Keywords

1 The History of TAMSAT

The African continent is a region where rainfall information is of utmost importance. This is especially the case in semi-arid regions such as the Sahel and southern Africa where the majority of livelihoods are highly sensitive to even small deviations in the expected rainfall patterns, resulting in adverse socioeconomic problems that may persist for many years (e.g., Boyd et al. 2013; Funk et al. 2013; Shiferaw et al. 2014).

With the exception of a few countries, the real-time African rain gauge network is extremely sparse compared to other inhabited continents (see Fig. 22.1), making it difficult to monitor real-time changes in rainfall using rain gauges alone. Several countries also have poor historical rain-gauge records, making long-term assessment of changes in rainfall challenging. As such, the use of satellite-derived rainfall information over areas where there are few or no rain gauges is of great value.

Fig. 22.1
figure 1

The WMO World Weather Watch global distribution of the Regional Basis Synoptic Network weather stations providing SYNOP (surface synoptic observations) reports during October 2002. (WMO 2003). The colour denotes each station’s reporting rate

Full size image

The “Tropical Applications of Meteorology using SATellite and ground-based observations” (TAMSAT) Group, was created in the late 1970s with the introduction of the European Organisation for the Exploitation of Meteorological Satellites (EUMETSAT)‘s flagship Meteosat programme. The group, based within the Department of Meteorology at the University of Reading (United Kingdom) – a world-leading centre for weather and climate research – have pioneered the use of thermal infra-red (TIR) satellite imagery in rainfall estimation over Africa since the 1980s. Since this time, TAMSAT have been producing rainfall estimates operationally for Africa, based on the use of satellite-derived cold cloud duration (CCD), which has its roots in the work of Arkin and Meisner (1987) and the subsequent GOES Precipitation Index (GPI) rainfall product. The TAMSAT method was originally devised in response to the need for rapid in-season detection of below-average rainfall and potential losses in crop yield across the Sahel, but has since been extended to all Africa and is now used in a range of applications. Although the TAMSAT approach for estimating rainfall was significantly revised in 2017, the essence of the estimation method has remained unchanged in over 35 years.

The changes to the TAMSAT estimation approach in 2017 (used to generate TAMSAT v3.0), aimed at addressing shortcomings in the previous methodology and made possible by the availability of long (30+ years) daily rain gauge records for many stations, have resulted in vast improvements to the rainfall estimates. In particular, it has eliminated spatial discontinuities in the rainfall estimates that arose from boundaries between large climatologically similar “calibration zones” that were used in previous versions, and has reduced a large dry bias that was characteristic in previous estimates (Maidment et al. 2017; Dinku et al. 2018).

There is currently a variety of satellite-based rainfall datasets that provide coverage for Africa (see Table 22.1). Each has varying characteristics (including data inputs, temporal and spatial coverage/resolution, and latency) leading to different levels of skill for different applications. There is therefore, no one perfect dataset for all end-user applications. For example, some products that incorporate multiple sensor inputs might be more suited to event-based estimation (see the ‘Primary Function’ column in Table 22.1), but the incorporation of multiple sensors leads to relatively short records which limits their use for rainfall monitoring. The TAMSAT product falls in a different classification of multi-decadal products which also provide near-real time updates, making them useful for seasonal monitoring. Today, the TAMSAT Group are very active in ensuring the evolution of TAMSAT products using new technologies and data, and continue to promote TAMSAT products and related-research by regular attendance at the biennial EUMETSAT Africa User Forum and other relevant meetings.

Table 22.1 Summary of widely used satellite rainfall datasets providing coverage for Africa
Full size table

2 TAMSAT Products

TAMSAT produces Africa-wide rainfall estimates at the daily, pentadal (5-day), dekadal (10-day), monthly and seasonal time-steps from January 1983 to the present on a regular 0.0375° × 0.0375° (~ 4 km) grid. The primary product is the pentadal rainfall total (pentads are defined as 1st–5th, 6th–10th, 11th–15th, 16th–20th, 21st–25th and 26th-last day of the month). Daily rainfall estimates are derived by disaggregating the pentadal rainfall total to daily rainfall totals by using the daily CCD fractions (described in Maidment et al. 2017), while the dekadal, monthly and seasonal estimates are derived by summing the respective pentadal estimates.

