Study On Remote Sensing Of Water Quality In Lakes And Reservoirs And Data Assimilation For Hydrodynamic And Water Quality Model

Water is an indispensable resource for human survival and society development.With the increase of global population,economic development and urbanization,water environmental problems become more and more serious.Observation analysis and model simulation are two important tools in water environment management.In-situ observations and remote sensing are two main observation methods.Although in-situ observations have high monitoring accuracy,they are time-consuming,labor-intensive,limited frequency and can only obtain water quality status at gauges.It cannot be suitable for dynamic management of water environment.Remote sensing can quickly map the spatial and temporal distribution of water quality in lakes and reservoirs,and has been widely applied in water environment monitoring and management of lakes and reservoirs.However,the frequency of remote sensing monitoring is also limited due to finite remote sensing images and poor transplantability of remote sensing retrieval models.Water quality models are important tools for water environment management.The models can simulate spatial-temporal variation of water quality parameters,but the simulation accuracy of the models should be further improved due to the uncertainties associated with model parameters,inputs and model structures.Data assimilation can achieve complementary advantages of observation and simulation through reasonably integrating multi-source observation data into water quality models.It can modify model simulation results,synchronously update model parameters,and improve model simulation accuracy.This study focuses on remote sensing of water quality in lakes and reservoirs,hydrodynamic and water quality simulation and assimilation.We modified the algorithms for remote sensing of water quality and developed well-balanced and efficient hydrodynamic and water quality models.In addition,observed water stages and water quality parameters were integrated into the hydrodynamic and water quality models via a Particle filter-based data assimilation algorithm to dynamically update model simulation results and model parameters.The main results of the study are as follows:(1)The development of the framework of data assimilation for hydrodynamic and water quality models.We developed the framework of data assimilation for hydrodynamic water quality models based on a modified particle filter-based data assimilation with local weighting procedure(MPFDALW).MPFDA-LW data algorithm considers the spatio-temporal variability of hydrodynamic and water quality model parameters.In MPFDA-LW data algorithm,particles involve water flow and water quality states as well as key parameters for model simulation.Observed water stages and concentration of water quality parameters can be assimilated into hydrodynamic and water quality models to modify simulated water stages,concentration of water quality parameters and model parameters via particle filter.MPFDA-LW data assimilation can adaptively update hydrodynamic and water quality models with observations.(2)The development of ensemble modeling methods for remote sensing of water quality.A variety of retrieval models have been proposed for remote sensing of water quality.Different retrieval models have different retrieval accuracies.In order to improve water quality retrieval accuracy,two deterministic ensemble methods,Entropy Weight-based Ensemble Model(EW-EM),Set Pair Analysis-based Ensemble Model(SPA-EM),and a probabilistic ensemble modeling model,Bayesian Model Averaging-based Ensemble Model(BMA-EM),as well as a Game Theory-based Ensemble Model(GT-EM)were developed for water quality retrieval.The ensemble models were used to retrieve Chlorophyll a(Chl-a)of Panjiakou and Daheiting reservoirs and the performances of the ensemble models in Chl-a retrieval were tested.The results indicated that ensemble modeling could improve Chl-a retrieval accuracy through integrating multiple Chl-a retrieval models.In addition,the BMA-EM could estimate uncertainty intervals for Chl-a retrieval.When the weights of multiple models adopted in different ensemble models were significantly different,game theory could determine comprehensive weights of the Chl-a retrieval models based on the weights determined by different ensemble modeling methods.The GT-EM based on the comprehensive weights may reduce the subjectivity in the selection of multiple ensemble models.(3)The development of a modified discrete binary particle swarm optimization-partial least squares(MDBPSO-PLS)model for water quality retrieval.We modified discrete binary particle swarm optimization(DBPSO)by combining with catastrophe strategy,called MDBPSO.MDBPSO has stronger global searching ability than DBPSO.Then MDBPSO was used to select partial least squares(PLS)modeling bands for water quality retrieval,then a Modified Discrete Binary Particle Swarm Optimization-Partial Least Squares(MDBPSO-PLS)was proposed.MDBPSO-PLS was respectively used to retrieve Chl-a,total suspended matter(TSM)and turbidity based on HJ-1A HSI imagery.The results indicated that MDBPSO could select the sensitive bands of PLS modeling for the three water quality parameters retrieval.MDBPSO-PLS has higher retrieval accuracy than PLS model.