黄土高原表土花粉与植被覆盖度定量模型的建立及应用  

Establishment and application of a quantitative model of modern pollen and vegetation covers on the Chinese Loess Plateau

作　　者：李雪银 孙爱芝 吴双双[1] LI Xueyin;SUN Aizhi;WU Shuangshuang(College of Earth and Planetary Sciences,University of Chinese Academy of Sciences,Beijing 100049;Beijing Yunshan Earth Critical Zone National Research Station,University of Chinese Academy of Sciences,Beijing 101408;Key Laboratory of Computational Geodynamics,Chinese Academy of Sciences,Beijing 100049)

机构地区：[1]中国科学院大学地球与行星科学学院,北京100049 [2]中国科学院大学,北京燕山地球关键带国家野外观测研究站,北京101408 [3]中国科学院计算动力学重点实验室,北京100049

出　　处：《第四纪研究》2023年第5期1309-1327,共19页Quaternary Sciences

基　　金：国家自然科学基金项目(批准号:41972198);中央高校基本科研业务费专项项目(批准号:E1E40408X2)共同资助。

摘　　要：黄土高原是一个气候变化敏感区,亦是一个生态环境脆弱区,更是我国人口-资源-环境矛盾最为集中的区域之一。因此,构建适合于黄土高原表土花粉与植被覆盖度定量模型对了解过去植被覆盖度的演变以及人与环境之间的关系具有重要意义。本工作基于黄土高原443个表土花粉数据和2000~2015年的归一化植被指数(NDVI),采用逐步回归法(Stepwise)、增长回归树法(BRT)、局部加权加权平均法(LWWA)、加权平均偏最小二乘法(WA-PLS)和最佳类比法(MAT),结合植被覆盖度(FVC)和年均降水量(MAP)、年均气温(Tann)、 1月月均温(T1)、 7月月均温(T7)、有效湿度(MAP/ET0)这5个气象要素数据,构建表土花粉与植被覆盖度定量模型;通过普通检验、留一交叉检验(LOO)、自助法检验(Bootstrap)和检验空间自相关性的h-block检验方法对定量模型进行检验,进而筛选出最优模型;最后,利用31个全新世地层花粉数据重建黄土高原过去植被覆盖度,探讨其变化特征。结果表明:1)FVC、 MAP、 T1、 T7和MAP/ET0是黄土高原环境变化的5个重要因子,其中,FVC的第一排序轴特征值/第二排序轴特征值(λ1/λ2)最大(0.47),单一解释变量最高(8.75),其与花粉组合的相关性最强(0.66),可用于构建表土花粉与植被覆盖度定量模型;2)采用Stepwise、 BRT、 LWWA、 WA-PLS和MAT这5种方法分别建立表土花粉-植被覆盖度定量模型,根据观测值与预测值的决定系数(R2)、最大偏差(Max.bias)和预测均方根误差(REMSP)评估构建的模型,结果显示BRT模型表现更优,可用于黄土高原过去植被覆盖度的定量重建;3)全新世FVC重建结果显示,早、中全新世黄土高原FVC较高,其中7.0 cal.ka B. P.时达到峰值(平均值为0.68);但自4.0 cal.ka B. P.开始,植被覆盖度下降,特别是3.0~2.0 cal.ka B. P.下降明显,平均值低于0.30。通过与研究区内其他气候指标和人类活动强度指标的对比,认为早、中全新�The Chinese Loess Plateau(CLP)is a climate-sensitive area,an ecologically fragile area,and one of the areas with the most concentrated population-resource-environment contradictions in China.Therefore,the establishment and application of a quantitative model of modern pollen and vegetation cover on the CLP is of great importance for understanding the evolution of past vegetation cover and the human-environment relationship on the CLP.Based on 443 modern pollen data and the Normalized Difference Vegetation Index(NDVI)during 2000~2015 on the CLP(33°41'~41°16'N,100°52'~114°31'E),the research uses Stepwise Regression(Stepwise),Boosted Regression Trees(BRT),Local Weighted Weighted Average(LWWA),Weighted Averaging Partial Least Squares(WA-PLS)and Modern Analogy Technique(MAT),combined with Fractional Vegetation Coverage(FVC)and Mean Annual Precipitation(MAP),Mean Annual Temperature(Tann),January average monthly temperature(T1),July average monthly temperature(T7)and Effective Humidity(MAP/ET0),to establish a quantitative model of modern pollen and vegetation cover on the CLP.In addition,the performance of the established quantitative models were tested by ordinary test,Leave-One-Out Cross Validation(LOO),Bootstrap test(Bootstrap)and h-block test to assess the influence of spatial autocorrelation,and the optimal model was selected by comparing the square of the correlation coefficient between the predicted value and the observed value(R2),the maximum deviation(Max.bias)and the root mean square error of the prediction(REMSP).Then,we used fossil pollen data from 31 sites,together with the optimal model,to reconstruct the Holocene vegetation covers of the CLP.The results show that(1)FVC can be used to establish quantitative models of modern pollen and vegetation cover,which has the largest value of Axis1 eigenvalue/Axis2 eigenvalue(λ1/λ2)(0.47),and has the highest value of single explanatory variable(8.75).(2)FVC has the strongest correlation with pollen assemblages(0.66).Then,we use above five methods to establish q

关 键 词：黄土高原 表土花粉 植被覆盖度 定量重建模型 

分 类 号：P941.74[天文地球—自然地理学] Q914.86[天文地球—古生物学与地层学]