分类回归树算法在土壤水分估算中的应用

针对特征复杂的大尺度区域作物根区土壤水分信息获取难的问题,提出了一个基于CART算法的土壤水分估算模型。模型综合了多源环境变量信息,通过对特征复杂的数据空间进行划分,得到一系列特征单一、建模容易的数据子空间,通过对子空间的土壤水分估算实现全局范围的土壤水分信息获取。采用TVDI、AMSR2土壤水分产品、实测数据3种方式对中国北方土壤水分估算结果进行验证。验证结果表明,虽然TVDI和AMSR2 2种方法的验证效果并不理想,但是不同深度层的站点实测值和估算值之间却有着良好的关系,相关系数均大于0.4且极显著,平均相对误差均小于7.3%,均方根误差均小于0.05m3/m3。该研究说明CART算法能够有效应用于估算大尺度区域的作物根区土壤水分。     

Sentinel-1和Sentinel-2协同反演地表土壤水分

土壤水分是水文循环、生态环境、气候变化等研究中的关键参数,获取高分辨率长时间序列的土壤水分信息对农业管理、作物生长监测等具有重要的意义,同时也是研究的难点。基于时间序列(2019年至2020年)的Sentinel-1雷达数据和Sentinel-2光学数据,构建了地表土壤水分的雷达与光学数据协同反演模型,即裸土条件下地表土壤水分的变化检测方法,并利用归一化植被指数对植被影响进行校正,实现了青藏高原多年冻土区(五道梁)100 m空间分辨率的土壤水分反演。与地面实际观测的土壤水分进行对比验证,结果表明土壤水分反演结果与地面实测数据的相关系数介于0.672与0.941之间,无偏均方根误差介于0.031 m³/m³与0.073 m³/m³之间,土壤水分变化与区域降水事件和特征密切相关,验证了本文提出的考虑植被物候的变化检测方法在地势平坦、植被稀疏的青藏高原地区具有极高的适用性。     

像元尺度土壤水分监测网络及其对L波段土壤水分产品的初步验证结果

土壤水分是作物生长、地—气水热交换及全球水循环过程中的关键变量,对于旱情监测、水文陆面过程及气候变化的研究具有重要的意义。被动微波遥感凭借对于土壤水分的敏感性已经成为监测土壤水分的主要手段。研究中针对吉林省农田下垫面,利用土壤水分传感器网络监测数据,开展了SMAP(Soil Moisture and Active and Passive)和SMOS(Soil Moisture and Ocean Salinity)被动微波土壤水分产品的真实性检验研究,得出了以下结论:(1)与实测数据相比较,SMOS L3(升降轨)和SMAP L3被动微波土壤水分产品存在低估现象,伴随降雨事件会出现高于实测土壤水分的情况;两种被动微波土壤水分产品的无偏均方根误差(unRMSE)都大于0.07m3/m3,但SMAP L3被动微波土壤水分产品数据的ubRMSE略低,为0.078m3/m3;(2)由于L波段的感应深度要浅于传感器的探测深度5cm,降雨后土壤表层的变干现象导致土壤水分的垂直不均匀性,这是SMOS和SMAP被动微波土壤水分产品低估土壤水分的原因之一;(3)SMOS与SMAP亮温分布范围对比结果表明:由于电磁射频干扰(RFI)的影响,RFI对于SMOS的影响更为严重,这或许是SMOS土壤水分产品的RMSE高于SMAP被动微波土壤水分产品的原因。     

基于多频被动微波遥感的土壤水分反演—以黑河上游为例

土壤水分是陆地生态系统中最重要的组成部分,如何有效地得到高精度的土壤水分产品成为当前研究最为关注的问题。被动微波遥感具有监测面积大、重访周期短、对土壤水分敏感等优点,成为反演土壤水分最有潜力的方式。基于SMOS（Soil Moisture and Ocean Salinity）和AMSR2（The Advanced Microwave Scanning Radiometer-2）数据,通过研究L波段与C波段融合亮度温度在土壤水分反演中的潜力,发展多频率土壤水分反演算法,并对黑河上游4个像元开展土壤水分反演研究。结果表明：①利用L/C组合亮温反演结果与实测数据较为吻合,长时间内变化趋势一致,相关系数为0.841,均方根误差为0.063 m³/m³。②通过与SMOS和AMSR2官方土壤水分产品比较发现,AMSR2土壤水分产品存在明显的低估,SMOS土壤水分产品缺失值较多,无法得到较为完整的土壤水分时间序列;利用L/C多频率组合反演得到的结果明显优于官方土壤水分产品。融合L与C波段亮温数据,可有效提高反演土壤水分精度,实现高精度土壤水分的获取。     

