Ensemble Kalman Filter Optimization for Convection-Permitting Forecasts in Complex Terrain
Abstract
Weather forecasting in complex terrain remains challenging due to mountainous topography, which generates high spatial variability, local atmospheric circulations, and rapidly evolving convective systems. This paper focuses on the optimization of the Ensemble Kalman Filter (EnKF) to enhance convection-permitting numerical weather prediction in complex terrain. A high-resolution ensemble forecasting framework is used to analyze the effects of key EnKF parameters, including ensemble size, covariance localization, and inflation. Controlled experiments are used to compare baseline and optimized configurations with respect to meteorological parameters, including temperature, wind, humidity, and precipitation. The analysis shows that ensemble size affects the stability of forecast-error covariance estimation, whereas covariance localization is required to control unrealistic spatial correlations caused by finite ensemble sizes. Ensemble inflation helps ensure sufficient ensemble dispersion to avoid overconfidence in the background forecast. The findings show that effective EnKF optimization must be based on a trade-off among uncertainty representation, observation impact, and computational efficiency, rather than maximizing a single parameter. The integrated system offers a more consistent approach to producing atmospheric initial conditions for convection-permitting forecasts. The paper identifies the potential of systematic EnKF optimization to enhance short-range forecasting in mountainous areas and provide more useful data for precipitation and weather-hazard applications.
Keywords
Download Options
Introduction
Precise weather prediction in complex terrain is a major scientific challenge because mountainous areas create a high degree of spatial and temporal variability in atmospheric events [1]. Orographic lifting, valley circulations, boundary-layer interactions, and rapidly developing convection create rainfall and wind patterns that cannot be adequately modeled in numerical weather prediction (NWP) models. Convection-permitting forecasts, with numerical model resolutions typically of a few kilometres, give a better representation of convective processes by explicitly resolving most of the deep-convective structure, instead of relying wholly on parameterization [2]. Nevertheless, errors in forecasts are also significant due to uncertainties in the initial atmospheric conditions, model physics, and observations. Data assimilation has provided a significant means of minimizing such uncertainties by combining short-range model predictions with observations to produce better initial conditions. The Ensemble Kalman Filter (EnKF) is especially beneficial because it employs an ensemble of model states to represent forecast uncertainty and estimate the evolving error covariance [3]. This enables observations to influence atmospheric variables based on dynamically changing relationships within the ensemble. However, EnKF performance can be compromised in complex terrain when ensemble spread fails, observation errors are poorly defined, localization is unsophisticated, or the ensemble fails to capture small-scale convective processes.
This study explores Ensemble Kalman Filter optimization of convection-permitting forecasts in complex terrain. The study aims to enhance the assimilation setup to generate more reliable initial atmospheric configurations and thereby increase short-range forecast performance. Particular emphasis is placed on aspects such as ensemble size, covariance localization, inflation settings, and observation assimilation, which may have a significant impact on EnKF performance. The quality of the forecast is then evaluated based on relevant meteorological parameters, such as temperature, precipitation, and wind errors. The research is important because better convection-permitting forecasts can enable earlier and more accurate prediction of high-impact weather in mountainous areas. The study aims to contribute to more effective strategies for data assimilation and enhanced numerical weather prediction in complex terrain and convection-related conditions through a systematic study of EnKF optimization.
Conclusion
This study was concerned with optimization of an Ensemble Kalman filter (EnKF) to enhance forecasts within complex terrain using convection-permitting models. The study touched upon the difficulties related to prediction of atmospheric conditions in mountainous areas where steep elevation gradients and orographic action, local circulations, and quickly formed convection may result in significant forecast uncertainty. An ensemble-based data assimilation (along with a high-resolution numerical weather prediction framework) was also regarded to study the effects of the various EnKF configurations on the forecast. Three major assimilation factors, ensemble size, covariance localization, and ensemble inflation were analyzed. It is shown that there is an interdependent relationship among these parameters that affect the performance of the data-assimilation system. Ensuring the representation of covariance of the forecast-error is more stable with an increasing ensemble size, but the computational demands also rise significantly. In complex terrain, covariance localization is especially crucial, as the use of the wrong spatial correlations may bias correction to the unrealistic, and over-localization may constrain useful observational data. Ensemble inflation can also be significant in the sense that it ensures there is enough ensemble dispersion so that the system of assimilation becomes not too self-confident in its background state.
