The Inversion Of Snow Cover And Snow Surface Temperature Based On Multi-source Remote Sensing Data

As the snowmelt runoff model input parameters, the accuracy of the snow parameterswill directly affect the snowmelt runoff modeling and forecasting results. Because of themacro, comprehensive, dynamic and fast advantages, remote sensing has become animportant means for snow study. At present, due to the influence of the sensor and applicationrequirements, remote sensing is often used for mesoscale and large-scale snow study but littlefor watershed scale snow study. With the in-depth study on the forecasts of snowmelt runoffsimulation, there is an urgent need to obtain accurate snow data in river basin during thesnowmelt period.This study is based on the above requirements, select the JunTanghu River basin as atypical study area, and choose HJ-1A/1B and MODIS data used for snow cover and snowsurface temperature inversion study.The main research contents include the following aspects:Explore the spectral characteristics of the snow in the visible, near infrared and thermalinfrared band. Analysis of the spectral characteristics and influencing factors of snow in thesnowmelt period, to provide the physical basis of remote sensing quantitative retrieval ofsnow parameters for the study area. The results show that, during the snowmelt, the snowphysical state rapidly changes will cause obvious change of snow reflectance, which willaffect the snow parameters quantitative inversion.Some commonly used atmosphere correction methods are used to corrected theatmospheric distortion for HJ-1A/1B data and discussed the feasibility of these methods.According to the characteristics that observation angle is biger in HJ-1A/1B, usingMODTRAN model building atmosphere correction parameter look-up table in differentobservation condition. The results indicate that, the method of look-up table is best inatmosphere correction. If the demand for accuracy is not high, can use the FLAASH model for atmospheric correction.According to the characteristics of the study area and the type of data source, to carry outthe snow cover extraction method research, and build snow cover extraction methods fordifferent data sources and snowmelt state. Fusion inversion forest data by TM image for snowcover extraction, the method can improve the snow cover extraction accuracy in forest area.Research the optimum band combination of the snow cover index and discriminationthreshold in different snowmelt state for HJ-1A/1B image. If only use CCD data, textureassisted SVM method can improve the snow cover extraction accuracy. In addition, for theMODIS data, to build a segmented snow cover fraction inverse model according to the statusof the study area and proposed use NDSI threshold to revision model coefficients. Themethod significantly improves the retrieval accuracy of snow cover on a watershed scale, andby the rapid adjustment of the model coefficients, can adapt the fraction of snow coverinversion under the different snowmelt state.To carry out the snow surface temperature inversion method study based on the HJ-1BIRS and MODIS data. Using IRS data to retrieved the snow surface temperature, and carriedout the research on QK&B and JM&S two single-channel algorithms and the parametersacquisition methods. Atmospheric radiative transfer model MODTRAN was used to carry outthe validation of the two algorithms errors and sensitivity analysis at low temperatures. Theresults show that in low temperature, if do not take into account the parameter errors, the twoalgorithms will produce to1K error. QK&B algorithm error rapidly increasedwith the snowsurface temperature increases, the error that caused by JM&S algorithm is little change. Ingeneral, both algorithms can accurately reflect the space distribution trends and thedifferences of the snow surface temperature, but both generally overestimated the snowsurface temperature, the accuracy of the JM&S algorithm slightly higher than the QK&Balgorithm. For MODIS image, firstly, simulate the linear coefficient of the Plank function inlow temperature environment, then to deduce the sub-pixel snow surface temperature inversion formula, and based on the snow cover fraction and forest areas and then proposed asurface average emissivity estimation methods. the result indicate that, compared to thetraditional component temperature inversion method this algorithm is simple and easy tooperate and it can improve the MODIS snow surface temperature retrieval accuracy in acertain extent, especially for the mixed pixel area, such as forest areas.