High water turbidity in aquatic ecosystems is a global challenge due to its harmful impacts. A cost-effective manner to rapidly and accurately measure water turbidity is thus of particular useful in water management with limited resources. This study developed a novel framework aiming to predict water turbidity in various aquatic ecosystems. The framework predicted water turbidity and quantified the uncertainty of the prediction through Bayesian modeling. To improve model performance, a model-update method was implemented in the framework to update the model structure and parameters once more measured data were available. 120 paired records (an image from smartphone and a measured water turbidity value by standard turbidimeters for each record) were collected from rivers, lakes and ponds across China to evaluate the performance of the developed framework. Our cross-validation results revealed a well prediction of water turbidity with Nash-Sutcliffe efficiency (NS) >0.87 (p<0.001) during the training period and NS>0.73 (p<0.001) during the validation period. The model-update method (in case of more measured data) for the developed Bayesian models in the framework resulted in a decreasing trend of model uncertainty and a stable mode fit. This study demonstrated a high value of the Bayesian-based framework in predicting water turbidity in a robust and easy manner.