The combined analysis suggests that the optimum EnKF setup ought hence to aim at obtaining a minimum trade-off between ensemble variance, spatial localization, and uncertainty modelling as opposed to maximizing a specific parameter. These optimizations can help to produce more stable analyses of the atmosphere and enhance initial conditions forecasts using convection-permitting models. The possible advantages are of interest especially to precipitation prediction where errors in temperature, humidity, and wind can be very strong contributors to convective activity and distribution of rain. In general, the research shows that systematic EnKF optimization is applicable to high-resolution forecasting in complex terrain. The results give a methodological basis on how to better short-range numerical weather prediction and assist other uses that involve heavy rainfall and other weather hazards. Further studies are needed to optimize the configuration in various seasons, topography, and extreme-weather conditions, to add extra observations and in-depth research on adaptive localization and inflation methods. Computational efficiency should also be subject to further research so that enhancements in quality of the forecasts are possible within the realistic operational limits.
References
- Byun, J., Hyeon-Joon, K., Kang, N., Yoon, J., Hwang, S., & Jun, C. (2024). Optimizing temporal weighting functions to improve rainfall prediction accuracy in merged numerical weather prediction models for the Korean Peninsula. Remote Sensing, *16*(16), 2904.
- Gu, J., Zhao, C., Xu, M., Feng, J., Li, G., Zhao, Y., Hao, X., Chen, J., & An, H. (2024). Global convection-permitting model improves subseasonal forecast of plum rain around Japan. Environmental Research Letters, *19*(10), 104021.
- Hamed, N., Lagerwall, G. L., Jayantha, O., & Liu, J. (2026). Machine learning in climate downscaling: A critical review of methodologies, physical consistency, and operational applications. Water, *18*(2), 271.
- Harshwardhan, J., Singh, P., Bodo, A., & Juerg, S. (2025). Machine learning-based identification of key predictors for lightning events in the Third Pole Region. ISPRS International Journal of Geo-Information, *14*(8), 319.
- Hu, L., Ma, Z., & Huijin, F. (2026). A review and perspectives on wind speed forecasting for high-speed railways in China. Atmosphere, *17*(5), 464.
- Ikuta, Y., Fujita, T., Ota, Y., & Honda, Y. (2021). Variational data assimilation system for operational regional models at Japan Meteorological Agency. Journal of the Meteorological Society of Japan, *99*(6), 1563-1592.
- Ishida, J., Aranami, K., Kawano, K., Matsubayashi, K., Kitamura, Y., & Muroi, C. (2022). ASUCA: The JMA operational non-hydrostatic model. Journal of the Meteorological Society of Japan, *100*(5), 825-846.
- Jiping, G., Chen, J., Li, X., Mengting, L., & Zhang, M. (2025). Impacts of the assimilation of radar radial velocity data using the Ensemble Kalman Filter (EnKF) on the analysis and forecast of Typhoon Lekima (2019). Remote Sensing, *17*(13), 2258.
- Latifa, A. D., Moustadraf, J., Rachdane, M., Mohamed, E. S., Benjmel, K., Amraoui, F., Mohamed, A. E., Abdelaziz, A. S., & Hadri, A. (2023). Spatiotemporal assessment and correction of gridded precipitation products in North Western Morocco. Atmosphere, *14*(8), 1239.
- Li, J., Chen, C., Chen, X., Hongrang, H., Yongqiang, J., & Yanzhen, K. (2025). Comparative analysis of perturbation characteristics between LBGM and ETKF initial perturbation methods in convection-permitting ensemble forecasts. Atmosphere, *16*(6), 744.
- Li, X., Gu, J., Li, Y., Zheng, Y., Zhuang, Y., Sun, S., & Cui, Z. (2026). Integration of deep neural network-based boundary layer subgrid frictional effects into the momentum equation constraint of WRFDA EnVar to improve tropical cyclone forecasts. Journal of Advances in Modeling Earth Systems, *18*(7), 31.
- Liao, Y., Wang, Z., Yu, H., Gao, W., Zeng, Z., Li, X., & Lai, C. (2025). Accelerating urban flood inundation simulation under spatio-temporally varying rainstorms using ConvLSTM deep learning model. Water Resources Research, *61*(8), 25.
- Liu, W., Xiaolu, L., Li, C., Botao, H., & Haonan, X. (2025). Simulation and assimilation of CO2 concentrations based on the WRF-Chem model. Processes, *13*(12), 4010.
