Kushal Keshavamurthy
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Kushal Keshavamurthy has a total of 16 co-authors. Here are the top co-authors by number of shared publications:
- Chanh Kieu
Department of Earth and Atmospheric Sciences, Indiana University Bloomington, Bloomington, Indiana
3 shared publications Allison E Reed Harris
Tomorrow.io
1 shared publication- Ashley Elizabeth Payne
Department of Climate and Space Sciences and Engineering, University of Michigan, Ann Arbor, MI 48109, USA
1 shared publication Faith Groff
Department of Earth and Atmospheric Sciences Indiana University Bloomington Indiana USA
1 shared publication- Luke Conibear
Institute for Climate and Atmospheric Science, School of Earth and Environment, University of Leeds
1 shared publication Luke Thomas Peffers
Tomorrow.io
1 shared publication
Latest Publications
- Open Access
- 27 October 2023
- Published by Wiley
Post-processing using deep learning to create operational, high-resolution, and probabilistic weather forecasts
We present a machine learning model that post-processes high-resolution, deterministic forecasts to produce short to medium-range probabilistic forecasts for seven core weather variables. We developed and operationally implemented a multi-task neural network with a custom loss function, the Continuous Ranked Probability Score (CRPS), using input data from the High Resolution Rapid Refresh (HRRR) model. This combined high-resolution Numerical Weather Prediction (NWP) modeling with machine learning to generate more accurate deterministic and probabilistic forecasts. The results show deterministic forecast improvements in Root Mean Squared Error (RMSE) from 2.5% to 24.7% and probabilistic improvements in CRPS from 31% to 45% over the HRRR model. The multi-task neural network is cheap to train and fast to run operationally on hourly forecasts.
- 27 October 2023
- Published by Wiley
Open AccessPost-processing using deep learning to create operational, high-resolution, and probabilistic weather forecasts
We present a machine learning model that post-processes high-resolution, deterministic forecasts to produce short to medium-range probabilistic forecasts for seven core weather variables. We developed and operationally implemented a multi-task neural network with a custom loss function, the Continuous Ranked Probability Score (CRPS), using input data from the High Resolution Rapid Refresh (HRRR) model. This combined high-resolution Numerical Weather Prediction (NWP) modeling with machine learning to generate more accurate deterministic and probabilistic forecasts. The results show deterministic forecast improvements in Root Mean Squared Error (RMSE) from 2.5% to 24.7% and probabilistic improvements in CRPS from 31% to 45% over the HRRR model. The multi-task neural network is cheap to train and fast to run operationally on hourly forecasts.
- Open Access
- 1 February 2021
- Published by Wiley in Quarterly Journal of the Royal Meteorological Society
- Vol. 147 (736) , 1606-1625
Dependence of tropical cyclone intrinsic intensity variability on the large‐scale environment
This study examines the dependence of tropical cyclone (TC) intrinsic intensity variability on the large‐scale environment. Using idealized simulations with the Hurricane Research and Forecasting (HWRF) model, the growth of TC intensity errors and the subsequent saturation of intensity errors at the mature stage are quantified for a range of environmental conditions. Among several factors, it is found that sea‐surface temperature (SST) and tropospheric static stability are the two main ones affecting the growth rate and saturation of TC intensity errors, with a larger error saturation at maximum intensity equilibrium and a faster error growth rate during rapid intensification for a warmer SST or less stable troposphere. Results from this study highlight the fundamental characteristics of TC intensity predictability, which appears to vary with environmental conditions, instead of being a universal measure for TCs in all ocean basins. These results also suggest a limit on TC intensity forecast accuracy that is governed by the large‐scale environment, which needs to be taken into consideration for future TC model development.
- 1 February 2021
- Published by Wiley in Quarterly Journal of the Royal Meteorological Society
- Vol. 147 (736) , 1606-1625
Open AccessDependence of tropical cyclone intrinsic intensity variability on the large‐scale environment
This study examines the dependence of tropical cyclone (TC) intrinsic intensity variability on the large‐scale environment. Using idealized simulations with the Hurricane Research and Forecasting (HWRF) model, the growth of TC intensity errors and the subsequent saturation of intensity errors at the mature stage are quantified for a range of environmental conditions. Among several factors, it is found that sea‐surface temperature (SST) and tropospheric static stability are the two main ones affecting the growth rate and saturation of TC intensity errors, with a larger error saturation at maximum intensity equilibrium and a faster error growth rate during rapid intensification for a warmer SST or less stable troposphere. Results from this study highlight the fundamental characteristics of TC intensity predictability, which appears to vary with environmental conditions, instead of being a universal measure for TCs in all ocean basins. These results also suggest a limit on TC intensity forecast accuracy that is governed by the large‐scale environment, which needs to be taken into consideration for future TC model development.
