About the Nowcasting Portal

This portal is designed to help forecasters in Sub-Saharan Africa to predict severe convective storms over the next few hours using near-real time satellite data — a process known as nowcasting. It has been developed by UKCEH scientists working in partnership with forecasters and researchers in Africa and the UK. The portal has been developed under a series of projects funded by the Natural Environment Research Council (NERC), UK Research and Innovation (UKRI), and the Foreign, Commonwealth and Development Office (FCDO). The underpinning idea of the nowcasting scheme arose from the Nowcasting FLood Impacts of Convective storms in the Sahel (NFLICS) project (NERC/FCDO), a collaboration between UKCEH and the Agence Nationale de l'Aviation Civile et de la Météorologie (ANACIM) in Senegal. Engagement with forecasting services across the wider West African region was enabled through the GCRF African SWIFT project (UKRI) led by the National Centre for Atmospheric Science. Since 2021, portal development and further in-country training and engagement visits has come from the GCRF and Newton Fund Consolidation Accounts (2022–23) and NERC National Capability support. The extension of the domain to include Southern Africa is funded by the WISER-EWSA project (FCDO). Data from the portal comes primarily from imagery provided by the Meteosat Second Generation series of satellites, delivered by Eumetsat. Additional derived products on the portal have been produced by the Land and Hydrology Satellite Application Facilities.

The availability of products on the portal is evolving, in terms of new products, the spatial extent of the products and methods used to generate the products. As well as near-real time information, the portal includes an archive of historical imagery and products dating back to 2020 (but this is not necessarily complete and may use several product versions). At the core of the portal are probabilistic ("NFLICS") nowcasts developed for Senegal during the Sahel wet season months of June to September starting in 2020. These have been extended to cover all of West Africa in time for the 2023 season. We hope to run this West African service over the full annual cycle from October 2023 onwards. We also plan to test the NFLICS approach over Southern Africa from late 2023. The spatial extent and historical availability of other satellite data (cloud top temperatures, convective cores, land surface temperatures etc.) broadly follows the evolution of the NFLICS nowcasts extent, but from June 2023 onwards, we aim to provide these products in near-real time over our full Sub-Saharan Africa domain.

Observations

Cloud-top temperature

This product identifies areas where high cloud is present, based on thermal infra-red images (10.8 μm) every 15 minutes from Meteosat Second Generation (MSG). Colder temperatures are associated with higher cloud-tops and therefore indicate more intense storms. This product is useful for nowcasting because deep convective clouds, responsible for the vast majority of intense rain across tropical Africa, have cloud-top temperatures typically well below -40°C. However, the presence of cold cloud-tops does not always imply convective rainfall. Typical long-lived convective storms create extensive areas of cold cloud around them which may produce light rain, or indeed no rain. Under these circumstances, the convective core product can help to identify the coldest (and most convectively-active) parts of the cloud shield.

Convective cores

This product identifies areas within large storm clouds where cloud-top temperatures are particularly cold. This is based on an automated spatial analysis of thermal infra-red images (10.8 μm) provided every 15 minutes by Meteosat Second Generation (MSG). "Convective cores" are colder than the surrounding cloud field and are identified from a spatial filtering process which distinguishes the most convectively active parts of a cold cloud system (the cores) from slightly warmer regions (for example the stratiform cloud shield). The core areas are responsible for a large fraction of intense rain rates in West Africa, particularly in the Sahel. Because the cores are based only on infra-red information, there is no degradation of their information content at night. The spatial filtering applied identifies features on spatial scales of 10–50km. This means that the emergence of a new convective core only becomes apparent once the cold cloud-top has expanded (typically 15–30 minutes). Also relatively rare cases of extremely large areas (>100 km) of intense convection will show up as having cores around their edge but not at the centre.

Convective cores (current)

This shows the convective cores for the current time.

Convective cores (recent)

This shows the convective cores last 6 hours at 30-minute intervals to indicate how the cores have evolved.

Visible radiation

This product shows radiation in the visible channel reflected by the earth back into space. It is taken directly from channel 1 data (0.6 μm) from Meteosat Second Generation (MSG), with images updated every 15 minutes during daytime. Reflected radiation picks out the evolution of clouds. In particular, this field identifies areas of shallow clouds which are not easily detected using cloud-top temperature. Visible radiation is therefore useful for detecting regions where deep convection may develop within the next hour. During the daytime, convergence lines associated with the Intertropical Discontinuity, cold pools, and other features, can be identified before deep convection develops.

