Due to technical limitations and the high cost of hazardous waste site clean up, there has been a shift toward risk-based long-term management of sites, where contamination is left in place. This study demonstrates how integrating all available site data can improve long-term monitoring, operation, and stewardship (LTMOS) decision making and provide cost savings. A learning machine is used to integrate historic and current data from the 317/319 Area phytoremediation site at Argonne National Lab-East (ANL-E). The learning machine uses these data and daily weather data to build a model to forecast groundwater head levels. Development of the learning machine framework provides a method for integrating the diverse data sources available at the site and using that information to determine the importance of each data source in achieving monitoring objectives. Future work will determine how long the historical record will retain its accurate predictive capability and whether the value of the surrogate data (continuous samples and rainfall) increases over time. In this preliminary study, the entire historical quarterly dataset was shown to be the most important data source, which could be used to predict future water levels with far more accuracy than the most recent quarterly dataset alone.