北方典型区域遥感和模型土壤水分产品的对比及评估

多种土壤水分产品的综合评估有助于了解产品的特性与差异,对产品的算法改进及合理应用有重要意义。从空间分布,站点评估,土地覆盖类型及干湿分类等多方面对2010—2011年中国北方典型区域遥感土壤水分产品(SMOS＿L3、AMSR-E＿LPRM、ESACCI v04.5)和模型土壤水分产品(ECMWF＿ERA5、GLDAS＿Noah v2.1、GLDAS＿CLSM v2.2)进行差异性及适用性分析,并从多角度讨论了影响土壤水分产品准确性的可能原因。结果表明：(1)在年尺度上,各产品均能有效表征西部干旱区土壤水分分布情况。在季节尺度上,ESACCI和3种模型产品夏秋季土壤水分较高且空间分布相似。(2)在站点评估方面,ERA5产品整体性能最优,平均相关系数R值最高为0.582,无偏均方根误差ubRMSE最低为0.045 m³/m³。模型产品在ubRMSE和R方面均优于遥感产品,能有效刻画站点观测的动态特征,但容易出现干湿偏差。ESACCI产品在遥感产品中准确性最高。AMSR-E与观测值之间的偏差最小(-0.015 m³/m     

祁连山区不同植被类型上的SMOS遥感土壤水分产品质量评估

为了分析SMOS遥感土壤水分产品在祁连山区的真实性和可靠性,利用祁连山区内布设于7种主要植被类型上的34个实测站点的实测土壤水分数据对其进行质量评估。首先挑选与实测值相对应的SMOS数据,进而依次计算每个站点上遥感产品与实测值的相关系数R、Bias和均方根误差RMSE,从而得到SMOS数据在不同植被类型上不同尺度(年和季节)的反演精度。结果表明:SMOS遥感土壤水分产品在研究区内是可信的,但低估了研究区土壤水分值,且未能达到产品预期目标0.04m~3/m~3。SMOS产品对于植被辐射反演效果好于土壤辐射反演,导致其在植被覆盖度越高的区域与实测值的拟合程度越高。SMOS产品在湿润条件下性能优于干旱条件,在变异性小的地区性能优于变异性大的地区。在季节尺度上,SMOS遥感产品与实测值拟合程度在夏、秋两季远好于春季。     

AMSR2土壤水分产品在黑河流域中上游的验证

本项工作在黑河流域中上游区域内,利用地下4cm深度的地面实测土壤水分数据验证了2012年7月至2014年12月期间AMSR2的两种算法产品——日本宇航局标准算法土壤水分产品(JAXA产品)和阿姆斯特丹自由大学联合美国宇航局开发的陆表参数反演模型算法土壤水分产品(LPRM产品)。验证结果显示:与地面实测数据相比,所有验证像元上两种土壤水分产品的均方根误差RMSE(Root Mean Square Error)普遍超过了0.1m~3/m~3。JAXA产品动态变化范围较小,升轨产品的总体精度略高于降轨,相比地面实测数据均存在明显的低估,在冻季与实测数据比较接近。LPRM产品动态范围较大,降轨产品在冻季不可用,在未冻季升轨产品精度高于降轨,相比地面实测数据有高估的倾向。同时,还进一步讨论并分析了两种算法对土壤温度和植被的不同处理方式对土壤水产品精度的可能影响,指出了算法可能的改进方向。     

