Prediction Of Key Soil Attributes Using VIS-NIR Spectroscopy In A Spartina Alterniflora Invaded Coastal Wetland

Coastal wetlands are located in the trainsition areas between land and sea.They are one of the most economically valuable ecosystems,but they are extremely vulnerable to the invasion of alien species.Spartina alterniflora,for example,is a popularinvasive species in coastal wetlands in China.This species has presented an important impact on the soil system of the coastal salt marshes.Consequencely,soil monitoring in a rapid and accurate way will provide unique knowledge for coastal wetlands management under the oncidtion of exotic species invasion.The study area is located in the coastal wetland of Yancheng City,Jiangsu Province.This region has experienced S.alterniflora invasion over thirty years.A total of 45 samples from 15 sites were collected from three soil layers(0-30,30-60 and 60-100 cm)using the space-for-time substitution method.The maximum invasive age of the soil samples was identified as 17 years.In the library,ten key soil physicochemical properties(organic carbon,inorganic carbon,total nitrogen,p H,salinity,water content,bulk density,clay,silt,sand)were determined.The spectral reflectance of soil samples were measured using ASD.Preprocession and transformation on spectrum was conduncted to analysis their influence on prediction,resulting in six types of spectral transformation types(original spectral reflectance R,reciprocal 1/R,first-order differential R' Second-order differential R ",reciprocal first-order differential(1/R)' and second-order differential(1/R)").Based on these data,selection of spractral features were performed using the correlation analysis.Additionally,the age of plant invasion and soil depth were used as auxiliary variables included in the model.Models were calibrated using partial least square regression(PLSR)and and random forest(RF)for each soil property.The major objective was to test the predictive capacity of spectral reflectance by exploring the effects of different modeling methods,band selection,and auxiliary variable selection on model accuracy.Results shows that the invasion of Spartina alterniflora has resulted in a significant increase in the content of organic carbon and total nitrogen in the topsoil,and a decrease in salinity.Soil organic carbon and salt content can be properly predicted using PLSR(RPD> 2),and the prediction accuracy of water content is lowest(RPD = 1.49),.For RF modeling,predictions of clay and salt content(RPD> 1.9)outperform other soil properties,and soil water content,silt and sand content cannot be accurately predicted(RPD <1.4).Our results show that PLSR is preferable when dealing with a small sample size.RF can avoid overfitting and provide stable predictions,but it is sensitive to the presence of variables.The results also highlight the valuable of spatio-temporal auxiliary variables in soil prediction.This study tested the suitability of the soil spectroscopy for soil prediction in invaded coastal wetlands colonized by Spartina alterniflora using partial least squares and random forest model combined with environmental auxiliary variables.In such areas,the spectral thechnique can be routely used as an efficient method in soil monitoring,and better predictions can be achieved by incorporating spatio-temporal auxiliary variables.