Estimation of Tree Species Diversity in Warm Temperate Forests via GEDI and GF-1 Imagery DOI Open Access
Lei Zhang, Yang Liu,

Jinhua Sun

et al.

Forests, Journal Year: 2025, Volume and Issue: 16(4), P. 570 - 570

Published: March 25, 2025

Estimates of tree species diversity via traditional optical remote sensing are based only on the spectral variation hypothesis (SVH); however, this approach does not account for vertical structure a forest. The relative height (RH) indices derived from GEDI spaceborne LiDAR provide vegetation information through waveform decomposition. Although RH have been widely studied, optimal index estimation remains unclear. This study integrated GF-1 imagery and data to estimate in warm temperate First, random forest plus residual kriging (RFRK) was employed achieve wall-to-wall mapping GEDI-derived indices. Second, recursive feature elimination (RFE) applied select relevant features. (RF), support vector machine (SVM), k-nearest neighbor (kNN) methods were subsequently data. results indicated that multisource achieved greater accuracy (average R2 = 0.675, average RMSE 0.750) than single-source 0.636, 0.754). Among three learning methods, RF model (R2 0.760, 2.090, MAE 1.624) significantly more accurate SVM 0.571, 2.556, 1.995) kNN 0.715, 2.084, 1.555) models. Moreover, mean_mNDVI, mean_RDVI, mean_Blue identified as most important features, whereas RH30 RH98 crucial features establishing models diversity. Spatially, high west low east area. highlights potential integrating modeling emphasizes indicative middle- lower-canopy

Language: Английский

Estimation of Tree Species Diversity in Warm Temperate Forests via GEDI and GF-1 Imagery DOI Open Access
Lei Zhang, Yang Liu,

Jinhua Sun

et al.

Forests, Journal Year: 2025, Volume and Issue: 16(4), P. 570 - 570

Published: March 25, 2025

Estimates of tree species diversity via traditional optical remote sensing are based only on the spectral variation hypothesis (SVH); however, this approach does not account for vertical structure a forest. The relative height (RH) indices derived from GEDI spaceborne LiDAR provide vegetation information through waveform decomposition. Although RH have been widely studied, optimal index estimation remains unclear. This study integrated GF-1 imagery and data to estimate in warm temperate First, random forest plus residual kriging (RFRK) was employed achieve wall-to-wall mapping GEDI-derived indices. Second, recursive feature elimination (RFE) applied select relevant features. (RF), support vector machine (SVM), k-nearest neighbor (kNN) methods were subsequently data. results indicated that multisource achieved greater accuracy (average R2 = 0.675, average RMSE 0.750) than single-source 0.636, 0.754). Among three learning methods, RF model (R2 0.760, 2.090, MAE 1.624) significantly more accurate SVM 0.571, 2.556, 1.995) kNN 0.715, 2.084, 1.555) models. Moreover, mean_mNDVI, mean_RDVI, mean_Blue identified as most important features, whereas RH30 RH98 crucial features establishing models diversity. Spatially, high west low east area. highlights potential integrating modeling emphasizes indicative middle- lower-canopy

Language: Английский

Citations

0