基于Triple-Collocation方法的微波遥感土壤水分产品不确定性分析及数据融合

微波遥感可以获取大范围的地表土壤水分信息,以及由此得到全球尺度的土壤水分产品。但由于传感器观测配置和反演方法等诸多因素的影响,使得不同的土壤水分产品在精度和可靠性方面存在差异。基于Triple-Collocation(TC)方法,在青藏高原那曲地区的0.25°×0.25°和1.0°×1.0°两个空间尺度上对AMSR2、SMAP和SMOS 3种土壤水分遥感产品进行不确定性分析,开展基于随机误差的数据融合算法研究。研究结果表明:不同遥感产品间的随机误差在空间分布上存在显著的不一致性,使得应用传统的算术平均方法进行数据融合不具有普适性。基于此不确定性,对3种产品配赋相应的权重进行融合,相比于3种土壤水分原始数据集,融合产品不仅具有更丰富的数据量,也会对数据精度有所改善。当遥感产品间的随机误差接近时,等权重和优化权重的融合结果非常接近;当遥感产品间的随机误差差异较大时,基于不确定性的数据融合方法相比等权重方法可以明显的提高融合数据的精度。     

基于Sentinel-1雷达数据的青藏高原地区土壤水分反演研究

土壤水分是地—气能量交换和全球水循环的重要参数之一,也是水文、气象、农业等研究中的关键参数。高空间分辨率的土壤水分在探讨区域水文过程、生态环境保护及农业水资源管理等方面具有重要意义。基于Sentinel-1雷达数据发展了青藏高原地区高空间分辨率土壤水分反演算法,并获取了区域尺度空间分辨率为20 m的土壤水分。该算法首先基于地面数据、Sentinel-1雷达数据和MODIS归一化植被指数对水云模型进行了参数优化,其次利用优化后的水云模型构建了模拟数据库,利用人工神经网络算法对模拟数据进行训练,构建了基于神经网络的土壤水分反演算法。为了检验该算法,利用Sentinel-1雷达数据反演了青藏高原站点区域土壤水分值,并使用站点实测土壤水分数据对其进行了验证。结果表明：土壤水分反演值与站点实测值有良好的一致性,其相关系数为0.784—0.82,均方根误差为0.052 m³/m³—0.064 m³/m³。土壤水分反演值在时间序列上能够捕捉到土壤水分实测值的变化趋势。该研究可为青藏高原地区高空间分辨率的土壤水分监...     

基于理论参数空间的遥感蒸散模型构建及验证

遥感数据提供了估算区域蒸散的重要数据源,基于VFC/LST参数空间构建了改进的遥感蒸散模型(EML)。在EML中,每个像元构建其专属的理论参数空间,并基于水分亏缺指数(WDI)实现目标像元的蒸散估算。使用SMACEX实验观测数据对模型进行评估,区域尺度的模型评估使用来自于Landsat 7ETM+的遥感参数。评估结果表明EML可以实现可靠的区域通量估算。潜热通量估算的平均绝对误差(MAD)和均方根误差(RMSD)分别为62.20和74.17 W/m~2,感热通量估算的MAD和RMSD分别为43.37和49.02 W/m~2。使用传统的梯形参数空间模型(TIM)与EML进行对比,结果表明EML模型克服了TIM模型的主观性和不确定性。研究结果表明:EML模型能够实现可靠的蒸散估算,且优于传统的梯形参数空间模型TIM,并适用于非均匀下垫面的蒸散估算。     

