Nowcasting thunderstorm hazards using machine learning: the impact of data sources on performance

Abstract

Abstract. In order to aid feature selection in thunderstorm nowcasting, we present an analysis of the utility of various sources of data for machine-learning-based nowcasting of hazards related to thunderstorms. We considered ground-based radar data, satellite-based imagery and lightning observations, forecast data from numerical weather prediction (NWP) and the topography from a digital elevation model (DEM), ending up with 106 different predictive variables. We evaluated machine-learning models to nowcast radar reflectivity (representing precipitation), lightning occurrence, and the 45 dBZ radar echo top height that can be used as an indicator of hail, producing predictions for lead times up to 60 min. The study was carried out in an area in the northeast United States, where observations from the Geostationary Operational Environmental Satellite 16 are available and can be used as a proxy for the upcoming Meteosat Third Generation capabilities in Europe. The benefits of the data sources were evaluated using two complementary approaches: using feature importance reported by the machine learning model based on gradient boosted trees, and by repeating the analysis using all possible combinations of the data sources. The two approaches sometimes yielded seemingly contradictory results, as the feature importance reported by the gradient boosting algorithm sometimes disregards certain features that are still useful in the absence of more powerful predictors, while at times it overstates the importance of other features. We found that the radar data is overall the most important predictor, the satellite imagery is beneficial for all of the studied predictands, and the lightning data is very useful for nowcasting lightning but of limited use for the other hazards. The benefits of the NWP data are more limited over the nowcast period, and we did not find evidence that the nowcast benefits from the DEM data.

Seamless Lightning Nowcasting with Recurrent-Convolutional Deep Learning

“…For instance, the upcoming Meteosat Third Generation satellites will provide higher-resolution geostationary observations for Europe, potentially helping CNNs extract more information. Furthermore, the importance of different data sources, previously examined by Zhou et al (2020) and Leinonen et al (2022b), is yet to be quantified in this context, but is necessary in order to understand the expected performance of the network in, for example, regions where ground-based radar observations are not available.…”

“…Using the results of Leinonen et al (2022b) as a guideline, we selected various features that pertain to the occurrence of deep convection from among the COSMO model outputs. These were the convective available potential energy with respect to the most unstable level (CAPE-MU), the convective inhibition (CIN), the height of the 0 • C isotherm (HZEROCL), the lifting condensation level (LCL), the moisture convergence (MCONV), the vertical velocity of air in pressure coordinates (OMEGA), the surface lifted index (SLI), the soil type, and the temperatures at the surface and at 2 m height (T-SO and T-2M, respectively).…”

“…Therefore, we expected that it would be important to include information about the elevation in the analysis despite the results of Leinonen et al (2022b) that suggested that the DEM does not contribute significantly to the prediction skill. We used a set of DEM data derived from the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) global DEM (Abrams et al 2020).…”

“…Grid-based nowcasting avoids the complexity of the storm detection and tracking algorithms (including a the somewhat arbitrary cell definition and the problems with merging and splitting cells) at the cost of requiring more advanced machine learning methods, especially if it is also desired to predict the motion of thunderstorm cells. In the object-based category, recently published research includes the studies of Shrestha et al (2021) and Leinonen et al (2022b). Among studies on grid-based ML prediction of lightning, Lin et al (2019), Zhou et al (2020), andGeng et al (2021) used deep-learning techniques using convolutional neural networks (CNNs), while Blouin et al (2016) and La Fata et al (2021) used methods based on decision trees.…”

“…The elimination of data duplication allows us to greatly reduce the memory footprint or, conversely, use a larger amount of data within the limits of the available memory.We divided the dataset into a training set used to train the model, a validation set used for performance evaluation at training time and for hyperparameter tuning, and a test set used for final evaluation. Following the strategy ofLeinonen et al (2022b), entire days were set aside for the training set and the testing set in order to minimize overlap between the three subsets of data. First, days were randomly selected until at least 10% of the available time steps had been assigned into the validation set.…”