Spatio-temporal Reconstruction For Globally Remotely Sensed Total Ozone Production

Remotely sensed total ozone data play an important role in the monitoring of the spatial distribution and temporal change of the atmospheric total ozone.However,due to factors such as ground swath width limit(across-track direction)and instrument anomaly,the existing total ozone data often appear a lot of information defect,it makes remotely sensed time series of total ozone product suffering spatio-temporal discontinuous and incomplete,which then severely hinders the subsequent application and analysis.Since late 2006,total ozone production from OMI(ozone monitoring instrument)aboard on Aura satellite named OMTO3e(ozone monitoring total ozone level 3 expanded)began to existing defect.Resulting from OMI instrument anomaly,the missing parts accounting for one third or even more of one set data,which is a serious obstacle for its normal usage.Total ozone production from TOMS(total ozone monitoring sensor)abroad on Earth Probe satellite named TOMSM3L3 appeared missing gaps in the equatorial region.This is mainly due to the height of the satellite orbit is reduced,it results in orbital space covering area is reduced,and then the missing information accordingly appears in the equatorial region.Therefore,it is of great significance to develop a set of feasible and efficient reconstruction technology to reconstruct the missing data which is a common phenomenon in the remotely sensed total ozone data and then to generate a set of global total ozone production with high precision,long time sequence,and spatio-temporal continuum.In this paper,the main work is summarized as the following aspects:(1)Focusing on the phenomena of missing gaps commonly existing in satellite total ozone product in the equatorial region,this paper introduces several kinds of methods for reconstruction the missing information of remotely sensed data.It also summarizes their insufficient in practical application to put forward the aim of this study.(2)Proposing the reconstruction technology based on the spatio-temproal weighted regression.This method is aiming at reconstructing the missing information existing in the level 3 inversion production of OMI sensor named OMTO3e.The study area is arectangular area which includes the mainland of China.In order to achieve spatio-temporally seamless total ozone production in the research area,this paper establishes spatio-temporal relationship between the target data and multi-temproal auxiliary data using locally spatio-temporal weighted regression method to guarantee spatio-temporal correlation and nonstationary of the total ozone data.The method adopts a strategy of pixel by pixel for reconstructing the missing pixels.Firstly,selects data set before and after the target data as auxiliary data sets;and then establish linear regression relationship of the target pixel from the target pixel and each auxiliary data,where regression equation coefficients are obtained.by choosing suitable reference pixels;the final prediction of the target pixel is obtained by weighted average of the two predictions predicted by the two auxiliary data sets.Simulation experiments,comparing with other traditional methods,verify that the proposed method has higher feasibility and more superiority for missing information reconstruction.(3)Proposing the residual error correction method by considering heterogeneous distribution of total ozone data.The method is on the basis of the prediction by the spatio-temproal reconstruction technology based on the spatio-temproal weighted regression,taking into account the different changing characteristics of total ozone along the latitude and longitude direction,to correct the above prediction by establishing a residual error correction model.The research object is the worldwide OMTO3e data.Firstly,makes use of the spatio-temproal reconstruction technology based on the spatio-temproal weighted regression to obtain a preliminary predicted value of the target pixel;secondly,computes the preliminary predictions of all the reference pixels in the target data by applying the first step to the reference pixels,and calculates the residuals by comparison of the real values and predictions;thirdly,through the establishment of anisotropy variogram model of reference pixels residuals,the evaluation of the residual value of the target pixel can be obtained by kriging interpolation technology.Simulation experiment and real experiment results show that the proposed method is able to reconstruct the spatio-temporal continuous total ozone data with global coverage and high precision.On this basis,we generated a spatially continuous and daily global total zone product(2004-2004).(4)Proposing the reconstruction and correction method based on long time series of total ozone production from multisensor.Monitoring spatial and temporal variation trends of total ozone often need long time series of total ozone data,while data from single sensor is usually limited,so combining data from multiple sensors is an effective way.However,due to different sensor having different observation system and inversion algorithm,the inversion production from different sensors inevitably has relative differences.In the ozone optical inversion process,the ozone absorption coefficient is calculated by a standard ozone effective temperature,without territoriality and seasonality,which can bring errors into the total ozone observation.This study aims to establish a normalized model to unify total ozone data from multisensors and a correction model to correcting the inversion error bringing by ignoring ozone absorption coefficient on the dependence of atmospheric temperature.To guarantee the robustness of the normalized model,M-estimation robust regression methodis used to eliminate the relative error between different sensors.The correction model,introducing effective ozone temperature data,is constructed by using wide field regression neural network(GRNN)to improve the inversion precision of remotely sensed total ozone data.Experiments show that the M-estimation robust regression method weakens the bad effect of the singular data on coefficient estimation,so it has higher estimation accuracy than the traditional least square method.Using ten-crossed validation method verifies that GRNN method can promote the inversion precision of remotely sensed total ozone data.