Study Of Algal Bloom Occurrence Probability Prediction Model Based On Remote Sensing Data And Meteorological Factors In Dianchi Lake

Algal bloom is a phenomenon in which phytoplankton floats and accumulates to the water surface under suitable conditions.A large number of plankton algae float on the water surface and change water color.The occurrence of algal bloom is an inevitable result of eutrophication.It is not only induced by the algae biomass,but also related to meteorological conditions,nutrients concentration.However,due to the still unclear mechanism of bloom occurrence,current models are not enough to determine the relationship between these parameters and bloom occurrence precisely.The long-term remote sensing data recorded provide an effective method for identifying the quantitative relationship between bloom occurrence and its related possible inducements.As one of the super eutrophic lakes in China,Dianchi Lake has suffered from high-frequency blooms and so do-Taihu Lake and Chaohu Lake.The governments have input plenty of resources and finances in the treatment of Dianchi Lake,but concentrations of nitrogen and phosphorus and suitable climatic conditions have caused algal blooms to occur almost all year round,seriously affecting the water ecological environment.Therefore,predicting the occurrence probability of algal bloom in Dianchi Lake is an urgent need for water ecological environment management.Base on the satellite remote sensing data from 2002 to 2018,this research firstly analyzed the relationship between the bloom occurrence and meteorological factors,and determining the driving factors for the bloom occurrence in Dianchi Lake.Then,an estimation model for the column-integrated biomass(CIB)over Dianchi Lake was established.Finally,based on meteorological factors and different algae biomass parameters,prediction models of the bloom occurrence probability on the lake scale and pixel scale were constructed,which can provide technical and data support for relevant departments.The research contents and conclusions are as follows:(1)Study on temporal and spatial distribution characteristics of algal blooms and CIB over Dianchi Lake.Firstly,the temporal and spatial distribution characteristics of algal bloom in Dianchi Lake were extracted through FAI based on MODIS images from 2002to 2018.Algal blooms occur throughout the year and particularly serious in late summer and early autumn.The algal blooms are concentrated in the north of Dianchi Lake during the dry seasons but extended to the whole lake in the wet seasons.Secondly,a CIB estimation model under non-bloom conditions was established based on the 72measured chlorophyll-a concentration data and the remote sensing reflectance difference at 859nm and 443nm.The RMSE,MAPE,and MRPE were 36.93mg/m~2,12.56%,and-5.63%,respectively.The areas near the edge of Dianchi Lake present higher CIB than that of the center;CIB in the west lake is higher than that in the east lake.CIB shows an obvious seasonal increase from winter to autumn.In addition,the comparison of spatiotemporal distribution between algal blooms and CIB indicates waters near the blooms have a higher CIB,and CIB on the non-bloom day has a stable range.There is a significant negative correlation between the bloom area on bloomday and the total algal biomass(R~2=0.43,p<0.001)in Dianchi Lake,especially when the bloom areas greater than 25km~2.(2)Study on the relationships between algal bloom occurrence and meteorological factors.Firstly,the statistical features between the meteorological variables on the bloomday and non-bloomday reveals blooms are more likely to occur under the weather conditions of low wind speed,low sunshine hours,high air pressure,high temperature and high humidity.Large bloom areas are more likely to occur under wind directions of the south,southeast,and south-southeast.Then,based on relative importance analysis and collinearity analysis,five meteorological variables,ws＿mean＿b0(the mean wind speed on the prediction day),shs＿b0tob3sum(the sum of sunshine hours in the previous 3 days),prsmean＿b0(the mean air pressure on the prediction day),atmean＿b0tob7sum(the sum of mean air temperature in the previous 7 days),and rhumean＿b0(the mean relative humidity on the prediction day)were considered to be the driving factors for the bloom occurrence in Dianchi Lake.(3)Probability prediction model construction of blooms occurrence at lake scale based on the Logistic method.The dependent variable was blooms presence or absence(1/0),the independent variables were continuous meteorological variables(wsmean＿b0,shs＿b0Tob3sum,prsmean＿b0,atmean＿b0Tob7sum,rhumean＿b0),and the output is the blooms occurrence probability of Dianchi Lake.The AUC(area under the ROC curve)of the training dataset is 0.81,and the percentage of correctly classified instances(CCI%)of the validation dataset is 75.86%.Next,a Logistic-based occurrence probability prediction model of the extended bloom in Dianchi Lake was established.The dependent variable was extended blooms presence or absence(1/0),and the independent variables were wsmean＿b0 and shs＿b0Tob3sum.The AUC of the training dataset is 0.78,and CCI%of the validation dataset is 85.71%.The logistic-based predictive models are easy to implement,however,it regards Dianchi Lake as a whole,and cannot predict the occurrence probability of algal bloom at specified locations.(4)Bloom occurrence probability models(NB＿FAI and NB＿CIB)were constructed at pixel scale using different bloom indicator variables(FAI and CIB)based on Bayesian theory.Firstly,meteorological factors(wd＿b0,wsmean＿b0,shs＿b0Tob3sum,prsmean＿b0,atmean＿b0Tob7sum,rhumean＿b0)and bloom indicator variables were classified and assigned.Then,the conditional probabilities were calculated under bloom and non-bloom conditions,respectively,the bloom occurrence probability were selected at monthly scale as the prior probability,and the look-up table of the posterior probability of algal bloom occurrence was calculated.Both NB＿FAI and NB＿CIB have more than 90%of the pixels with AUC values between 0.8 and 1,and the CCI%of most pixels in the verification dataset are between 0.5 and 1,indicating that the model achieved good prediction results.Moreover,the prediction result on the same day of NB＿FAI and NB＿CIB does not show a significant difference.A slight overestimation has been appeared in NB＿FAI,especially on the non-bloom pixels.However,owing to the sophisticated calculation process of CIB,it is recommended to use FAI to construct the na?ve Bayes model for blooms occurrence probability in inland lakes.The model framework established in this study can be extended to other eutrophic lakes,which can provide technical and data support to deal with algal bloom emergencies for the management department.