Support vector regression snow-depth retrieval algorithm using passive microwave remote sensing data

Xiao, Xiongxin; Zhang, Tingjun; Zhong, Xianqiong; Shao, Wei; Li, Xiaodong

doi:10.1016/j.rse.2018.03.008

Cited by 109 publications

(116 citation statements)

References 71 publications

Supporting

Mentioning

114

Contrasting

Order By: Relevance

“…Several studies pointed out that passive microwave remote sensing has a limited ability to detect wet snow during the snow melt season, which may underestimate the D d [87][88][89][90]. Meanwhile, misclassification and error in deriving snow cover were attributed to relatively coarse spatial resolution, as well as the complexity of snow characteristics and topography [91][92][93][94]. Combined with optical remote sensing, passive microwave remote sensing and a land surface model can effectively improve the monitoring accuracy of snow phenology and snow depth [95].…”

Section: Limitation and Outlookmentioning

confidence: 99%

Changes in Snow Phenology from 1979 to 2016 over the Tianshan Mountains, Central Asia

Yang

Ahmad

et al. 2019

Remote Sensing

View full text Add to dashboard Cite

Snowmelt from the Tianshan Mountains (TS) is a major contributor to the water resources of the Central Asian region. Thus, changes in snow phenology over the TS have significant implications for regional water supplies and ecosystem services. However, the characteristics of changes in snow phenology and their influences on the climate are poorly understood throughout the entire TS due to the lack of in situ observations, limitations of optical remote sensing due to clouds, and decentralized political landscapes. Using passive microwave remote sensing snow data from 1979 to 2016 across the TS, this study investigates the spatiotemporal variations of snow phenology and their attributes and implications. The results show that the mean snow onset day (Do), snow end day (De), snow cover duration days (Dd), and maximum snow depth (SDmax) from 1979 to 2016 were the 78.2nd day of hydrological year (DOY), 222.4th DOY, 146.2 days, and 16.1 cm over the TS, respectively. Dd exhibited a spatial distribution of days with a temperature of <0 °C derived from meteorological station observations. Anomalies of snow phenology displayed the regional diversities over the TS, with shortened Dd in high-altitude regions and the Fergana Valley but increased Dd in the Ili Valley and upper reaches of the Chu and Aksu Rivers. Increased SDmax was exhibited in the central part of the TS, and decreased SDmax was observed in the western and eastern parts of the TS. Changes in Dd were dominated by earlier De, which was caused by increased melt-season temperatures (Tm). Earlier De with increased accumulation of seasonal precipitation (Pa) influenced the hydrological processes in the snowmelt recharge basin, increasing runoff and earlier peak runoff in the spring, which intensified the regional water crisis.

show abstract

Section: Limitation and Outlookmentioning

confidence: 99%

Changes in Snow Phenology from 1979 to 2016 over the Tianshan Mountains, Central Asia

Yang

Ahmad

et al. 2019

Remote Sensing

View full text Add to dashboard Cite

show abstract

“…For SVM-based models, parameter optimization is indispensable and it can significantly affect the obtained results [16,24]. Furthermore, many previous studies reported that parameter optimization has positive effects.…”

Section: Influence Of Parameter Optimizationmentioning

confidence: 99%

“…Some researches show that SVM can successfully achieve higher classification accuracy than traditional methods [9,[19][20][21][22][23][24]. Nevertheless, two challenges for SVM remain to be addressed in the remote sensing field [15,16]: (1) to select the kernel function and parameter [25]; and, (2) to optimally obtain key parameters, including kernels and penalty parameter C [22,[26][27][28].…”

Section: Introductionmentioning

confidence: 99%

Fine Land Cover Classification in an Open Pit Mining Area Using Optimized Support Vector Machine and WorldView-3 Imagery

Chen

Wang

2019

Remote Sensing

View full text Add to dashboard Cite

Fine land cover classification in an open pit mining area (LCCOM) is essential in analyzing the terrestrial environment. However, researchers have been focusing on obtaining coarse LCCOM while using high spatial resolution remote sensing data and machine learning algorithms. Although support vector machines (SVM) have been successfully used in the remote sensing community, achieving a high classification accuracy of fine LCCOM using SVM remains difficult because of two factors. One is the lack of significant features for efficiently describing unique terrestrial characteristics of open pit mining areas and another is the lack of an optimized strategy to obtain suitable SVM parameters. This study attempted to address these two issues. Firstly, a novel carbonate index that was based on WorldView-3 was proposed and introduced into the used feature set. Additionally, three optimization methods—genetic algorithm (GA), k-fold cross validation (CV), and particle swarm optimization (PSO)—were used for obtaining the optimization parameters of SVM. The results show that the carbonate index was effective for distinguishing the dumping ground from other open pit mining lands. Furthermore, the three optimization methods could significantly increase the overall classification accuracy (OA) of the fine LCCOM by 8.40%. CV significantly outperformed GA and PSO, and GA performed slightly better than PSO. CV was more suitable for most of the fine land cover types of crop land, and PSO for road and open pit mining lands. The results of an independent test set revealed that the optimized SVM models achieved significant improvements, with an average of 8.29%. Overall, the proposed strategy was effective for fine LCCOM.

