Convection is a complex spatiotemporal process, which has made it a particularly attractive application for deep learning, which excels at both spatial and temporal reasoning. We have developed deep learning models for predicting the occurrence of hazards caused by convective storms, so that this information may be used by forecasters, emergency services and infrastructure managers to respond to the threats caused by these hazards.

Our network is based on a recurrent-convolutional architecture that can process input data at multiple resolutions. It issues probabilistic predictions of hazard occurrence, currently up to 1 hour to the future. As inputs, we use data from weather radars, geostationary satellites, ground-based lightning detections, numerical weather predictions and digital elevation models. We have studied the importance of each data source to the quality of the predictions, finding that radar-based inputs contribute most to the prediction quality; however, some hazards can be well predicted also without radar, indicating that it is plausible to create warning systems for these hazards in areas where radar networks are not available.

In this presentation, we will describe the model architecture and case studies, as well as our experiences so far in bringing the model to real-time use by forecasters and automated warning systems at MeteoSwiss. We will also discuss future directions of this research.

How to cite: Leinonen, J., Hamann, U., Sideris, I., and Germann, U.: Predictive hazards from convective systems with deep learning, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-9488, https://doi.org/10.5194/egusphere-egu23-9488, 2023.

Deep convection plays a fundamental role in the climate system by transporting from the lower layers of the atmosphere to the free troposphere, air, water and momentum. Although its study has been the subject of intense and rich scientific activities for decades, our ignorance of the vertical distribution of convective movements in the heart of convective cells is today an important scientific and operational obstacle. Only space-borne observations can meet the needs in documentation necessary to progress on the science of the water and energy cycle and simultaneously improve numerical forecasting systems. Pending the emergence (hypothetical) of microwave missions in geostationary orbit with high repeatability (~ 1 minute), an approach based on satellite constellations in convoys could provide a first response.

The “Convective Core Observations through MicrOwave Derivatives in the trOpics”, or C²OMODO for short, proposes to rely on 2 passive microwave radiometers with a multispectral sampling of the 183 and 325 GHz lines in a mini-train of 2 satellites. The time-spacing of 60 to 180sec between the 2 swaths encompasses information on the updraft motions of hydrometeors, and is thus used to characterize the intensity and the size of individual updrafts in deep convective systems.

We will present this original observational strategy, associated to the NASA / AOS general framework, as well as its expected added-value for the characterization of deep convection.

How to cite: Brogniez, H., Roca, R., Chaboureau, J.-P., Auguste, F., Gharbi, I., Lefebvre, T., Fiolleau, T., and Bouniol, D.: Observing the dynamics of deep convection using a tandem of spaceborne microwave radiometers, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-2324, https://doi.org/10.5194/egusphere-egu23-2324, 2023.

The NASA TROPICS Earth Venture (EVI-3) CubeSat constellation mission will provide nearly all-weather observations of 3-D temperature and humidity, as well as cloud ice and precipitation horizontal structure, at high temporal resolution to conduct high-value science investigations of tropical cyclones. TROPICS will provide rapid-refresh microwave measurements (median refresh rate of approximately 60 minutes for the baseline mission) over the tropics that can be used to observe the thermodynamics of the troposphere and precipitation structure for storm systems at the mesoscale and synoptic scale over the entire storm lifecycle. The TROPICS constellation mission comprises four 3U CubeSats (5.4 kg each) in two low-Earth orbital planes. Each CubeSat comprises a Blue Canyon Technologies bus and a high-performance radiometer payload to provide temperature profiles using seven channels near the 118.75 GHz oxygen absorption line, water vapor profiles using three channels near the 183 GHz water vapor absorption line, imagery in a single channel near 90 GHz for precipitation measurements (when combined with higher resolution water vapor channels), and a single channel at 205 GHz that is more sensitive to precipitation-sized ice particles. TROPICS spatial resolution, measurement sensitivity, and calibration accuracy and stability are all comparable with current state-of-the-art observing platforms. Two launches for the TROPICS constellation mission are planned for the Summer of 2023. Data will be downlinked to the ground via the KSAT-Lite ground network. NASA's Earth System Science Pathfinder (ESSP) Program Office approved the separate TROPICS Pathfinder mission, which launched on June 30, 2021, in advance of the TROPICS constellation mission as a technology demonstration and risk reduction effort. The TROPICS Pathfinder mission continues to yield excellent data over 18+ months of operation and has provided an opportunity to checkout and optimize all mission elements prior to the primary constellation mission. This presentation will describe the on-orbit results for the successful TROPICS Pathfinder precursor mission and will describe the recent development progress for the TROPICS constellation mission and discuss recent activities to improve the data latency and generation of near-real-time products for forecasting applications.

