studies similar to Koster et al. (2002) and Schlosser and Milly (2002) are planned to further understand the role of soil moisture variations on the Australian climate variability in future Bureau of Meteorology Research Centre (BMRC) model simulations (e.g., Koster et al. 2003).
5. Discussion and conclusions
This study has been focused on assessing the influence of land surface processes such as soil moisture on Australian surface temperature variations. It augments efforts in testing the hypothesis that soil storage of water and energy can act as another slow-varying boundary process for the atmosphere and can therefore affect the atmospheric mean state at monthly or longer time scales, modulate the atmospheric responses to SST forcing, and increase the predictability of climate systems (e.g., Chen and Kumar 2002). It has comprised both observational analysis and AGCM modeling studies.
A large number of previous studies have concentrated on examining the influence of ENSO and other SST forcing on Australian surface temperature variations and exploring the possible mechanism in explaining such influences (e.g., Jones 1998, 1999; Power et al. 1998; Simmonds and Hope 1998). In this study, using the Australian Bureau of Meteorology observational datasets of monthly averaged daily surface temperature maximum (Tmax), minimum (Tmin), and monthly precipitation, simple partial and lag correlations have been calculated to try to isolate the contributions from SST forcing and those related to land surface processes in affecting the observed surface Tmax and Tmin variations. Taking the Southern Oscillation Index (SOI) as the first-order estimation of large-scale SST forcing, such impacts are excluded from the observed time series by linear regressions. High autocorrelations of Tmax residuals (Tmax′) have demonstrated that besides the influence of slow-varying large-scale SST forcing, which can lead to the persistence of Tmax anomalies, a large part of the observed persistence of Tmax is from processes not represented by SOI. Some of the processes are from the Indian Ocean, particularly in the winter season in the southwest Australian region. Nevertheless, it is evident from this study that soil moisture is one of the chief candidates. This has been illustrated by significant lag correlations between Tmax′ and precipitation residual (P′). As the only mechanism through which the signal of precipitation forcing can remain in the climate system at a monthly or longer time scale is by the soil storage of precipitation water, the lagged correlation of P′ to Tmax′ has largely been attributed to the impacts from the soil moisture memory of precipitation forcing. A similar but much weaker relationship has been seen in the observed Tmin variations, as there are other important processes (e.g., cloud coverage, etc.) in determining the Tmin variations (Power et al. 1998; Jones 1999; Jones and Trewin 2000; Bonan 2001).
In the second part of the study, results from 16 AMIP2 model simulations have been analyzed to explore the impacts of land surface modeling on the model-simulated surface climate variability and predictability in the Australian region. The modeling analysis has been focused on whether soil moisture memories have different features among the models and whether such differences affect the model-simulated surface climate variations. Exploring the causes of different soil moisture memories in the climate models has not been the primary focus of this study, even though it is an import scientific issue warranting further carefully designed studies (e.g., Koster and Suarez 1996, 2001; Schlosser and Milly 2002).
Lag-correlation analysis has revealed that the characteristics of climatic “memory” from land surface processes (e.g., soil moisture) differ among the 16 models. Models with simple bucket-type schemes tend to show rapid decay rates in soil moisture anomalies, leading to much weaker lag correlations between soil moisture conditions and surface climate anomalies. On the other hand, for most of the nonbucket models, introducing extra physical processes in the model parameterizations, such as including extra resistance in the pathway surface evaporation (Zhang et al. 2001), considering soil water vertical movement through soil water diffusion and root extraction processes, and effectively imposing different time scales of soil water variations at different soil layers (e.g., Viterbo and Beljaars 1995; Noilhan and Mahfouf 1996), all contribute to slower soil moisture variations and longer soil moisture memories, and consequently produce larger influences on the model integrations over a longer time period.
Analyzing areally averaged results has clearly demonstrated the close linkage between model soil moisture parameterization and the model-simulated impacts of soil moisture on surface climate variations. Comparison of a number of bucket and nonbucket models that have the same or very similar field capacity tends to suggest that the size of soil water reservoir (i.e., water-holding capacity) used in the models cannot largely explain the differences in the model-simulated soil moisture memories. Furthermore, for models using nonbucket schemes, the number of soil layers in soil hydrological modeling does not affect the results much. In contrast, it is the functionality between soil moisture and surface evaporation that significantly influences the model-simulated soil moisture memory; models having higher simultaneous correlations between soil moisture and surface evaporation show a more rapid decay rate of soil moisture memory.
