基于地理加权回归的草原产草量遥感估算模型研究

发布时间：2018-04-11 08:16

本文选题：地理加权回归模型 + 产草量估算　；参考：《辽宁工程技术大学》2013年硕士论文

【摘要】：产草量是草地生产力的体现,也是制定畜牧业生产管理策略的基础。利用RS技术估算产草量已经成为草场生产力研究和草场管理的重要途径之一。产草量可以通过遥感数据驱动,采用生物-物理模型来估算,也可以通过建立经验模型来估算。基于生物-物理的估算模型多用于大尺度宏观估算,但是估算精度不高；经验模型主要基于遥感数据与地面实测数据建立统计模型,简单易行。常用的经验模型一般都忽略了产草量的空间异质性,导致估算精度比较低。产草量与地理位置有关,邻近地理位置间的产草量具有空间相关性。本研究充分考虑产草量的空间相关性,将地理加权回归(Geographically Weighter Regression,简称GWR)的思想引入产草量遥感估算建模中,基于GWR模型,利用与产草量密切相关的植被指数、气象因子、草地类型等因子构建产草量估算模型。论文以三江源区为试验区,首先分析植被指数、气象因子、草地类型数据与地面实测产草量数据的相关性,利用相关性较高的因子和GWR模型构建试验区的产草量估算模型。然后利用验证点对模型进行精度评价,并与多元线性回归模型进行比较,说明已构建模型的可用性及优越性。最后以三江源区果洛藏族自治州为例,构建区域产草量GWR模型,并使用国产环境卫星影像估算整个区域的产草量。本文主要取得了如下的进展和贡献： (1)将地理加权回归的思想引入产草量估算建模过程中,构建了基于GWR的产草量遥感估算模型。试验结果证明基于GWR的产草量估算模型可以提高产草量估算模型的拟合优度,模型的拟合r2从不足0.3提高到0.8以上,估算精度明显优于多元线性回归模型,可以提高20%左右。 (2)通过试验发现与三江源区产草量遥感估算密切相关的因子包括：5-8月累积降水量、5-8月干燥度、修正型土壤植被指数(Modified Soil Adjusted Vegetation Index,简称MSAVI)。本研究基于GWR构建的三江源区产草量模型中,r2达到0.858,调整r2为0.772,可信度比较高,实地验证精度为71.61%。该估算模型的参数少,可直接由HJ影像和气象数据获得,应用方便,可直接用于三江源区实际产草量的估算。 (3)本研究在构建基于GWR的产草量估算模型的基础上,建立了利用国产环境卫星数据估算区域产草量的技术方法。以果洛藏族自治州为例,其中产草量估算GWR模型的r2为0.845,调整r2为0.727,并达到P值小于0.001的显著水平,精度为67.47%。利用2010年8月的环境卫星数据和5-8月的气象数据和已建好的模型,估算得到果洛藏族自治州的总产草量(鲜重)为3260.01×104t,产草量空间呈现自东向西逐渐减少的趋势。
[Abstract]:The yield of grass is the embodiment of grassland productivity and the basis of formulating the strategy of animal husbandry production and management.The estimation of grass yield by RS technology has become one of the important ways to study grassland productivity and grassland management.The yield of grass can be estimated by remote sensing data, biophysical model and empirical model.Bio-physical estimation models are mostly used in large-scale macroscopic estimation, but the estimation accuracy is not high, and the empirical model is mainly based on remote sensing data and ground measured data to establish statistical model, which is simple and easy to carry out.The spatial heterogeneity of grass yield is neglected in common empirical models, which leads to low estimation accuracy.The yield of grass was related to geographical location, and there was a spatial correlation between the yield of grass near geographical location.In this study, the spatial correlation of grass yield was fully considered, and the idea of geographical weighted regression was introduced into the remote sensing estimation modeling of grass yield. Based on the GWR model, the vegetation index and meteorological factors, which were closely related to grass yield, were used.Grassland type and other factors were used to estimate the yield of grass.Firstly, the correlation between vegetation index, meteorological factors, grassland type data and the measured grass yield data was analyzed, and the estimation model of grass yield in the experimental area was constructed by using the highly correlated factors and GWR model.Then the accuracy of the model is evaluated by the verification point, and compared with the multivariate linear regression model, the usability and superiority of the constructed model are illustrated.Finally, taking Guoluo Tibetan Autonomous Prefecture in Sanjiangyuan region as an example, the GWR model of regional grass yield is constructed, and the total grass yield of the whole region is estimated by using domestic environmental satellite images.The main achievements of this paper are as follows:1) the idea of geographical weighted regression is introduced into the modeling process of grass yield estimation, and the remote sensing estimation model of grass yield based on GWR is constructed.The experimental results show that the estimation model based on GWR can improve the goodness of fit of the estimation model. The fitting R2 of the model is improved from less than 0.3 to more than 0.8, and the estimation accuracy is obviously better than that of the multivariate linear regression model, and it can be increased by about 20%.(2) it was found that the factors closely related to the remote sensing estimation of grass yield in the source region of the three rivers included the accumulated precipitation in May to August and the drying degree in May to August, and the modified Soil Adjusted Vegetation Index (MSAVIX).In this study, the model of grass yield in Sanjiangyuan region based on GWR was 0.858, and adjusted to 0.772.The reliability of the model was high, and the accuracy of field verification was 71.611.The estimated model has few parameters and can be obtained directly from HJ image and meteorological data. It is convenient to be applied and can be directly used to estimate the actual grass yield in the source region of the three Rivers.3) based on the GWR model, a technical method for estimating regional grass yield using domestic environmental satellite data was established.Taking Guoluo Tibetan Autonomous Prefecture as an example, the estimated yield of grass in GWR model is 0.845, the adjusted R2 is 0.727, and the P value is less than 0.001, and the precision is 67.47.Based on the environmental satellite data of August, 2010, meteorological data of May-August and established models, the total grass yield (fresh weight) of Guoluo Tibetan Autonomous Prefecture is estimated to be 3260.01 脳 10 ~ 4 t, and the space of grass yield is gradually decreasing from east to west.
【学位授予单位】：辽宁工程技术大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：S812;P237

【参考文献】