TAMSAT also produce rainfall anomalies for each time-step (except daily) against a 30-year climatology (1983–2012) and quick-look images for all rainfall estimates. The data files are released within 2 days after each pentad and available on the TAMSAT website (http://www.tamsat.org.uk) in netCDF format. The calibration (described in the next section) and estimation process is summarised in Fig. 22.2, with an example of a monthly estimate and anomaly given in Fig. 22.3.

Fig. 22.2
figure 2

Schematic summarising the TAMSAT calibration and rainfall estimation process. Squares denote inputs or outputs and ovals denote processes. (Adapted from Maidment et al. 2017)

Full size image
Fig. 22.3
figure 3

TAMSAT v3.0 rainfall total (left) and rainfall anomaly (right) for August 2017

Full size image

3 The TAMSAT Rainfall Estimation Approach

3.1 Overview

TAMSAT rainfall estimates are derived primarily from two inputs; Meteosat TIR imagery from EUMETSAT, and in situ rainfall measurements from rain gauges, used for calibration. The premise of the TAMSAT rainfall estimation methodology is based on the detection of precipitating storm clouds, characterised by their cold cloud tops, using TIR. Such clouds are readily identified from the warmer land surface or shallow non-precipitating clouds below (this is illustrated in Fig. 22.4).

Fig. 22.4
figure 4

Full disc (background) Meteosat-8 thermal-infrared image captured at 1800 UTC on 12 August 2005. The grey scale assigned signifies warmer (colder) surfaces as the darker (whiter) regions. A subset (foreground) of this image projected onto a longitude-latitude grid over West Africa giving the brightness temperature (scale in K) of the scene observed

Full size image

The assumptions behind the TAMSAT method are:

  1. 1.

    Rainfall predominantly comes from convective cumulonimbus storm clouds.

  2. 2.

    These clouds will only start precipitating once their tops have reached a certain height. As temperature decreases with height throughout the atmosphere, the height of the cloud top can be inferred by the temperature from the TIR imagery. This is the threshold temperature (Tt), and can be defined as the cloud top temperature that best distinguishes between rain and no rain (illustrated in Fig. 22.5).

  3. 3.

    The amount of rainfall can be determined by the length of time the cloud top temperature is colder than Tt. This is the cold cloud duration or CCD.

  4. 4.

    Rainfall is linearly related to CCD, provided sufficient averaging of the data in space and/or time.

Fig. 22.5
figure 5

Schematic illustrating the assumptions in the TAMSAT algorithm. Clouds with tops colder than the threshold temperature (Tt) are assumed to be raining, while clouds with tops warmer than Tt are assumed not to be raining (schematic by David Grimes)

Full size image

The TAMSAT method can thus be categorised as a cloud indexing approach, in that TIR imagery is used to identify areas of cold cloud, and in turn, maps of CCD. To convert CCD to rainfall, we apply the following equations:

$$ {\displaystyle \begin{array}{c} If\ \left({CCD}_T=0\right),R=0\\ {} If\left({CCD}_T>0\right),R={a}_0+{a}_1{CCD}_T\end{array}} $$
(22.1)

where R is rainfall, CCDT is the cold cloud duration at a given threshold temperature, and a0, a1 are the offset and slope of the linear regression between R and CCDT.

The regression coefficients (a0 and a1) and the threshold temperature (Tt) at which the CCDT is calculated are derived during the calibration stage. Where CCDT is equal to zero, it is assumed rainfall is also zero.

The long-standing ethos of the TAMSAT system has been to provide a fully automated, and pre-calibrated algorithm. As such, the TAMSAT calibration compares gauges and CCD to derive the threshold temperature and regression coefficients, in the form of Africa-wide calibration maps, which are applied in near-real time to estimate rainfall from satellite imagery. This approach overcomes the need for contemporaneous rain gauge records, simplifying the operational implementation of the method.

3.2 Calibration Method

The calibration process, which is carried out periodically when additional rain gauge records become available, can be summarised as follows:

  • For all available daily gauge records in the TAMSAT archive, pixel-scale daily CCD values at multiple threshold temperatures (213 K/−60 °C to 243 K/−30 °C at 10 K intervals) are extracted. Gauge-CCD pairs for CCD values at 1 K intervals between 213 and 243 K are then derived by interpolating between the existing CCD thresholds.

