| Forest biomass is a distinctive fundamental data for the study of forest ecosystem function, which is also an important index for evaluation of the forest carbon storage and the study of global climate change. Research methods, including remote sensing and mathematical model, were employed to estimate forest biomass in the Xiangjiang watershed. Dynamic change of forests in Xiangjiang watershed should be quickly and accurately estimated by taking the watershed as the basic unit of ecosystems. The work can provide basic data for assessment of ecosystem services such as the soil conservation, water conservation, and the carbon sink increase and carbon emission abatement under the background of rapid urbanization in the watershed. The results can also provide a scientific base for making strategies and reasonable countermeasures for the protection of ecological security of the watershed. The results should be a good guidance for the scientific development planning of Xiangjiang watershed.This study was conducted in the Xiangjiang watershed and based on the data from the 782 plots in 2009 and the remote sensing image from Landsat 5 in the same period. Forest biomass in the watershed was estimated with the inversion model using the multiple linear regression model, logistic regression model and kNN algorithm model. The inversion results of the above 3 models were validated by a cross validation method, with comparisons of the accuracy of forest biomass estimation from the three models above. The spatial characteristics of forest biomass in the watershed were analyzed based on the estimation. The main results are as follows:(1) Using the stepwise regression method, five remote sensing modeling factors, which were significantly correlated with the aboveground biomass in the watershed, were selected. These factors are normalized difference vegetation index (NDVI), two band ratio vegetation index (simple two-band ratios, SRij) SR23 and three band ratio vegetation index (simple three-band ratios SRijk) SR415, SR546 and SR625. The correlation coefficient between NDVI and the biomass was obtained the maximum of 0.681, while negative correlation were -0.282 and -0.401 for SR546 and SR625, respectively. And the correlation coefficient was 0.389 and 0.646 for SR23 and SR415, respectively. The correlation between the five remote sensing variables and measured samples biomass was statistically significant at the 0.001 level.(2) We compared the estimation of above-ground forest biomass (AGB) at sample plots scale by three forest biomass models. Based on the root mean square error (RMSE) in the biomass estimation, the order was followed as multiple linear regression model (31.55 t/ha), kNN simulation (31.87 t/ha), and logistic model (32.43 t/ha). the mean AGB by the logistic regression model was 64.52 t/ha, which was more close to the measured biomass. Some predicted results from multiple linear regression model were negative. The logistic model and kNN model were more suitable for the forest biomass estimation.(3) kNN method was more suitable for mapping the spatial distribution of forest biomass in Xiangjiang watershed. The method can estimate more accurate AGB in the watershed, especially the AGB is null on the location of Xiangjiang River.(4) AGB in the study area showed obvious spatial distribution. The estimation of AGB averaged by 48.79 t/ha, while the high value area (120-140 t/ha) mainly distributed in the northeast region of the lower reaches of Xiangjiang River. And the middle value area (80-100 t/ha) mainly occured in the southwest region of the upper reaches, while the low values (1-20 t/ha) are mainly distributed in the middle reaches. Remote sensing inversion showed a clear mapping the change trend of the higher values of AGB in the upper and the lower reaches, and the lower values of AGB in the lower reaches of Xiangjiang watershed, while these results were very difficult to get with the forest resource inventory method. |
PDF Full Text Request