Continuous fusion algorithm analysis for multi-source remote sensing soil moisture data based on cumulative distribution fusion

Abstract: As an important grain-producing region, Songnen Plain located in Northeast China has been significantly affected by drought in recent years. Remote sensing soil moisture is one of the important indices for monitoring agricultural drought in large-scale farmland area. The time series length and update speed of remote sensing data are 2 important factors affecting its application. In 2009, satellite called SMOS (soil moisture and ocean salinity) was launched. As the first satellite dedicated to monitoring soil moisture of earth, daily updated SMOS soil moisture data have been proven to be suitable for the application in real-time drought monitoring and evaluation in many researches. In the field of agricultural drought management, drought characteristics and frequency analysis are basic contents of these researches. However, it is impossible to analyze the drought frequency and characteristic evolution by SMOS data, due to their short time series. CCI (climate change initiative) soil moisture data, which have a long time series (1979-2013), was combined with a variety of C-band scattered data and multi-frequency radiometer data. As a kind of historical data, CCI soil moisture product can make up for SMOS data to analyze the agricultural drought characteristics. Because of the difference of the sensors and the inversion methods, remote sensing data from different sources cannot be directly compared and analyzed. Therefore, data fusion becomes a hotspot and key issue in the application research of remote sensing data nowadays. Based on cumulative distribution matching principle, the key of data fusion is to establish the correlation between cumulative probability curves of different data. The work amount of traditional piecewise linear fusion method is proportional to the fusion accuracy. This linear method is difficult to process a number of data in batches with high precision. Unary interpolation can establish this correlation between any quantile on different cumulative probability distribution curves. Therefore, a continuous fusion algorithm of multi-source remote sensing soil moisture was built in this study. Using this continuous fusion method, SMOS and CCI data were fused to real-time remote sensing soil moisture data product with long time series characteristics with the Songnen Plain as the case. This study compared the fusion accuracy between this continuous fusion and piecewise linear fusion method. And the time series of original SMOS data and fused SMOS data was also analyzed. The analysis results indicate that this unary interpolation continuous fusion method can improve the fusion accuracy of multi-source remote sensing soil moisture significantly. Data segment of the cumulative probability distribution curve with low water content can characterize agricultural drought. By the piecewise linear fusion method, data segment of the cumulative probability distribution curve with low water content yet has some errors, which will lead to the inaccuracy of drought evaluation. By this new continuous fusion method, fused SMOS data and CCI data are completely coincident at each quantile in the low-value region of the curve. Through the accurate evaluation of drought events, the fused SMOS data can reflect local drought conditions. Through time series analysis, the range of fused SMOS data is closer to the CCI data, and the relative change pattern of original SMOS data still remains. This remote sensing fusion data combining the advantages of CCI and SMOS data can provide reliable data support for the next study of agricultural drought evaluation.