How to cite: Blackwell, W.: 18+ Months of Tropical Cyclone and Convective Storm Observations with the NASA TROPICS Pathfinder Satellite, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-10494, https://doi.org/10.5194/egusphere-egu23-10494, 2023.

Moisture convergence, latent heat, upper-level divergence all contribute to the genesis and growth of convective updrafts. In order to characterize the morphology of this evolution, and identify its constituent modes, we analyzed a large data set of synthetic updrafts simulated using a convection-resolving differential-equation solver run at high spatial and temporal resolutions (respectively 100 meters and 10 seconds). The analysis started by fitting each simulated updraft with a 6-parameter analytic representation, so that the joint statistics of the 6 parameters and of their evolution in time can be quantified. The first result is that an effective 6-parameter representation does exist and approximates the vertical profiles with a residual relative error whose r.m.s. value is smaller than 10% for 59% of all cases, and smaller than 20% for 89% of all cases. The r.m.s value of the absolute error is smaller than 0.4 m/s for 97% of all cases. Having established the suitability of this approximation, the variability of the 6 parameters for the 2-minute average Wa of a profile W was quantified, as was the variability of the evolution of W – Wa over a two-minute interval. The analysis reveals that 4 scalars suffice to capture the bulk of the variability of the evolution of convective updrafts. The modes (spanning the range of values of these 4 scalars) turn out to be related to the maximum amplitudes of w and to the heights at which they are achieved. This description paves the way toward the characterization of the environmental determinants of updraft evolution and, in turn, the determination of the effects of updraft characteristics on upper-level air density, divergence and the resulting anvil clouds.

How to cite: Haddad, Z., Prasanth, S., van den Heever, S., Marinescu, P., Freeman, S., and Posselt, D.: Parametrizing the evolution of convective updraft vertical velocities, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-10858, https://doi.org/10.5194/egusphere-egu23-10858, 2023.

Data assimilation techniques can improve the simulation of regional precipitation and temperature over complex regions. Nowadays, regional climate simulations using convection-permitting scales are becoming available, but the number of those simulations including an additional data assimilation scheme is rather small because of their high computational costs. Hence, it is important to evaluate the effect of data assimilation schemes for such convection-permitting simulations, and to determine if data assimilation produces any improvement in the simulation of temperature or precipitation fields.

To investigate this, we employ the Weather Research and Forecasting model (WRF; version 3.8.1) to dynamically downscale the state-of-the-art ERA5 reanalysis over Western Europe. A 3 km spatial resolution grid is employed, together with 51 vertical levels. The temporal resolution of the WRF outputs is one hour. Two model configurations are tested in two experiments spanning the period 2010-2020 after a one-year spin-up. In the first experiment (NoDA), after the initialization of the model, the boundary conditions drive the model. The second experiment (DA) is configured the same way as NoDA, but the additional 3DVAR data assimilation step (WRFDA) is run every six hours (00, 06, 12 and 18 UTC – analysis times). Observations obtained from the PREPBUFR dataset (NCEP ADP Global Upper Air and Surface Weather Observations) are employed, and only those included inside a 120 min window around analysis times were assimilated. For DA, monthly varying background error covariance matrices were created. In both cases, the model uses the Noah-MP land surface model, and high-resolution daily-varying SST fields from the NOAA OI SST v2 data set instead of the SST field from ERA5.

