基于遥感数据的盐渍农田四维水盐信息提取

发布时间：2018-02-26 02:31

  本文关键词： 土壤 水分 盐分 高光谱 参数反演　出处：《武汉大学》2016年博士论文　论文类型：学位论文

【摘要】：农田水分的高效利用对于解决我国水资源短缺和粮食安全问题非常重要,并且这一需求在干旱、半干旱地区尤其突出。然而已有的有关农田水盐运移的研究多为侧重于土壤的控制实验或模拟,在作物生长情况下,受气象条件、土壤条件以及人类活动的多重影响,农田水盐状况在时间和空间上都会发生变化,即农田四维水盐信息发生变化,这些变化是干旱、半干旱地区盐渍农田节水灌溉与粮食增产的重要依据。但考虑到传统水盐监测技术耗时长、花费大等不足,目前仍然难以在区域上对这一时空变化进行准确描述。因此有必要开展盐渍农田水盐时空变化的研究,充分利用遥感数据,发展适用于区域尺度的提取四维水盐信息的有效方法。本文首先研究了高光谱遥感定量提取表层土壤水盐信息的方法,通过研究发现光谱数据前处理(如归一化变换、导数变换等)对于提取敏感光谱信号非常重要；选择少数敏感波段建立定量提取土壤水盐的模型不仅精度较高,而且模型简单稳定,物理意义明显。通过基于主成分分析的评价系统(PCAr),本文选择了对土壤水盐敏感的14个波段,即440、540、570、1390、1430、1460、1740、1870、1900、1940、2010、2270、2350和2410rim。通过综合分析土壤水分、盐分、质地三个主要因素和光谱的关系,发现土壤盐分信息提取受土壤水分和质地的影响较大,并建立了土壤水分和盐分的联合提取模型,基于提取的水分信息获取盐分信息。与直接提取盐分信息相比,联合提取模型可以显著提高土壤盐分信息的提取精度。具体而言,针对室内土壤样本,直接估计土壤盐分时,决定系数R2为0.47,相对均方根误差rRMSE为0.380；采用联合模型后,精度有明显提高,0.63R20.95、0.132rRMSE0.366。以上遥感技术提取的水盐数据可以补充野外样本数据,从而在区域尺度上较准确的提取土壤水盐信息。利用土壤表层和根区水盐的关系,可以研究水盐的空间变异关系。但是这种基于空间分析的方法仅能提取三维土壤水盐信息,而通过水动力学模型可以模拟土壤水分在包括时间维度上的四维变化情况。这种模拟方法会受水动力参数选取的影响。本文提出了一种全局参数反演模型,通过反演全局优化因子,修正水动力参数偏差,并控制参数在允许范围内变化。本文共建立了三个反演模型,GW5、GW10和GW30,全局优化因子的数量分别为5、10和30。相对于正演模型预测的剖面水分精度R2=0.23,本文建立的GW5、GW10和GW30反演模型的预测精度R2分别为0.313、0.765、0.998。使用修正的赤池信息量准则AICc综合评价模型的复杂度和模型拟合的优良性,避免过拟合情况,发现结构最简单的GW5模型精度最高(AICc=-467.367),这主要是因为该研究区域质地比较均一,水动力参数的异质性不明显。根据数据情况的不同,基于GW5模型的结构建立一组模型S GW5,并通过两年的野外实验研究和验证,发现该组模型在研究区域的实用性良好。其中仅使用遥感数据预测的水分作为目标函数可建立V(RS)模型,使用2013年的野外数据校正V(RS)模型,R2=0.892,RMSE=0.01,使用2014年的野外数据进行验证,R2=0.738, RMSE=0.034。结果表明结合遥感数据与S GW5反演模型,预测作物生长各个阶段的土壤水分数据的精度显著高于正演模型。本文建立的土壤水盐的联合提取模型,可以广泛运用于定量提取易产生混合光谱的各种物质。通过遥感数据与全局参数反演模型结合,本文提出了区域农田四维水盐信息提取的方法,可广泛运用于研究不同环境下土壤水盐信息的时空变化。
[Abstract]:The efficient use of water is very important to solve the shortage of water resources and food security in our country, and the demand in arid and semi arid area in particular. However, research on farmland water and salt transport have been mostly focused on soil control experiments or simulation, in the growth of crops, affected by weather conditions, multiple effects of soil conditions and human activities, water salt status will change both in time and space, namely the farmland four-dimensional water salt information changes, these changes are an important basis for drought, half arid area of salinized farmland water-saving irrigation and crop yield. But considering the traditional water salt monitoring techniques are time-consuming, lack of cost so, it is still hard to in the region of the spatial and temporal changes were described accurately. So it is necessary to study the spatial variation of the saline water salt, make full use of the remote sensing data, development The effective method for information extraction of water and salt for four-dimensional regional scale. This paper studies the quantitative method of hyperspectral remote sensing information extraction of water and salt in surface soil, through research found that the spectral data pretreatment (such as normalized transformation, derivative transformation etc.) for the extraction of sensitive spectral signal is very important; not only the precision of the model is high sensitive bands to establish a quantitative extraction of soil water and salt, and the model is simple and stable, clear physical meaning. Through the evaluation system based on principal component analysis (PCAr), this paper chose to soil water salt sensitive 14 waves, namely 4405405701390143014601740187019001940201022702350 and 2410rim. through the comprehensive analysis of soil moisture, salinity, the