Abstract：Soil salinity is an important index to evaluate soil quality， and it is also one of the main environmental factors affecting the growth of Suaeda salsa (S.salsa) in the coastal wetland of Liaohe Estuary. It is necessary to propose a real-time， accurate and large-scale monitoring method for the soil salinity of S.salsa community and surrounding tidal flats. In order to reduce the influence of the atmosphere on the model， ground hyperspectral data was used to simulate the reflectance of Landsat 8 OLI satellite and a stepwise regression analysis method based on cross-validation was introduced to construct a soil salt inversion model. The result are shown as follows. 1) The soil salinity of S.salsa samples are significantly lower than that of tidal flats. The soil salinity in the region of Hainan San is lower than that in Yuanyang Gou and Bijialing regions， but the plant height and biomass are higher than those in Yuanyang Gou and Bijialing regions， which indicated the effect of soil salinity on the growth of S.salsa to a certain extent. 2) The correlation between multispectral indices constructed by the simulated satellite reflectance and soil salinity is improved as a whole compared with the correlation between the single band and soil salinity. NDVI and RVI have the high correlation with soil salinity， the values reach—0.689 and—0.683. 3) The stepwise regression analysis method based on cross-validation was used to construct a soil salinity inversion model. The independent variables of the model were RVI， SAVI and SI3. The fitting accuracy of modeling set R2 is 0.684， with the root mean square error (RMSE) of 3.45， the validation set RMSE of 1.88， and the relative analysis error(RPD) of 2.28， suggesting that the model has good inversion accuracy and inversion ability. In order to further verify the accuracy of the model， a multivariate regression inversion model based on the factors screened by stepwise regression analysis is compared and analyzed. The results indicate that the R2and RMSE of the cross-validation stepwise regression model are better than those of the multivariate regression model. Meanwhile， the scatter plot of soil salinity inversion value and measured value are closer to the 1∶1 line. It provides technical and data support for the inversion of soil salinity factors of S.salsa community and tidal flats in the northern Liaodong Bay.
2022, Vol. 56 