Application of Remote Sensing Data in WRF-Chem Model Simulating Sandstorm

Sand emission process of sandstorm is a fundamental part of sand-dust cycle. Sand emission process simulating accuracy plays a crucial role in correctly simulating sand transporting and settling process. As one of the most widely used sandstorm models, WRF-Chem (Weather Research and Forecasting with Chemistry) is used to simulate the sandstorm happened during March 26 and March 28, 2018 in northern China in this study. It is reported that uncertainties in underlying surface and soil moisture initial status in WRF-Chem can lead to great bias in its simulating results. Remote sensing products like land cover and soil moisture products have been widely accepted for their higher accuracy, which provides an opportunity for WRF-Chem simulating sandstorms better. Therefore, to examine the effects of initial field uncertainties on sandstorm simulating, we simulated a sandstorm using WRF-Chem by replacing the underlying surface and soil moisture initial field with new version soil database, MODIS (Moderate Resolution Imaging Spectroradiometer) land cover products and AMSR2 (the Advanced Microwave Scanning Radiometer 2) soil moisture products. Four experiments were carried out, including a control experiment and three contrast experiments. The three contrast experiments are organized by only replacing the soil moisture initial field, only replacing land cover and soil texture, and replacing both. After replacing traditional initial field with remote sensing data, the simulation accuracy all has improved. Among the three contrast experiments, replacing all three parameters (land cover, soil texture and soil moisture) has the greatest improvement: the correlation coefficient of PM10 increases by 0.30, the average deviation reduces by 31.18 μg/m³, the root mean square error reduces by 21.7 μg/m³, the correlation coefficient of AOD (Aerosol Optical Depth) improves by 0.14, the average deviation reduces by 0.29, the root mean square error reduces by 0.18. The contrast experiment which only replacing soil moisture performs the second, followed by only replacing land cover and soil texture which does not improve the simulation results much. In conclusion, the simulation accuracy of sandstorm is improved by introducing the remote sensing products.     

Watershed scale temporal and spatial stability of soil moisture and its role in validating satellite estimates

Watershed scale soil moisture estimates are necessary to validate current remote sensing products, such as those from the Advanced Microwave Scanning Radiometer (AMSR). Unfortunately, remote sensing technology does not currently resolve the land surface at a scale that is easily observed with ground measurements. One approach to validation is to use existing soil moisture measurement networks and scale these point observations up to the resolution of remote sensing footprints. As part of the Soil Moisture Experiment 2002 (SMEX02), one such soil moisture gaging system in the Walnut Creek Watershed, Iowa, provided robust estimates of the soil moisture average for a watershed throughout the summer of 2002. Twelve in situ soil moisture probes were installed across the watershed. These probes recorded soil moisture at a depth of 5 cm from June 29, 2002 to August 19, 2002. The sampling sites were analyzed for temporal and spatial stability by several measures including mean relative difference, Spearman rank, and correlation coefficient analysis. Representative point measurements were used to estimate the watershed scale (∼25 km) soil moisture average and shown to be accurate indicators with low variance and bias of the watershed scale soil moisture distribution. This work establishes the validity of this approach to provide watershed scale soil moisture estimates in this study region for the purposes of satellite validation with estimation errors as small as 3%. Also, the potential sources of error in this type of analysis are explored. This study is a first step in the implementation of large-scale soil moisture validation using existing networks such as the Soil Climate Analysis Network (SCAN) and several Agricultural Research Service watersheds as a basis for calibrating satellite soil moisture products, for networks design, and designing field experiments.     

Pixel-scalesoil Moisture Monitoring Network and Its Preliminary Validation of L-band Soil Moisture Products

Soil moisture is a key variable in the process of crop growth,ground-air water heat exchange and global water cycle,which plays an important role in drought monitoring,hydrological land surface processes and climate change.Passive microwave remote sensing has become the main means of monitoring soil moisture with the sensitivity to soil moisture.In this study,the authenticity test of SMAP（Soil Moisture and Active and Passive） and SMOS（Soil Moisture and Ocean Salinity）passive microwave soil moisture products using the soil moisture sensor network monitoring data carried out against the underlying surface of farmlands in Jilin Province was carried out.The following conclusions were obtained：（1）Compared with the in situ measured data,SMOS L3(ascending and descending overpasses) and SMAP L3 passive microwave soil moisture products generally underestimated the ground data,but With the occurrence of rainfall events,there will be the phenomenon which is the value of soil moisture products is higher than the in situ data; although the unbiased root mean square error (unRMSE) of the two soil moisture products was greater than 0.07 m3/m3,the unRMSE of SMAP passive microwave soil moisture product data which was 0.078 m3/m3 was slightly lower;（2）Since the depth of induction of the L-band is lighter than the depth of detection of the sensor(5cm),and the dryness of the soil surface after rainfall causes the vertical inhomogeneity of soil moisture,which is one of the reasons why SMOS and SMAP passive microwave soil moisture products underestimate soil moisture; （3）SMOS has a higher value than the range of SMAP brightness temperature,which may be caused by radio frequency interference (RFI),which makes the error of soil moisture Retrieval and affects the validation accuracy.The comparison of bright temperature distribution of SMOS and SMAP shows that the effect of RFI on SMOS is more serious due to the influence of electromagnetic radio frequency interference (RFI),which may be the reason why the RMSE of soil moisture product of SMOS is higher than that of passive microwave soil moisture product of SMAP.     

