Soil Salinity Retrieval And Dynamic Analysis Based On Spectral Intercalibration Of Multi-sensor Data

Salinization is one of the most damaging soil problems in the Yellow River Delta,seriously restricted the development of regional economy.Monitoring soil salinity with remote sensing technique is in urgent need.Remote sensing data of single sensor type can not meet the data demands of soil salinization dynamic analysis.However,the main threat of time series data from different sensors lies in the diversity of band width and the lack of data continuity.A band spectral inter-calibration method setting the latest OLI as target sensot type is investigated in this study to simulate OLI time series data based on four Landsat-5 TM,EO-1 ALI,Landsat-8 OLI time series images and Hyperion data within 2000-2016.PLSR(Partial Least Square Regression),MLR(Multiple Linear Regression)and BPNN(Back Propagation Neural Networks)models were developed to quantify and retrieve soil salinity.Dynamic analysis on soil salinity were finally detected through dynamic index,transition matrix and overlay analysis.Following conclusions were obtained from the study:(1)TM,ALI and OLI data were simulated from Hyperion data resampling.The latest OLI sensor was set as the target sensor type.Spectral band inter-calibration coefficients were calculated from simulated TM,ALI and OLI data with a statistical regression method.Simulated OLI time series images were finally constructed by transforming TM,ALI to OLI images with the coefficients.Results show that band correlation between homologous spectral bands of two sensors can not be improved significantly while band correlation between newly increased spectral bands of OLI and TM or ALI are remarkably increased through spectral band inter-calibration method.Data continuity of multi-sensor time series images has been improved through spectral band inter-calibration,guaranteeing the longtime saline soil monitoring.(2)PLSR,MLR and BPNN models were developed to quantify the predictive relationships between soil spectra and soil salinity parameters(electrical conductivity,EC or soil saliniy content,S).Modelling results show that PLSR model performs better than MLR when fixing the dependent variable and models which set EC as dependent variable perform better than those set S in PLSR or MLR model.Non-linear regression BPNN model exhibits a higher calibration precision than linear regression models but is prone to over-fitting status which indicates a fairly low validation precision.Among all tested models,PLSR-EC model performs optimally with coefficient of determination Rc2=0.716 for calibration,Rv2=0.700 for validation and relative error K=33.56%.(3)The optimal predictive model(PLSR-EC)was applied to time series images to map soil salinity in the year of 2000,2008,2012 and 2016.The test set for validating retrieval results consist of 31 soil samples collected in 2012 and 96 soil samples collected in 2016.Compared with measured EC with image retrieval EC,mapping accuracy yields correlation of 0.690 for 2012and 0.795 for 2016.The amount of abnormal soil salinity retrieval values in four time series images is less than 10%.The overall soil salinity retrieval results are reliable while ECs of high salinity soil may be underestimated.(4)Soil salinity dynamic analysis based on soil salinity spatial distribution maps of four years was detected through statistics of salinity soil area,variation index,transition matrix and overlay analysis.Land cover transition matrix shows that completely saline soil and slightly saline soil throughout the research area are of poor stability and convert with other soil types frequently.The variation index calculation indicates lightly and moderate saline soil increase rapidly while highly saline soil has been migrating from 2012 to 2016.It is observed from GIS overlay analysis results that non-soil land cover mainly convert to slightly and moderate saline soil in Gudao town and the south part of Xianhe town and convert to moderate and severe saline soil in the east part of Yongan town.Non-saline soil land cover convert dispersedly to saline soil,mainly to severe saline soil particular in the central and south part of typical research area.