show abstract

“…Long-term snow cover records are crucial for climate studies, hydrological applications and weather forecasts over the Northern Hemisphere (Gong et al, 2007;Derksen et al, 2012;Safavi et al, 2017;Tedesco et al, 2016;. A key parameter is the snow water equivalent (SWE), which describes the amount of water stored in the snowpack as a product of snow depth and mean snow density (Dressler et al, 2006;Kelly et al, 2009;Foster et al, 2011;Xiao et al, 2018;Takala et al, 2017;Tedesco et al, 2016). Fortunately, passive microwave (PMW) signals can penetrate snow cover and provide snow depth estimates through volume scattering of snow particles in dry snow conditions.…”

Section: Introductionmentioning

confidence: 99%

“…CC BY 4.0 License. application in remote sensing fields is promising (Liang et al, 2015;Bair et al, 2018;Xiao et al, 2018;Xiao et al, 2019). ML techniques can reproduce the nonlinear effects and interactions between variables without assumptions of a functional form.…”

Section: Introductionmentioning

confidence: 99%

Real-Time Snow Depth Estimation and Historical Data Reconstruction Over China Based on a Random Forest Machine Learning Approach

Yang¹,

Jiang²,

Luojus³

et al. 2019

Preprint

View full text Add to dashboard Cite

Abstract. Snow depth data time series are valuable for climatological and hydrological applications. Passive microwave (PMW) sensors are advantageous for estimating spatially and temporally continuous snow depth. However, PMW estimate accuracy has several problems, which results in poor performances of traditional snow depth estimation algorithms. Machine learning (ML) is a common method used in many research fields, and its early application in remote sensing is promising. In this study, we propose a new and accurate approach based on the ML technique to estimate real-time snow depth and reconstruct historical snow depth from 1987–2018. First, we trained the random forest (RF) model with advanced microwave scanning radiometer 2 (AMSR2) brightness temperatures (TB) at 10.65, 18.7, 36.5 and 89 GHz, land cover fraction (forest, shrub, grass, farm and barren), geolocation (latitude and longitude) and station observation from 2014–2015. Then, the trained RF model was used to retrieve a reference dataset with 2012–2018 AMSR2 TB data as the accurate snow depth. With this reference snow depth dataset, we developed the pixel-based algorithm for the Special Sensor Microwave/Imager (SSM/I) and Special Sensor Microwave Imager Sounder (SSMI/S). Finally, the pixel-based method was used to reconstruct a consistent 31-year daily snow depth dataset for 1987–2018. We validated the trained RF model using the weather station observations and AMSR2 TB during 2012–2013. The results showed that the RF model root mean square error (RMSE) and bias were 4.5 cm and 0.04 cm, respectively. The pixel-based algorithm’s accuracy was evaluated against the field sampling experiments dataset (January–March, 2018) and station observations in 2017–2018, and the RMSEs were 2.0 cm and 5.1 cm, respectively. The pixel-based method performs better than the previous regression method fitted in China (RMSEs are 4.7 cm and 8.4 cm, respectively). The high accuracy of the pixel-based method can be attributed to the spatial dynamic retrieval coefficients and accurate snow depth estimates of the RF model. Additionally, the 1987–2018 long-term snow depth dataset was analyzed in terms of temporal and spatial variations. On the spatial scale, daily maximum snow depth tends to occur in Xinjiang and the Himalayas during 1992–2018. However, the daily mean snow depth in Northeast China is the largest. For the temporal characteristics, the February mean snow depth is the thickest during snowy winter seasons. Interestingly, the January mean snow depth represents the annual mean snow depth, which plays an important role in snow depth prediction and hydrological management. In conclusion, through step-by-step validation using in situ observations, our pixel-based approach is available in real-time snow depth retrievals and historical data reconstruction.

show abstract

Support vector regression snow-depth retrieval algorithm using passive microwave remote sensing data

Cited by 109 publications

References 71 publications

Changes in Snow Phenology from 1979 to 2016 over the Tianshan Mountains, Central Asia

Changes in Snow Phenology from 1979 to 2016 over the Tianshan Mountains, Central Asia

Fine Land Cover Classification in an Open Pit Mining Area Using Optimized Support Vector Machine and WorldView-3 Imagery

Real-Time Snow Depth Estimation and Historical Data Reconstruction Over China Based on a Random Forest Machine Learning Approach

Contact Info

Product

Resources

About