The results of this study show that both experiments produce similar monthly precipitation patterns to those from observational data sets such as IMERG and CHIRPS, or the reanalysis ERA5. However, in general, and particularly during summer months, DA produces larger amounts of precipitation than NoDA. These amounts are in line with those from CHIRPS. In terms of temperature, DA show colder temperatures than NoDA in most of the months, which again are similar to those from observational data sets such as CRU or EOBS. The monthly temperature patterns of both experiments are similar to those from both observational data sets. These results highlight the fact that NoDA already is able to generate reliable precipitation and temperature fields compared to diverse gridded observational data sets, but the 3DVAR data assimilation can additionally improve the performance of the regional model when convection-permitting scales are employed.

How to cite: González-Rojí, S. J. and Raible, C. C.: The effect of 3DVAR data assimilation and convection-permitting scales on the simulation of precipitation and temperature over Western Europe, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-6378, https://doi.org/10.5194/egusphere-egu23-6378, 2023.

The development of atmospheric deep moist convection has been a challenging topic for numerical weather prediction, due to its chaotic nature of the development with multi-scale physical interactions. We recently found that greater than 20-km scale (as commonly known as meso-α (2000-200 km) and meso-β (200-20 km) scales) initial features helped to constrain the general location of convective activity with a few hours of lead time, but meso-γ (20-2 km) or even smaller scale features with less than 30-minute lead time were identified to be essential for capturing the spatiotemporal features of individual convection. To examine the potentials of ensemble-based data assimilation in capturing the individual convective development, as well as the subsequent development of severe weather events, we have conducted large ensemble convection-permitting data assimilation experiments with all-sky infrared satellite radiances from the latest-generation geostationary satellites. We found that the greater number of ensembles more effectively suppressed the spurious correlation for convective-scale data assimilation. However, the exact signals of convective development were not clearly captured in covariances even with thousands of ensemble members. These results suggest the potential limitation of the traditional “Eulerian” (i.e. physical grid-based) ensemble approach in convective-scale data assimilation.

How to cite: Minamide, M. and Posselt, D.: Predictability of moist convection through ensemble-based convective-scale data assimilation, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-12321, https://doi.org/10.5194/egusphere-egu23-12321, 2023.

Convection-permitting data assimilation requires observations with high spatial density and high temporal frequency to provide information on appropriate scales for high resolution forecasting. Those observation types (e.g., geostationary satellite data) were found to exhibit strong spatial error correlations. Explicitly introducing correlated error statistics in the assimilation may increase the computational complexity and parallel communication costs of the matrix-vector multiplications with the observation precision matrices (the inverse observation error covariance matrices). Therefore, without suitable approaches we cannot take full advantage of the new observation uncertainty estimates. In this work, we present a new numerical approximation method, called the local SVD-FMM, which is developed based on a particular type of the fast multipole method (FMM) using a singular value decomposition (SVD), and a domain localization approach. The basic idea of the local SVD-FMM is to divide the observation domain into boxes of (approximately) equal size and then separates the calculations of the matrix-vector products according to the domain partition. These calculations can be done in parallel with very low communication overheads. Moreover, the local SVD-FMM is easy to implement and applicable to a wide variety of the precision matrices. We applied the local SVD-FMM in a simple variational data assimilation system and found that the computational cost of the variational minimisation was dramatically reduced while preserving the accuracy of the analysis. This new method has the potential to be used as an efficient technique for practical data assimilation applications where a large volume of observations with mutual error correlations needs to be assimilated in a short period of time.

How to cite: Hu, G. and Dance, S. L.: A Novel Numerical Approximation Method for Computations with Spatially Correlated Observation Error Statistics in Data Assimilation, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-14476, https://doi.org/10.5194/egusphere-egu23-14476, 2023.

