Abstract: Based on the Landsat TM and Envisat ASAR HH/HV imagery data and by using the GPS data of red-crowned crane nesting sites (n=28) at Zhalong National Nature Reserve of Northeast China, the models of the breeding habitat selection of red-crowned crane at the Reserve were established by binary Logistic regression to identify the key variables for the habitat selection at eight spatial scales (30-240 m). The relative performance of the two models based on the Landsat TM and Envisat ASAR HH/HV databases was compared, and the prediction capacity of the models across the eight scales was approached. The overall precisions of the two models were satisfactory (≥ 69.0%). At scale 30 m, only variable TCA_2 entered with negative value into the model based on Landsat TM database, which indicated that the crane at this scale avoided selecting higher density reed marshes. At scales 60-120 m, the variable PCA_2 entered with positive value into the two models, indicating that the crane at these scales had higher demand of high density reed marshes to improve its concealment. At scale 90 m, the variable HV backward scatting coefficient also entered into the combined model, which indicated that water condition was the important factor for the habitat selection of the crane at this scale. At scales >120 m, the texture information of the two satellite sensors started to be involved into the two models, indicating that at larger scales, the crane had decreasing demand on the vegetation features for its breeding habitat selection but increasing sensitivity to the anthropogenic disturbance factors. The introduction of ASAR variables into the models increased the prediction accuracy of the models markedly at all scales.

Key words: Landsat TM, Envisat ASAR, Logistic regression, multi-scale, red-crowned crane, breeding habitat