Furthermore, among the 16 models, the extent to which soil moisture memory is translated to climate variability and predictability is different. A number of models have higher soil moisture memory but weaker lag correlations between soil moisture and surface evaporation and surface temperature. Results suggest that there are other factors that are also important in affecting model results, such as the different control of soil moisture conditions on model surface evaporation calculations, and different coupling strengths between the land surface and the atmosphere due to model boundary layer and convection parameterizations. The model intercomparison results prompt the need for further research in understanding the mechanisms involved, such as the strength of coupling between the land surface and the atmosphere in current climate models and the degree of translation of soil moisture predictability to the predictability of atmospheric variables (e.g., Koster et al. 2002; Schlosser and Milly 2002).
Both the results from observational analysis and AMIP2 model simulations have demonstrated that land surface modeling, in particular, soil hydrological modeling, is an important area that warrants more efforts in order to improve current climate models in seasonal- and longer-scale climate forecasts and in the study of climate variability and predictability. It is also important to take into account the land surface conditions in statistical seasonal forecasting schemes. Indeed, Zheng and Renwick (2003) have shown the improvement of their statistical forecasting scheme in forecasting the New Zealand surface temperature anomalies when the rainfall anomalies in the previous season(s) are included.
Acknowledgments
This study forms part of the AMIP2 diagnostic subproject 12. The author deeply appreciates the discussions with the team members (Prof. A. Henderson-Sellers, Drs. T. Phillips, P. Irannejad, S. Sharmeen, and others). Discussions with Drs. N. Nicholls and C. Frederiksen and comments from Dr. S. Power and Prof. A. Henderson-Sellers on an early version of the manuscript are acknowledged. Dr. D. Jones is thanked for providing the Bureau of Meteorology's observational data. Efforts from the 16 modeling groups and PCMDI/LLNL in making the model simulations available for this analysis are also acknowledged. Comments and suggestions from Dr. M. A. Brunke and another anonymous reviewer are appreciated.
REFERENCES
Bonan, G. B., 2001: Observational evidence for reduction of daily maximum temperature by croplands in the midwest United States. J. Climate, 14 , 2430–2442.
Chambers, D. P., B. D. Tapley, and R. H. Stewart, 1999: Anomalous warming in the Indian Ocean coincident with El Niño. J. Geophys. Res, 104 , 10523–10533.
Chen, J., and P. Kumar, 2002: Role of terrestrial hydrologic memory in modulating ENSO impacts in North America. J. Climate, 15 , 3569–3585.
Delworth, T. L., and S. Manabe, 1988: The influence of potential evaporation on the variability of simulated soil wetness and climate. J. Climate, 1 , 523–547.
Desborough, C. E., A. J. Pitman, and P. Irannejad, 1996: Analysis of the relationship between bare soil evaporation and soil moisture simulated by 13 land surface schemes for a simple non-vegetated site. Global Planet. Change, 13 , 47–56.
Drosdowsky, W., 1993: Potential predictability of winter rainfall over southern and eastern Australia using Indian Ocean sea-surface temperature anomalies. Aust. Meteor. Mag, 42 , 1–6.
Drosdowsky, W., and L. Chambers, 2001: Near-global sea surface temperature anomalies as predictors of Australian seasonal rainfall. J. Climate, 14 , 1677–1687.
Frederiksen, C. S., and R. C. Balgovind, 1994: The influence of the Indian Ocean/Indonesian SST gradient on the Australian winter rainfall and circulation in an atmospheric GCM. Quart. J. Roy. Meteor. Soc, 120 , 923–952.
Gedney, N., P. M. Cox, H. Douville, J. Polcher, and P. J. Valdes, 2000: Characterizing GCM land surface schemes to understand their responses to climate change. J. Climate, 13 , 3066–3079.
Hammer, G. L., N. Nicholls, and C. Mitchell, Eds.,. 2000: Applications of Seasonal Climate Forecasting in Agricultural and Natural Ecosystems: The Australian Experience. Kluwer, 469 pp.
IOCIP, 2000: Towards understanding climate variability in southwestern Australia: Research reports on the first phase of the Indian Ocean Climate Initiative. Indian Ocean Climate Initiative, Perth, Australia, 237 pp.
Jones, D. A., 1998: The prediction of Australian land surface temperatures using near global sea surface temperature patterns. BMRC Research Rep. 70, Bureau of Meteorology, Melbourne, Australia, 44 pp.
Jones, D. A., 1999: Characteristics of Australian land surface temperature variability. Theor. Appl. Climatol, 61 , 11–31.
Jones, D. A., and B. C. Trewin, 2000: On the relationships between the El Niño–Southern Oscillation and Australian land surface temperature. Int. J. Climatol, 20 , 697–719.
Koster, R. D., and M. J. Suarez, 1996: The influence of land surface moisture retention on precipitation statistics. J. Climate, 9 , 2551–2567.
