This paper analyses budget-constrained, nonpoint source (NPS) pollution control with costly information acquisition and learning. To overcome the inherent ill-posed statistical problem in NPS pollution data the sequential entropy filter is applied to the sediment load management program for Redwood Creek, which flows through Redwood National Park in northwestern California. We simulate the dynamic budget-constrained management model with informationacquisition and learning, and compare the results with those from the current policy. The analysis shows that the manager can reallocate resources from treatment effort to information acquisition, which in turn increases overall treatment effectiveness, and reduces sediment-related damage.