| Taihu Lake is the largest freshwater lake in the Yangtze River Delta, China, its increasing water pollution and eutrophication problem seriously hindered the economic development and the health of local residents. Estimating the content of water components and monitoring water quality in Taihu lake using the remote sensing technology is of great convenience and significance. Chlorophyll-a is the main indicator of water eutrophication, also one of the most important parameters for water quality monitoring.In order to improve the performance and validation accuracy of existing estimation models, this study take the temporal and spatial variation of water quality into account and promote a new method to estimate chlorophyll a concentration (Chla) in Taihu lake. Based on19field experiments from2004-2012in Taihu lake,18typical Chla estimation models were validated and model residuals’variation with time and position was analyzed. This study proposed a rule for water quality division using the input of month and lake region, and constructed a new Chla estimation model that considers temporal and spatial variation, which can improve the accuracy of Chla estimation when validated by other datasets. The main contents and conclusions are as follows:(1) Validation of typical Chla estimation models using data in Taihu lake and analysis of the model residuals.Based on several different datasets,18existing Chla estimation models was validated and model residuals’ variation with month/season and lake district, as well as its influencing factors were analyzed. The results show that:the applicability of R-NIR algorithms in Taihu lake is better than the fluorescence algorithms, and the accuracy of model validation can be greatly improved after reparameterization of model parameters (RMSE is less than20mg/m3) or tuning of band position. Model accuracy validated by different datasets varies a lot, and the model residuals change with the sampling time and space, indicating that the accuracy of inversion model is correlated with the spatial and temporal variation of water body.(2) Estimation of Chla considering seasonal variation and model improvement.This paper studied the data transformation method for improving the accuracy of model application based on adjacent monthly data, and constructed a new Chla index in water based on data in summer and autumn. Then the Chla estimation model considering the seasonal variation was constructed. The results show that:logarithmic transformation and spectral kernel regression smoothing can be used to improve the homogeneity of model residual and the accuracy of model validation, for example, when a model built by data in July2004was validation by data in August2004, the RMSE reduced from33.56mg/m3before smoothing to25.60mg/m3after smoothing. A new chlorophyll index (NCI=(R690/R550-R675/R700)/(R690/R550+R675/R700)) was built based on data in summer and autumn, which presented better accuracy when validated by multiple datasets than the existing three band and four-band model. The accuracy and residual distribution can be partially improved by constructing Chla estimation model using data after seasonal devision.(3) Analysis of spatial and temporal difference of water quality and its division rule.Based on the spatial and temporal difference of water quality of Taihu lake and the spatial and temporal distribution of water quality parameters, this study promoted a division rule of water quality using the input of month and lake district and the method of CRT and C5.0decision tree. The investigated water was divided into three types according to the Chla level, and the division rule is as follows:(Dlf the month∈(3,5)|(month∈(4,8,9,11)®ion∈(centeral lake)), then it was divided into type Ⅰ;②If the month(6,7,10)|(month∈(4,11)®ion∈(Meiliang Bay, Zhushan Bay)), then it was divided into type Ⅱ;③If the month∈(8,9)®ion∈(Meiliang Bay, Zhushan Bay), then it was divided into type Ⅲ.The remote sensing reflectance and optical properties of these three types of water bodies are able to distinguish from each other, representing the water that suspended sediment dominated, both suspended sediment and phytoplankton dominated, and phytoplankton dominated, respectively. The distinguishability of optical characteristics among the three types divided by the new rule, as well as its correlation with Chla is better than those divided by Chla grade or spectral classification.(4) Construction of Chla estimation model in Taihu lake considering the spatial and temporal difference.The modeling and validation datasets were divided based on the constructed rule. Construct the band ratio, three-band, four-band and NCI model using the modeling dataset, and determine the best model expression and parameters for each type in the new model by comparing model precision and residual distribution. The expression of the new model is Chla=exp(ax2+bx+c), and x is R701/R677、(1/R686-1/R695)×R710、 NCI for type Ⅰ、Ⅱ、Ⅲ. The validation accuracy of the new model (R2=0.75, RMSE=11.54mg/m3) is better than the model after seasonal division (R2=0.56, RMSE=17.63mg/m3) and the reparameterized bands ratio, three-band, four-band and NCI (R2<0.61; RMSE>16.79mg/m3). Model application based on HJ1/HSI data show that the temporal and spatial distribution of Chla calculated by the proposed model is comparable to existing investigation. The Chla estimation model that considering the spatial and temporal variation built in this study can improve the accuracy of model application, thus can be applied to estimate chlorophyll a concentration in Taihu Lake. |
PDF Full Text Request