Search

The CORDEX Flagship Pilot Study ELVIC (climate Extremes in the Lake VICtoria basin) was recently established to investigate how extreme weather events will evolve in this region of the world and to provide improved information for the climate impact community. Here we assess the added value of the convection-permitting scale simulations on the representation of moist convective systems over and around Lake Victoria. With this aim, 10-year present-day model simulations were carried out with five regional climate models at both PARameterized (PAR) scales (12-25km) and Convection-Permitting (CP) scales (2.5-4.5km), with COSMO-CLM, RegCM, AROME, WRF and UKMO. Most substantial systematic improvements were found in metrics related to deep convection. For example, the timing of the daily maximum in precipitation is systematically delayed in CP compared to PAR models, thereby improving the agreement with observations. The large overestimation in the total number of rainy events is alleviated in the CP models. Systematic improvements were found in the diurnal cycle in Top-Of-Atmosphere (TOA) radiation and in some metrics for precipitation intensity. No unanimous improvement nor deterioration was found in the representation of the spatial distribution of total rainfall and the seasonal cycle when going to the CP scale. Furthermore, some substantial biases in TOA upward radiative fluxes remain. Generally our analysis indicates that the representation of the convective systems is strongly improved in CP compared to PAR models, giving confidence that the models are valuable tools for studying how extreme precipitation events may evolve in the future in the Lake Victoria basin and its surroundings.

The tracks of the Zanzibar and Lindi tropical cyclones on April 1872 and 1952, respectively. The approximate landfall position of cyclone Kenneth is also noted as ~12°S to highlight that this occurred much further south than the two historical cyclones presented here.

Current-climate precipitation and temperature extremes have been identified by decision makers in West Africa as among the more impactful weather events causing lasting socioeconomic damage. In this article, we use a plausible future-climate scenario (RCP8.5) for the end of the twenty-first century to explore the relative commonness of such extremes under global warming. The analysis presented considers what a typical day in the future climate will feel like relative to current extrema. Across much of West Africa, we see that the typical future-climate day has maximum and minimum temperatures greater than 99.5% of currently experienced values. This finding exists for most months but is particularly pronounced during the Boreal spring and summer. The typical future precipitation event has a daily rainfall rate greater than 95% of current storms. These findings exist in both a future scenario model run with and without parameterised convection, and for many of the Coupled Model Inter-comparison Project version 5 ensemble members. Additionally, agronomic monsoon onset is projected to occur later and have greater inter-annual variability in the future. Our findings suggest far more extreme conditions in future climate over West Africa. The projected changes in temperature and precipitation could have serious socioeconomic implications, stressing the need for effective mitigation given the potential lack of adaptation pathways available to decision makers.

The Hadley circulation and tropical rain belt are dominant features of African climate. Moist convection provides ascent within the rain belt, but must be parameterized in climate models, limiting predictions. Here, we use a pan-African convection-permitting model (CPM), alongside a parameterized convection model (PCM), to analyze how explicit convection affects the rain belt under climate change. Regarding changes in mean climate, both models project an increase in total column water (TCW), a widespread increase in rainfall, and slowdown of subtropical descent. Regional climate changes are similar for annual mean rainfall but regional changes of ascent typically strengthen less or weaken more in the CPM. Over a land-only meridional transect of the rain belt, the CPM mean rainfall increases less than in the PCM (5% vs 14%) but mean vertical velocity at 500 hPa weakens more (17% vs 10%). These changes mask more fundamental changes in underlying distributions. The decrease in 3-hourly rain frequency and shift from lighter to heavier rainfall are more pronounced in the CPM and accompanied by a shift from weak to strong updrafts with the enhancement of heavy rainfall largely due to these dynamic changes. The CPM has stronger coupling between intense rainfall and higher TCW. This yields a greater increase in rainfall contribution from events with greater TCW, with more rainfall for a given large-scale ascent, and so favors slowing of that ascent. These findings highlight connections between the convective-scale and larger-scale flows and emphasize that limitations of parameterized convection have major implications for planning adaptation to climate change.

