QuickBird image-based estimation of tree stand density using local maxima filtering method: A case study in a Beijing forest

The stand density of trees affects stand growth and is useful for estimating other forests structure parameters. We studied tree stand density in Jiufeng National Forest Park in Beijing. The number of spectral local maxima points (NSLMP) calculated within each sample plot was extracted by the spectral maximum filtering method using QuickBird imagery. Regression analysis of NSLMP and the true stand density collected by ground measurements using differential GPS and the total station were used to estimate stand density of the study area. We used NSLMP as an independent variable and the actual stand density as the dependent variable to develop separate statistical models for all stands in the coniferous forest and broadleaf forest. By testing the different combination of Normalized Difference Vegetation Index (NDVI) thresholds and window sizes, the optimal selection was identified. The combination of a 3 × 3 window size and NDVI ≥ 0.3 threshold in coniferous forest produced the best result using near-infrared band (coniferous forest R2 = 0.79, RMSE = 12.60). The best combination for broadleaf forest was a 3 × 3 window size and NDVI ≥ 0.1 with R2 = 0.44, RMSE = 9.02 using near-infrared band. The combination of window size and NDVI threshold for all unclassified forest was 3 × 3 window size and NDVI ≥ 0.3 with R2 = 0.70, RMSE = 11.20 using near-infrared band. A stand density planning map was constructed using the best models applied for different forest types. Different forest types require the use of different combination strategies to best extract the stand density by using the local maximum (LM). The proposed method uses a combination of high spatial resolution imagery and sampling plots strategy to estimate stand density.