Three Dimensional Variation And Risk Assessment Of Soil Salinity In A Coastal Saline Land

With the rapid development of industrialization and urbanization,the contradiction between the demand and the supply of cultivated land is outstanding day by day.Most coastal provinces look upon the tidelands as an important reserved land resource to be developed for industry and urban construction.Since the founding,more than 200 000 ha of the tidelands in Zhejiang Province in south-east China has been reclaimed for agriculture.As a result,reclaiming the tidelands is becoming a more and more important method to implement the implementation of Cultivated Land Requisition-Compensation Balance Policy.However,the soil of the reclaimed land is still too saline and the salt can be brought to the soil surface by capillary transport.It retards the growth of crops and constrains agricultural production.Thus it is significant to illuminate the spatial variability of soil salinity and assess the risk of soil salinization for scientific amelioration,management of saline-soil resources and the improvement of the quality in the salinization farmlands.In this study,modern sensor such as the Geonics EM38 conductivity meter was employed to record the apparent electrical conductivity,ECa at batches of points in a paddy rice field in reclaimed coastal saline land in the north of Zhejiang Province and south of Hangzhou Gulf of the Yangtze Delta in China.Besides,we explored the inversion of ECa by a linear model and modeled electrical conductivity profiles in study area.Based on it,a trend was revealed in the result.Linear Mixed Model,LMM was then used to separate the data with two parts:a fixed effect of the trend and a random residual to estimate the parameters of the model by residual maximum likelihood,REML.Then by empirical best linear unbiased predictor(E-BLUP)we predicted the salinity at the nodes of a fine grid for mapping.Furthermore,we employed the indicator kriging,IK to predict and explore the risk or probability that the soil salinity exceeded a critical threshold at each location.Detailed study results include following four parts.1.Inversion of the ECa data and the descriptive statistics of soil salinityThe apparent electrical conductivity data at different depth layers was inversed to be the data source by using a linear model and the results of descriptive statistics showed that the apparent soil electric conductivity had a well correlation with salinity in the study area.Based on the descriptive statistics of soil salinity,it is apparent that the salt contents in all soils are all moderate in spatial variability.2.Estimation of the fixed effects of the trend and the random residuals and pseudo three-dimension digital mapping of soil salinityA strong trend was discovered and was coped with RK and LMM respectively.Once we have determined the most suitable structure for the LMM,the parameters were estimated by REML.Then we used E-BLUP to predict the ECa on a regular three-dimensional grid.We evaluated the coefficients of the trend and the parameters of the covariance of the residuals by REML and gained a more gratified result.For comparison,we both employed spherical model and exponential model to find the exponential model fitted somewhat better.Then we compared the predictions of E-BLUP and RK by cross-validation by the leave-one-out technique and calculated the mean errors(ME),mean squared errors(MSE)and mean squared deviation ratios(MSDR).In the result,all combinations of models and kriging methods gave small average errors,as expected;kriging is after all unbiased.And the differences among the MSEs are small.If we judge the goodness of a combination from the mean squared deviation ratio,the MSDR,we concluded that our predictions by LMM with an exponential variogram are the most valid.The combination was taken to map the predicted ECa and associated errors.3.Study on three-dimension spatial variability of soil salinity by 3D kriging methodThough we modeled the pseudo three-dimensional variation in ECa as a series of correlated two-dimensional regionalized variables,the trend components in the vertical direction was ignored,that appeared to be a general increasing trend in salinity with increasing depth.So we turned to model the full three-dimensional variation in salinity,taking into account both the lateral and vertical trends.And both an isotropic variogram function and a geometric anisotropic variogram function were chosen to predict the salinity in the three dimensions by kriging.Then we compared the predictions by cross-validation.As a result,the summary validation statistics for the geometric anisotropic model appeared to be acceptable.The mean square deviation ratio,MSDR,was 1.05,and the estimated result seemed better than the isotropic model's under the selection of AIC.Then the combination was taken to visualize the predicted ECa data in three dimensions.From the map of three-dimensional kriged predictions,a trend was evident in the topsoil layers;whereas at the southern end of the field the topsoil seemed much more saline than that at northern end of the field where ECa was only weakly saline.The spatial distribution of ECa can be attributed to several factors.Refer to the remote sensing image,the high soil salinity within the field was located in the south-east,where there was a pool.And the shape of soil salinity profile was downward flowed presenting a certain appearance like a backward "V".It revealed that the soil salinity increased with depth in the field and accumulated in the subsoil.4.Predictive distribution of probability prediction of soil salinity and risk assessment of crop plantingIn view of the fact that soil salinity has severely disrupted local agricultural production,we determined to explore the risk-or probability that the soil salinity exceeded a critical threshold in the study area.As the FAO suggested that soil salinity equivalent to an ECa of 123 mS/m would be likely to lead to a 25%reduction in rice yield compared with non-saline soil,the study thus set the threshold as 123 mS/m.Since the kriging predictor yielded both a prediction of ECa and an estimation of the uncertainty of this prediction at each point in the field,we can easily determine the probability that this threshold was exceeded.In the results,high probability was concentrated mainly in the south-east part of the studied region,while low probability in the north-west.