【作者】 张静；

【导师】 张丽； 梁勇；

【作者基本信息】 山东农业大学 ， 农业信息化（专业学位）， 2017， 硕士

【摘要】 研究草地生物量和环境因子的关系以及准确预测草地生物量是草地资源科学利用的一项重要课题。准确预测草地的生物量能掌握生长季草地的生长规律,有计划地预报草地的生长状况,对草地资源的保护和规划利用,放牧强度及载畜量的确定具有重要的指导意义。目前生物量预测方面的研究很少,相关研究人员主要采用积分回归的方法建立回归方程来预测生物量,通过对历史数据拟合趋势线,并按照趋势对未来某一时刻的草地资源状况进行预测,难以全面准确预测草地生物量。本文为了预测未来生物量的分布状况,提供动态的草场长势信息和草场生物量数据,首先构建不同时间序列模型进行精度比较,最终选取最优时间序列模型建立草场生物量动态预测模型,利用支持向量机模型(Support Vector Machine,SVM)生成青海湖流域2015年生长季预测生物量数据集(8天,500米),预测分析青海湖流域生物量的空间分布格局及其变化趋势特征。本文主要的研究工作和结论如下:(1)首先提取青海湖流域站点数据构建模型,以青海湖流域2000-2014年归一化植被指数(Normalized Difference Vegetation Index,NDVI)作为输入数据(8天,500米),利用时间序列模型生成预测NDVI。选取多个时间序列模型:趋势移动平均法、二次指数平滑法、差分自回归滑动平均模型(ARIMA)预测2015年NDVI的空间分布并进行精度比较。基于以上模型的构建与对比,发现ARIMA模型有更高的精度,能够真实反映青海湖流域2015年NDVI分布状况。(2)基于预测NDVI数据集,结合气候数据建立回归优化模型并进行精度分析,选取不同修正方案,获取最优NDVI修正方案生成面上预测NDVI数据集。实验分为三个方案(方案一:NDVI利用ARIMA模型生成预测NDVI;方案二:基于方案一经回归模型优化生成新的预测NDVI;方案三:利用ARIMA模型分别生成预测的NDVI、温度和降水,三者经回归模型优化生成新的预测NDVI)。实验发现:方案三精度(R2=0.8912,RMSE=0.0263)高于方案二(R2=0.8867,RMSE=0.0267)和方案一(R2=0.8879,RMSE=0.0271)。(3)基于优化的预测NDVI面数据,利用SVM生成青海湖流域预测生物量数据集(2015年,8天,500米)。生物量的分布主要集中在青海湖的北部、南部以及流域中部的高寒草甸地区,2015年研究区预测生物量均值主要分布在120~180g/m~2。更多还原

【Abstract】 It is one of the most important issues for scientific use of grassland resources to study the relationship between grassland biomass and environmental factors,and to predict grassland biomass accurately.Accurate prediction of grassland biomass can control the growth of grassland in growing season.Forecast the growth of grassland systematically,which is of great significance for grassland resource protection and planning utilization,grazing intensity and stocking capacity.At present,there are few studies on biomass prediction,mainly using the method of integral regression,through the historical data to fit the trend line,and in accordance with the trend of a future time to predict the status of grassland resources,this method is difficult to predict grassland biomass fully and accurately.In order to forecast the distribution of future biomass,to provide dynamic grassland growth information and grassland biomass data,to realize the healthy development of grassland ecosystem and the scientific management of animal husbandry,in this paper,firstly,we compared different time series models.Then,we used support vector machine(SVM)model developed the optimal aboveground biomass model for Qinghai Lake Basin and generate the biomass data with 500 meter resolution and 8 day interval in 2015.Eventually,we analyzed the grassland trend in Qinghai Lake basin.The main content and results of this paper are as follows:(1)We were using the synthetic NDVI series(8 days,500 meters,2000-2014)as the input data,and using the time series model to generate the prediction NDVI(8 days,30 meters,2015).Select multiple time series model: including trend moving average method,second exponential smoothing method(ES),self-adaptive filtering method,modified exponential curve method and auto regressive integrated moving average(ARIMA)were used to predict NDVI in 2015.Based on the above models,it was found that the ARIMA model had higher precision and can reflect the distribution of NDVI in Qinghai Lake basin in 2015.(2)Based on the forecast NDVI dataset,the regression model was established and the accuracy analysis was carried out.The NDVI data sets were obtained by obtaining the optimal NDVI correction scheme.Based on the forecasted NDVI dataset and the climate data,we established a regression optimization model and compared.Different correction schemes were selected to obtain the NDVI dataset.There were three schemes :1)we use ARIMA model to generate prediction NDVI;2)We use ARIMA model to generate prediction NDVI,and then optimized by regression model to generate new prediction NDVI 3)We use ARIMA model to generate prediction NDVI,precipitation and temperature data,then optimized by regression model to generate new prediction NDVI.The results show that the accuracy of scheme III(R2 = 0.8912,RMSE = 0.0263)is higher than that of scheme II(R2 = 0.8867,RMSE = 0.0267)and Scheme 1(R2 = 0.8879,RMSE = 0.0271).(3)Based on the optimized NDVI surface data,we generated the forecast biomass data set of Qinghai Lake Basin(2015,8 day,500m)by using SVM.The distribution of biomass is mainly concentrated in the northern part of Qinghai Lake,the southern part and the alpine meadow area of the basin.The forecasting biomass in the study area in 2015 is mainly distributed in 120 ~ 180 g / m~2.更多还原