| Surface reflectance and optical remote sensing indices have been taking an important role in the information extraction of land cover,and widely serving the monitoring of ecology,agriculture,forestry,urban and other communities.The data acquired by a single satellite sensor has rarely been able to meet the growing demand for quantitative remote sensing accuracy.While the increasing number of satellite-borne sensors provides more chance to acquire multisource,complementary and time series of satellite-based data,for long-term dynamic monitoring.However,various sensors,due to diversified program tasks,are mounted on different platforms and designed special bands,which imply differences of orbit attitudes,number of bands,center wavelength,band width and observation geometry.Thus,it is essential to further understand the quantitative difference exsiting among various sensors for effective multi-sensor data fusion in many satellite-based applications.Though research on multi-source data fusion of satellite-borne sensors keeps increasing,the lack of studies on the influence of sensors SRF(Spectral Response Function)on reflectance still exists.This paper aims to quantitively invesitgate effects induced by SRF and observation geometry of dominant optical sensors on surface reflectances,and examine its propagation into frequent used remote sensing indices.The following work centered this objective is developed:(1)Orgnize vegetation,rock,mineral,soil,water,and artificial coating materials hyperspectrum collected from filed measurements(3759)or open spetrum databases(999),and as well as 4 Landsat scenes at two time-phase,collect 35 SRF of popular used sensors mounted respectively on satellites Landsat,Terra/Aqua,Sentinel-2,NOAA,GF,ZY-3,and CBERS etc.,to cross-analyze the hyperspectral features of typical surface targets and explore charateristics of various SRFs.(2)Study the generation,transfer,and response mechanism of sensor’s optical signal,split the disturbance from non-SRF factors in terms of the mechanism in collecting sensor-based signals,investigate the generation model of equivalent spectrum simulated from hyperspectrum,evaluate on the simulation model through the cross-evaluation with previous studies to ensure the quality of simulation for 35 selected sensors and as well as 12 adopted remote sensing indices.(3)Construct required senarios for the exploration of features in the equivalent reflectance spectrum at visible,near infrared,and shortwave bands of 35 selected sensors,quantitatively investigat the reflectance difference induced by SRF and its propagation in 12 adopted indices,analyze potential angle effects introduced by observation geometries or SRF copupled with observation direcitons into the reflectance and related indices,understand the effects propagation enhanced or supressed by construction scheme of various indices,and discuss on an empirical model for the correction of SRF-induced differences.The following research remarks are concluded:(1)The analysis on features of 35 selected optical SRF shows the frequent heredities on the number,positions,band width,and signal response percentage of designed channels from identical satellite program,such as TM and ETM+ of Landsat,MSI of Sentinel-2A and Sentinel-2B,PMS1 and PMS2 of GF,AVHRR1-3 of NOAA 7-18,and increased differences among satellite programs,as that MODIS has narrower band at visible,near infrared and shortwave bands than TM/ETM+,and further elevated differences on band width and positions of center wavelength of special bands compared with that of GF PMS/WFV and NOAA AVHRR.(2)According to the optical signal acquisition mechanism of the sensor and the comparison with the previous simulation methods,it is confirmed that the spectral response function and the hyperspectral convolution coupling method are effective to obtain the equivalent spectrum.An example is also used to prove the error in the simulation of NOAA-8 AVHRR1 in the paper by Alexander et al.(2002 & 2009),ensuring the correct simulation of 4758 multi-band equivalent spectra of eight typical features under different scenarios with 35 sensors.(3)The analysis of the equivalent spectrum of each sensor in different scenarios shows that the difference in the spectral response function can cause different degrees of captureable absolute difference and relative difference in the reflection spectrum of the surface object.Under the same observation geometry,the influence of the spectral response function on each band is different.In general,the similarity between TM and ETM+,ETM+ and OLI,OLI and MSI sensors in the NIR band is high,and the overall average similarity is up to 0.9919.The relative difference between NIR and SWIR1 band is less than 3%,and in the visible bands are close,from small to large are Green(8.14%),Blue(10.63%)and Red(12.03%).The SWIR2 band is most affected by the difference of the spectral response function,and the relative difference can reach 20.60%.When the difference of spectral response function is coupled to angle effect,the relative bias of OLI/MODIS in SWIR2 band can be as high as-97.41%,and TM/MODIS can reach-42.6% in Green band.(4)The difference in the reflection spectrum induced by the difference of spectral response function of the sensors can be further transferred to the selected commonly used vegetation indices(NDVI,EVI,SAVI,RVI),Char Soil Index(CSI),Normalized Burn Ratio(NBR),water monitoring indices(NDMI,NDWI)and shortwave infrared related remote sensing indices.The relative difference of the difference index of visible band such as NDVI and EVI is no more than3%,which is less affected by the spectral response function;while the relative difference of the ratio or weighted difference index of RVI,MIRBI,SWIR2/NIR and SWIR2/SWIR1 exceeds10%,and the difference is significant;SWIR2/SWIR1 can even reach 26.1%,which cannot be ignored.(5)The angle effect of the observation geometry can cause large difference in the reflectance and remote sensing indices,and it is far greater than the uncertainty induced by the difference in the corresponding spectral response function.Analysis from the two observation levels of pixel scale and multi-angle on the surface shows that: the absolute difference between the sensor’s spectral response function is proportional to the change of the view zenith angle,and it has a hot spot effect in the backward observation,and the relative difference is different in different bands.Among them,the absolute difference and relative difference of the visible band are proportional to the change of the view zenith angle.The absolute difference of the NIR band is positively correlated with the view zenith angle,and the relative difference is negatively correlated.The absolute difference of individual sensors in the SWIR band can still see the law of positive correlation,but the law of relative difference with the observation geometry is not obvious.In addition,when the view zenith angle changes within the range of ±65°,the relative difference of the 12 remote sensing indices caused by the observation geometry generally exceeds 20%,and the difference also fluctuates with the growth period.The maximum relative errors of NBR and NDMI of each sensor reached the above-mentioned highest values [99.5%,115.8%] and[161.6%,206.7%],respectively.(6)Studies have shown that the difference index form can suppress part of the angle effect to a certain extent,while the ratio and linear weighted form of the index have the potential to enhance the difference.Therefore,when multi-sensor data fusion is used in the research of satellite remote sensing monitoring application,full consideration should be given to the satellite-borne reflection and remote sensing index data with small differences or within the tolerance range in spectral response function and angle effect of the research problem(such as MSI and OLI).And avoid data series with large differences and potentially disruptive effects on the conclusions(such as the difference of blue band between MODIS and GF/HJ series),so as to avoid introducing non-negligible systematic errors into subsequent research.There are 53 figures,31 tables,and 139 references. |
PDF Full Text Request