(4)The development of a method for remote sensing of water quality in macrophytic lakes considering aquatic vegetation phenology.Remote sensing is difficult to retrieve water quality in macrophytic lakes where aquatic vegetation grows due to mixed pixels.Aiming at this problem,we proposed a method for remote sensing of water quality in macrophytic lakes considering aquatic vegetation phenology: a case study in Weishan Lake.Weishan Lake was divided into the water overlying aquatic vegetation and the water area.We qualitatively monitored TSM and turbidiy in the water overlying aquatic vegetation by the water quality indicator of Potamogeton lucens,Myriophyllum spicatum and Potamogeton crispus in different phenological phases.Meanwhile,retrieval models were established based on HJ-1A/1B and GF-1 multiple-spectral images to quantitatively retrieve TSM concentration and turbidity in the water area.The method combined qualitative and quantitative monitoring was used to monitor spatial-temporal variation of TSM and turbidity in Weishan Lake.(5)The development of a 2-D hydrodynamic model based on adaptive grids and OpenMP parallel computation.Hydrodynamic models based on traditional adaptive grids are difficult to preserve C-property.To solve the problem,we modified adaptive grid technology by combining a new slope criterion.The girds generated by the modified adaptive grid technology can maintain C-property.In order to further improve model computational efficiency,we parallelized a 2-D hydrodynamic model based on Open Multi-Processing(OpenMP)project and proposed a 2-D Hydrodynamic Model based on Adaptive girds and Parallel computation(HydroM2D-AP).Flume experiments,physical models and actual cases were used to test the performance of the proposed model.The results indicated that HydroM2D-AP model can accurately and efficiently simulate different hydrodynamic processes over different terrain.(6)Uncertainty analysis and data assimilation for a two-dimensional hydrodynamic modelTwo uncertainty analysis methods,SCEM-UA and LHS-GLUE were used to analyze the uncertainty of Manning's roughness coefficient based on two likelihood functions in the case study on Toce River physical modelling.Meanwhile,the sensitivity of simulated water stages to Manning's roughness coefficient was analyzed.The results indicated that the uncertainty analysis depended on the likelihood function.SCEM-UA and GLUE with the same likelihood function provided similar estimates of normal posterior distributions of Manning's roughness coefficient.Simulated water stages were sensitive to Manning's roughness coefficients and the sensitivities showed obvious spatial-emporal variability.In addition,we developed two particle filter-based data assimilation algorithms for the HydroM2D-AP model with global(PFDA-GW)and local(MPFDALW)weighting procedures,respectively.MPFDA-LW and PFDA-GW for the HydroM2D-AP model respectively adopted spatially non-uniform and uniform Manning's roughness coefficients.We compared the performances of the two PF-based data assimilation algorithms.The results indicated that MPFDA-LW could significantly improve the simulation accuracy of water stages at all gauges simultaneously.By contrast,PFDA-GW could only slightly improve the simulation accuracy of water stages at a few gauges.Overall,MPFDA-LW is more suitable for data assimilation for two-dimensional hydrodynamic models.(7)The development of 2-D water quality models based on adaptive grids and OpenMP parallel computation.We combined the contaminant convection-diffusion equation with the HydroM2D-AP model to propose a two-dimensional hydrodynamic-pollutant transport model based on adaptive grids and OpenMP parallel computation(HydroPTM2D-AP).Flume experiments,physical models and actual cases were used to test the performance of the HydroPTM2D-AP model in simulating pollutant transport under different flow conditions.The results indicated that the HydroPTM2D-AP model can accurately simulate the characteristic of pollutant transport.In addition,we proposed a 2-D hydrodynamic and water quality model based on adaptive grids and parallel computation(HydroWQM2D-AP).The HydroWQM2D-AP model considers the interaction between ammonia nitrogen,nitrate nitrogen,phosphate,phytoplankton,carbonaceous biochemical oxygen demand,dissolved oxygen,organic nitrogen and organic phosphorus based on the principle of the Water Quality Analysis Simulation Program(WASP).The HydroWQM2D-AP model was used to simulate the spatial-temporal variations of water quality parameters in Poyang Lake.The results show that the HydroWQM2D-AP model can simulate the spatial-temporal variations of water quality parameters such as dissolved oxygen,ammonia nitrogen,total nitrogen,total phosphorus and chlorophyll a concentration in Poyang Lake.(8)Particle filter-based data assimilation for a 2-D water quality modelWe developed a modified particle filter-based data assimilation with local weighting procedure(MPFDA-LW)for the HydroWQM2D-AP model,a case study in Poyang Lake.The MPFDA-LW algorithm was used to integrate in-situ observed and retrieved Chl-a concentration data into the the HydroWQM2D-AP model to adjust simulation results of Chl-a concentration and update the model parameter for Chl-a simulation.The performance of the MPFDA-LW assimilation algorithm was investigated.The verification results show that the MPFDA-LW could adjust the simulation results of Chl-a concentration and update the model parameter at gauges through assimilating in-situ observations of Chl-a into the HydroWQM2D-AP model.The simulation accuracy of Chl-a at gauges was improved via data assimilation.The MPFDA-LW could modify the spatial distribution of Chl-a and estimate the spatial variability of the model parameter via assimilating Chl-a retrieval results into the HydroWQM2D-AP model.The spatial distributions of Chl-a derived from the MPFDA-LW were more consistent with the spatial distributions of remote sensing observations than those derived from the HydroWQM2D-AP model.The spatial distributions of Chl-a derived from the MPFDA-LW could provide more accurate initial conditions of Chl-a at assimilation seconds.The MPFDA-LW could integrate multi-source observations of Chl-a into the HydroWQM2D-AP model to modify simulation results of Chl-a and update the model parameter to improve simulation and prediction accuracy of Chl-a.