Spaceborne GNSS-R Soil Moisture Retrieval Model Considering Vegetation Water Content

As an important parameter in seasonal climate change research,soil moisture plays a very important role in hydrological and agricultural processes,arid area development and river discharge,and agricultural irrigation.Global Navigation Satellite System(GNSS)L-band signals have the advantages of low cost,strong penetration,and high spatial and temporal resolution.Therefore,GNSS-R technology based on GNSS reflected signals has become a new hot spot in the field of current Earth observation research.Due to the influence of various factors such as vegetation coverage and surface roughness,the reflected signal will inevitably carry common information of various surface coverages such as vegetation water content and soil moisture.Therefore,based on two kinds of spaceborne GNSS-R data,this thesis selects two study areas with different surface cover types in the Argentine Pampas grassland and southern China,and studies the construction of soil moisture retrieval model assisted by BP neural network and Deep Belief Network algorithm.The main research results are:(1)The soil moisture retrieval model assisted by BP neural network and DBN algorithm was constructed.Experiments show that the overall accuracy of results based on the DBN algorithm model is slightly better than the BP model and the operating efficiency of the DBN algorithm is much higher than that of the BP algorithm.(2)The performance of soil moisture retrieval based on TDS-1 and CYGNSS data was compared and analyzed.Before and after the reflectivity correction,the retrieval accuracy of the TDS-1 data is better than that of the CYGNSS.Based on the CYGNSS data,the highest R in the Pampas steppe region obtained by the parameter combination input is 0.698,which is nearly 46% higher than that of the single-parameter input model,and the highest R in southern China is 0.408,an increase of nearly 50%.The best R for the Pampas region based on the input of the TDS-1 data parameter combination is 0.717,and the best R for southern China is 0.442.(3)The spaceborne GNSS-R soil moisture inversion model based on the watercloud model to correct the surface reflectivity and introduce roughness was studied.The parametric water-cloud model was used to eliminate the contribution of vegetation water to the surface reflectivity,so as to reduce the influence of surface vegetation coverage and further improve the soil moisture retrieval accuracy in the two regions.After several experiments in two areas with different surface coverage,the values of water cloud model parameters A and B that are more suitable for the two different surface coverage areas were obtained.Experiments show that in the Pampas grassland area,the accuracy of the single input model before and after the reflectivity correction has no significant difference,but the parameter combination model has a significant improvement effect.For the TDS-1 data,the best R in the Pampas region was improved by nearly 0.03 compared with the original best model;the R of the best inversion model in southern China was improved by nearly 0.11 compared with the original model.For the CYGNSS data,the best inversion model R for the Pampas steppe is about 0.03 higher than the original model;the best R for southern China is nearly 0.1 higher than the original model,and the improvement ratio reaches 25%.(3)Considering the effect of seasons on vegetation coverage,the differences in soil moisture retrieval accuracy in different months based on CYGNSS data were analyzed.Retrieval results based on CYGNSS data in different months show that for the selected pampas grassland and southern China,the correction of surface reflectance and the introduction of roughness based on water cloud model can improve the inversion accuracy of soil moisture in the two places,especially in southern China.Before the surface reflectance correction,the inversion effect of pampas area is the best in August,with R of 0.928,and that of southern China is the best in December,with R of 0.582.After the reflectance correction and the introduction of roughness,the inversion accuracy in pampas area increased the most in December,with the proportion of R increased by 7%.The inversion improvement effect in southern China was the best in September,with the proportion of R increased by 80%.