Bayesian Inverse Modeling For Solute Transport And Transformation In Soil

To better understand the environmental behaviors of soil solutes(e.g.,nutrients and pesticides),numerical models have been widely applied to simulate solute transport and transformation.The solute transport and transformation processes are essentially deterministic in soil but subject to various uncertainties,due to the scarcity of observations and the spatial heterogeneities of soil physical,chemical and biological properties.To cope with complicated natural conditions,transport and transformation modeling not only needs to identify the correct models(model structural uncertainty),but also to obtain proper parameters for the models by considering their uncertainties.To accurately simulate the solute transport and transformation in soil and understand the inherent mechanisms,it is important to effectively infuse information from noisy observations,and identify the correct model structure and optimal parameters to reduce the associated uncertainties.Bayesian inference is a probabilistic method,which has been widely applied in numerical modeling studies.The Bayesian inverse modeling incorporates prior knowledge and observations into a posterior probabilistic description of quantities of interest.The current work combined Bayesian method with numerical models to inversely identify the key processes and their associated parameters of nitrogen and Pentachlorophenol(PCP)transport and transformation in soils and characterized the contribution of each process in their dissipations.The details of our findings are listed as follows:(1)Bayesian inference for kinetic models using a generalized rate equationThe selection of a kinetic model is important in the simulation of solute transport and transformation in soil.Nevertheless,it is difficult to determine an appropriate form of the rate equation in the kinetic model.To reduce the model structural uncertainty in selecting a kinetic model,we developed an integrated Bayesian approach to simultaneously perform model selection and parameter estimation by using a generalized rate equation.In the approach,the model hypotheses were represented by discrete parameters and the rate constants were represented by continuous parameters.Then Bayesian inference of the kinetic model was then solved by a Markov chain Monte Carlo(MCMC)algorithm for parameter estimation with the mixed(i.e.,discrete and continuous)priors.The validity of this approach was illustrated through a synthetic case and a complex nitrogen transformation experimental study.In the synthetic case,the proposed approach could precisely identify the underlying dynamic model and estimate the parameters in the presence of different levels of observational errors and observation deficiencies.In the nitrogen transformation experimental study,the approach successfully identified the optimal model from 19683(39)candidates,and the simulated results accurately matched with the observations.(2)Identification of the dominate N2O formation pathway at different temperature and water-filled pore space(WFPS)conditions in a paddy soilFurthermore,we used the complex nitrogen transformation model to simulate nitrogen cycle under different temperature and WFPS treatments.MCMC was applied for model selection and parameter estimation.The simulated results matched well with the concentrations of NH4+,NO3-and NO2-in all the treatments,as well as with the N2O emission rates in the treatments of 20T90W(T = 20 ? and WFPS = 90%,similarly hereinafter),20T120W,30T90W and 30T120W.It was inferred that the dominant N2O formation pathways under 30T120W treatment were ONH4-(ae)-N2O(s).(from autotrophic nitrification)and RNO2-(an)-N2O(s)(from heterotrophic denitrification or dissimilatory nitrate reduction to ammonium(DNRA)).and ONH4+(ae)-N2O(s)+ was the dominant N2O formation pathway in treatments of 30T90W,20T90W and 20T120W.N2O could emit from soil in the treatments of high temperatures and high moisture contents,but it was almost entirely reduced to N2 under other treatments.(3)Bayesian model averaging(BMA)for the inference of nitrogen transformation pathways in a paddy soilAlthough a complex model was used in the above study,the simulations of N2O emission in the treatments of 5T(T = 5 ?)or 60W(WFPS = 60%)could not match the observations well.To fully consider the uncertainty in model structure and avoid the potential bias with a single model,we constructed 12 nitrogen transformation models with different combinations of nitrogen transformation pathways.BMA was then employed to integrate the above models by calculating the model weights.More reasonable simulations for the treatments of lower temperature and lower moisture content were then obtained and a comprehensive uncertainty analysis was implemented.The results showed that BMA was able to integrate the candidate models to fully consider the model structural uncertainty,which resulted in better predictions.(4)Bayesian inverse modeling of PCP dissipation at the aerobic-anaerobic interface of a paddy soilTo verify the applicability of the inverse modeling approach in more complicated scenarios,we applied a reactive transport model to simulate PCP dissipation in the aerobic-anaerobic interfacial region at the soil-water interface of a paddy soil.MCMC was applied to inverse unknown parameters of the reactive transport model,which considers the diffusion,sorption and degradation processes of PCP and its metabolite at the soil-water interface.The simulation from maximum-a-posteriori(MAP)estimation precisely demonstrated the PCP dissipation at the aerobic-anaerobic interface of a paddy soil and characterized the contribution of each process in PCP dissipation.The results indicated that the most reactive zone for PCP dissipation occurred in the 0-6.0 mm layer where degradation in solid phase dominated the PCP dissipation,while diffusion process transported PCP from the deeper layer(2.4-4.8 mm)to the active degradation zone(0-2.4 mm)to facilitate PCP dissipation.By considering the coupled reactive transport of PCP and Cl-,the average degree of PCP dechlorination in each layer was estimated from the corresponding total concentrations of PCP and Cl-.The degree of PCP dechlorination in the ponding water was highest,while 2,3,4,5-TeCP and 3,4.5-TCP were identified as the main dechlorination products in the soil.In summary,the approaches developed in this study can greatly facilitate model selection and parameter optimization using limited observation data.Our methods can effectively reduce uncertainties in modeling,help to reveal more mechanistic insights of the solute transport and transformation processes in soil,and can be used as a powerful tool for data mining in studying complex soil processes.