Conventionally, radar-based nowcasting methods assume that the short-term evolution of precipitation can be obtained by simply shifting the last observed precipitation along a stationary motion field. A major limitation of this assumption arises from neglecting lifecycles of precipitation cells, for example, the decay of mature cells and the initiation of new ones. In this perspective, P2 aims to improve the prediction skill of advection-based nowcasting by considering statistical properties and life cycles of precipitation.
In a first step, trends of intensity, size, and shape of precipitation cells observed during previous time steps will be extrapolated in time. In a second step, polarimetric radar signatures indicative for potential changes in precipitation generation will be exploited for refinements of the approach. In both steps, P2 will derive an ensemble of nowcasted precipitation fields based upon the precipitation fields provided by P1 while using the 3D multi-sensor composite (radar-satellite) provided by C1. The Quantitative Precipitation Nowcasting (QPN) ensembles will serve as input for seamless prediction (P3) and flood prediction (P4).

TCWV can be observed from satellites under clear-sky conditions with a high accuracy and has the potential to improve monitoring of pre-convective environments and convective initiation before the onset of clouds and precipitation. Thus it may provide complementary information for radar-based QPN, particularly for nowcasting the emergence of new precipitation cells. Novel satellite-based TCWV datasets derived from geostationary observations are particularly valuable for monitoring water vapor distributions and associated meteorological phenomena across scales: from synoptic down to local. These high-temporal-resolution observations support the advancement of observation-based nowcasting techniques, assessment of Numerical Weather Prediction (NWP) models, and enable more detailed process studies.

We develop and use TCWV retrievals from passive imagers onboard 3 different satellites to assess the added value of TCWV for improving (convective) nowcasting. TCWV represents vertically integrated water vapor, a single variable, and the NIR-based TCWV retrievals are particularly sensitive to boundary layer moisture content, which is a key ingredient in convective development. Comparisons between satellite-derived TCWV and model forecast fields often reveal substantial differences. This might be related to spatial misalignment of large-scale weather features or, at local scales, from the model's inability to capture small-scale dynamics in the boundary layer, such as moisture pooling. It becomes clear that the satellite-based observations do provide an important observational update of the TCWV field, relevant for both model- and observation-based nowcasting.



Current Status

With TCWV retrievals from OLCI onboard polar-orbiting Sentinel-3 satellites (Preusker et al, 2021), we set up the straightforward CAPE sharpening method (Carbajal Henken et al, 2025), where we integrate the observed high-resolution (300 m) TCWV field with model data to compute instability indices such as CAPE (Convective Available Potential Energy) and CIN (Convective Inhibition) to assess convective potential (WP-P2-5). By increasing the spatial resolution of these indices, the method enables improved characterization and monitoring of the pre-convective environment, e.g. , enhanced detection of local hotspots of convective potential. We demonstrate this with a set of case studies in Germany during summertime with isolated to scattered convective development in the early afternoon (WP-P2-4). Results show that the CAPE sharpening approach effectively enhances convective potential features and areas with convective development within several hours after satellite overpass align well with areas of elevated sharpened CAPE values at satellite overpass time. Limitations of using the CAPE sharpening approach with OLCI and ERA5 data are following: OLCI only has a couple of morning overpasses, thus offering only single snapshots of the environment without information on temporal evolution. This restricts statistical assessments and excludes many cases with extensive cloud cover in the morning time. Also, the CAPE sharpening model relies on the model atmospheric vertical profiles and if the model fields are far from reality, also no meaningful update with the CAPE sharpening can occur.

To address these shortcomings, there is a need to incorporate the temporal evolution of water vapor distributions in order to better capture localized moisture increases (e.g., in convergence zones) and rising convective potential, signaling upcoming convective initiation and development. Therefore, this more physically grounded CAPE sharpening approach will also be applied and assessed using MTG FCI TCWV retrievals and ICON-D2 model data to evaluate its capability to improve monitoring of pre-convective environments and convective potential.



