关键词: 有效叶面积指数, 针叶树种, 森林类型, 卫星遥感, 空间分布规律

Abstract:

Objective: The aim of this study was to study develop the rapid, accurate and macroscopic methods for obtaining effective leaf area index (LAI_e) of different forest types, and to explore their spatial distribution, so as to provide new ideas for the development of medium and small-scale forest LAI_e remote sensing products, and offer scientific and reliable technological means for precision forestry monitoring and simulation of forest ecosystem carbon and water cycles. Method: Using Changbai Mountain as the research area, this study extracts the spatial distribution of four coniferous forest types (Larix olgensis, Pinus sylvestris var. mongolica, Pinus koraiensis and Picea koraiensis ) based on Sentinel-2A multispectral images through a three-dimensional convolutional neural network. It adopts 2 schemes, differentiated forest types and full sample, to analyze the correlation between field-measured LAI_e and 7 vegetation indices [enhanced vegetation index (EVI), inverted red-edge chlorophyll index (IRECI), modified simple ratio (MSR), normalized difference water index (NDWI), normalized difference vegetation index(NDVI), soil adjusted vegetation index (SAVI), simple ratio (SR)]. The optimal vegetation index corresponding to each tree species was used to construct regression models for LAI_e and vegetation index of differentiated forest type and full sample, and the accuracy performance of differentiated forest type model, full sample model and PROSAIL model in LAI_e inversion was compared based on validation sample data. Subsequently, spatial pattern and change rule of LAI_e of 4 tree species were analyzed by combining geographical factors. Result: 7 vegetation indices in all sample groups were significantly correlated with relative LAI_e values (P<0.01), except for EVI and LAI_e of Pinus koraiensis , SR and LAI of Picea koraiensis, the correlation coefficients were all greater than 0.6, while the correlation between LAI_e and different vegetation indexes was significantly different among groups. LAI_e of Pinus koraiensis, Larix olgensis and Pinus sylvestris var. mongholica had the highest correlation with IRECI, while LAI_e of Picea koraiensis and Pinus sylvestris var. mongholica had the highest correlation with EVI and MSR, respectively. Compared with the full-sample model, the R² value of the 4 different forest types model increased by more than 12.7%, and RMSE decreased by 34.5%. The LAI_e of the 4 forest types in the study area ranged from 0.37 to 5.86, and the average LAI_e from high to low was Pinus koraiensis, Larix olgensis, Pinus sylvestris var. mongholica. and Picea koraiensis. Pinus koraiensis was the most sensitive to changes in altitude, slope and aspect, followed by Picea koraiensis and Pinus sylvestris var. mongholica, while Larix olgensis was the least affected. Conclusion: There were significant differences in the correlation between LAI_e and remote sensing vegetation index among different forest types. The inversion accuracy of LAI_e can be improved by constructing specific regression models for different forest types. The accuracy of the fitted linear model after distinguishing forest types was higher than that of the PROSAIL model and the full-sample model, but in the area with high LAI_e value performed unstable compared with the PROSAIL model. LAI_e of the 4 tree species varied greatly in reflecting the changes of geographical factors. This study can provide a scientific reference for the selection of LAI remote sensing inversion models for coniferous forests in small and medium scales under tree species differentiation.

Key words: effective leaf area index (LAI_e), conifer species, forest type, satellite remote sensing, spatial distribution