一种提高积雪深度估算精度的D-InSAR方法

A D-InSAR method to improve snow depth estimation accuracy

积雪深度(雪深)是反映积雪时空变化规律的重要参数,是全球及区域气候变化与水文循环等研究不可或缺的观测变量.差分合成孔径雷达干涉(differential interferometric synthetic aperture radar, D-In SAR)技术利用降雪前后微波穿透积雪层形成的差分干涉相位与雪深建立几何函数关系,被广泛地应用于区域小尺度雪深估算研究.然而,其估算精度受干涉像对相干性、局地地形和积雪介电常数等因素的影响.本研究基于高分辨率Sentinel-1 SAR数据,通过引入高相干系数区域、Sentinel-2光学影像、Google Earth影像和地表覆盖类型选取地面控制点优化雪深差分干涉相位解缠精度,利用卫星局地入射角和实测积雪密度降低斜距-相位关系模型经验误差,估算得到青藏高原东北部八宝河流域2021年消融期雪深时空分布,同时根据卫星同步积雪实测资料详细探讨了雪深估算误差来源.122个地面雪深实测数据(气象站点+野外测量)验证结果表明,优化后的D-In SAR差分干涉处理能提高雪深估算精度, RMSE为3.9 cm, MAPE为20.03%, R^(2)达到0.92.但雪深估算值总体存在低估现象(MBE%=-16.8%),雪深最大低估误差为9.1 cm.进一步研究发现,雪深估算精度除了D-In SAR系统失相干影响因素外,还受积雪微波穿透性和剖面结构、温度及湿度等积雪参数的影响.本方法能快速监测厘米级的雪深变化,且由于积雪异质性会限制差分干涉估算雪深的能力,更适用于干燥均质的积雪层,可为D-In SAR雪深估算研究提供科学的参考依据.

Snow depth is an important parameter that reflects the law of spatiotemporal variation in snow cover and is an indispensable observation variable for studying global and regional climate change and the hydrological cycle. Traditional manual field measurements and ground station observations have limited scopes and cannot fully reflect regional-scale snow conditions. Satellite remote sensing can simultaneously observe surface information over a large area and has been widely used in snow monitoring since the 1970 s. Optical remote sensing has difficulty obtaining snow depth information directly through the visible light band and is limited by atmospheric conditions. Microwave remote sensing can penetrate clouds and fog and can effectively capture changes in snow depth. Differential interferometric synthetic aperture radar(DIn SAR) technology uses the differential interferometric phase formed by microwave penetration through the snow layer before and after snowfall to establish a geometric function relationship with the snow depth;it is widely used in regional small-scale snow depth estimation research. However, its estimation accuracy is affected by many factors, such as interferometric image pair coherence, local topography, and snow dielect capability at microwave wavelengths.Based on high-resolution Sentinel-1 SAR data, this study optimizes the snow depth differential interferometric phase unwrapping accuracy by introducing Sentinel-2 optical images and high coherence coefficient regions to select ground control points, which are phase correction benchmarks. Furthermore, introducing field-measured snow parameters such as snow density and satellite local terrain incidence angles based on digital elevation model(DEM) data reduces the error of the differential interferometric phase-slope distance relationship model, thereby enabling estimations of the spatiotemporal distribution of snow depth in the Babao River Basin in the northeastern Qinghai-Tibet Plateau during the 2021 ablation period. Simultaneously, the accuracy of the snow depth estimation ability is evaluated based on the synchronous field measurement data of snow cover satellites. The main factors affecting the accuracy of snow depth estimation are discussed and analysed. Data from 122 ground snow depth measurements(meteorological stations + field measurements) are used to verify the results. The results show that the optimized D-In SAR differential interferometry can improve the snow depth estimation accuracy. The RMSE is 3.9 cm, the MAPE is 20.03%, and the R2 is 0.92. However, the estimated snow depth is generally underestimated(MBE%=-16.8%), and the maximum underestimation error of the snow depth is 9.1 cm. In addition to the influencing factors of D-In SAR system interferometric decorrelation, the estimations are affected by the penetration ability of microwaves in the snow and snow parameters such as stratigraphy structure, temperature and humidity. This limits the ability of differential interferometry to estimate snow depth and is more suitable for dry and homogeneous snow cover.This method allows for more precise and faster monitoring of centimetre-level snow depth changes;at the same time, the underlying surface of the study is relatively simple, and the influence of forest and other vegetation coverage is not considered. In follow-up research, snow cover should be optimized with multi-property and multi-environment interferometric models. An In SAR scattering model with a layered factor is introduced to improve the propagation information inside the snow layer then combined with the multiangle radar measurement of the ascending and descending orbits;experiments are carried out using the penetration ability of more bands in microwaves to reduce interferometric error sources and improve the accuracy of snow depth estimation. This study can provide a scientific reference for DIn SAR snow depth estimation research.