Figure 1: Illustration of the CAPE sharpening approach, which incorporates OLCI TCWV retrievals into model-based CAPE and CIN computations. The base model used is the ERA5 reanalysis forecast product and for reference, the COSMO-D2 analysis field is also shown. The OLCI TCWV field reveals parallel bands of alternating low and high moisture, indicating the presence of horizontal convective rolls, as well as a convergence zone with increased moisture along the coastal area. These features are clearly reflected in both the TCWV difference plot and the sharpened CAPE field, but less in the model fields. Notably, regions of elevated sharpened CAPE align well with observed convective initiation and subsequent precipitation development later on.


As a proxy for the upcoming MTG FCI TCWV product, we use TCWV retrievals from MSG SEVIRI measurements (El Kassar et al. 2021). These retrievals are based on thermal infrared measurements and, while they offer coarser spatial resolution (~3 km x 6 km over Germany) and higher uncertainty compared to NIR-based retrievals, they provide valuable high-temporal-resolution observations (every 15 min). This enables assessment of regional moisture trends before the onset of convective clouds and precipitation. We have prepared a dataset combining SEVIRI TCWV fields for clear-sky pixels with cloud type and convective initiation (CI) and rapid developing thunderstorm (RDT) product derived using the NWCSAF software (www.nwcsaf.org). The dataset covers a German domain and consists of daytime 15-minute temporal resolution files for selected days from several months in the spring and summertime. This dataset has been prepared and made available for integration into a deep learning framework such as PredRNN (but more state-of-the-art architecture types are currently being explored), to investigate the added value of satellite-derived TCWV and cloud products for observation-based precipitation nowcasting (WP-P2-5). In particular, the goal is to investigate whether these additional observations can help improve the detection and prediction of newly emerging convective cells, which remain a challenge for more traditional QPN systems. While the work is still in an exploratory stage, initial testing and development has been done.



Figure 2 illustrates SEVIRI TCWV and cloud type fields at 15-minute intervals for a day during daytime hours.


In parallel, we adapted our NIR-based TCWV retrieval framework, that was previously set up for OLCI and other satellite instruments with similar band settings, to the new MTG FCI measurements, taking into account its specific technical challenges (WP-P2-3). Although there were significant delays in the provision of fully calibrated Level 1 data due to onboard calibration issues, initial testing was possible using an OLCI/SLSRT synergy dataset and early FCI test data (El Kassar et al., 2024). After processing several months of fully calibrated FCI data from this year, including spring and summer months with typically higher moisture amounts in Europe and where we have a dense ground-based reference network for comparison, we started a statistically robust validation study.





Figure 3 illustrates the observational capabilities of MTG-FCI for monitoring TCWV fields in a (pre-)convective environment and the added detail it provides compared to the prior information (here from ERA5).



References

• Preusker, R., C. Carbajal Henken, J. Fischer, 2021: Retrieval of daytime total column water vapour from OLCI measurements over land surfaces. Remote Sens., 13(5), 932, https://doi.org/10.3390/rs13050932

• El Kassar, J., C. Carbajal Henken, R. Preusker, J. Fischer, 2021: Optimal Estimation MSG-SEVIRI Clear-Sky Total Column Water Vapour Retrieval Using the Split Window Difference. Atmosphere, 12(10), 1256, https://doi.org/10.3390/atmos12101256.

• El Kassar, J., C. Carbajal Henken, X. Calbet, P. Rípodas, R. Preusker, and J. Fischer, 2024: Optimal Estimation Retrieval Framework for Daytime Clear-Sky Total Column Water Vapour from MTG-FCI Near-Infrared Measurements, EGUsphere [preprint], https://doi.org/10.5194/egusphere-2024-3605.

• Carbajal Henken, C.K., J. El Kassar, R. Preusker, 2025: CAPE sharpening: an enhanced view on pre-convective environments in Germany using OLCI TCWV retrievals. Submitted.

Although the SPROG model statistically outperforms pure advection-based nowcasting, it tends to smooth excessively small-scale convective precipitation. Therefore, the SPROG approach has been refined into its localized version: the SPROG-LOC approach (Reinoso-Rondinel et al., 2022). The SPROG-LOC approach uses a 2-dimensional autoregressive process to model the temporal evolution of each cascade level considering the inherent heterogeneity of observed precipitation fields. Figure P2-1 and figure P2-2 illustrate how the SPROG approach rapidly smooths small convective cells with increasing lead-time whereas the SPROG-LOC model is more reluctant to smooth such areas of convective precipitation. The SPROG-LOC model shows its advantage over the SPROG model on the convective band with small-scale structures of high precipitation (red, yellow and orange areas).