This study investigates the impact of high frequency, such as 3-hourly and 1-hourly satellite microwave radiances, in global atmospheric data assimilation. To understand the impact of such a high-frequency satellite radiances data assimilation, we designed an observing system simulation experiment (OSSE) using the global NICAM-LETKF system at 56 km horizontal resolution. A free run was conducted with the NICAM model and treated as the reference (Nature) for the OSSE experiments. With the NICAM-LETKF system, we conducted five experiments, without data assimilation (NoDA), with only conventional data assimilation but not satellite radiances (NoSat), 6-hourly (6H), 3-hourly (3H), and 1-hourly (1H) satellite clear-sky radiances assimilation. The results showed that satellite microwave radiances assimilation improved the forecast of air temperature and wind over the global ocean compared to NoSat experiments. With the increase in the assimilation frequency of the satellite radiances, the air temperature and winds showed improvement in their representation over the ocean but degraded over land. Over the ocean, microwave radiances assimilation improved the typhoon eyewall wind intensities and its structure for 1H satellite radiances assimilation compared to 6H. These improvements in the wind intensities are prominent during the landfall stage of the typhoon. Forecasting landfall storms' strong winds are essential for disaster prevention and mitigation.

How to cite: Konduru, R. T., Liang, J., and Miyoshi, T.: High-frequency microwave satellite radiances data assimilation using NICAM-LETKF in the OSSE framework, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-10561, https://doi.org/10.5194/egusphere-egu23-10561, 2023.

The European Space Agency (ESA)’s Aeolus mission, launched in August 2018, provides the first global horizontal line-of-sigh (HLOS) wind profile measurements. Many Numerical Weather Prediction (NWP) centres, including ECMWF, DWD, Météo-France, Met-Office and ECCC, have shown that assimilating Aeolus winds improves overall forecast skill, especially in the tropics and data-sparse regions. To better characterize the locations and drivers of improved skill from Aeolus, we use a series of Observing System Experiments (OSEs) with the ECCC Global Deterministic Prediction System (GDPS) covering the period July to September 2019 and December 2019 to March 2020. Three experiments are used: CNTRL, CNTRL+Aeolus, and CNTRL-winds. All the observations assimilated in the GDPS are included in the CNTRL experiment. The Aeolus winds are added in the CNTRL+Aeolus experiment and the operational wind observations are withheld in the CNTRL-wind experiment. The impact of the operational winds and Aeolus are quantified by comparing the forecast error of the CNTRL-winds and CNTRL experiments with the CNTRL and CNTRL+Aeolus experiments. 

As expected, the operational winds improve the tropospheric forecast over the tropics the most, with a normalized forecast error of 8% for the wind field. By adding the Aeolus winds, which account for less than 1% of the observations, the tropospheric forecast further improves by 0.7-0.9% over the tropics and the Arctic, and by 0.5-0.6% over the data-sparse Southern Hemisphere extra-tropics. The added value of Aeolus winds is further highlighted when its impact on forecasts as a function of length scale is investigated, using a spherical harmonic decomposition. The impact is measured as the difference of the 250-hPa kinetic energy forecast error spectra between experiments. The impact of operational winds and Aeolus is dominated by the transient component whose impact is nearly four times greater than the impact on the mean component. The operational winds largely improve the forecast of global scale to intermediate scale in the short-range forecasts. The impact then decreases as forecast range increases. On the other hand, the impact of Aeolus is mostly seen in the intermediate to large scale range with a peak around spherical harmonics of degree 9 (scales about 4000 km), and is the smallest on day 1 and increases until days 4 to 5. This analysis suggests that Aeolus winds provide estimates of the wind state that are valuable and complementary to that provided from current operational winds.

How to cite: Chou, C.-C. (., Kushner, P. J., and Laroche, S.: Assessing the added value of Aeolus winds in the ECCC forecast system, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-10418, https://doi.org/10.5194/egusphere-egu23-10418, 2023.