| As the main body of the terrestrial ecosystem,the forest ecosystem is the largest carbon pool in the terrestrial ecosystem.Forest ecosystem plays an irreplaceable role in keeping the global carbon balance,alleviating the rise of greenhouse gas concentration in the atmosphere and regulating the global climate.As an important component of the terrestrial cycle,forest net primary productivity（NPP）can directly reflect the production capacity of forests under natural environmental conditions,and is one of the important indicators for evaluating the sustainable development of forest ecology,and is the main factor for judging the carbon sink of ecosystems and regulating ecological processes.Therefore,in the context of global warming,the study of regional and global forest NPP can not only provide a theoretical basis for terrestrial carbon cycle research,but also provide a certain scientific basis for medium-long term sustainable management plan of regional forests.In this study,Shaoguan City,Guangdong Province was taken as the study region,the Landsat-5 TM in 1997,Landsat-7 ETM+SLC-on in 2002,Landsat-7 ETM+SLC-off in 2007,Landsat-7ETM+SLC-off in 2012,Landsat-8 OLI images in 2017 and corresponding years of data of national forest resources continuous inventory are collected as the main information sources.Using three models of random forest,multiple linear regression and artificial neural network method,and estimate the NPP of the study region and analyze the spatial and temporal dynamic change and the driving factors of the NPP,and identify the low function forest in the study region based on the productivity.The study included the following three steps:Firstly,to conduct the forest NPP statistics in different forest types and species,and to conduct the forest NPP statistics for tree forest age groups.Secondly,different methods were used to select the base variables for different models,and the accuracy of different models was evaluated and validated to obtain the optimal model and estimate the NPP of forests in the study region.Second,to analyze the spatial and temporal dynamics of the NPP of forests in the study region and the driving factors.The results of the study indicated that:（1）The NPP of the forests in the study region showed an increasing trend but the forest species structure and age group structure need to be further optimized.Different forest types were statistically analyzed in the sample area of the study region,and the size of the NPP was bamboo forest>broad-leaf mixed forest>coniferous mixed forest>broad-leaf pure forest>coniferous mixed forest>coniferous pure forest>shrub forest.Different forest species were counted,and the NPP of forests in descending order was special-purpose forests,timber forests,protection forests,fuel-wood forests,and economic forests.The NPP of tree forest sample plots was divided into different age groups,and the size of the NPP was in the order of middle-aged forest,near-mature forest,young-aged forest,mature forest and over-mature forest.（2）Area statistics by forest type,the area of coniferous forest accounted for the largest proportion in 1997,and the area of coniferous forest kept decreasing from 1997 to 2002,and the area of broad-leaf forest and broad-leaf mixed forest kept increasing,and in 2002 broad-leaf forest became the the forest type with the largest area share,and from 2002 to 2017,the area of broad-leaf forest decreased and the area of broad-leaf mixed forest still increased,and in 2017,broad-leaf mixed forest became the forest type with the first area share.The area statistics were conducted by age group and the area share of young and middle-aged forest in the study region has been dominant from 1997 to 2017,and even in an increasing state.（3）Texture characteristics,topographic factors,and canopy density of images are important for modeling estimates of NPP of forests.The random forest model and back propagation neural network used variable screening based on variable importance ranking method,and multiple linear regression used stepwise regression method for variable screening,and among the selected variables,contrast and mean variables in texture features were important for each model.（4）When the three models were compared,the random forest model had higher accuracy for estimating the NPP of forests in the study region,and the coefficient of determination of the prediction accuracy ranged from 0.492 to 0.660.Based on the random forest model for spatial mapping and hierarchical processing of forest NPP in the study region,the results showed that the proportion of low grade and medium grade forest NPP in the study region was the highest in 1997,and the area of low grade and medium grade in the study region kept decreasing and the area of medium and high grade kept increasing from 1997 to 2017,but the area of low grade,low grade and medium-grade areas still accounted for the major share.（5）The study area showed an increasing trend in NPP of forests in 1997,2002,2007,2012 and2017,with mean values of 5.66 t·hm-2·a-1、7.68 t·hm-2·a-1、8.17 t·hm-2·a-1、8.25 t·hm-2·a-1 and10.52 t·hm-2·a-1.Meanwhile spatial aggregation increased from 1997 to 2017 and the center of gravity shifted from the mid-west to the southwest.The forest NPP growth area in the study region accounted for 69.86%and was mainly located in the mountainous and hilly areas.The negative growth area accounted for 30.12%and was mainly in urban areas.The spatial and temporal dynamics of NPP of forests in the study region are strongly influenced by forestry policies,anthropogenic activities and social development.（6）The stand factor had the greatest influence on the NPP of the forest in the study region,followed by the understory factor and the environmental factor.Among them,stand and environmental factors were positively correlated with forest NPP while understory factors were negatively correlated because of the competition between understory vegetation and forest vegetation for nutrients and space for growth.The NPP of the forest in the study region had the highest correlation with the average diameter at breast height and the average tree height among the stand factors,the highest correlation with the soil texture among the understory factors,and the highest correlation with the slope among the environmental factors. |
PDF Full Text Request