- Molina-Almaraz, M., Solís-Sánchez, L. O., Bañuelos-García, L. E., Castañeda-Miranda, C. L., Guerrero-Osuna, H., & García-Sánchez, E. (2025). Efficient neural modeling of wind power density for national-scale energy planning: Toward sustainable AI applications in Industry 5.0. Applied Sciences, *15*(24), 13000.
- Nakashita, S., & Enomoto, T. (2025). Flow-dependent large-scale blending for limited-area ensemble data assimilation. Tellus, *77*(1), 1-19.
- Naresh, C., & Mathew, A. (2026). Integrating climate modelling, downscaling, and bias correction for drought studies: Current approaches and future directions. Water Resources Management, *40*(10), 486.
- Onyejuruwa, A., Hu, Z., Anosike, C., Islam, A. R. M. T., & Madhushanka, D. (2025). Assessment of WRF model simulations of extreme rainfall events in West Africa: A comprehensive review. Environmental Monitoring and Assessment, *197*(4), 486.
- Oruç, S., Tuğrul, T., & Hınıs, M. A. (2026). Enhancing drought forecasting with CNN-TQWT and metaheuristic hybrids: Evidence from Norway. Scientific Reports, *16*(1), 25536.
- Piero, S., Antonio, R., D'Amico Cristiano, & Rossella, F. (2026). Advancing cyclone tracking with HIMPACT: High-resolution multilevel Python-based algorithm for cyclones' centroid tracking. Atmosphere, *17*(9), 862.
- Pryor, S. C., Barthelmie, R. J., Cadence, J., Dellwik, E., Hasager, C. B., Kral, S. T., Reuder, J., Rodgers, M., & Veraart, M. (2022). Atmospheric drivers of wind turbine blade leading edge erosion: Review and recommendations for future research. Energies, *15*(22), 8553.
- Ruisheng, H., Jiaqi, D., Yang, J., Difu, S., Zengliang, Z., Zhao, J., Hongze, L., & Junqiang, S. (2026). A coordinate-based framework for sea surface wind speed reconstruction from sparse multi-source observations. Remote Sensing, *18*(16), 2709.
- Shi, J., Liu, Y., Yang, L., Liu, Y., Roux, G., Shi, L., & Fan, X. (2022). Wind speed forecasts of a mesoscale ensemble for large-scale wind farms in Northern China: Downscaling effect of global model forecasts. Energies, *15*(3), 896.
- Siou-Ying Jiang, Jing-Shan, H., & Ben Jong-Dao Jou. (2021). Blend global analyses into the cyclic convective-scale data assimilation system to improve short-term QPF. TAO: Terrestrial, Atmospheric and Oceanic Sciences, *32*(5), 6-647.
- Sofia, G., Yang, Q., Shen, X., Mahjabeen, F. M., Patlakas, P., Chaniotis, I., Kallos, A., Alomary, M. A., Alzahrani, S. S., Christidis, Z., & Anagnostou, E. (2024). A nationwide flood forecasting system for Saudi Arabia: Insights from the Jeddah 2022 event. Water, *16*(14), 1939.
- Talapatra, A., & Rana, N. K. (2026). Recent advances in flood monitoring and prediction methods: A systematic review. Environmental Science and Pollution Research, *33*(2), 325-343.
- Wang, L., Chen, Z., Zhang, W., Lu, Z., Cheng, Y., Qu, X., Gul, C., & Yang, Y. (2023). The causes and forecasting of icing events on power transmission lines in Southern China: A review and perspective. Atmosphere, *14*(12), 1815.
- Wang, Y., Ningxin, Y., Siyu, Z., & Yang, H. (2026). Characterizing short-duration summer rainstorms in Nanjing, China, using multi-source remote sensing and explainable AI. Remote Sensing, *18*(13), 2212.
- Whittaker, T., Taylor, S., Cardoso-Bihlo, E., Di Luca, A., & Bihlo, A. (2026). Learning vertical coordinates via automatic differentiation of a dynamical core. Journal of Advances in Modeling Earth Systems, *18*(8), 24.
- Xia, X., Feng, J., Wang, K., Sun, J., Gao, Y., Jin, Y., Ma, Y., Gao, Y., & Wan, Q. (2023). Performance of a hybrid gain ensemble data assimilation scheme in tropical cyclone forecasting with the GRAPES model. Atmosphere, *14*(3), 565.
- Xu, Y., Liu, C., Yu, Q., Zhao, C., Quan, L., & Hu, C. (2023). Study on a hybrid hydrological forecasting model SCE-GUH by coupling SCE-UA optimization algorithm and general unit hydrograph. Water, *15*(15), 2783.