- Open Access
- 1 July 2018
- Published by Wiley in Quarterly Journal of the Royal Meteorological Society
- Vol. 144 (715) , 1803-1819
On the growth of intensity forecast errors in the operational hurricane weather research and forecasting (HWRF) model
This study examines the growth of tropical cyclone (TC) intensity forecast errors and related intensity predictability for the NOAA operational Hurricane Weather Research and Forecasting (HWRF) model. Using operational intensity forecasts during the 2012 to 2016 seasons, two conditions for a limited range of TC intensity predictability are demonstrated, which include (a) the existence of an intensity error saturation limit, and (b) the dependence of the intensity error growth rate on storm intensity during TC development. By stratifying intensity errors based on different initial intensity bins, it is shown that TC intensity error growth rate is relatively small (∼0.3 kt h−1) at the early stage of TC development, but it quickly increases to ∼1 kt h−1 during TC intensification. Of further importance is that the intensity error saturation varies in the range of 14–18 kt in different ocean basins, thus suggesting the potential dependence of the intensity predictability on large‐scale environment. Additional idealized experiments with the HWRF model confirm the saturation of intensity errors, even under a perfect model scenario. The existence of the intensity error saturation together with the finding of a faster error growth rate for higher intensity suggests that the TC dynamics possesses an inherent limited predictability, which prevents us from reducing the intensity errors in TC dynamical models below a certain threshold.
- 1 July 2018
- Published by Wiley in Quarterly Journal of the Royal Meteorological Society
- Vol. 144 (715) , 1803-1819
Open AccessOn the growth of intensity forecast errors in the operational hurricane weather research and forecasting (HWRF) model
This study examines the growth of tropical cyclone (TC) intensity forecast errors and related intensity predictability for the NOAA operational Hurricane Weather Research and Forecasting (HWRF) model. Using operational intensity forecasts during the 2012 to 2016 seasons, two conditions for a limited range of TC intensity predictability are demonstrated, which include (a) the existence of an intensity error saturation limit, and (b) the dependence of the intensity error growth rate on storm intensity during TC development. By stratifying intensity errors based on different initial intensity bins, it is shown that TC intensity error growth rate is relatively small (∼0.3 kt h−1) at the early stage of TC development, but it quickly increases to ∼1 kt h−1 during TC intensification. Of further importance is that the intensity error saturation varies in the range of 14–18 kt in different ocean basins, thus suggesting the potential dependence of the intensity predictability on large‐scale environment. Additional idealized experiments with the HWRF model confirm the saturation of intensity errors, even under a perfect model scenario. The existence of the intensity error saturation together with the finding of a faster error growth rate for higher intensity suggests that the TC dynamics possesses an inherent limited predictability, which prevents us from reducing the intensity errors in TC dynamical models below a certain threshold.
- 8 May 2017
- Published by American Geophysical Union (AGU) in Geophysical Research Letters
- Vol. 44 (9) , 4313-4323
Large‐scale control of the lower stratosphere on variability of tropical cyclone intensity
Recent studies of tropical cyclones (TC) have suggested intricate impacts of the lower stratosphere layer (LSL) on TC development at the high‐intensity limit. This study examines the potential realization of the impacts of the LSL interannual variability on TC intensity. By minimizing the effects of sea surface temperature and outflow temperature based on the potential intensity framework, partial correlation analyses show a negative correlation between the power dissipation index (PDI) for intense TCs and the tropopause height in the northwestern Pacific basin but a weaker relationship in the north Atlantic basin. Similar analyses for the LSL stratification also reveal signals of negative correlations between the LSL stratification and PDI in both basins, corroborating recent modeling studies about the impacts of the LSL on TC development. Due to the complexity of the relationships and data limitations, however, all correlations are statistically insignificant, thus rendering the impacts of LSL on TC intensity inconclusive at present.
- 8 May 2017
- Published by American Geophysical Union (AGU) in Geophysical Research Letters
- Vol. 44 (9) , 4313-4323
Large‐scale control of the lower stratosphere on variability of tropical cyclone intensity
Recent studies of tropical cyclones (TC) have suggested intricate impacts of the lower stratosphere layer (LSL) on TC development at the high‐intensity limit. This study examines the potential realization of the impacts of the LSL interannual variability on TC intensity. By minimizing the effects of sea surface temperature and outflow temperature based on the potential intensity framework, partial correlation analyses show a negative correlation between the power dissipation index (PDI) for intense TCs and the tropopause height in the northwestern Pacific basin but a weaker relationship in the north Atlantic basin. Similar analyses for the LSL stratification also reveal signals of negative correlations between the LSL stratification and PDI in both basins, corroborating recent modeling studies about the impacts of the LSL on TC development. Due to the complexity of the relationships and data limitations, however, all correlations are statistically insignificant, thus rendering the impacts of LSL on TC intensity inconclusive at present.