Eastern Africa’s fast-growing population is vulnerable to changing rainfall and extremes. Using the first pan-African climate change simulations that explicitly model the rainfall-generating convection, we investigate both the climate change response of key mesoscale drivers of eastern African rainfall, such as sea and lake breezes, and the spatial heterogeneity of rainfall responses. The explicit model shows widespread increases at the end of the century in mean (~40%) and extreme (~50%) rain rates, whereas the sign of changes in rainfall frequency has large spatial heterogeneity (from −50% to over +90%). In comparison, an equivalent parameterized simulation has greater moisture convergence and total rainfall increase over the eastern Congo and less over eastern Africa. The parameterized model also does not capture 1) the large heterogeneity of changes in rain frequency; 2) the widespread and large increases in extreme rainfall, which result from increased rainfall per humidity change; and 3) the response of rainfall to the changing sea breeze, even though the sea-breeze change is captured. Consequently, previous rainfall projections are likely inadequate for informing many climate-sensitive decisions—for example, for infrastructure in coastal cities. We consider the physics revealed here and its implications to be relevant for many other vulnerable tropical regions, especially those with coastal convection.

The West African monsoon (WAM) is the dominant feature of West African climate providing the majority of annual rainfall. Projections of future rainfall over the West African Sahel are deeply uncertain, with a key reason likely to be moist convection, which is typically parameterized in global climate models. Here, we use a pan-African convection-permitting simulation (CP4), alongside a parameterized convection simulation (P25), to determine the key processes that underpin the effect of explicit convection on the climate change of the central West African Sahel (12°–17°N, 8°W–2°E). In current climate, CP4 affects WAM processes on multiple scales compared to P25. There are differences in the diurnal cycles of rainfall, moisture convergence, and atmospheric humidity. There are upscale impacts: the WAM penetrates farther north, there is greater humidity over the northern Sahel and the Saharan heat low regions, the subtropical subsidence rate over the Sahara is weaker, and ascent within the tropical rain belt is deeper. Under climate change, the WAM shifts northward and Hadley circulation weakens in P25 and CP4. The differences between P25 and CP4 persist, however, underpinned by process differences at the diurnal scale and large scale. Mean rainfall increases 17.1% in CP4 compared to 6.7% in P25 and there is greater weakening in tropical ascent and subtropical subsidence in CP4. These findings show the limitations of parameterized convection and demonstrate the value that explicit convection simulations can provide to climate modelers and climate policy decision makers.

The representation of convection remains one of the most important sources of bias in global models, and evaluation methods are needed that show that models provide the correct mean state and variability, both for the correct reasons. Here we develop a novel approach for evaluating rainfall variability due to convectively coupled Kelvin waves (CCKWs) in this region. A phase cycle was defined for the CCKW cycle in OLR and used to composite rainfall anomalies. We characterize the observed (TRMM) rainfall response to CCKWs over tropical Africa in April and evaluate the performance of regional climate model (RCM) simulations: a parameterized convection simulation (P25) and the first pan-Africa convection-permitting simulation (CP4). TRMM mean rainfall is enhanced and suppressed by CCKW activity, and the occurrence of extreme rainfall and dry days is coupled with CCKW activity. Focusing on regional differences, we show for the first time that there is a dipole between West Africa and the Gulf of Guinea involving onshore/offshore shifts in rainfall, and the transition to enhanced rainfall over west equatorial Africa occurs one phase before the transition over east equatorial Africa. The global model used to drive the RCMs simulated CCKWs with mean amplitudes of 75%–82% of observations. The RCMs simulated coherent responses to the CCKWs and captured the large-scale spatial patterns and phase relationships in rainfall although the simulated rainfall response is weaker than observations and there are regional biases that are bigger away from the equator. P25 produced a closer match to TRMM mean rainfall anomalies than CP4 although the response in dry days was more closely simulated by CP4.