relationship between three main factors and the texture spectrum. Found that the soil salinity information extraction is affected by soil moisture and texture, and the establishment of the soil water and salt The combined extraction model, water information extraction based on information get salt. Compared with the direct extraction of salt, combined with extraction model can significantly improve the extraction accuracy of soil salinity information. Specifically, in the indoor soil samples, direct estimation of soil salinity, determination coefficient R2 is 0.47 and the relative root mean square error of rRMSE is 0.380; the combined model, accuracy is obviously improved, the water salinity data above 0.63R20.95,0.132rRMSE0.366. remote sensing technology to extract can complement the field sample data, information extraction of soil water and salt and at the regional scale accurately. The relationship between soil surface and soil water salt spatial variability, relationship study on water and salt. But the method of spatial analysis based on can only extract three-dimensional information of soil water and salt, and the water dynamics model can simulate the soil moisture in the four-dimensional time dimension variable including Situation. This method will be affected by the selection of hydrodynamic parameters. This paper presents a global parameter inversion model, by inversion of global optimization factor, correction of hydrodynamic parameter deviation, and control parameters in the allowable range. This paper has established three inversion model, GW5, GW10 and GW30, the number of global optimization the factors were 5,10 and 30. R2=0.23 moisture profile accuracy compared with the forward prediction model, this paper established the GW5 complexity, excellent prediction accuracy of R2 GW10 and GW30 0.313,0.765,0.998. respectively. The inversion model using the Akaike information criterion AICc comprehensive evaluation model and modified model fitting, avoid over fitting, found the most simple structure GW5 model of the highest accuracy (AICc=-467.367), this is mainly because the study area is relatively uniform texture and heterogeneity of hydrodynamic parameters is not obvious. According to the data situation is different, The structure of GW5 model to establish a set of model S based on GW5, and through the field experiment and verification for two years, found that the model group in good practical research area. The only use of remote sensing data to predict water as objective function can be established V (RS) model, V correction field data using the 2013 (RS) model R2=0.892, RMSE=0.01, field data verify the use of the 2014, R2=0.738 RMSE=0.034., the results show that the combination of remote sensing data and S GW5 inversion model, soil moisture data prediction of crop growth stages were significantly higher than the accuracy of the forward model. The combined extraction model of soil water and salt in this paper, can be widely used in quantitative extraction of various substances easily mixed spectrum. Through the combination of remote sensing data and global parameter inversion model, this paper puts forward a method of extracting four regional farmland water salt information, can be widely used in the study The temporal and spatial variation of soil water and salt information in the same environment.

【学位授予单位】：武汉大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：S156.4;S127
，

本文编号：1536188