地理研究

地理研究 ›› 2011, Vol. 30 ›› Issue (2): 278-290.doi: 10.11821/yj2011020008

• 地球信息科学 • 上一篇    下一篇

湿地植被地上生物量遥感估算模型研究——以洪河湿地自然保护区为例

李爽1, 张祖陆1, 周德民2   

  1. 1. 山东师范大学人口·资源与环境学院,济南 250014;
    2. 首都师范大学资源环境与旅游学院,北京 100048
  • 收稿日期:2010-03-07 修回日期:2010-07-29 出版日期:2011-02-20 发布日期:2011-02-20
  • 通讯作者: 张祖陆(1949-),男,汉,上海市人,教授,博士生导师,主要从事自然地理和水文环境研究。 E-mail: zulzhang@126.com
  • 作者简介:李爽(1984-),女,汉,山东济南人,博士研究生。主要从事遥感和地理信息系统应用方面的研究。 E-mail: ls8412519@163.com
  • 基金资助:

    国家863资助项目(2007AA12Z176)

An estimation of aboveground vegetation biomass in a national natural reserve using remote sensing

LI Shuang1, ZHANG Zu-lu1, ZHOU De-min2   

  1. 1. College Population, Resource and Environment, Shandong Normal University, Jinan 250014, China;
    2. College Tourism, Resource and Environment, Capital Normal University, Beijing 100048, China
  • Received:2010-03-07 Revised:2010-07-29 Online:2011-02-20 Published:2011-02-20

PDF (PC)

赞

摘要： 以洪河湿地自然保护区的TM图像和29个实测样地生物量数据为数据源,采用单变量线性和非线性回归、多元线性逐步回归及人工神经网络(BP网络、RBF网络)技术,构建了研究区内典型湿地植被(草甸和沼泽)的地上生物量干重和湿重的遥感估算模型,并对比得到最优模型。主要结论有:(1)RBF神经网络模型及多元非线性模型是研究区内湿地植被地上生物量遥感估算的最优模型,生物量干重估算值的平均相对误差为2.795%,生物量湿重估算值的平均相对误差为3.399%。(2)比较2004年8月、2006年8月和2008年8月研究区内草甸和沼泽总生物量可得,总生物量干重呈上升趋势,而总生物量湿重呈下降趋势。(3)研究区内生物量极高值和极低值分布较少,且主要集中于混合像元分布的地方,如岛状林、灌丛的周边地区或是沼泽内含水较多的地区。

关键词: 湿地, 生物量, 遥感, 回归模型, 神经网络

Abstract: Wetland vegetation is an important component of wetland. The biomass of vegetation is an essential index to describe the wetland ecosystem and reflects its health status. Therefore, the investigation of wetland vegetation biomass has important practical significance.In this paper, the Honghe National Natural Reserve (HNNR) was selected as the study area. The TM images on August 19th, 2008, August 30th, 2006, August 1st, 2004 and 29 samples of biomass data in the same period were used as the data source to establish the estimation models. The correlations between the remote information (reflectivity, vegetation index) and measured biomass were analyzed in this paper. The estimation models were established based on the method of regress model and artificial neural network (ANN). The models included the linear regression models, the curve regression models, the stepwise regression models, and ANN models (BP network and RBF network). In comparison of all the models, the best estimation models were obtained. The accuracy of the dry biomass models and the humid biomass models were compared. Then, the total biomass of meadow and marsh in HNNR were estimated. Finally, the total biomass spatial distribution maps of 2004, 2006 and 2008 were made, and the trend of the biomass was analyzed in this paper. The conclusions of the research were as follows. (1) The correlations between the wetland vegetation aboveground biomass and RS information were good, and the correlation between the dry biomass and RS information was better than that between the humid biomass and RS information. The estimation models based on the RS information can estimate the wetland vegetation aboveground biomass relatively well.(2) The performance of the models based on RBF network was better than that based on regression models and BP network. With the method of RBF, the mean relative error (MRE) of estimated dry biomass was 2.795% and the MRE of estimated humid biomass was 3.399%. The dry biomass models were better than the humid biomass models in comparison with them.(3) In this study area, dry biomass was mainly between 300 g/m2 and 900 g/m2 and the humid biomass was mainly between 600 g/m2 and 1800 g/m2. By analyzing the total biomass of the three years, the total dry biomass showed an upward trend, and the total humid biomass showed a downward trend. There was little extreme high or low biomass, which was mainly distributed in the places where a lot of mixed-pixels existed, such as the edge of the forest and bush, or the marsh with a lot of water.

Key words: Wetland biomass, RS, estimated model, artificial neural network