Article Open Access Published: 08 November 2019 Nowcasting lightning occurrence from commonly available meteorological parameters using machine learning techniques Amirhossein Mostajabi, Declan L. Finney, Marcos Rubinstein & Farhad Rachidi npj Climate and Atmospheric Science volume 2, Article number: 41 (2019) Cite this article Article metrics 71 Altmetric Metrics details Abstract Lightning discharges in the atmosphere owe their existence to the combination of complex dynamic and microphysical processes. Knowledge discovery and data mining methods can be used for seeking characteristics of data and their teleconnections in complex data clusters. We have used machine learning techniques to successfully hindcast nearby and distant lightning hazards by looking at single-site observations of meteorological parameters. We developed a four-parameter model based on four commonly available surface weather variables (air pressure at station level (QFE), air temperature, relative humidity, and wind speed). The produced warnings are validated using the data from lightning location systems. Evaluation results show that the model has statistically considerable predictive skill for lead times up to 30 min. Furthermore, the importance of the input parameters fits with the broad physical understanding of surface processes driving thunderstorms (e.g., the surface temperature and the relative humidity will be important factors for the instability and moisture availability of the thunderstorm environment). The model also improves upon three competitive baselines for generating lightning warnings: (i) a simple but objective baseline forecast, based on the persistence method, (ii) the widely-used method based on a threshold of the vertical electrostatic field magnitude at ground level, and, finally (iii) a scheme based on CAPE threshold. Apart from discussing the prediction skill of the model, data mining techniques are also used to compare the patterns of data distribution, both spatially and temporally among the stations. The results encourage further analysis on how mining techniques could contribute to further our understanding of lightning dependencies on atmospheric parameters.

An observed decline in the Eastern African Long Rains from the 1980s to late 2000s appears contrary to the projected increase under future climate change. This “Eastern African climate paradox” confounds use of climate projections for adaptation planning across Eastern Africa. Here we show the decline corresponds to a later onset and earlier cessation of the long rains, with a similar seasonal maximum in area-averaged daily rainfall. Previous studies have explored the role of remote teleconnections, but those mechanisms do not sufficiently explain the decline or the newly identified change in seasonality. Using a large ensemble of observations, reanalyses and atmospheric simulations, we propose a regional mechanism that explains both the observed decline and the recent partial recovery. A decrease in surface pressure over Arabia and warmer north Arabian Sea is associated with enhanced southerlies and an earlier cessation of the long rains. This is supported by a similar signal in surface pressure in many atmosphere-only models giving lower May rainfall and an earlier cessation. Anomalously warm seas south of Eastern Africa delay the northward movement of the tropical rain-band, giving a later onset. These results are key in understanding the paradox. It is now a priority to establish the balance of mechanisms that have led to these trends, which are partially captured in atmosphere-only simulations. Springtime drying trends in East Africa can be explained by shifts in rainfall seasonality and not reductions in daily rainfall. The East African “Long Rains” — which fall during March, April and May — deliver the majority of precipitation to the region. Since the mid-1980s, observations reveal a reduction in rainfall which some climate models fail to capture. Using a suite of observations, reanalyses, and atmospheric simulations, Caroline Wainwright from the University of Reading, UK, and colleagues investigate the mechanisms behind this change. The reduction in Long Rain precipitation is characterised by a later onset and earlier cessation — and thereby a shortening — of the rainy season, attributed to a more rapid movement of the rain band associated with warmer SSTs off East Africa. Understanding these mechanisms will benefit climate projections and adaptation measures in an already-vulnerable region.