  • The daily gauge-CCD pairs are then spatially binned into 1° × 1° grid squares across Africa to determine the Tt for each grid. A 1° × 1° grid is deemed large enough to contain sufficient gauge-CCD pairs to derive stable calibration parameters, yet small enough to capture local variations in the rainfall climate. For each threshold temperature, a contingency table is constructed (see Table 22.2) which compares the occurrence of rainfall and CCD for all gauge-CCD pairs in each grid box (by considering the hits, misses, false alarms and correct negatives). The Tt is the threshold temperature which maximises agreement and minimises disagreement between rainfall and CCD. In practice, Tt is therefore the threshold temperature whose occurrence of misses and false alarms are roughly balanced, thus minimising any underestimation or overestimation of rainfall occurrence. Given the lack of gauge observations in places, this is only done for boxes where at least 100 gauge-CCD pairs exist. The derived Tt values are then spatially interpolated to produce Africa-wide maps for each calendar month (see Fig. 22.6).

  • The derivation of a0 and a1 is based on 5-day sums of the daily gauge-CCD pairs. For each 1° × 1° grid box, the gauge-CCD pairs are binned (by CCD value) and regressed by comparing the mean rainfall against the mean CCD for each bin, to obtain a0 and a1 respectively. This is repeated for each CCD threshold value across all grid boxes and months. All regression coefficients are then modelled as a function of Tt to produce maps of a0 and a1 based on the Tt map for the respective month. Finally, these maps are then adjusted by the Climate Hazards Group’s Precipitation Climatology (CHPclim) pentadal rainfall climatology (described in Funk et al. 2015a, b) to account for geographical rainfall artefacts (e.g., orographic enhancement) that the CCD and available rain gauge measurements alone do not capture. This results in a set of 72 maps (one for each pentad) of a0 and a1.

Table 22.2 Contingency table for determining Tt. The occurrence threshold is set at zero for both rainfall (mm) and CCD (hours). All gauge-CCD pairs are split into one of four possible groups (n11:n22) with the counts for each group recorded
Full size table
Fig. 22.6
figure 6

TAMSAT v3.0 threshold temperature (Tt) maps for January (left) and August (right) respectively

Full size image

A novel feature of the TAMSAT approach that differs from other existing TIR-based algorithms is that the parameters Tt, a0 and a1 vary both spatially and temporally to account for the strong geographical and seasonal changes in the rainfall climate across the African continent. While some other TIR-based approaches use fixed calibration parameters (e.g., GPI), several studies have indicated that a single rainfall-CCD relationship is inadequate over Africa, especially for products that depend entirely on satellite imagery (e.g., Dugdale et al. 1991; Todd et al. 1999, 1995; Dybkjær 2003; Chadwick et al. 2010). This is evident in Fig. 22.6 for example, which shows the Tt map used in the TAMSAT v3.0 calibration for January and August respectively. Across southern West Africa for example, the Tt varies by around 30 K between the dry season (January) and peak of the West African Monsoon (August), and also by a similar magnitude between the Guinea Coast and the Sahel (between 10°W and 5°E) during August. These drastic changes over southern West Africa during the rainy season are a result of a large influx of low-level moisture from the Gulf of Guinea and enhanced mid-tropospheric subsidence towards the coast (Omotosho 1988) which promotes the development of considerably shallower rainfall systems (Schumacher and Houze 2003; Liu and Zipser 2009; Mülmenstädt et al. 2015; Young et al. 2018), and necessitates the need for a warmer threshold. Conversely, during the dry season, rainfall typically only occurs when convective systems are very deep, reflected by a colder threshold. This is also the case as one moves northward towards the Sahel where moisture availability becomes more limited, and only the deepest of storm systems precipitate.

3.3 Strengths and Limitations

Many studies that have evaluated TAMSAT rainfall estimates using rain gauges, in conjunction with other satellite products, have generally demonstrated that TAMSAT data performs well and has comparable, or better skill than other datasets (Laurent et al. 1998; Thorne et al. 2001; Tucker and Sear 2001; Dinku et al. 2007, 2018; Chadwick et al. 2010; Jobard et al. 2011; Maidment et al. 2013, 2017; Ayehu et al. 2018; Young et al. 2014), although performance does vary depending on the time-step and spatial aggregation of the data, as well as region and season in question. Of the studies evaluating TAMSAT v3.0 against v2.0, the results indicate that v3.0 is a major improvement over v2.0 and compares favourably with other satellite products, particularly in regards to reducing bias (Maidment et al. 2017; Dinku et al. 2018). The skill of the TAMSAT data is notable given that no rain gauge measurements are directly included in the TAMSAT estimates, indicating that a locally calibrated, TIR-based approach can provide reliable rainfall information.