Precipitation rate (HSAF H60B)

This product is produced in near-real time by the Hydrology Satellite Applications Facility (HSAF) from a combination of Meteosat Second Generation (MSG) 10.8 μm thermal infra-red data and imagery from polar-orbiting satellites with passive microwave sensors onboard. The MSG thermal infra-red data provide 15 minute updates of the location of cold cloud-tops, which give an indication of where precipitation may be falling. The passive microwave data is sensitive to hydrometeors within the cloud, not simply the cloud-top properties, and therefore provides more reliable estimates of rainfall rate. Together, the two sources exploit the high temporal frequency of the thermal infra-red data and the higher information content of the passive microwave imagery. An accumulated rainfall layer is also provided over a range of periods from 1 hour to 3 days.

Land Surface Temperature (LST) Anomaly

The LST anomaly is a daily product derived from daytime, clear-sky Meteosat Second Generation (MSG) observations. The product is useful for nowcasting because convection is sensitive to land surface fluxes of heat and moisture. When evapotranspiration is limited by soil moisture deficit, both LST and sensible heat flux increase. Such conditions are met throughout the year in the semi-arid regions of Africa. In the days after rain, areas which have wetter top soil produce high rates of evapotranspiration, accompanied by low sensible heat flux, and low LST compared to climatological values. Areas with negative LST anomalies are less likely to experience deep convection than warmer neighbouring areas.

Every 15 minutes, the Land Surface Analysis Satellite Applications Facility (LSA SAF) produce an LST image from MSG channels 9 and 10 in the infra-red. For each 15-minute image, we apply additional cloud-screening and compare with a climatology based on data for the same pixel and month from the period 2004–2015. We then compute a daytime mean LST Anomaly (LSTA) using all cloud-screened data between the hours of 0700 and the current time or 1700 (local time), whichever is the earlier. Over the first few hours of the morning, the LSTA patterns emerge as more cloud-free data become available. Where there is no information about LSTA, the pixel is shown in grey. During the Sahel wet season for example, such conditions predominate in Southern West Africa. Note that outside of the wet season, when soils are dry, LSTA anomaly patterns are strongly influenced by atmospheric features (e.g. dust, cloud) rather than soil moisture. Such patterns tend to be rather smooth and should be ignored for nowcasting convection.

For more information about how LST anomalies can be used for nowcasting convection, see key papers.

Land modification factor

This product provides a quantitative estimate of how the likelihood of intense convective rainfall is modified by the land surface. A value of 30% indicates that convection is 30% more likely given the current state of the land surface. It is derived from a combination of daytime Land Surface Temperature (LST) anomalies, observed from Meteosat Second Generation (MSG), and historical data encapsulating the statistical relationships between convective cores and LST anomalies (LSTA). Its usage for nowcasting in the Sahel has been tested during September 2021 and due to its experimental nature, is only currently being provided for the months of the Sahel wet season (June to September).

To compute the land modification factor, we translate daily LSTA to a land modification factor by considering the climatological relationship between LSTA and convective cores at the given validity time of day. These statistics come from analysis of where convective cores occurred during the years 2004–2015 relative to LSTA values. The statistics are computed by month (June to September only) and within 3 latitude bounds (south of 12.5N, 12.5–15N and north of 15N). We determine for each pixel where today's LSTA value sits within the climatology for the month and latitude band in question and read off the associated probability of a convective core from the historical data given the strength of that LSTA. The land modification factor is provided at 3-hourly intervals between 1200 UTC and 0300 UTC. For validity times from 1800 onwards, we use the LSTA value 1 degree to the east of the target pixel as the statistics show that long-lived convective systems are most sensitive to the land surface upstream of the target. The land modification factor plotted represents the percentage increase (or decrease) in likelihood of a convective core given the land surface state. It takes no account of the current state of the atmosphere, and depends only on land conditions. Where cloud cover obscures the surface, the land modification factor is masked out. From 1000 UTC, the LSTA pattern is sufficiently robust to start producing a new set of land modification factors for the day.

Land modulation factors are strongest during the afternoon and early evening, when land surface fluxes are directly affecting the properties of the lower atmosphere, but still provide useful (though weak) predictability at 0300 UTC. The predictive skill also varies by month and latitude, driven primarily by the presence of vegetation cover; where vegetation is sparse, the relationship between LSTA and surface fluxes is strongest. For more information about how the land modification factor can be used for nowcasting convection, see key papers.

Rain over Africa

This layer provides a retrieval of rain rate every 15 minutes, based on a machine learning approach applied to Meteosat Second Generation imagery. It has been made available for the duration of the WISER-EWSA Testbed in Zambia (Jan 29 to Feb 9 2024) by one of its developers, Adria Amell Tosas at Chalmers University of Technology, Sweden. The method uses a convolutional neural network trained on the combined GPM DPR and GMI precipitation L2B product. Whilst the method produces a distribution of rain rates, providing a quantification of per-pixel uncertainty, we only show the posterior mean of the distribution.