Temporal persistence and stability of surface soil moisture in a semi-arid watershed

Satellite soil moisture products, such as those from Advanced Microwave Scanning Radiometer (AMSR), require diverse landscapes for validation. Semi-arid landscapes present a particular challenge to satellite remote sensing validation using traditional techniques because of the high spatial variability and potentially rapid rates of temporal change in moisture conditions. In this study, temporal stability analysis and spatial sampling techniques are used to investigate the representativeness of ground observations at satellite scale soil moisture in a semi-arid watershed for a long study period (March 1, 2002 to September 13, 2005). The watershed utilized, the Walnut Gulch Experimental Watershed, has a dense network of 19 soil moisture sensors, distributed over a 150 km² study region. In conjunction with this monitoring network, intensive gravimetric soil moisture sampling conducted as part of the Soil Moisture Experiment in 2004 (SMEX04), contributed to the calibration of the network for large-scale estimation during the North American Monsoon System (NAMS). The sensor network is shown to be an excellent estimator of the watershed average with an accuracy of approximately 0.01 m³/m³ soil moisture. However, temporal stability analysis indicated that while much of the network is stable, the soil moisture spatial pattern, as represented by mean relative difference, is not replicated by the network mean relative difference pattern. Rather, the network is composed of statistical samples. Geophysical aspects of the watershed, including topography and soil type are also examined for their influence on the soil moisture variability and stability. Soil type, as characterized by bulk density, clay and sand content, was responsible for nearly 50% of the temporal stability. Topographic effects were less important in defining representativeness and stability.     

基于地基微波辐射观测的土壤水分反演算法评估

当前常用的被动微波土壤水分反演算法有水平极化单通道算法、垂直极化单通道算法、双通道算法、微波极化差比值算法和扩展双通道算法,5种反演算法具有不同的差异,对这些反演算法进行系统的评估和分析将有助于反演算法的改进和星载高精度土壤水分产品的发布。为了避免直接采用卫星产品验证时的尺度匹配、空间异质性等问题,基于地基L波段微波辐射观测以及配套的土壤和植被参数测量数据,对这5种反演算法进行了实现、对比和分析,得出以下结论：①单通道算法具有最佳的反演性能,水平极化单通道算法反演结果具有最高的相关性(相关性系数R=0.83),垂直极化单通道算法反演结果具有最小的反演误差(均方根误差RMSE=0.028 m³/m³,偏差BIAS= -0.011 m³/m³),但单通道算法需要精确的植被含水量输入;②其余3种算法能脱离植被辅助数据的使用,性能略差但也能满足星载微波传感器的探测指标要求(小于等于0.04 m³/m³);其中,扩展双通道算法和微波极化差比值算法的土壤水分反演结果比双通道算法略差,但本例中扩展双通道算法在植被含水量反演方面更具优势。     

