EGI OpenIR
基于 WOFOST 模型的新疆棉花遥感估产研究
Alternative TitleStudy on cotton yield estimation in Xinjiang using remote sensing integrated with coupled WOFOST model
张杰云
Subtype博士
Thesis Advisor张清凌 ; 包安明
2019-06-30
Degree Grantor中国科学院大学
Place of Conferral北京
Degree Discipline理学博士
Keyword遥感 旱情 估产 WOFOST 模型 同化算法 Remote Sensing Drought Yield Estimation WOFOST model Assimilation Algorithm
Abstract棉花是我国重要的经济作物。及时准确的估测区域尺度棉花产量,对于合理安排生产,把握市场,增加农民收入,抵御或降低自然灾害等都具有重要的经济和科学意义。作物生长模型通过数学抽象的方式实现对作物生长发育生理过程的定量化描述,成为农作物长势监测、产量估测的重要手段,然而基于点尺度的作物生长模型受限于作物遗传参数、田间管理措施等信息的区域化获取手段匮乏,难以满足区域尺度的应用需求。作物模型本质上是对植株生长发育过程的数学模拟,其初始参数、物质能量转化的数学模拟过程等均存在不确定性,从而对模拟结果的准确性产生影响。 随着遥感信息在农业领域应用的深入,基于遥感技术获取区域尺度作物冠层状态,能够支撑作物生长模型的区域化应用。 通过遥感手段获取的农作物冠层状态信息作为对作物生长状态的客观反映,能够对作物模型的模拟过程进行约束,通过对模拟轨迹的矫正,使模拟结果更加接近实际生长状态。遥感信息与作物生长状态的耦合,能够最大程度的利用二者优势,实现区域尺度作物生长过程的客观表达。数据同化方法在农业领域的应用,为这一耦合过程提供了一个有效方法。针对国内外对遥感信息与 WOFOST 模型耦合研究中区域适用性及同化变量选取的问题,结合新疆棉花产量的区域化估算需求,开展了基于WOFOST 模型的新疆棉花遥感估产研究。主要研究成果如下:(1)基于棉田实验样点的实测光谱反射率数据与实测调查所获取的 LAI、地上生物量、土壤含水量、单产等地面数据,通过对作物遗传参数的全局敏感分析,选取了 SLATB2、 SLATB1、 TSUM1、 SPAN、 TMPFTB2 作为敏感参数集合,结合模型默认值和参考文献,标定 WOFOST 模型初始参数开展模拟,结果表明,模拟初花天数超过实测天数 5 天,相对误差约 7%,成熟天数超过实测天数 3 天,相对误差 2.1%;地上总干物质量及各器官干物质量的模拟方面,相对误差均小于 16%,地上总干物质量相对误差为 6.24%; LAI 模拟方面,模型对整个生长季LAI 模拟结果的线性拟合方程斜率为 0.998,具有较高精度;土壤含水量模拟方面,土壤含水量的模拟值与实测值拟合较好,但模拟值高于实测值,初步完成了WOFOST 模型在新疆棉花模拟的“本地化”工作。(2)选取了新疆棉花主产区沙湾县和阿拉尔市为研究区,综合运用时间序列谐波分析、粗糙集约简方法,通过基于多时相遥感影像的多特征优选,提取了研究区范围内棉花种植的空间分布情况,其生产精度达到 94.92%,用户精度达到 93.75%。(3)基于近红外(Near Infrared, NIR)和红光(Red)波段反射率特征,分析归纳了不同土壤含水量条件在 NIR-R 光谱中间中的分布规律,提出了一种新型的土壤含水量反演方法。在此方法基础上,对研究区土壤含水量进行遥感反演,结果表明,估算的沙湾县与阿拉尔市的土壤水分含量与实测值之间的 R2分别为 0.7186和 0.6776,均方根误差均小于 2%,可以满足模型同化对区域化土壤含水量产品的需求。另外,采用 NDVI 与 LAI 之间的指数关系模型,结合 Landsat-8 OLI 遥感影像,生成了生长季内多个时相棉田 LAI 产品,经地面样点实测数据验证,反演 LAI 与实测值之间的 RMSE 为 0.47 m2/m2。区域 LAI 整体格局符合棉花生长规律。反演结果可以作为同化变量在 WOFOST 模型模拟棉花生长过程中起到矫正作用。(4)在研究区棉花种植范围基础上,采用 SCE-UA 同化算法和 PF 同化算法,设计了两种同化方案以实现遥感反演的土壤含水量及 LAI 产品与本地化后WOFOST 模型的耦合,生成棉花生育期内的土壤含水量和 LAI 同化值,并与田间实测数据进行对比,验证同化精度。在基于 SCE-UA 算法的同化方案下,沙湾县和阿拉尔市的同化土壤含水量与实测值的 R2 分别达到 0.77 和 0.74, LAI 同化值与实测值的 R2 分别达到 0.73 和 0.71,同化产量与实际产量的 R2 分别为 0.56和 0.53, RMSE 分别为 418.57kg/ha 和 469.54kg/ha;在基于 PF 算法的同化方案下,沙湾县和阿拉尔市的同化土壤含水量与实测值的 R2 分别达到 0.75 和 0.71,LAI 同化值与实测值的 R2 分别达到 0.68 和 0.62,同化产量与实际产量的 R2 分别为 0.54 和 0.49。从区域尺度同化结果来看,基于 SCE-UA 同化方案下,沙湾县和阿拉尔市平均同化单产分别为 5634kg/ha 和 5487kg/ha;基于 PF 同化方案下,沙湾县和阿拉尔市平均同化单产分别为 5843kg/ha 和 5365kg/ha,与两研究区的产量统计数据相比,基于 SCE-UA 同化方案的产量估算更准确。
Other AbstractCotton is an important economic crop in China. Timely and accurate estimation ofcotton yield at regional scale is of great economic and scientific significance for rationalarrangement of production, grasping the market, increasing farmers'income andresisting or reducing natural disasters. Crop growth model can quantitatively describethe physiological process of crop growth and development through mathematicalabstraction, and it has become an important means of monitoring crop growth andestimating crop yield. However, the point-scale crop growth model is limited by thelack of regionalized access to information such as crop genetic parameters and fieldmanagement measures, which makes it difficult to meet the application needs at theregional scale. With the further application of remote sensing information in agriculture,the regional