Pan-Africa convection-permitting regional climate model simulations have been performed to study the impact of high resolution and the explicit representation of atmospheric moist convection on the present and future climate of Africa. These unique simulations have allowed European and African climate scientists to understand the critical role that the representation of convection plays in the ability of a contemporary climate model to capture climate and climate change, including many impact-relevant aspects such as rainfall variability and extremes. There are significant improvements in not only the small-scale characteristics of rainfall such as its intensity and diurnal cycle, but also in the large-scale circulation. Similarly, effects of explicit convection affect not only projected changes in rainfall extremes, dry spells, and high winds, but also continental-scale circulation and regional rainfall accumulations. The physics underlying such differences are in many cases expected to be relevant to all models that use parameterized convection. In some cases physical understanding of small-scale change means that we can provide regional decision-makers with new scales of information across a range of sectors. We demonstrate the potential value of these simulations both as scientific tools to increase climate process understanding and, when used with other models, for direct user applications. We describe how these ground-breaking simulations have been achieved under the U.K. Government’s Future Climate for Africa Programme. We anticipate a growing number of such simulations, which we advocate should become a routine component of climate projection, and encourage international coordination of such computationally and human-resource expensive simulations as effectively as possible.

Mean temperatures and their extremes have increased over Africa since the latter half of the 20th century and this trend is projected to continue, with very frequent, intense and often deadly heatwaves likely to occur very regularly over much of Africa by 2100. It is crucial that we understand the scale of the future increases in extremes and the driving mechanisms. We diagnose daily maximum wet bulb temperature heatwaves, which allows for both the impact of temperature and humidity, both critical for human health and survivability. During wet bulb heatwaves, humidity and cloud cover increase, which limits the surface shortwave radiation flux but increases longwave warming. It is found from observations and ERA5 reanalysis that approximately 30% of wet bulb heatwaves over Africa are associated with daily rainfall accumulations of more than 1 mm/day on the first day of the heatwave. The first ever pan-African convection-permitting climate model simulations of present-day and RCP8.5 future climate are utilised to illustrate the projected future change in heatwaves, their drivers and their sensitivity to the representation of convection. Compared to ERA5, the convection-permitting model better represents the frequency and magnitude of present-day wet bulb heatwaves than a version of the model with more traditional parameterised convection. The future change in heatwave frequency, duration and magnitude is also larger in the convective-scale simulation, suggesting CMIP-style models may underestimate the future change in wet bulb heat extremes over Africa. The main reason for the larger future change appears to be the ability of the model to produce larger anomalies relative to its climatology in precipitation, cloud and the surface energy balance.

The East African precipitation seasonal cycle is of significant societal importance, and yet the current generation of coupled global climate models fails to correctly capture this seasonality. The use of convective parameterization schemes is a known source of precipitation bias in such models. Recently, a high-resolution regional model was used to produce the first pan-African climate change simulation that explicitly models convection. Here, this is compared with a corresponding parameterized-convection simulation to explore the effect of the parameterization on representation of East Africa precipitation seasonality. Both models capture current seasonality, although an overestimate in September–October in the parameterized simulation leads to an early bias in the onset of the boreal autumn short rains, associated with higher convective instability and near-surface moist static energy. This bias is removed in the explicit model. Under future climate change both models show the short rains getting later and wetter. For the boreal spring long rains, the explicit convection simulation shows the onset advancing but the parameterized simulation shows little change. Over Uganda and western Kenya both simulations show rainfall increases in the January–February dry season and large increases in boreal summer and autumn rainfall, particularly in the explicit convection model, changing the shape of the seasonal cycle, with potential for pronounced socioeconomic impacts. Interannual variability is similar in both models. Results imply that parameterization of convection may be a source of uncertainty for projections of changes in seasonal timing from global models and that potentially impactful changes in seasonality should be highlighted to users.

The 2019 October-December rains over East Africa were one of the wettest seasons on record, with many locations receiving more than double the climatological rainfall, leading to floods and landslides across the region. Above average rainfall continued into January 2020. The persistently high rainfall also contributed to the locust plagues that affected much of East Africa in January 2020. Wet conditions in East Africa are typically associated with El Nino and/or positive Indian Ocean Dipole events. In October-December 2019 a warm anomaly was present in the western Indian Ocean while a cool anomaly was present in the eastern Indian Ocean (a positive Indian Ocean Dipole); conditions known to give above average rainfall over East Africa. The warm anomaly in the western Indian Ocean persisted into January 2020. Seasonal and monthly forecasts correctly predicted above average rainfall during the October-December season. January rainfall is found to be correlated with sea surface temperatures over the western Indian Ocean. Climate model projections suggest that strong positive Indian Ocean Dipole events and wet October-December seasons may become more frequent under future climate change, with associated increased risks of floods.