The success of the TAMSAT method can be attributed to several reasons:

  • Where rainfall is primarily convective in nature, CCD can act as a useful proxy for rainfall. As the duration of cold cloud increases with the number of rain storms that passes over a given point, the association between rainfall and CCD increases. As such, the TAMSAT method is very well-suited for tropical regions where precipitation is predominantly convective in origin, such as the flat, semi-arid inland regions of Africa such as the Sahel and central southern Africa.

  • As TAMSAT is locally calibrated, it is tuned to the expected CCD-rainfall relationship for a given region.

  • As the Meteosat satellites follow a geostationary orbit, the on-board sensor is able to provide full spatial coverage of the African continent (see Fig. 22.4) and at frequent intervals (every 30-min for Meteosat First Generation of satellites and every 15-min since the introduction of the Meteosat Second Generation in 2002). Such coverage and sampling enable the CCD product to adequately capture the rapid evolution of such systems anywhere across the continent.

Despite the skill, there are some notable shortcomings to the TAMSAT method that users need to consider:

  • Persistent cirrus, which is cold enough to be mistaken as precipitating storm clouds, can lead to an overestimation of rainfall. In practice though, this is typically not a significant problem across most months and locations.

  • Extreme rainfall events are often underestimated as there is not enough information in the CCD product to capture intense rain rates over short periods. In practice, this is only a concern for daily rainfall amounts.

  • The influence of large bodies of water (i.e. around lakes and near the coast) and presence of complex topography (e.g., Ethiopian highlands) can create microclimates that modulate the CCD-rainfall relationship.

  • Long-lived shallow precipitating systems (e.g., stratiform rainfall) are non-distinguishable from non-raining cloud, leading the TAMSAT algorithm to have considerably lower skill during such events.

4 Usage and Applications

The TAMSAT Group have built up strong relationships with many African National Meteorological and Hydrological Services (NMHS) and other stakeholders, across a range of sectors, and have established themselves as a dependable source of reliable rainfall information. TAMSAT have worked closely with many African agencies – by developing TAMSAT products during TAMSAT-led workshops, and by providing capacity building through training Met Service staff to operate TAMSAT rainfall estimation software and applying TAMSAT rainfall estimates appropriately. In recent years, the bond between TAMSAT and African NMHS has strengthened further by a WMO-led programme supporting forecasters to study at the University of Reading and work with the TAMSAT Group. The value of such relationships is paramount not only to the effective use of TAMSAT data, but also to the development and evolution of TAMSAT rainfall estimates and related products, through improved access to climate records and trialling newly developed tools.

TAMSAT data plays a key role in supporting climate services in many African countries, for example, through the various agencies that access the data through EUMETSAT’s EUMETCast distribution service, as well as the Enhancing National Climate Services (ENACTS) (Dinku et al. 2016) and Rainwatch (Tarhule et al. 2009) programmes, that provide value-added services using TAMSAT data. The data is also well used by the research community, for example, by furthering our understanding of rainfall variability and trends across Africa (e.g., Maidment et al. 2015; Dunning et al. 2016; Cattani et al. 2018).

In recent years, the focus of TAMSAT has broadened to encompass a range of stakeholders in the commercial, humanitarian, agricultural and financial sectors. The temporal consistency and longevity of the TAMSAT record makes it a valuable dataset for risk assessment. In particular, as TAMSAT’s primary rainfall product does not incorporate contemporaneous rain gauge measurements, its skill is unaffected by changes in gauge coverage that can introduce time-dependent biases into the rainfall record (Maidment et al. 2015). As an example of TAMSAT’s adoption for risk management, TAMSAT rainfall is now widely used in weather index insurance – a form of insurance that pays out in the event of a weather index being breached (Black et al. 2016a, b). A temporally consistent record is especially important in the insurance industry where premium rates are often calculated using a historical burn analysis of long time-series. In 2017–2018, TAMSAT rainfall estimation enabled over 1 Million farmers to be insured using weather index insurance.

The myriad of publicly available Earth Observation and reanalysis products provide many opportunities to improve drought risk assessment. This raises the question of how users can assess risk in the face of multiple sources of information. The development of the TAMSAT-ALERT system (Brown et al. 2017; Asfaw et al. 2018) meets this need by deriving objective agricultural risk assessments based on seasonal weather, the historical climatology, the condition of the land surface and meteorological forecasts. Over the next years, TAMSAT will work with the humanitarian and agricultural sectors, as well as NMHS to implement TAMSAT-ALERT across Africa.