Braided rivers switch between quiet and active periods of bedload transport while their planform changes quickly. This makes both simple descriptive indices and heavy morphodynamic models hard to use in practice. This thesis offers a practical middle path: it treats channel change as a sequence of morphological states read from planform images and models how the river switches between them with a continuous-time Markov model. The aim is to turn images into probabilistic forecasts of sediment transport with stated uncertainty.

From flume imagery, binary water masks are compared with two complementary measures that capture edge movement and area overlap. We reduce these pairwise differences and cluster the images to obtain a small, readable set of recurring states, ranging from narrow and simple to wide and partitioned. The time the river stays in a given state is well described by an exponential law, which allows us to estimate transition rates and jump probabilities for the Markov model.

The resulting ensemble recovers means, variances, extremes, and main time scales, and it shows clear morphological control of transport, including a negative link with wetted width and with a braiding index. Splitting the variance indicates that differences between states explain a meaningful part of the instant variability, while the rest arises within states. The learned states and transitions remain stable across independent runs.

For image-only cases, two variants extend the method: one that preserves the long-term mean, and another that uses stream power from images to scale state-wise means. Weighting by how long the river stays in each state keeps the overall mean accurate when direct bedload data are missing. Overall, the framework provides a clear path from images to forecasts, explains intermittency as switching among states with different export capacity, and enables practical predictions with quantified uncertainty.