Global climate models struggle to simulate both the convection and cloud ice fundamental to lightning formation. We use the first convection‐permitting, future climate simulations for the lightning hot spot of Africa, at the same time utilizing an ice‐based lightning parametrization. Both the model and observations show that lightning over Africa's drier areas, as well as the moist Congo, have more lightning per rainfall than other regions. Contrary to results in the literature, the future projection shows little increase in total lightning (~107 flashes (or 2%) per degree warming). This is a consequence of increased stability reducing the number of lightning days, largely offsetting the increased graupel and updraft velocity driving an increase in lightning per lightning day. The next step is to establish if these results are robust across other models and, if combined with parametrized‐convection models, whether ensemble‐based information on the possible responses of lightning to climate change can be investigated. Plain Language Summary Lightning depends on ascending air in thunderstorms and the collision of cloud ice particles, which charge the thundercloud. Many climate models have too coarse a resolution to reliably capture these processes. We focus on Africa, which has some of the most frequent lightning in the world. We use a model that is much higher resolution than usual, and this allows us to explicitly simulate the deep convection associated with thunderstorms as well as provide more detailed representation of the distribution of cloud ice particles. Our results show that in drier regions, as well as the much wetter Congo, there is relatively more lightning per kilogram of surface rainfall than there is in other parts of the continent. Lightning does increase across the continent under climate change, but by a relatively small amount. This is despite the number of days with lightning decreasing as the lower atmosphere becomes more stable. On days with lightning, there are more lightning flashes because there is an increase in cloud ice and intensity of convection. This study gives much more detailed information about African lightning than previous work. However, it is a single simulation. Future research should look at these results across other climate models.

Extreme rainfall is expected to increase under climate change, carrying potential socioeconomic risks. However, the magnitude of increase is uncertain. Over recent decades, extreme storms over the West African Sahel have increased in frequency, with increased vertical wind shear shown to be a cause. Drier midlevels, stronger cold pools, and increased storm organization have also been observed. Global models do not capture the potential effects of lower- to midtropospheric wind shear or cold pools on storm organization since they parameterize convection. Here we use the first convection-permitting simulations of African climate change to understand how changes in thermodynamics and storm dynamics affect future extreme Sahelian rainfall. The model, which simulates warming associated with representative concentration pathway 8.5 (RCP8.5) until the end of the twenty-first century, projects a 28% increase of the extreme rain rate of MCSs. The Sahel moisture change on average follows Clausius–Clapeyron scaling, but has regional heterogeneity. Rain rates scale with the product of time-of-storm total column water (TCW) and in-storm vertical velocity. Additionally, prestorm wind shear and convective available potential energy both modulate in-storm vertical velocity. Although wind shear affects cloud-top temperatures within our model, it has no direct correlation with precipitation rates. In our model, projected future increase in TCW is the primary explanation for increased rain rates. Finally, although colder cold pools are modeled in the future climate, we see no significant change in near-surface winds, highlighting avenues for future research on convection-permitting modeling of storm dynamics.