Figure P2-1: Comparison between the QPE fields (left), the SPROG model (centre) and the SPROG-LOC model (right) based on a three hours nowcast. The two nowcasts have been performed on the 2019-07-20 from 14:00 to 17:00 UTC.



Figure P2-2: Comparison between the QPE fields (left), the SPROG model (centre) and the SPROG-LOC model (right) based on the last frame of a three hours nowcast performed on the 2019-07-20 from 14:00 to 17:00 UTC

Lastly, the STEPS approach is currently being refined into its localized version, namely the STEPS-LOC approach, to better represent the stochastic perturbations and therefore the spatio-temporal representation of the ensemble-members. As a next step, the skill of the SPROG-LOC approach and the ensemble spread of the STEPS-LOC approach will undergo an in-depth evaluation within a joint effort of the entire RealPEP research group. RealPEP-P1 will apply the most recent hybrid precipitation algorithms developed (Chen et al., 2021) to a 4 months national polarimetric C-band radar data set, which serve as the data base for STEPS-LOC. These new QPE products use an optimized combination of reflectivity 𝑍𝐻, specific attenuation 𝐴𝐻, and specific differential phase 𝐾𝐷𝑃 and also take into account when the radar beam monitors within or above the melting layer. The resulting benchmark data set will also be used for a robust estimation of QPE uncertainties using the national gauge network and thus to further optimize the ensemble generation in STEPS-LOC.

Finally, vertically extensive enhancements of differential reflectivity, so-called 𝑍𝑑𝑟-columns, are investigated to further improve observation-based nowcasting. They are polarimetric signatures indicative of the location of updrafts and thus precipitation development. In order to exploit this information content for the refinement of our nowcast, algorithms for the detection and tracking of 𝑍𝑑𝑟-columns have been developed. Preliminary results (Reinoso-Rondinel et al., 2021 and Evaristo et al., 2021) suggest columns lifetimes of 20 to 30 min and a lag time up to 30 min for the precipitation increments at the surface. Statistical relationships between the features of the 𝑍𝑑𝑟-columns and its given intensification of precipitation will be used during the Phase II of P2 to increase the skill of our nowcast approach.

References

• Evaristo, R., Reinoso-Rondinel, R, Trömel, S., Simmer, C., 2021: Validation of Wind Fields Retrieved by Dual-Doppler Techniques Using a Vertically Pointing Radar. 2021 21st International Radar Symposium (IRS), Berlin, Germany, 2021, 1-7, https://doi.org/10.23919/IRS51887.2021.9466173

• Reinoso-Rondinel, R., Rempel, M., Schultze, M., Trömel, S., 2022: Nationwide Radar-Based Precipitation Nowcasting – A localization filtering approach and its Application for Germany. IEE J. Select. Topics Appl. Earth observation. and Remote Sensing, (15), 1670-1691, https://doi.org/10.1109/JSTARS.2022.3144342

• Reinoso-Rondinel, R., Evaristo, R., Schmidt, M., Crijnen, F., Trömel, S., 2021: Storm Cell Observation And Prediction Using Polarimetric Weather Radars. 2021 21st International Radar Symposium (IRS), Berlin, Germany, 2021, 1-7, https://doi.org/10.23919/IRS51887.2021.9466168

• Chen, J., Trömel, S., Ryzhkov, A., Simmer, C., 2021: Assessing the Benefits of Specific Attenuation for Quantitative precipitation Estimation with a C-band Radar Network. J. of hydrometeor., (22)(10), 2617-2631, https://doi.org/10.1175/JHM-D-20-0299.1

Contribution of Free University of Berlin


P2 will derive an ensemble of nowcasted precipitation fields based upon the precipitation fields provided by P1 while using the 3D multi-sensor composite (radar-satellite) provided by C1. The predictive potential of satellite-based information on convective cell initiation and its integration into the observation-based nowcasting method is investigated using machine learning methods. The QPN ensembles will serve as input for seamless prediction (P3) and flood prediction (P4).

