Drought Modeling Consistency and Discrepancies in Predicting Drought in the Future

Many models of soil moisture, drought indices, and precipitation-minus evaporation predict increased drought in the twenty-first century. Furthermore, precipitation, stream flow, and drought indices have demonstrated increased aridity since the 1950s over land. Still, there are major differences between observed data and model predictions. Sea surface temperature has been shown to influence land precipitation. However, coupled models have not reproduced recent regional precipitation changes in their predictions, which may be due to a lack of observed sea surface temperature data in these model stimulations. Dai (2012) demonstrates that the models reproduce the effects of El Niño-Southern Oscillation and the observed data on global mean aridity for 1923 through 2010. According to Dai, natural variations in tropical sea surface temperature not accounted for in the models cause the regional differences in observed and model-simulated aridity changes. Furthermore, he concludes that the observed global aridity changes through 2010 are consistent with model predictions, thus validating predictions of increasing severity and frequency of droughts over the next century. —Hilary Haskell
                  Dai, A. 2012. Increasing drought under global warming in observations and models. Nature Climate Change 3, 52–58.

                  Dai reconciles historical data and model projections of increasing aridity and drought in order to gain a more comprehensive understanding of global climate change’s effects on drought patterns. A variety of different drought indices quantify drought, yet yield different results, especially on smaller geographical scales. Dai uses precipitation, stream flow, and soil moisture fields to quantify meteorological, hydrologic, and agricultural drought, respectively. For historical soil-moisture data, the author used the Palmer Drought Severity Index (PDSI) Penman–Monteith (sc_PDSI_pm) equation to calculate the potential evapotranspiration of water out of the leaves of vegetation, given there is enough water for evapotranspiration to occur. The PDSI calculation is based on water-balance models of soil moisture forced with observed precipitation and temperature. This index is widely used in monitoring drought and paleoclimate reconstructions. The PDSI_PM is regarded as more accurate in its analysis than the PDSI_th calculation, because it includes temperature, precipitation, radiation, wind-speed, and humidity data in its calculation of potential evapotranspiration. Therefore, it provides a more comprehensive analysis of global warming scenarios. The sc_PDSI_pm calculation is used to determine the relative impacts of different factors (humidity, precipitation, etc.) affecting drought. By comparing the results of forced sc_PDSI_pm calculations that include or exclude various drought factors, each factor’s impact on drought can be determined. The author used two coupled climate model simulations based on future GHG emission scenarios: Coupled Model Intercomparison Project phase 3 (CMIP3), which was used in the Intergovernmental Panel on Climate Change Fourth Assessment Report and the new phase 5 (CMIP5).
                  For 1950–2010, Dai found that observed annual precipitation data and sc_PDSI_pm data had similar linear trends. Furthermore, these trends were similar to stream flow trends since 1948 for the world’s main river basins. The author noted some regional and quantitative differences between observed annual precipitation, sc_PDSI_pm, and stream flow data. However, these variations are still closely related. The similar linear trends indicate increased aridity over most of Africa, Southeast Asia, Eastern Australia, and Southern Europe, while there is increased wetness over the central U.S., Argentina, and northern high-latitude areas. This consistency in linear trends across independent measurements of precipitation and stream flow data indicate that these trends are accurate reflections of projected hydroclimatic variations in the future. Furthermore, this comparison also verifies that the sc_PDSI_pm is a reliable method for monitoring changes in aridity.
                  Warming trends since the 1980s have globally impacted the upward trend in drought areas, increasing these areas by about eight percent. This drying is attributable to warming patterns that cause increased evapotranspiration, especially over northern mid-to-high latitudes. However, increased aridity in Africa, Southeast Asia, Eastern Australia, and Southern Europe is mainly due to precipitation decreases. This decrease in precipitation is mainly caused by variations in sea surface temperature. Dai used The Hadley Centre Sea Ice and Sea Surface Temperature data set (SST) over varying periods of time to study the effects of (SST) on global drought. SST trends over the long-run are included in global warming predictions. However, over the course of a few decades, SST variations are absent in greenhouse gas (GHG) and aerosol-coupled model simulations. The absence of SST variations means that these natural variations are excluded from model predictions, albeit their timing and spatial patterns may be dependent on the initial conditions of the other factors included in the models. The lack of SST trends in current models makes the effects of SST changes irreproducible in models’ future predictions.
                  There are consistencies across soil moisture predictions for the coupled models and the sc_PDSI_pm calculations. Fourteen of the CMIP5 models analyzed in this study demonstrated decreases in soil moisture content in the top 10 cm layer of soil for most of the Americas, Europe, Southern Africa, most of the Middle East, Southeast Asia and Australia during the twenty-first century. The multi-model average predicts further decreases of 5 to 15% by 2080–2099. The sc_PDSI_PM using the same multi-model data reproduced the same increased aridity in soil moisture. However, the sc_PDSI_pm yields larger increases in wetness over central and eastern Asia and northern North America. In comparison to the CMIP5 model, the CMIP3 models, with some regional differences, also predicts the same increasing soil aridity for all seasons.
