与站点统计降尺度插值和动力降尺度相比,高精度曲面建模(HASM)降尺度,具有不需大尺度预报因子,直接从GCM结果构建区域上高空间分辨率的未来气候模拟曲面的优势。HASM降尺度将未来气候,分为历史观测拟合的气候基准值和GCM未来气候变化值进行模拟,精度明显高于传统方法,但常系数全局拟合的气候基准值忽略了降水分布的空间非平稳性,导致降水模拟受到较大影响。为增强降水降尺度的气候背景值的描述能力,通过分析全国尺度降水的非线性非平稳性特点,提出耦合空间变系数气候基准值的HASM空间变系数降尺度模型(HASM-SVDM)以改进HASM对非平稳要素的降尺度能力,并以1961-2010年全国气温降水观测数据结合地形特征信息,利用HASM降尺度方法对HadCM3的A1Fi、A2a和B2a3种情景的1961-1990、2010-2039、2040-2069和2070-2099时段的全国未来气温与降水进行降尺度模拟。分析表明,耦合全局线性模型的HASM常系数降尺度模型适合全国气温的降尺度模拟,而耦合空间变系数拟合的HASM-SVDM增强了空间非平稳背景值的描述能力,模拟的空间分布更能体现降水总体的非均匀分布趋势,适合全国降水的降尺度模拟。
Compared with statistical downscaling methods and dynamical downscaling methods, HASM- based downscaling methods, which do not need large-scale predictor, can directly create high-resolution climatic surfaces under GCM scenarios. HASM downscaling methods separate future climate elements in- to climate base value and prospect climatic change value. This method is termed HASM-Constant Coeffi cient Downscaling Model (HASM-CDM) because climatic base value is fitted by global constant regression model and climatic change value is interpolated by HASM. Although HASM can obtain higher accuracy than other classical methods, precipitation base value fitted by (HASM-CDM) lost spatial non-stationary features of precipitation, which decreases the accuracy of precipitation simulation. The relationship be- tween precipitation and auxiliary variables such as DEM and some topographical factors may change ac- cording to geographical location which can not be represented by HASM-CDM. HASM-Spatially Variable Coefficient Downscaling Model (HASM-SVDM) was developed which integrated with spatially variable coefficient regression model and data transformation in this paper. HASM-SVDM uses variable coefficient regression and data transformation to solve accuracy problem of climatic base value. The mean annual temperature (MAT) and mean annual precipitation (MAP) are constructed under different scenarios of HadCM3 AIFi, A2a and B2a during the periods T1 (1961 - 1990), T2 (2010 - 2039), T3 (2040 - 2069) and T4 (2070- 2099) by HASM downscaling models. The results show that HASM-constant coefficient downscaling model integrated with global linear model is applicable to the temperature downscaling simu- lation, while HASM-spatially variable coefficient downscaling model improves spatial non-stationary base value, and is appropriate for the precipitation downscaling modeling at the national level.