[Study on artificial neural network combined with multispectral remote sensing imagery for forest site evaluation]

Area of Science:

• Forestry
• Remote Sensing
• Artificial Intelligence

Background:

• Classic forest site quality evaluation relies solely on ground survey data, underutilizing rich multispectral remote sensing information.
• Existing methods lack comprehensive data integration for precise site quality assessment.

Purpose of the Study:

• To develop a more effective site quality evaluation system by integrating multispectral remote sensing data with traditional survey data.
• To improve the accuracy of sublot site quality evaluation for larch using a novel neural network approach.

Main Methods:

• Developed an improved back propagation artificial neural network (BPANN) model.
• Combined multispectral remote sensing spectra factors with sublot survey data and site index relations.
• Utilized sensitivity analysis to exclude irrelevant factors and simplify the neural network for improved training efficiency.

Main Results:

• The optimal site index prediction model achieved an accuracy of 95.36%.
• This represents a 9.83% improvement over models using only classic sublot survey data.
• The integrated approach demonstrated superior predictive accuracy for larch site index.

Conclusions:

• Integrating multispectral remote sensing data with small plot survey data significantly enhances the accuracy of forest site index prediction.
• The developed neural network model offers a more effective and superior method for forest site quality evaluation.