Abstract: Through introducing self-organizing map (SOM) neural network into forest health assessment, and combining with geographic information system (GIS), a quantitative assessment was conducted on the health status of different forest types (broadleaved mixed forest, broadleaf-conifer mixed forest, and Larix olgensis forest) at subcompartment scale in the Baihe Forestry Bureau in Changbai Mountains, and a comparison was made on the health status of the sub-compartments with different average age, different average tree height, and different canopy density. The results showed that SOM neural network would be a more advanced approach for the automated and quantitative assessment of forest health. Its greatest advantage in the assessment of forest health was no need to know the priori knowledge about classification categories, and no need of the assessment indicators’ weights beforehand determined. As a result, SOM neural network could effectively overcome the interference of subjective factors, and let the classification results become more objective and accurate. The health level of test forest types was in the order of broadleaved mixed forest subcompartments Ⅲ> Ⅱ> Ⅰ> Ⅳ> Ⅴ, broadleaf-conifer mixed forest subcompartmentsⅡ> Ⅳ> Ⅰ> Ⅲ> Ⅴ, and Lalix olgensis forest subcompartments Ⅰ> Ⅱ> Ⅲ>> Ⅴ> Ⅳ. Relatively, the forest subcompartments that had greater average age and higher average tree height and canopy density would have higher level forest health. This study could provide theoretical support for the sustainable management and multifunctional use of the forests in Baihe Forestry Bureau.

Key words: Earthworm, Dynamics of soil C and N, Crop yield, Crop residue application, Upland rice winter wheat rotation