Koster, R. D., and P. C. D. Milly, 1997: The interplay between transpiration and runoff formulations in land surface schemes used with atmospheric models. J. Climate, 10 , 1578–1591.
Koster, R. D., and M. J. Suarez, 2001: Soil moisture memory in climate models. J. Hydrometeor, 2 , 558–570.
Koster, R. D., P. A. Dirmeyer, A. N. Hahmann, R. Ijpelaar, L. Tyahla, P. Cox, and M. J. Suarez, 2002: Comparing the degree of land–atmosphere interaction in four atmospheric general circulation models. J. Hydrometeor, 3 , 363–375.
Koster, R. D., Z. Guo, and P. A. Dirmeyer, 2003: First results from GLACE. GEWEX News, Vol. 13, No. 4, International GEWEX Project Office, 2–3.
Livezey, R. E., and W. Y. Chen, 1983: Statistical field significance and its determination by Monte Carlo techniques. Mon. Wea. Rev, 111 , 46–59.
Manabe, S., 1969: Climate and the ocean circulation. I: The atmospheric circulation and the hydrology of the earth's surface. Mon. Wea. Rev, 97 , 739–774.
McBride, J. L., and N. Nicholls, 1983: Seasonal relationships between Australian rainfall and the Southern Oscillation. Mon. Wea. Rev, 111 , 1998–2004.
Nicholls, N., 1989: Sea surface temperatures and Australian winter rainfall. J. Climate, 2 , 965–973.
Nicholls, N., and W. Drosdowsky, 2001: Is there an equatorial Indian Ocean SST dipole independent of the El Niño–Southern Oscillation? Preprints, Symp. on Climate Variability, the Oceans, and Societal Impacts, Albuquerque, NM, Amer. Meteor. Soc., 17–18.
Noilhan, J., and J-F. Mahfouf, 1996: The ISBA land surface parameterisation scheme. Global Planet. Change, 13 , 145–159.
Phillips, T. J., and Coauthors, 2000: Validation of land-surface processes in AMIP models: A pilot study. PCMDI Rep. 63, PCMDI, Livermore, CA, 22 pp.
Power, S., F. Tseitkin, S. Torok, B. Lavery, R. Dahni, and B. McAvaney, 1998: Australian temperature, Australian rainfall and the Southern Oscillation, 1910–1992: Coherent variability and recent changes. Aust. Meteor. Mag, 47 , 85–101.
Richter, H., A. W. Western, and F. H. S. Chiew, 2004: The effect of soil and vegetation parameters in the ECMWF land surface scheme. J. Hydrometeor., in press.
Saji, N. H., B. N. Goswami, P. N. Vinayachandran, and T. Yamagata, 1999: A dipole mode in the tropical Indian Ocean. Nature, 401 , 360–363.
Schlosser, C. A., and P. C. D. Milly, 2002: A model-based investigation of soil moisture predictability and associated climate predictability. J. Hydrometeor, 3 , 483–501.
Shao, Y., and A. Henderson-Sellers, 1996: Validation of soil moisture simulation in land surface parameterisation schemes with HAPEX data. Global Planet. Change, 13 , 11–46.
Simmonds, I., and P. Hope, 1998: Seasonal and regional responses to changes in Australian soil moisture conditions. Int. J. Climatol, 10 , 1105–1139.
Timbal, B., S. Power, R. Colman, J. Viviand, and S. Lirola, 2002: Does soil moisture influence climate variability and predictability over Australia? J. Climate, 15 , 1230–1238.
Viterbo, P., and A. M. Beljaars, 1995: An improved land surface parameterization scheme in the ECMWF model and its validation. J. Climate, 8 , 2716–2748.
Zhang, H., and C. S. Frederiksen, 2003: Local and nonlocal impacts of soil moisture initialization on AGCM seasonal forecasts: A model sensitivity study. J. Climate, 16 , 2117–2137.
Zhang, H., A. Henderson-Sellers, and K. McGuffie, 1996: Impacts of tropical deforestation. Part I: Process analysis of local climate change. J. Climate, 9 , 1497–1517.
Zhang, H., A. Henderson-Sellers, A. J. Pitman, J. L. McGregor, C. E. Desborough, and J. Katzfey, 2001: Limited-area model sensitivity to the complexity of representation of the land surface energy balance. J. Climate, 14 , 3965–3986.
Zhang, H., A. Henderson-Sellers, P. Irannejad, S. Sharmeen, T. Phillips, and K. McGuffie, 2002: Land-surface modelling and climate simulations: Results over the Australian region from sixteen AMIP2 models. Bureau of Meteorology Research Centre Rep. 89, 51 pp.