Further details can be found at:

Soil moisture anomalies

This product uses estimates of near-surface soil moisture derived from Advanced Scatterometer (ASCAT) measurements from the Metop series of satellites, provided by the Hydrology Satellite Applications Facility. Data from the morning and preceding evening overpasses cover much of the region in broad swaths, and are presented as anomalies from a long-term climatology. The product is useful for nowcasting because soil moisture can control the surface fluxes of heat and moisture, to which convection is sensitive. Throughout the year in semi-arid Africa, and outside of the rainiest months further south, soil moisture deficit limits evapotranspiration. Under these conditions, negative soil moisture anomalies tend to favour deep convection by allowing higher LSTs that heat up the overlying air relative to moister surrounding regions. For more information about how land surface anomalies can be used for nowcasting convection, see key papers.

Nowcasts (0–6h)

SII-NowNet intensification

A probabilistic nowcast over Zambia representing the likelihood of convection intensification in 1 hour. For this product, convection intensification is considered as a rapid decrease in the cloud top temperature. This machine learning model was trained over Sumatra (Indonesia) and is being applied to Africa as a test run.

SII-NowNet initiation

A probabilistic nowcast over Zambia representing the likelihood of convection initiation in 1 hour. For this product, convection initiation is considered as a new convective cell that appears in the cloud top temperature field. This machine learning model was trained over Sumatra (Indonesia) and is being applied to Africa as a test run.

Nowcast probability of convection

This product aims to provide rapidly updated probability nowcasts of convective structures (cores) occurring over the next 6 hours. These convective structures are associated with heavy rainfall events so are useful for short-term forecasting and warning. Note, nowcasts are only produced if convective structures are present in the latest cloud-top temperature image. Also the version and method used for the nowcast product has evolved over time and further updates are planned.

Nowcasts are produced every 15 minutes, out to 6 hours and have an hourly timestep. The forward (Clock rotate right icon) and backward (Clock rotate left icon) buttons on the legend allow users to move through the hourly nowcast images. Probability time-series graphs are available for specific named points of interest by clicking on the blue circles or for any location by clicking on the map. The probabilities (0–100%) represent the chance of a convective structure occurring within a given spatial scale of the coloured pixel. This spatial scale has been optimised using past data and attempts to balance having useful forecast skill against the increased spatial uncertainties in storm locations at longer lead-times.

All nowcasts are produced using Step 1 below.

Step 1

  • A "conditional climatology" approach is used to produce the nowcasts. The nowcasts use the convective structures (cores) identified at the start of a nowcast to "look up" the likely future locations of convective structures based on a historical analysis over a relevant historical period (e.g. JJAS 2004–2019). The conditional climatology (i.e. what has happened in the next 6 hours given there is a convective structure at this location) is calculated for each time of day and across a grid of source area locations. As a result, the NFLICS nowcasts can potentially allow for common decay and growth sequences. The outputs are probabilities of convective structures occurring rather than ensembles of possible future storms. The NFLICS products do not explicitly infer or advect recent storm trajectories.

Additional Land Surface Temperature information can be used to further refine the nowcasts by applying a second step to the nowcast calculation.

Step 2

  • Step 2 uses recent Land Surface Temperature anomalies to modify the Step 1 nowcast probabilities. The probability adjustments are based on historical analysis that shows cool/wet areas are less favourable for convection and heavy rain, whilst warm/dry areas are more favourable.

Nowcast versions

The nowcast versions are labelled vA_B_C.

A indicates the nowcast configuration used, with a particular value assigned for a given combination of domain, scale selection, historical data used for "conditional climatology", months of the year it is available for and other elements of the nowcasting method not covered by B and C below (e.g. convective core identification method or if coastal/sea source areas are considered).

  • 1, 2 This is the original NFLICS domain focussed on Senegal. It is only available for the summer rainy season (June, July, August, September (JJAS)). The conditional climatology uses JJAS from 2004–2019, it is pooled across JJAS so does not vary month-to-month.
  • 3 An extended West African domain. Over the original NFLICS domain this is identical to 2.
  • 4 Identical to 3 except nowcasts are now initiated from an updated version of convective cores calculated on the MSG grid.
  • 5 Southern Africa domain with nowcasts initiated, and the conditional climatologies calculated, from an updated version of convective cores calculated on the MSG grid. This preliminary product is available for the month of February (using a conditional climatology of February 2004–2019). Note that this development version is unverified, and uses a fixed spatial scale of around 75km for all nowcast lead-times.

