| Leaf area index (LAI) is strongly related to forest productivity and stand structure. As a result of the great interest in ecological modeling at stand, regional, and global scales, much attention is given to LAI which is considered to be a key parameter of ecosystem processes. Various ecophysiological processes of a forest ecosystem are strongly controlled by LAI:interception of light and precipitation, gross productivity, transpiration, and soil respiration by means of litter return. Evergreen broad-leaved forest play a significant role in the whole ecosystem, which is also widely distributed in southwest Sichuan.The study of remote sensing estimation of LAI and the volume are less in this region. On the other hand, the model of remote sensing in estimating forest parameters is different because of the different regions and species composition. Therefore,the study for estimating LAI and the volume in this region are significant for understanding the eological processes in mountainous evergreen broad-leaved forest.In this study, Shangli, a township in Yaan county, was chosen as the study area. SPOT-5data is the primary source of data, combined with ground survey data. Through remote sensing image orthorectification, atmospheric correction, classification and extraction of various characteristics factors of the image, the relationship between the characteristic factors and LAI and also the relationship between the characteristic factors and volume were analysed. Using stepwise multiple regression and partial least squares regression, the remote sensing estimation models of LAI and volume were established separately. On this basis, a thematic maps of LAI and the volume of evergreen broadleaved forest in the studied area were obtained.The main conclusions are as follows:(1) Using the method of maximum likelihood to classify the SPOT-5image, the overall classification accuracy is81.42%, which achieved the accuracy requirements.The distribution of land types of the studied area was obtained. By extracting evergreen broadleaved woodland, its distribution map in the studied area was gained.(2) The shadow fraction (SF) obtained by pixel unmixing have a higher correlation with LAI and volume(R2=0.678/0.739/p<0.01), using linear and nonlinear combinations of each band, eight vegetation indices such as SLAVI were extracted,which approved that all vegetation index are highly correlated. Energy and five other texture features were extracted in the full color image using a3×3window size. Altitude and other two topographical features were extracted from DEM. A total of22characteristics were extracted as independent variables. (3) Among the22extracted alternative modeling factors, LAI and the volume have most closely relationships with the specific leaf area vegetation index (SLAVI), and relatively good fitting relationships with the shadow fraction (SF) and other vegetation indices,which was most appropriate using the quadratic polynomial to express. However, there were not a good fitting relationship between LAI, volume and the single-band data, texture features and topographical factors(except elevation).(4) The remote sensing estimation models of LAI in the studied area were established using stepwise multiple regression and partial least squares regression respectively. Accuracy test with independent samples of the two models showed that the accuracy of partial least squares regression was higher than that obtained by the method of stepwise multiple regression, the overall accuracy of which was88%.The VI3in the model was most important for the model projection. The expression of the model is as follows: Y=-0.000367*BAND4+0.520719*SF+0.428499*TNDVI+0.030734*VI3+0.004151*RVI+0.158799*SAVI+0.799497*SLAVI+0.238784*NDVI-2.452356(5) Two remote sensing estimation models for volume in the studied area was established using stepwise multiple regression and partial least squares regression respectively. Accuracy test with independent samples of the two models showed that the accuracy of partial least squares regression was higher than that obtained by the method of stepwise multiple regression, the overall accuracy of which was85%. The VI3in the model was most important for the model projection. The expression of the model is as follows: Y=-0.050523*BAND4+54.052904*SF+121.309755*TNDVI+2.775407*VI3+0.280446*P VI+1.494319*RVI+0.000932*DVI+45.648556*SAVI+74.062619*SLAVI+68.558577*ND VI-313.335171(6) The Pearson correlation coefficient of LAI and the volume was0.862, using both to linearly regress, the coefficient of determination was0.7423. Fitting the linear model of the two, the accuracy of volume predicted by the Remote sensing estimation of LAI was70%. In practical applications, on the condition that accuracy requirements are not high, they can be interchangeable through a linear relationship. |
PDF Full Text Request