深圳市建成区森林碳分布估测方法比较研究

Comparison of Landsat 8 images based city forest carbon modeling for Shenzhen built-up area

城市森林是森林生态系统的重要组成部分,在改善城市生态环境方面发挥着重要作用。研究城市森林碳分布对于陆地生态系统碳循环的模拟和城市的生态建设规划具有重要意义。以深圳市建成区森林为研究对象,分别采用逐步回归、协同克里格和K个最靠近邻居(KNN)3种方法对深圳市的森林碳分布进行模拟。结果表明:3种方法估测结果各有其优缺点。逐步回归和KNN法能够较准确地反映森林碳分布的信息(如行道树、公园等),但逐步回归的预测结果偏小,KNN法的预测结果偏大。协同克里格法则可以较准确的反映建成区森林碳分布并能导出总体估测值,但对森林碳空间分布的细节反映不如逐步回归和KNN。

City forest is one of important components of terrestrial ecosystems and plays a critical role in improving city ecosystem and environment. Thus, it is very significant to study urban forest carbon distribution for simulating carbon cycling in terrestrial ecosystem and setting out urban ecological construction. By taking Shenzhen built-up area as studied object, the urban forest carbon distribution of the area were simulated with the methods of stepwise regression modeling, Cokriging and K-nearest neighbors （KNN）. The results show that these three methods had their own advantages and disadvantages; The stepwise regression and KNN can more accurately revealed the detailed information of forest carbon spatial distribution （such as trees along streets, parks, and so on） than Cokriging; However, the regression led to underestimation, while KNN resulted in overestimation; The Cokring produced more accurate estimation value of forest carbon for Shenzhen city urbanized areas, but the obtained spatial distribution of forest carbon was less detailed compared to those from the regression and KNN.