application of crop growth model can be supported by the acquisition ofcrop canopy status at regional scale based on remote sensing technology. Crop modelis essentially a mathematical simulation of plant growth and development process.There are uncertainties in its initial parameters and mathematical simulation process ofmaterial and energy conversion, which leads to the uncertainty of simulation results.Crop canopy status information obtained by remote sensing as an objective reflectionof crop growth status can restrict the simulation process of crop model. By correctingthe simulation trajectory, the simulation results are closer to the actual growth status.The coupling of remote sensing information and crop growth state can maximize theiradvantages and achieve objective expression of crop growth process at regional scale.The application of data assimilation in agriculture provides an effective method for thiscoupling process. Aiming at the problem of regional applicability and assimilationvariable selection in the coupling research of remote sensing information and WOFOSTmodel, combined with the regionalized estimation demand of cotton yield in Xinjiang,the research on cotton remote sensing estimation based on WOFOST model wasprocessed. The main research results are as follows:(1) Based on the measured spectral reflectance data of cotton field experimentalsamples and the ground data of LAI, aboveground biomass, soil water content and yieldobtained from the survey, SLATB2 and SLATB1 were selected through globalsensitivity analysis of crop genetic parameters. TSUM1, SPAN, TMPFTB2 as sensitiveparameter sets. Combined with the model default values and references, the initial parameters of the WOFOST model were calibrated and the cotton growth process wassimulated. The results showed that the simulated initial flowering days exceeded themeasured days by 5 days, the relative error was about 7%, the mature days exceededthe measured days by 3 days, and the relative error was 2.1 %; the relative error of thetotal dry matter quality and the dry matter quality of each organ, the relative error isless than 16%, and the relative error of the total dry matter mass on the ground is 6.24%;in the LAI simulation, the slope of the linear fitting equation of the model for the LAIsimulation results of the whole growing season It has a high precision of 0.998. In termsof soil water content simulation, the simulated and measured values of soil watercontent are well fitted with the measured values, but the simulated values are higherthan the measured values, and the “localization” of the WOFOST model in Xinjiangcotton simulation is initially completed.(2) Shawan County and Alar City, the main cotton producing areas in Xinjiang,were selected as the research area. The time series harmonic analysis and rough setreduction method were used to extract the spatial distribution of cotton using multifeature optimization based on multi-temporal remote sensing images. The results showthat the production accuracy is 94.92% and the user accuracy is 93.75%.