Probabilistic Streamflow Forecasts for Mississippi River Navigation Risk and Parametric Insurance

Jennifer Nakamura , Xander Nakamura, Upmanu Lall

Mississippi River navigation is disrupted by both low and high water extremes, yet shippers, ports, and terminals lack a simple instrument to transfer that risk. We test whether probabilistic forecasts of monthly streamflow can provide useful navigation-risk information. Using pretrained Chronos-2 at three major Mississippi River subbasin outlets, we compute navigation risk probabilities, parametric insurance premiums, and basis risk using land surface, atmospheric, and ocean predictors. Land surface and moisture-budget predictors provide the strongest forecast, with a probabilistic forecast skill relative to climatology declining from 0.30 at one month to 0.08 at four months. A perfect-future experiment that supplies the observed future land state while withholding discharge retains much greater skill, 0.44 at one month and 0.34 at four months, showing that uncertainty in the future land conditions is the main source of loss of streamflow predictability. Low-, high-, and combined-water probabilities are evaluated at 1-4 months, to assess whether forecast-based risk pricing can send price signals for anticipatory action. Results support a 1-4 month decision time line and motivate initiating seasonal forecasts coupled to a land-surface or hydrologic model.