INVERSION OF A FOREST BACKSCATTER MODEL USING NEURAL NETWORKS

Citation
Ds. Kimes et al., INVERSION OF A FOREST BACKSCATTER MODEL USING NEURAL NETWORKS, International journal of remote sensing, 18(10), 1997, pp. 2181-2199
Citations number
39
Categorie Soggetti
Photographic Tecnology","Remote Sensing
ISSN journal
01431161
Volume
18
Issue
10
Year of publication
1997
Pages
2181 - 2199
Database
ISI
SICI code
0143-1161(1997)18:10<2181:IOAFBM>2.0.ZU;2-X
Abstract
A neural network approach was used to develop acccurate algorithms for inverting a complex forest backscatter model. The model combines a fo rest growth model with a radar backscatter model. The forest growth mo del captures natural variations of forest stands (e.g., growth, regene ration, death, multiple species and competition for light). This model was used to produce vegetation structure data typical of transitional /northern boreal hardwood forests in Maine. These data supplied inputs to the radar backscatter model which simulated the polarimetric radar backscatter (C, L, P, X bands) above the forests. Using these simulat ed data, various neural networks were trained with inputs of different backscatter bands and output parameters of above ground biomass, tota l number of trees, mean tree height and mean tree age. These trained n eural networks act as efficient algorithms for inverting the complex f orest backscatter model. The accuracies (r.m.s. and R-2 values) for in ferring various parameters from radar backscatter were above ground bi omass (1.6 kg m(-2), 0.94), number of trees (48 ha(-1), 0.94), tree he ight (0.47 m, 0.88) and tree age (24.0 years, 0.83). The networks that used only AIRSAR bands (C, L, P) had a high degree of accuracy. The i nclusion of the X band with the AIRSAR bands did not seem to increase significantly the accuracy of the networks. The networks that used onl y the C and L bands still had a relatively high degree of accuracy for all forest parameter (R-2 values from 0.75 to 0.91). Modest accuracie s (R-2 values from 0.65 to 0.84) were obtained with networks that used only the L band and poor accuracies (R-2 values from 0.36 to 0.46) we re obtained with networks that used only the C band. Several networks were shown to be relatively insensitive to the addition of random nois e to radar backscatter. The results demonstrate that complex, forest b ackscatter models can be efficiently inverted using neural networks th at use only radar backscatter data.