Zheng, X., and J. A. Renwick, 2003: A regression-based scheme for seasonal forecasting of New Zealand temperature. J. Climate, 16 , 1843–1853.
A schematic diagram showing the physical processes that were studied in the observational analysis
Citation: Journal of Climate 17, 21; 10.1175/JCLI3141.1
One-month autocorrelations of monthly averaged daily Tmax using the BoM's observational detrended data for the period of 1950– 99. For instance, results in Jul are the correlation between observed Tmax anomalies in Jun and Jul. Local correlation coefficients at the 95% significance level are shaded based on 500 Monte Carlo runs. The number in the title of each diagram is the total number of grid points passing the local significant test at the 95% confidence level. The correlation field is statistically significant above the 95% confidence level if there are more than 28 points passing the local significant test, following the approach of Livezey and Chen (1983)
Citation: Journal of Climate 17, 21; 10.1175/JCLI3141.1
As in Fig. 2, but for 1-month-lag correlations between monthly averaged daily Tmax and observed SOI. For instance, results in Jul are the correlation between observed Tmax anomalies in Jul with observed SOI in Jun
Citation: Journal of Climate 17, 21; 10.1175/JCLI3141.1
Partial autocorrelations of monthly averaged daily Tmax after removing the SOI component in the observed time series using linear regression as described in the text, that is Tmax′. Results show the persistence of Tmax anomalies that are not due to SST forcing represented by SOI
Citation: Journal of Climate 17, 21; 10.1175/JCLI3141.1
Partial lag correlations of monthly averaged daily Tmax and precipitation after removing the SOI component in the observed time series using linear regression as described in the text, that is, Tmax′ and P′. For instance, results in Jul are the correlation between the Tmax anomaly residual in Jul and the observed precipitation anomaly residual in Jun. Results show the persistence of Tmax anomalies due to soil moisture responses to anomalous precipitation forcing
Citation: Journal of Climate 17, 21; 10.1175/JCLI3141.1
As in Fig. 4, but for monthly averaged daily Tmin. The local and field significance levels are derived from a separate set of Monte Carlo random calculations and the correlation field is statistically significant above a 95% confidence level if there are more than 28 points passing the local significant test
Citation: Journal of Climate 17, 21; 10.1175/JCLI3141.1
As in Fig. 5, but for monthly averaged daily Tmin
Citation: Journal of Climate 17, 21; 10.1175/JCLI3141.1
The variation of soil moisture simulated in three models over the location 25°S and 135°E. The linear trend in model D has been removed
Citation: Journal of Climate 17, 21; 10.1175/JCLI3141.1
Three-month autocorrelations of total soil moisture anomalies for the different models. The seasonal cycle and linear trend have been removed in the correlation calculations. The value of 0.14 corresponds roughly to the 95% confidence level
Citation: Journal of Climate 17, 21; 10.1175/JCLI3141.1
As in Fig. 9, but for the lag correlation between total soil moisture and precipitation forcing, with soil moisture lagging precipitation
Citation: Journal of Climate 17, 21; 10.1175/JCLI3141.1
As in Fig. 9, but for the lag correlations between surface evaporation and total soil moisture, with soil moisture leading evaporation
Citation: Journal of Climate 17, 21; 10.1175/JCLI3141.1
As in Fig. 9, but for the lag correlations between monthly averaged daily Tmax and total soil moisture, with soil moisture leading Tmax
Citation: Journal of Climate 17, 21; 10.1175/JCLI3141.1
(a) Areally averaged autocorrelations (×10) of total soil moisture over the Australian region, with a 0- to 12-month lag. The x axis is the lag time and the y axis represents the models in the same order as in Table 1. Correlations with magnitudes exceeding 0.15 are shaded. (b) as in (a), but for lag correlations between total soil moisture anomalies and surface evaporation anomalies, with soil moisture leading evaporation; (c) as in (a), but for soil moisture and daily averaged surface air temperature, with soil moisture leading surface air temperature
Citation: Journal of Climate 17, 21; 10.1175/JCLI3141.1
The relationship between (a) areally averaged instantaneous correlations (r(t+0)) of surface evaporation and soil moisture from 16 AMIP2 models and the soil moisture autocorrelation 1-month decay rate (r(t+0) − r(t+1)); (b) areally averaged soil moisture 2-month autocorrelation decay rate (r(t+0) − r(t+2)) and that of surface evaporation; and (c) soil moisture and surface temperature. Model codes are the same as in Table 1
Citation: Journal of Climate 17, 21; 10.1175/JCLI3141.1
Model codes and features of the 16 AMIP2 models analyzed in this report