Extreme weather events, like heavy precipitation, heat waves, droughts and wind storms have a detrimental impact on East African societies. The Lake Victoria Basin (LVB) is especially vulnerable, since nightly storms on the lake catch fishermen by surprise. As the frequency and intensity of climate extremes is projected to further increase substantially with climate change, so do the risks, with potentially major consequences for livelihoods and policy in the LVB.The ultimate aim of the ELVIC CORDEX-FPS is to investigate how extreme weather events evolve in the future in the LVB and to provide improved probabilistic information to the impact community. ELVIC (Climate Extremes in the Lake Victoria Basin) brings together different research groups that perform simulations with multiple high-resolution regional climate models operating at the convection-permitting scale (CPS) (https://ees.kuleuven.be/elvic).As a first step towards this overall goal, the added value of the CPS on the representation of deep convective systems in Equatorial Africa was assessed. For this purpose, 10-year present-day model simulations were carried out with five regional climate models both at the CPS and at the scale where convection was parameterized, namely COSMO-CLM, RegCM, HCLIM-AROME, WRF and the Met Office Unified Model. From a comparison of model output with different observational products, no robust improvement was found for seasonal average meteorological variables. Moreover, the change in the seasonal precipitation when going to CPS differs between the models. A robust improved performance was found for deep convection, reflected in an improved representation of the daily precipitation cycle. Preliminary results also point towards an improvement in the representation of extreme precipitation. This suggests that regional climate model simulations at the convection-permitting scale are indeed relevant to assess the climate sensitivity of extreme precipitation in the Lake Victoria Basin.

Lightning strongly influences atmospheric chemistry1–3, and impacts the frequency of natural wildfires 4 . Most previous studies project an increase in global lightning with climate change over the coming century1,5–7, but these typically use parameterizations of lightning that neglect cloud ice fluxes, a component generally considered to be fundamental to thunderstorm charging 8 . As such, the response of lightning to climate change is uncertain. Here, we compare lightning projections for 2100 using two parameterizations: the widely used cloud-top height (CTH) approach 9 , and a new upward cloud ice flux (IFLUX) approach 10 that overcomes previous limitations. In contrast to the previously reported global increase in lightning based on CTH, we find a 15% decrease in total lightning flash rate with IFLUX in 2100 under a strong global warming scenario. Differences are largest in the tropics, where most lightning occurs, with implications for the estimation of future changes in tropospheric ozone and methane, as well as differences in their radiative forcings. These results suggest that lightning schemes more closely related to cloud ice and microphysical processes are needed to robustly estimate future changes in lightning and atmospheric composition. It has been suggested that lightning activity will increase with anthropogenic warming. However, the use of a physically based lightning parameterization—incorporating cloud ice fluxes—reveals global flash rates in 2100 may decrease by 15% under RCP8.5.

Thawing Arctic permafrost, and release of its stored carbon, is a known amplifier of global warming. Now research suggests an increase in Arctic lightning could speed up the permafrost’s demise.

The precipitation and diabatic heating resulting from moist convection make it a key component of the atmospheric water budget in the tropics. With convective parameterization being a known source of uncertainty in global models, convection-permitting (CP) models are increasingly being used to improve understanding of regional climate. Here, a new 10-yr CP simulation is used to study the characteristics of rainfall and atmospheric water budget for East Africa and the Lake Victoria basin. The explicit representation of convection leads to a widespread improvement in the intensities and diurnal cycle of rainfall when compared with a parameterized simulation. Differences in large-scale moisture fluxes lead to a shift in the mean rainfall pattern from the Congo to Lake Victoria basin in the CP simulation—highlighting the important connection between local changes in the representation of convection and larger-scale dynamics and rainfall. Stronger lake–land contrasts in buoyancy in the CP model lead to a stronger nocturnal land breeze over Lake Victoria, increasing evaporation and moisture flux convergence (MFC), and likely unrealistically high rainfall. However, for the mountains east of the lake, the CP model produces a diurnal rainfall cycle much more similar to satellite estimates, which is related to differences in the timing of MFC. Results here demonstrate that, while care is needed regarding lake forcings, a CP approach offers a more realistic representation of several rainfall characteristics through a more physically based realization of the atmospheric dynamics around the complex topography of East Africa.

A parameterisation to incorporate the effects of frozen soil on modelled hydrology is described and implemented within a land surface model, the Joint UK Land Surface Environment Simulator. It is shown to generally improve the modelled flow of Siberian rivers compared to observations, specifically in seasons of freezing or thawing soil. Most noticeably, the revised model increases the snowmelt flow peak by 26–100% compared to the control model thereby better matching observed flows. The model physics resulting in the changes to river flow are discussed and attention is given to the effect of inaccuracies in snowfall driving data which can hinder the comparison of new model processes.

