Correcting Systematic and State‐Dependent Errors in the NOAA FV3‐GFS Using Neural Networks. (31st October 2022)

• Record Type: Journal Article
• Authors: Chen, Tse‐Chun; Penny, Stephen G.; Whitaker, Jeffrey S.; Frolov, Sergey; Pincus, Robert; Tulich, Stefan
• Is Part Of: Journal of advances in modeling earth systems. Volume 14:Number 11(2022)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2022-10-31
• Subjects: neural networks -- model error -- data assimilation -- numerical weather prediction -- machine learning; Geological modeling -- Periodicals; Climatology -- Periodicals; Geochemical modeling -- Periodicals; 551.5011
• DOI: 10.1029/2022MS003309
• Languages: English
• ISSNs: 1942-2466

Abstract

Abstract: Weather forecasts made with imperfect models contain state‐dependent errors. Data assimilation (DA) partially corrects these errors with new information from observations. As such, the corrections, or "analysis increments, " produced by the DA process embed information about model errors. An attempt is made here to extract that information to improve numerical weather prediction. Neural networks (NNs) are trained to predict corrections to the systematic error in the National Oceanic and Atmospheric Administration's FV3‐GFS model based on a large set of analysis increments. A simple NN focusing on an atmospheric column significantly improves the estimated model error correction relative to a linear baseline. Leveraging large‐scale horizontal flow conditions using a convolutional NN, when compared to the simple column‐oriented NN, does not improve skill in correcting model error. The sensitivity of model error correction to forecast inputs is highly localized by vertical level and by meteorological variable, and the error characteristics vary across vertical levels. Once trained, the NNs are used to apply an online correction to the forecast during model integration. Improvements are evaluated both within a cycled DA system and across a collection of 10‐day forecasts. It is found that applying state‐dependent NN‐predicted corrections to the model forecast improves the overall quality of DA and improves the 10‐day forecast skill at all lead times. Plain Language Summary: Computer models used for operational weather prediction are not perfect—they are naturally only simplifications of the true atmosphere. Such imperfections result in reduced forecast quality. Weather forecast systems routinely correct the forecasts by pulling them closer to observations, thus providing some information about the errors present in the forecast model. Here, a neural network (NN) is trained to correct National Oceanic and Atmospheric Administration's operational weather forecast model, FV3‐GFS, by "learning" the relation between the forecasts and the estimated model errors. The learned NN correction is then fed back into the weather model to improve the quality of the best guess state of the atmosphere and the subsequent 10‐day forecasts. By analyzing how the NN output depends on its input forecast, we gain some insight about the model errors, which may be helpful for future atmospheric model development and improvements to future error‐correcting NNs. Key Points: A neural network (NN) trained to infer analysis increments from model forecasts learns to correct systematic errors in the FV3‐GFS model Sensitivity analysis of the NN reveals physically consistent error characteristics that may be used to improve the NN architecture Applying online corrections from NN improves the accuracy of sequential data assimilation and extended free forecasts