Summaries - Research
Back Improving Tactical Environmental Support in Data Denied Areas: Applications of Machine Learning (ML)
|Division||Research & Sponsored Programs|
|Department||NPS Naval Research Program|
Murphree, James T.
Nuss, Wendell A.
|Sponsor||NPS Naval Research Program (Navy)|
|Summary||Battlespace decisions rely on: (a) accurate forecasts of environmental conditions (winds, clouds, waves, etc.) and the operational impacts of those conditions; and (b) information about the uncertainty or probability that the conditions will occur as forecasted. Dynamical models of the atmosphere and ocean are the primary tools used to produce operational forecasts. However, in data denied regions, where adversaries limit access or observations do not exist, dynamical models may have limited accuracy due to a shortage of observational data to ingest into the models. This lack of data also reduces our ability to assess and correct forecast errors via post-processing of the model outputs. Post-processing involves comparing many prior forecasts to observations to identify characteristic errors, and then adjusting future forecasts. Identifying the characteristic errors, and the most effective and computationally efficient ways to correct them, is challenging. This is particularly problematic when very limited observations are available to assess errors. ML or AI methods can be used to predict forecast uncertainty and quantify the probabilities of occurrence of specific weather or climate regimes. These methods work best when ample data is available to train the model, but some methods can be effective with limited training data. We will develop and test ML/AI methods for use in improving forecasts and uncertainty information in data limited regions. Bayesian multivariate multiple linear regression has shown promise with limited training data, if intelligent choices are made about the training data. Other methods also show promise and will be tested. The primary outcome from this project will be the identification of methods for operationally producing more accurate and useful environmental information for use in battlespace planning and other operational decision making. Primary deliverables will be a final report detailing results, scientific publications, and code.|
|Publications||Publications, theses (not shown) and data repositories will be added to the portal record when information is available in FAIRS and brought back to the portal|