The amount of water vapor in the atmosphere plays a key role in convective initiation processes. Here, we specifically investigate the potential of satellite-based total column water vapor (TCWV) for improving observation-based nowcasting of convective initiation (CI). The main idea is that these clear-sky satellite observations of water vapor fields enable the monitoring and characterization of (pre-)convective environments before convective cloud development and the onset of precipitation.
To this end, we first developed and evaluated two new satellite-based TCWV products. High-resolution spatial information of water vapor fields is obtained from the OLCI TCWV product (Preusker et al., 2021), while information on temporal evolution of water vapor fields comes from the SEVIRI TCWV product (El Kassar et al., 2021). Then, a multi-annual set of OLCI and SEVIRI TCWV fields is merged with NWC/SAF products related to in-cloud CI and thunderstorm development (www.nwcsaf.org), based on SEVIRI observations. Statistical assessments are performed to identify suitable TCWV metrics for the characterization of pre-convective environments in Germany. In next steps, we will apply machine learning methods (specifically predictive recurrent neural network) to merge satellite-based CI information with radar-based QPN, with the aim to increase lead times.Moreover, to benefit from the newest observational capabilities of Meteosat Third Generation (MTG) in the very near future, we can build on the TCWV retrieval framework set up for the synergy measurements from OLCI and SLSTR. The new MTG Flexible Combined Imager (FCI) based TCWV product is expected to enable improved spatio-temporal monitoring and characterization of (pre-)convective environments.



Figure P2-3: An example of a high resolution (~300 m) OLCI TCWV field in the morning time.



Animation P2-1: An example of a time series of SEVIRI TCWV fields and collocated NWC/SAF-SEVIRI observations of convective cloud development in the morning time and early afternoon.

References

• Preusker, R., C. Carbajal Henken , J. Fischer, 2021: Retrieval of daytime total column water vapour from OLCI measurements over land surfaces. Remote Sens., 13(5), 932, https://doi.org/10.3390/rs13050932.

• El Kassar, J., C. Carbajal Henken, R. Preusker, J. Fischer, 2021: Optimal Estimation MSG-SEVIRI Clear-Sky Total Column Water Vapour Retrieval Using the Split Window Difference. Atmosphere, 12(10), 1256, https://doi.org/10.3390/atmos12101256.

The conventional advection-based nowcasting is being improved by applying the filtering approach known as the Spectral Prognosis (S-PROG) method in order to model the spatio-temporal properties of precipitation. It is based on the assumption that large spatial structures of precipitation have longer lifetimes than smaller ones. The rain field is decomposed in the spectral domain into a multifractal geometry with K cascade levels. The temporal evolution of each cascade level is then managed by a 1-dimensional autoregressive model of order p. This way, the AR coefficients control the evolution of each cascade level consistent with their expected lifetime.
To consider the uncertainties, the probabilistic STEPS approach (Bowler et al., 2006) has been considered as well. It is based upon the SPROG approach but perturbs each cascade level with a Gaussian white noise that is spatially correlated with the precipitation such that the statistical properties of nowcast errors (spatial and temporal correlations) are simulated. Thanks to this approach, it has been possible to generate ensembles, for instance, 20 ensemble-members every 5 min with a lead time of 2 hr at 5 min time steps. The STEPS approach has been applied, among others, to three rain events that produced floods in the Mehlemer Bach region and resulting nowcast ensembles have been used as input for flash flood prediction in P4.
Parallel to the application of STEPS, object-oriented nowcasting of convective cells is being developed based on Kalman filtering. As a first task, a 2D storm cell identification and tracking algorithm have been implemented and tested on the Radolan Reflectivity product. As an example, Figure P2-1 indicates the tracks of 42 identified storm cells during a 2 hr period on July 25 2017.

Figure P2-1: Different colors and symbols indicate tracks of 42 storm cells identified in Radolan RW during a 2 hr period on July 25 2017.





References

• Bowler, N., C. E. Pierce, and A. W. Seed: 2006: STEPS: A probabilistic rainfall forecasting scheme which merges an extrapolation nowcast with downscales NWP, Q. J. R. Meteorol. Soc., 132, 2127–2155.