| The forest structural parameters including stand average tree height, stand average breast diameter and stand average stock volume could describe the basic situation of the horizontal and vertical structure of forest. Optical remote sensing data could provide abundant surface reflectance. However it was enormously affected by atmosphere and forest canopy. Due to the all-weather work, all-time work and strong penertrating advantages of SAR(synthetic aperture radar),the extraction of forest structural parameters by SAR and optical remote sensing has been an research focus.The study area located in Jiangle forest farm, Samming, Fujian province. The forest structural parameters was estimated by multi spectral data and SAR, and high resolution data was used to implement classification and mapping. The optical data Landsat8 OLI, rada data ALOS PALSAR, high resolution data Worldview-2 were used as the remote sensing data. Landsat8 OLI and ALOS PALSAR were used to build the estimated models. To improve the estimation accuracy of forest structural parameters with ALOS PALSAR data, we introduced adjusted entropy (ENTadj) which represents the complexity of stand structure for the estimation. Thus, the interference of radar backscattering coefficient caused by stand structure could be eliminated. Firstly, the ENTadj of stand was defined by the measured tree heights in sample plots of the field. And then, the ENTadj based on pixels was calculated by linear regression model established with the integration of the ENTadj of stand and Landsat 8 OLI band 6. Commonly, the relationship between forest structural parameters and ALOS PALSAR backscattering coefficient could be simulated by a logarithm regression model. In this research, ENTadj based on pixel was introduced as a new independent variable to improve the original logarithm model. Three types of improved models were established for stand mean tree height, stand mean DBH and stand stock volume respectively. The original model and three improved models were used to estimate the above forest structural parameters for Cunninghamia lanceolata stand, Pinus massoniana stand, broadleaf stand and mixed stand. Ultimately, optimal estimating models for each stand structural parameter in the four types of stands were selected by comparing R2 (coefficient of determination) with totally 12 results. The Worldview-2 was used to classify the stand type based on object-oriented method,which provide unique zones for different stands. The result of object-oriented classification provideed species distribution boundary for mapping forest structural parameters.The results showed:(1) The R2 of models in radar estimating forest structural parameters increased after considering the influence of forest stand structure,which P. massoniana stand increased the most. The stand structure impacted P. massoniana stand most.(2) The sensitivity of forest type to the forest structure was quite different. According to R2, C.lanceolata stand, mixed stand, P. massoniana stand had a high sensitivity to the mean forest height with the stand structure interference. Broadleaf stand had a high sensitivity to the mean stock volume with forest stand structure interference.(3)The results of accuracy examination revealed that there were desired precisions for estimating tree height (RMSE:0.74-2.51 m), DBH (RMSE:2.61-5.61 cm) and stock volume (RMSE:21.71-30.92 m3/hm2).(4) The relative error of the above three forest structural parameters were within 30%. There was no direct linear relationship between the relative errors and backscattering coefficient.The average relative error changed up and down around 10%.This study explored the potential of applying stand structure information in forest structural parameters, and increased the ability to estimate the stand structural parameters by combining the optical and radar remote sensing data. |
PDF Full Text Request