B indicates the steps used in nowcast probability of convection calculation.

  • 1 for step 1 only,
  • 2 for steps 1 and 2,
  • 0 Versions used for testing are given a value of B equals 0. When used these nowcasts are temporary and will be re-created using the appropriate version at the earliest possibility.

C gives further details of the type of Land Surface Temperature information used

  • 0 not relevant as only step 1 is used,
  • 1 LST information using a prototype LST perturbation field,
  • 2 Land modification factors as detailed above.

Historically, the following versions have been used.

  • 01 June 2020 to 30 September 2020 — v1_1_0.
    Method fully described in Anderson et al (2023) under the key papers.
  • 01 June 2021 to 30 September 2021 — v2_2_1
  • 01 June 2022 to 30 September 2022 — v2_2_1
  • 01 June 2023 to 30 September 2023 — v3_2_1 (NFLICS domain) and 3_1_0 (outside of NFLICS domain)
  • 25 January 2024 to 28 February 2024 — v5_0_0

Nowcast Dakar flood risk

A nowcast of Flood Risk over Dakar is also provided for 22 communes across Dakar. This uses the Flood Risk Matrix approach that combines the potential impact (here the population at risk), and the likelihood of flooding, to give an overall flood risk classification (very low, low, medium, high).

The estimates of the potential impact are based on the population impacted by the 2009 Dakar floods but taking more recent interventions into account.

The likelihood of flooding occurring is formed from:

  1. the chance of convective activity occurring (as calculated from the probability of convection nowcast)
  2. the chance of flood producing rain occurring given convective activity occurs (based on historical relationships between convective cores and 24h raingauge data).

This can be calculated for different starting surface conditions:

  1. estimated surface (based on recent number of cores)
  2. wet surface conditions (40mm of water)
  3. dry surface conditions (0mm of water).

Key papers

The impact of soil moisture on the initiation and propagation of convective storms in West Africa is documented in the following papers.

Taylor, C M, Gounou, A, Guichard, F, Harris, P P, Ellis, R J , Couvreux, F, and De Kauwe, M. Frequency of Sahelian Storm Initiation Enhanced over Mesoscale Soil-Moisture Patterns. Nature Geosci 4, no. 7 (2011): 430–33.

Klein, C, and Taylor, C M. Dry Soils Can Intensify Mesoscale Convective Systems. Proceedings of the National Academy of Sciences 117, no. 35 (2020): 21132–37.

The value of land surface data for nowcasting in the Sahel has been demonstrated in the following paper.

Taylor, C M, Klein, C, Dione, C, Parker, D J, Marsham J, Diop, C A, Fletcher, J, et al. Nowcasting Tracks of Severe Convective Storms in West Africa from Observations of Land Surface State. Environmental Research Letters 17, no. 3 (2022/02/23 2022): 034016.

The use of cloud-top temperature data to identify convective cores is described in the following paper.

Klein, C, Belušić, D, and Taylor, C M. Wavelet Scale Analysis of Mesoscale Convective Systems for Detecting Deep Convection from Infrared Imagery. Journal of Geophysical Research: Atmospheres 123, no. 6 (2018): 3035–50.

The first version of the nowcasting method is described in the following paper.

Anderson, S R, Cole, S J, Klein, C, Taylor, C M, Diop, C A, and Kamara, M. Nowcasting convective activity for the Sahel: A simple probabilistic approach using real-time and historical satellite data on cloud-top temperature. Quarterly Journal of the Royal Meteorological Society (2023).

Funding acknowledgements

The following funding sources and grants are acknowledged as contributing to the development of the nowcasting products, the portal and engagement with national hydro-meteorological agencies and other users in Africa.

African SWIFT programme (2017–22) Global Challenges Research Fund (GCRF), Grant Number NE/P021077/1.

Nowcasting Flood Impacts of Convective storms in the Sahel (NFLICS, 2018–22) under the Science for Humanitarian Emergencies & Resilience (SHEAR) programme (Foreign and Commonwealth Development Office (FCDO/Natural Environment Research Council), Grant Number NE/S006087/1 and NE/S006087/2.

Land, Air, Water International Science (LAWIS, 2021–22), Natural Environment Research Council National Capability funding.

GCRF and Newton Consolidation Accounts (2022–23), Engineering and Physical Sciences Research Council, EP/X52797X/1.

International Science for Net Zero Plus (2022–26), Natural Environment Research Council National Capability International Programme, NE/X006247/1.

Weather and Climate Information Services - Early Warnings for Southern Africa (WISER-EWSA, 2023–2025), Foreign and Commonwealth Development Office.