(3) Based on the characteristics of near-infrared (NIR) and red-light (Red)reflectance, the distribution of different soil water content in the middle of NIR-Rspectrum is analyzed, and a new soil moisture monitoring index is proposed. Based onthis method, the soil moisture content in the study area was inverted. The results showthat the R2 between the estimated soil moisture content and the measured value ofShawan County and Alar City are 0.7186 and 0.6776, respectively, RMSE are both lessthan 2%, which can meet the needs of model assimilation for regionalized soil watercontent products. In addition, using the exponential relationship model between NDVIand LAI, combined with Landsat-8 OLI remote sensing image, the LAI products ofmultiple phase cotton fields in the growing season were generated, verified by themeasured data of ground samples. the RMSE between inversion LAI and measuredvalues is 0.47 m2/m2. The overall pattern of regional LAI is consistent with the law ofcotton growth. The inversion results can be used as an assimilation variable to correctthe cotton growth process in the WOFOST model.(4) Based on the cotton planting range in the study area, combined with remotesensing inversion of soil water content and LAI products, SCE-UA assimilationalgorithm and PF assimilation algorithm were used to design two assimilation schemes to coupling with WOFOST model. The soil water content and LAI assimilation valuesduring the growth period of the cotton were generated and compared with the fieldmeasured data to verify the assimilation accuracy. In the assimilation scheme based onSCE-UA algorithm, the R2 between assimilated soil water content and measured valuesof Shawan County and Alar City are 0.77 and 0.74, respectively, and the R2 betweenLAI assimilation value and measured are 0.73 and 0.71, respectively. The R2 betweenassimilation yield and actual yield are 0.56 and 0.53, respectively, and RMSE are418.57kg/ha and 469.54kg/ha, respectively. In the assimilation scheme based on PFalgorithm, the R2 between assimilated soil water content and measured values ofShawan County and Alar City are 0.75 and 0.71, respectively, and the R2 between LAIassimilation value and measured are 0.68 and 0.62, respectively. The R2 betweenassimilation yield and actual yield are 0.54 and 0.49, respectively. The results ofregional scale assimilation show that, the average assimilation yields of Shawan Countyand Alar City are 5634kg/ha and 5487kg/ha, respectively, based on the SCE-UAassimilation scheme. The results of regional scale assimilation show that, the averageassimilation yields of Shawan County and Alar City are 5843kg/ha and 5365kg/ha,respectively, based on the PF assimilation scheme. Compared with the statistic yield ofthe two study areas, the yield estimation based on the SCE-UA assimilation scheme ismore accurate.
Subject Area地图学与地理信息系统
Language中文
Document Type学位论文
Identifierhttp://ir.xjlas.org/handle/365004/15270
Collection中国科学院新疆生态与地理研究所
研究系统
Affiliation中国科学院新疆生态与地理研究所
First Author Affilication中国科学院新疆生态与地理研究所
Recommended Citation
GB/T 7714
张杰云. 基于 WOFOST 模型的新疆棉花遥感估产研究[D]. 北京. 中国科学院大学,2019.
Files in This Item:
There are no files associated with this item.
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[张杰云]'s Articles
Baidu academic
Similar articles in Baidu academic
[张杰云]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[张杰云]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.