                  According to Dai, SSTs have large influences on land precipitation and drought. To demonstrate this influence, the author used maximum covariance analysis (MCA) of global SSTs at latitudes (40–60° N) and sc_PDSI_pm calculations at latitudes (60–75°N) based on observations and the CMIP models. The author’s goal was to reproduce the observed relationships between SST and sc_PDSI_pm by using MCA modes, in order to conclude whether the models could stimulate the recent drying trend. MCA analysis uses a standard singular value, obtained through decomposition of variables or inputs into a model from two separate fields. By decomposing this singular value, comparisons can be made between the two fields. For this study, Dai used MCA to compare difference in the SSTs and sc_PDSI_pm from observations and the CMIP models. Dai’s comparison of leading MCA modes excluded some of the unforced, irreproducible natural variation in the modes.
The first leading MCA modes (MCA1) based on observations and models represent global warming in relation to sea surface temperature. The temporal coefficient in this analysis is strongly correlated (r=0.97) between the observed global mean surface temperature and the SST MCA1 patterns, which are similar to the observed warming patterns over the oceans. MCA1 from the models show similar nonlinear global warming trends, with widespread warming over the oceans. The sc_PDSI-PM demonstrates short-term variability in these models. Considering the models, the observed mean global warming mode is evident in the GHG-forced CMIP simulations for both SST and sc_PDSI_PM, with a correlation of r=0.86 and a regression coefficient of .09566 between global means of sc_PDSI_pm anomalies from MCA1 observations. This finding suggests that inclusion of GHG emissions as a factor in global aridity modeling is a valid methodology based on its demonstrated consistency across global aridity change models and historical observations.
The second MCA modes (MCA2) calculated from both observations and models demonstrate similarity across spatial patterns. Both the observations and the models represent El Niño-Southern Oscillation (ENSO) through SST anomaly patterns. Furthermore, the time-based coefficient is highly correlated to the ENSO index. El Niño modes vary considerably across decades. Since 1999, ENSO observations for the central and eastern Pacific have become cooler than the time period from 1977 to 1998.
The spatial patterns for the sc_PDSI_pm were from the various much different than the observations, trend maps, and long term MCA1 (1950–2009). This discrepancy is attributable to large intermodal variations in the MCA1 from 1923–2010 caused by unforced natural variations and weak GHG-forced indicators of precipitation. These results indicate that individual models and observed global warming modes have large natural variation patterns unrelated to previous calculations that did not include GHG data. The MCA does not completely separate GHG-caused changes in precipitation and the changes in the sc_PDSI_pm from other natural variations included in the predictions. This finding is supported by the fact that until the year 2010, GHG -attributable variation was not very strong (4–6 % of total variance), compared with the natural variations forced in the models.
                  Dai suggests that the large regional differences between observations and individual model runs, the differences over West Africa, the USA, Brazil, Southern Africa, and eastern Australia, are the result of sample errors between actual realization of hydroclimatic patterns and natural variation not seen in the CMIP models. The major differences between the U.S. and Sahel stand out. The Sahel’s drying trend since 1950 is mainly due to decreased summer rainfall caused by warming in the South Atlantic Ocean relative to the North Atlantic, along with the warming of the Indian Ocean. This warming, coupled with dynamic vegetation feedback (not included in the CMIP models) leads to the discrepancy between the U.S. and the Sahel. CMIP3 models would predict the opposite warming pattern in the Atlantic Ocean and increased precipitation over the Sahel in the twenty-first century in the face of GHG caused global warming. CMIP5 models reproduce the decline in rainfall over the Sahel from 1950s–1980s. However, the CMIP5 models predict the decline with decreased magnitude and consider sulphate aerosols as the main cause of the decline. Other CMIP5 models do not include the effect of sulphate aerosol in the twentieth century. However, for the twenty-first century, the GHG effect will dominate over the aerosol parameter, and therefore, the drought over the Sahel may not actually take place.
                  Projected increased wetness over the U.S. is the result of the upward trend from 1950–1990 in precipitation levels. After the 1990s, however, the U.S. has become drier. This variation across decades can be attributed to Interdecadal Pacific Oscillation (IPO), with the warm phase of above normal SSTs in the tropical Pacific occurring around the year 1977, and the cold period taking place around 1999. IPO greatly influences US precipitation and drought, especially in the southwest. Anthropogenic data forcing does not coincide with these cycles due to the fact that these natural cycles are not included in predictions that depend on initial conditions of the coupled models. Therefore, these cycles are not reproducible in climate change predictions.
                  Differences between temporal and spatial patterns of the MCA1 mode for SST and sc_PDSI_pm in observed data versus CMIP5 data are largely attributable to model deficiencies in representing the effect of sulphate aerosols in the twentieth century, natural SST variations not taken into account by the CMIP models, and sampling errors among different GHG induced changes in the sc_PDSI_pm that are still relatively weak. MCA1 and MCA2 demonstrate the only statistically significant resemblance. This finding indicates that the global warming mode from the observations will likely become part of the GHG-induced warming model. The models are able to reflect the GHG-induced trend mode (MCA1) seen in observations  and the main ENSO mode (MCA2), therefore increasing confidence in the model predications of increasing drought patterns for the Americas Southern Europe, Southern and Central Africa, Australia, and Southeast Asia as GHG emissions contribute to continuing warming in the twenty first century. Still, these models are not reliably able to stimulate the precipitation and PDSI changes for these regions with valid certainty. The MCA1 patterns for sc_PDSI_pm for the twenty-first century are stable due to the large forced trend in the context of natural variations in temperature, precipitation, and other variables. The MCA1 patterns for sc_PDS_pm predict severe drought conditions by the late half of the twenty-first century, especially for densely populated areas such as Europe, the eastern U.S., Southeast Asia, and Brazil. If the model’s regional predictions are correct, the effect on these populations will be drastic.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s