偏最小二乘(PLS)回归方法在中国东部植被变化归因研究中的应用

植被变化往往受到不同气候变量的综合作用,人类活动影响又使得植被对气候响应变得更为复杂,如何准确判别各种影响因素的相对重要性是植被变化归因研究中的一个关键点。研究基于偏最小二乘(PLS)回归方法,使用1982—2006年的归一化植被指数(NDVI)数据,分析了降水、气温、日照、相对湿度和风等气候变量对中国东部植被变化的相对影响,并选取了NDVI变化较为典型的区域,量化了农业活动对该地区植被变化的相对贡献。PLS回归方法兼具了主成分分析和多元回归的优点,克服了众多自变量之间存在强烈交互相关导致的多元共线性问题。研究结果表明:1 1982—2006年间,中国东部逐月NDVI的年际变化呈现出明显的南北差异。在12、1—5月,NDVI以显著上升为主,上升区域主要位于淮河以北。在6—10月,NDVI以显著下降为主,下降区域主要为淮河以南的部分区域,特别是6月江苏一带NDVI的大范围下降尤为明显。不过与NDVI发生显著变化的区域相比,更多区域的NDVI在各月并没有出现显著变化。2在NDVI显著上升的站点,对NDVI变化最具解释意义的气候变量为气温,特别是冬末春初(2—3月)的升温对黄淮海区域NDVI的显著上升具有主导控制作用。而对于NDVI显著下降的站点,多数都不能从气候角度解释这些区域的NDVI变化。3江苏省NDVI在6月出现的大范围显著下降,与农业种植结构的调整,主要是棉花种植面积的减少以及油菜面积的增加具有显著关系。

Application of Partial Least Squares (PLS) Regression Method in Attribution of Vegetation Change in Eastern China

Vegetation change is generally caused by the combined effects of various climate variables, which is further complicated by the impacts of human activities. Assessing the importance of each explanatory variable is critical for the study of vegetation change attribution. The responses of vegetation to temperature and precipitation in eastern China have been widely explored in previous studies. However, less attention has been paid to the influence of other climate variables in vegetation change. In this study, we introduced a statistical method called partial least squares (PLS) to investigate the relative importance of different climate variables. ThePLS regression, combining features of principal components analysis (PCA) and multiple regression, overcomes the multicollinearity problem which arises when two or more explanatory variables in a multiple regression model are highly correlated. Using GIMMS NDVI products and PLS method, we first investigated the relative effects of different climate variables (temperature, precipitation, sunrise, relative humidity, wind) on vegetation change in eastern China from the period 1982 to 2006. Then, the relative contribution of anthropogenic factors on the vegetation change was quantified in the region of Jiangsu Province where vegetation shows distinctive changes. The results indicated that: 1) there were distinct north -south differences among interannual variations of monthly NDVI in eastern China in the period of 1982-2006. A significant increase of NDVI was found in December through May in some areas north to the Huaihe River, while the drop of NDVI occurred in June through October in some areas south to the Huaihe River; 2) in the areas with significantly increased NDVI, the greatest contributor was temperature and it had the most significant effect on the increase of NDVI. In particular, the temperature rise could play a dominant role in driving the increase of NDVI in the Huang-Huai-Hai Plain in late winter and early spring (February-March). The decrease in NDVI, by contrast, might not be attributed to climate factors in many areas. However, it should be noted that there was no obvious change in NDVI trends in many parts of eastern China compared with the areas suffering significant NDVI change; 3) Jiangsu Province was mainly characterized by a significant decline of NDVI in June from 1982 to 2006. However, such large regional concentration of NDVI change was not observed in other months and regions. Statistical analysis showed that the agricultural structural adjustment played a key role in controlling the NDVI change in June in Jiangsu Province. The decline of NDVI in June was mainly attributed to the decrease in sown area of cotton across a large spatial extent.