Many meandering rivers migrate, at rates that vary both along-stream and inversely with the observation interval. Many numerical models have been developed to predict this migration; their success is usually evaluated statistically or by qualitative comparison to observations in map view. We propose a framework to test migration models that unites these statistical, spatial, and temporal perspectives. We measure model fit with a statistic that compares the magnitude and direction of migration between predictions and observations. Model fit is contextualized in space, using a dimensionless coordinate system based in the location along a half-meander bend; and in time, using a dimensionless observation interval that accounts for channel scale and migration rate. We applied this framework to test predictions for a curvature-driven model of channel migration, using data from seven rapidly migrating rivers in the Amazon Basin and 103 more slowly migrating rivers across the continental US, as reconstructed from a legacy data set. We find that across both datasets, channel migration rates peak slightly downstream of the bend apex. Migration rate underestimation/overestimation tends to occur when the observed rate is greater/less than its median along the channel. Predicted migration direction opposes observations for slowly migrating locations and upstream of the bend apex. Model forecasts break down if the channel migrates by more than its width. The analysis framework is portable to testing other models of channel migration, and can help improve predictions for the stability of infrastructure along rivers and for landscape change over geologic timescales.
