Improving Atmospheric River Forecasts With Machine Learning

Abstract

This study tests the utility of convolutional neural networks as a postprocessing framework for improving the National Center for Environmental Prediction's Global Forecast System's integrated vapor transport forecast field in the Eastern Pacific and western United States. Integrated vapor transport is the characteristic field of atmospheric rivers, which provide over 65% of yearly precipitation at some western U.S. locations. The method reduces full‐field root‐mean‐square error (RMSE) at forecast leads from 3 hr to seven days (9–17% reduction), while increasing correlation between observations and predictions (0.5–12% increase). This represents an approximately one‐ to two‐day lead time improvement in RMSE. Decomposing RMSE shows that random error and conditional biases are predominantly reduced. Systematic error is reduced up to five‐day forecast lead, but accounts for a smaller portion of RMSE. This work demonstrates convolutional neural networks potential to improve forecast skill out to seven days for precipitation events affecting the western United States.

Experimental Subseasonal‐to‐Seasonal (S2S) Forecasting of Atmospheric Rivers Over the Western United States

“…In addition, future work focusing on large ensemble operational hindcast generation would benefit the reliability and robustness of the statistical relationships suggested in this and related published work. Finally, exploiting postprocessing and machine learning techniques applied to S2S hindcast data (e.g., Chapman et al, ) has the potential to improve prediction skill of uncalibrated hindcast data from multiple operational centers.…”

On the Rivers in the Euro-Atlantic Sky

“…Thus, time-integrated (4 time-steps) daily ARs data has been generated from six-hourly reanalysis datasets using IVT (kgm -1 s -1 ), normalized IVT (kgm -1 s -1 K -1 ); IWV (mm) and normalized IWV (mmK -1 ) from the surface to 750 hPa, 500 hPa and 300 hPa. Temperature normalization is done to understand the change in the thermodynamic component of IVT and IWV using the Clausius-Clapeyron equation (3), which states that the water-vapour content of saturated air, q*, increases nearly exponentially with temperature T (Payne et al, 2020).…”

“…Data obtained from both satellites and statistical methods have limitations in forecasting the landfall and intensity of ARs well in advance. In recent times machine learning techniques(Chapman et al, 2019;Kashinath et al, 2020) have evolved as other alternatives. However, the average error in estimating the intensity of ARs through IVT is around 40-60 kgm -1 s -1 using different sources of data including data from reanalysis and amounts to 22% of mean observed ux(Chapman et al, 2019, Lavers et al, 2018.…”

A Deep Learning Trained Clear-Sky Mask Algorithm for VIIRS Radiometric Bias Assessment

“…In this section, nine days (11)(12)(13)(14)(15)(16)(17)(18)(19) March 2019) of S-NPP CSMs were generated by the FCDN-CSM model to evaluate the model performance by checking the stability and accuracy of the O-M biases. Figure 6 shows the O-M error bars with STDs for the five TEB M-bands, using ACSPO CSM and FCDN-CSM to identify clear-sky pixels.…”

“…With cutting-edge artificial intelligence (AI) evolving rapidly, deep learning (DL), one of the most popular AI methods, has made a remarkable difference in many science and engineering fields. Its application in remote sensing [13,14] and numerical weather prediction [15][16][17][18] is also being explored. Deep learning is constructed using artificial neural networks (ANNs), including more than one hidden layer, with a so-called "deep" neural network distinguished from a "shallow" neural network.…”