遥感资料在WRF-Chem沙尘模拟中的应用

沙尘暴的起沙过程是沙尘循环中的重要部分,起沙过程模拟的准确性对于输送和沉降过程的准确模拟十分重要。WRF-Chem (Weather Research and Forecasting with Chemistry)是目前应用最广泛的沙尘暴模拟模式之一,但目前WRF-Chem对于起沙量的模拟具有很大的不确定性,受下垫面和土壤湿度的影响较大。WRF-Chem模式中的下垫面数据比较老旧,且驱动WRF-Chem模式的资料中土壤湿度是偏高的。土地覆被和土壤水分等遥感产品的日趋成熟,为WRF-Chem模拟沙尘暴提供了新的选择和契机,因此,将AMSR2 (the Advanced Microwave Scanning Radiometer 2) 土壤湿度和MODIS (Moderate Resolution Imaging Spectroradiometer) 土地利用遥感产品及实地调查资料替换WRF-Chem下垫面,利用WRF-Chem模拟了2018年3月26日至28日发生在我国华北地区的一场沙尘过程,用于探究下垫面参数对WRF-Chem模式模拟沙尘暴的精度产生的影响。共开展了4组实验,包括1组控制实验和3组对照实验,3组对照实验分别是在控制试验基础上仅替换土壤水分初始场、仅替换土地覆被和土壤质地以及同时替换土地覆被、土壤质地及土壤水分初始场。在加入遥感数据之后,3组对照实验的模拟精度比控制实验均有所提高,其中同时替换土壤水分初始场、土地覆被和土壤质地对模式模拟结果改善最大。改善后PM10模拟的相关系数提高了0.30,平均偏差减少了31.18 μg/m³,均方根误差减少了21.70 μg/m³;AOD (Aerosol Optical Depth) 的相关系数提高了0.14,平均偏差减少了0.29,均方根误差减少了0.18,仅替换土壤水分初始场效果次之,仅替换土地覆被和土壤质地对于模拟结果改善不大。以上结果表明：加入遥感资料可以有效提高WRF-Chem对沙尘过程的模拟精度。     

激光遥感近期发展趋向

激光遥感就是用激光束作为光谱探头来探测大气成分。在过去的十五年中,这种激光雷达技术已被证明是一种测量大气中几种重要化学成分的有效探测方法,现在,它不但对我们了解大气起着重要的作用,而且正在     

微波遥感土壤湿度研究进展

发展实用的微波遥感土壤湿度卫星反演算法,以提供区域尺度上的土壤湿度信息,对水文学、气象学以及农业科学研究与应用至关重要。简要分析了主动微波遥感土壤湿度的研究进展情况,重点对被动微波遥感土壤湿度的原理、算法发展及研究趋势进行了详细论述。主动微波具有卫星数据空间分辨率高的特点,随着携载主动微波器的一系列卫星的发射,主动微波遥感土壤湿度将受到重视。被动微波遥感研究历史长、反演算法特别是卫星遥感算比较成熟,是今后区域尺度乃至全球尺度监测土壤湿度的重要手段。     

SMAP土壤水分产品在淮河流域的适用性评估

选取淮河流域为研究区域,利用2016年6月至2019年5月流域内的313个土壤水分观测站0～10 cm土壤体积含水量数据,使用多种指标分析SMAP卫星(Soil Moisture Active Passive)9 km分辨率土壤水分产品(L2_SM_P_E)精度的空间和时间(年、月、日尺度)特征,并讨论植被、土壤、地形...     

Error Characteristics of Microwave Soil Moisture Products based on Triple Collocation and Its Spatial-temporal Pattern

Soil moisture is an important state variable connecting the land surface-atmosphere system, and its information can be efficiently acquired by the new technique of microwave remote sensing. Accurate interpretation of the microwave soil moisture products qualities and in-depth understanding of their temporal and spatial distributions are important prerequisites for their successful application in earth science through data assimilation. In this study, three microwave soil moisture products, FengYun-3C(FY-3C), Soil Moisture Active Passive (SMAP) and Advanced Scatterometer(ASCAT), were evaluated over China based on the triple collocation (TC) method. The abilities of three products to obtain temporal and spatial variations of soil moisture were illustrated by Hovm?ller diagram. The results show that: (1) SMAP generally outperforms ASCAT and FY-3C, with highest TC-based signal-to-noise ratio(SNR) under different land use types. The TC-based SNRs are 1.668dB, -0.316dB and -2.182dB for SMAP, ASCAT and FY-3C respectively; and their correlation coefficients with ground observations are 0.514, 0.501 and 0.209, respectively. (2) The accuracies of FY-3C and ASCAT in Northwest China are overall higher than those in the southern China. All three products can capture the latitudinal and longitudinal gradients of soil moisture, whereas their seasonal fluctuations are higher than those of in-situ measurements. Among three products, FY-3C shows highest spatial gradient and strongest seasonal fluctuations. (3) FY-3C product performance is more susceptible to vegetation coverage than ASCAT and SMAP, but it outperforms ASCAT in barren areas. The results of our study could provide useful insights for assimilating microwave soil moisture products into land surface models to improve hydrological prediction.