The Applicability Assessment Of GCMs For The Loess Plateau Of China

As the core of the research on global change, climate change has aroused worldwide concerns. Supported by the rapid progress achieved in the field of computer technology, climate models have been strengthened and are playing an important role in measuring climate change and its impacts. General Circulation Models (GCMs) show a good performance in reproducing the basic characteristics of the general circulation of atmosphere at large-scaled regions, such as at global and continental scales. However, when it is used to capture regional climate patterns or changes, the results may be varied or even divergent for a given region due to models'diverse physical structures, parameterization and feedbacks. Thus, the applicability of GCMs for regions needs to be assessed before they are applied to provide the basic information on local climate change and the potential responses of ecosystem to climate change. Based on the idea of 'upscaling', in this research, three types of methods—directly inverse-distance-weighted interpolation (DIDW),'multiple linear regression + residual interpolation'(MLR), and'tendency surface simulation + residual interpolation'(TSS)—for spatializing metrological variables were employed, by which approaches a satisfied baseline values were produced to assess the applicability of GCMs for the Loess Plateau. The climate models involved here were CGCM3, FGOALS, ECHAM5, and CSMK3, which were all released by the 4th Assessment Report of IPCC. The potential changes in temperature and precipitation on the Loess Plateau were also predicted. An array of conclusions was arrived here:(1) According to the results of significant test and partial regression coefficient test, it was found that the effects of macro-geographical factors (including latitude, longitude, and elevation) on the spatial distribution of temperature on the Loess Plateau were rather strong, which can be reflected well by the high coefficients of determination (all were above 0.94). Moreover, elevation was the most effective factor to the local pattern of temperature. As for the precipitation pattern, the influences of the factors mentioned above were relatively weaker, with the coefficients of determination ranging from 0.598 to 0.828. Latitude appeared to have the strongest impact on the spatial distribution of precipitation in the study area, followed by the factor of longitude. Elevation, by contrast, was the weakest one among the three. In general, as elevation lifts and latitude shifts from south to north, local temperature tends to decrease gradually; and for precipitation, it decreases with the elevation of latitude and the drop of longitude and elevation.(2) In order to obtain ideal results of spatialized temperature and precipitation for the Loess Plateau, a certain number of stations were involved to test the performance of the three spatialization methods. The testing results (MAE values) for mean monthly temperature showed that the methods of TSS and MLR incorporated with three macro-geographical factors performed much better than DIDW method. The MAE values of the former two only changed from 0.242 to 0.509℃and 0.485 to 0.776℃, respectively, while the DIDW's were all above 1.0℃. This suggested that macro-geographical factors can help to improve the precision of spatializations for mean monthly temperature. In simulating local monthly precipitation pattern, there were no significant differences between the three methods. However, they can still be used to simulate the annual precipitation pattern for the Loess Plateau.(3) TSS method seems to be more reliable in spatializing mean monthly temperature for those regions with rough terrain. When it comes to the relatively flat regions, both methods, TSS and MLR, may yield ideal results. Stepwise analysis may help to optimize regression models and keep the crucial variables in the models, but the precision of its spatializations, to some extent, was declined.(4) By comparing the baseline of mean monthly temperature with models' projections for the period from 1971 to 2000 on the Loess Plateau, the results indicated that FGOALS performed better than CGCM3, and ECHAM5 was better than CSIRO. Moreover, the distribution of MAE values produced by ECHAM5 model was comparatively steady and most close to X axis, so ECHAM5 tends to be the one among the four which is more applicable in simulating temperature pattern for the Loess Plateau. For the local annual precipitation pattern, it was over-estimated by all the models remarkably. What delighted us was that, two models——ECHAM5 and CSIRO——could mimic the spatial changes of the precipitation generally, particularly the MAE values yielded by the later one were relatively lower. From that point, the model of CSIRO was the best performer among the other three.(5) Over the next thirty years, assuming a development similar to the emission scenarios of A2, temperature on the Loess Plateau is going to increase by approximately 1.23℃annually and those districts which are far from the east coast and with a relatively high latitude will bear much more increases in temperature. On the contrary, yearly precipitation in most regions of the Loess Plateau is likely to have a downward trend with the ratios changing between 4.08% and -13.51%.