Gridded Water Properties for Chesapeake Bay for Model Validation, Development, and Discovery
Extensive measurement networks exist to monitor salinity and temperature in Chesapeake Bay due to their control of Bay productivity, organic matter production, and phytoplankton community composition. Unfortunately, currently available data products may be of limited use for applications such as validation of next-generation Earth systems models (such as the DOE E3SM model), which aim to represent thermal, sediment, salinity, and nutrient exchanges between rivers and the land on a global basis. To satisfy this need, we developed a machine learning pipeline for generating daily, high-resolution, gridded datasets based on remote sensing imagery. A key innovation of our study is an augmentation of imagery with estimates of the freshwater influence of Bay tributaries.
The performance of salinity and temperature predictions was as good or better than prior efforts, and we found particular improvements in model performance in low-salinity tributary areas. We attribute these improvements to the realism introduced by our freshwater influence model features, which encourage physical consistency in upstream-downstream spatial patterning. Estimates produced by our models have a high potential to inform the calibration and evaluation of ESMs as well as in other applications, such as habitat suitability and sea level rise models.
Remotely sensed water properties are important for a variety of applications, including validation of Earth systems models (ESMs), habitat suitability models, and sea level rise projections. For the validation of next-generation, high or multi-resolution (30 to 60 km) ESMs in particular, the usefulness of operational forecasting products and directly observing satellite-based sensors for validation is limited due to their temporal availability and spatial resolution. To address this validation data gap, we developed a data-driven model to produce high-resolution estimates of temperature and salinity over decadal time scales as required by next-generation ESMs. Our
models include novel predictive features to capture the influence of flow and surface water exchange in upstream coastal watersheds. These features boosted the accuracy and performance of salinity and temperature models and, as a result, have a high potential to inform calibration and evaluation of ESMs as well as in other applications, such as habitat suitability and sea level rise models.