Lightning is an important natural source of nitrogen oxide especially in the middle and upper troposphere. Hence, it is essential to represent lightning in chemistry transport and coupled chemistry–climate models. Using ERA-Interim meteorological reanalysis data we compare the lightning flash density distributions produced using several existing lightning parametrisations, as well as a new parametrisation developed on the basis of upward cloud ice flux at 440 hPa. The use of ice flux forms a link to the non-inductive charging mechanism of thunderstorms. Spatial and temporal distributions of lightning flash density are compared to tropical and subtropical observations for 2007–2011 from the Lightning Imaging Sensor (LIS) on the Tropical Rainfall Measuring Mission (TRMM) satellite. The well-used lightning flash parametrisation based on cloud-top height has large biases but the derived annual total flash density has a better spatial correlation with the LIS observations than other existing parametrisations. A comparison of flash density simulated by the different schemes shows that the cloud-top height parametrisation has many more instances of moderate flash densities and fewer low and high extremes compared to the other parametrisations. Other studies in the literature have shown that this feature of the cloud-top height parametrisation is in contrast to lightning observations over certain regions. Our new ice flux parametrisation shows a clear improvement over all the existing parametrisations with lower root mean square errors (RMSEs) and better spatial correlations with the observations for distributions of annual total, and seasonal and interannual variations. The greatest improvement with the new parametrisation is a more realistic representation of the zonal distribution with a better balance between tropical and subtropical lightning flash estimates. The new parametrisation is appropriate for testing in chemistry transport and chemistry–climate models that use a lightning parametrisation.

Lightning is formed in the atmosphere through the combination of complex dynamic and microphysical processes. Lightning can have a considerable influence on the environment and on the economy since it causes energy supply outages, forest fires, damages, injury and death of humans and livestock worldwide. Therefore, it is of great importance to be able to predict lightning incidence in order to protect people and installations. Despite numerous attempts to solve the important problem of lightning prediction (e.g., [1]–[3]), the complex processes and large number of parameters involved in the problem lend themselves to the potential application of a machine learning (ML) approach.We recently proposed a ML-based lightning early-warning system with promising performance [4]. The proposed ML model is trained to nowcast lightning incidence during any one of three consecutive 10-minute time intervals and within a circular area of 30 km radius around a meteorological station. The system uses the real-time measured values of four meteorological parameters that are relevant to the mechanisms of electric charge generation in thunderstorms, namely the air pressure at station level (QFE), the air temperature 2 m above ground, the relative humidity, and the wind speed. The proposed algorithm was implemented using the data from 12 meteorological stations in Switzerland between 2006-2017 with a granularity of ten minutes. The stations were selected to be well distributed among different ranges of altitude and terrain topographies.The algorithm requires the filtering out of a portion of the data which are identified as outliers. However, the process of the automatic identification of outliers is a challenging task which could also affect the model’s performance. In this presentation, we discuss this problem and present approaches that can be used to optimize the process. References[1] D. Aranguren, J. Montanya, G. Solá, V. March, D. Romero, and H. Torres, “On the lightning hazard warning using electrostatic field: Analysis of summer thunderstorms in Spain,” J. Electrostat., vol. 67, no. 2–3, pp. 507–512, May 2009.[2] G. N. Seroka, R. E. Orville, and C. Schumacher, “Radar Nowcasting of Total Lightning over the Kennedy Space Center,” Weather Forecast., vol. 27, no. 1, pp. 189–204, Feb. 2012.[3] Q. Meng, W. Yao, and L. Xu, “Development of Lightning Nowcasting and Warning Technique and Its Application,” Adv. Meteorol., vol. 2019, pp. 1–9, Jan. 2019.[4] A. Mostajabi, D. L. Finney, M. Rubinstein, and F. Rachidi, “Nowcasting lightning occurrence from commonly available meteorological parameters using machine learning techniques,” npj Clim. Atmos. Sci., vol. 2, no. 1, p. 41, 2019.