Chin J Plant Ecol ›› 2012, Vol. 36 ›› Issue (10): 1106-1119.DOI: 10.3724/SP.J.1258.2012.01106

Previous Articles     Next Articles

Applying generalized additive model to integrate digital elevation model and remotely sensed data to predict the vegetation distribution

SONG Chuang-Ye1,*(), LIU Hui-Ming2, LIU Gao-Huan3, HUANG Chong3   

  1. 1State Key Laboratory of Vegetation and Environmental Change, Institute of Botany, Chinese Academy of Sciences, Beijing 100093, China
    2Satellite Eenvironment Center, Ministry of Environmental Protection, Beijing 100094, China
    3State Key Laboratory of Resources and Environmental Information System, Institute of Geographic Sciences and Nature Resources Research, Chinese Academy of Sciences, Beijing 100101, China
  • Received:2011-07-14 Accepted:2012-05-08 Online:2012-10-01 Published:2012-09-26
  • Contact: SONG Chuang-Ye

Abstract:

Aims Our objective was to integrate vegetation survey data, remotely sensed data and environmental data based on generalized additive model to map vegetation and investigate whether coupling remotely sensed data and environmental data could improve the performance of generalized additive model.
Methods Altitude, slope, nearest distance to coastline, nearest distance to the Yellow River and spectral variables were selected as predictive variables. Generalized additive model was employed to describe the relation between the vegetation and predictive variables. Deviance (D 2) was employed to test the goodness of curve fitting, and a probability map of each vegetation type produced by the fitted generalized additive model was assessed by cross-validation of receiver operating characteristic (ROC). Vegetation type of each grid cell was determined according to the maps of probability. We validated the final predicted vegetation type map by comparing the predicted vegetation type map with vegetation survey data.
Important findings Results indicated that integrating field sampling, digital elevation model and remotely sensed data based on generalized additive model was feasible to map vegetation. Coupling environmental variables and spectral variables improved the performance of generalized additive model. Prediction accuracy varied according to the plant community type, and plant community with dense cover was better predicted than sparse plant community. The selected generalized additive models for each vegetation type indicated that both environmental variables and spectral variables were important factors for predicting vegetation distribution. Drop contribution calculation indicated that the contribution of the same predictive variable for different vegetation types was different. This can be explained by the different spectral and environmental characteristics of vegetation types. The contribution of each predictive variable for the same vegetation type varied according to the modeling scene. This might be explained by the coupling effects of environmental variables and spectral variables.

Key words: area under curve (AUC), normalized difference vegetation index (NDVI), receiver operating characteristic (ROC)