Turbidity observation in a river is important for understanding sediment dynamics and the river environment. However, turbid water often occurs during flood events, which have risks such as damage to observation equipment and the safety of observers. If a non-contact method for observing turbidity can be developed, these risks can be avoided. In this study, we examined the possibility of estimating turbidity using artificial neural network (ANN) with color information obtained from images taken continuously in a river using time-lapse camera. The results showed that it is difficult to estimate turbidity based on color information alone. On the other hand, the addition of water level data to ANN improves the turbidity estimation over the conventional method based on the L-Q equation. Furthermore, the ANN with color information, water level, and light conditions suggests the possibility of accounting for hysteresis loops that cause differences between the rise in water level and the rise in turbidity during flood.