Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

• A Seasonal climate prediction is a probabilistic statement on the future state of the atmosphere over a season (i.e. 1, 2, 3 months or more).

​

• It is different from weather forecast, which provides a deterministic statement of the atmosphere over a day or two weeks, and not the same as climate changes projection, which provides probabilistic statement on climate condition over a long time (ranging from 20 to 40 years or more).

​

• Seasonal climate predictions usually offer the future seasonal climate information as departures from the climatic mean i.e. anomalies.

​

• Rainfall and temperature are the main climate variables communicated in seasonal climate forecasts.

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Importance of Seasonal Climate Prediction

• Timely and reliable prediction of seasonal climate can help reduce the damages caused by extreme climate events.

​

• The information from seasonal climate forecasts are needed for planning and risk management in various socio-economic sectors such as
• agriculture,
• health,
• water resources management,
• environmental management,
• engineering, e.t.c.

​

• Its usefulness also extends to insurance sectors where it assists in the operational task of preparing for major pay-outs.

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Figure 8.1: Scientists in Senegal working with local farming groups to communicate seasonal forecasts.

Photo © Dorian Speakman

​

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Figure 8.2: West Africa Climate Zones

The Zones are climatically homogenous during the wet season

(July – September (JAS))

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

In West Africa……….

• Seasonal predictability tends to be evident on larger scales

​

• Seasonal climate anomalies had a greater impact than everywhere else in the world

​

• The vast majority of people depend on rain‐fed agriculture for their livelihood and lives.

​

Therefore, the high variability of seasonal climate (e.g. rainfall) from year to year underscores the importance of a timely and reliable prediction of seasonal climate.

​

For example, a drier than average wet season in 1984 worsened a pre‐existing drought and led to a famine.

​

So what is the cause of events like the 1984 drought?

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

There is some predictability of the statistics of weather and climate on the seasonal timescale. In order to determine climate phenomena that might influence and or cause seasonal anomalies, like the 1984 drought observed in West Africa, we can specify some criteria that these phenomena will need to satisfy:

​

1. Any phenomenon that is considered to be associated with or perhaps a cause of droughts or other seasonal timescale anomalies would need to be active on similar timescales.

​

• The phenomenon should involve the transfer of sufficient energy to physically cause or respond to atmospheric changes related to droughts.

​

• The phenomenon should be located near enough to the affected anomaly in order to have a credible physical impact.

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

A phenomenon that fits all the above mentioned criteria is the pattern of large‐scale anomalies in sea surface temperature (SST).

​

• SST anomalies tend to last for months, unlike most atmospheric phenomena, which typically do not last for more than a few days.

​

• SST anomalies involve transfer of very large amounts of energy from the ocean to atmosphere and vice versa; this is easily on the same scale as energy changes associated with drought season or wet seasons in West Africa.

​

• Some of the most energetic SST anomalies are located in the tropics, not too far from West Africa.

​

• SSTs are known to modulate climate around the whole globe, particularly in the tropics through the El Nino Southern Oscillation (ENSO) phenomenon.

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Figure 8.3: Schematic illustration of SST teleconnections that often cause drier than average July-to-September rainfall totals over the Sahel and 'Guinea Coast'. Red ocean areas show where warmer than average SSTs are linked to drier than average seasons, and blue ocean areas show where cooler than average SSTs are linked to drier seasons. Note that the striped Atlantic region shows an overlap of areas that affect either the Sahel or Guinea Coast. Source: Contributing author, Rowell, DP

In terms of its variety of teleconnections from SSTs, the Sahel represents one of the most complex regions of the globe, being influenced by all three tropical oceans, and additionally by the Mediterranean.

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Nevertheless, SST is not the only variable that varies on seasonal timescales in a way that could explain droughts.

​

• Sea ice cover and snow cover over land also vary on seasonal timescales and involve considerable amounts of energy. However, these phenomena are located in the extra-tropics, a long way from West Africa.

​

• The role of land surface changes is arguably weaker, as there is less moisture involved and consequently less energy.

​

• Soil moisture levels can evolve on a seasonal timescale and affect vegetation and albedo. Soil moisture deficits do prolong or enhance droughts through a positive feedback of increasing albedo and consequent reduction in evaporation.

​

• Fontaine et al. (1999) argued that adding information about pre‐season moisture content and temperature in the West African region could improve on predictions made using pre‐season SST alone.

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Figure 8.4: Correlations of observed rainfall averaged over (a) the Sahel or (b) the Guinea Coast, against observed SSTs on a 2.5° x 3.75° grid. Data are July – September averages for the period 1922 to 1994.

Source: Contributing author, Rowell, DP.

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Inferences over Sahel show that…..

​

• Warm SSTs in the equatorial and South Atlantic lead to a decrease in the surface temperature gradient to the Sahel, weakening the West African monsoon flow and so weakening the surface moisture convergence over the Sahel.

​

• Warm SST anomalies over the Mediterranean often lead to an enhanced evaporation and the additional moisture is advected across the Sahara by the mean flow, where it enhances rainfall over the Sahel. This effect is then amplified by a variety of positive‐feedback mechanisms.

​

Besides the above mention regional SST influences, there are also remote influences on the Sahel from the Indian and Pacific Oceans through teleconnection mechanisms:

• a stationary Kelvin wave provides an eastward link from the East Pacific to the Sahel, and
• a Rossby wave provides a westward link from the West Pacific–Indian Ocean SST gradient (see more details in Section 7.1.4).

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Inferences over Guinea Coast show that………..

​

• SST fluctuations have little or no impact on this region before July, and after September

​

• Therefore, rainfall anomalies are primarily governed by chaotic variability.

​

Although these months partly coincide with a break in the rains over the Guinea Coast, a likely mechanism for this relationship is that

​

• a Rosby wave response to warmer‐than‐average SSTs produces anomalous ascent along the Guinean Coast, so increasing rainfall here.
• Conversely, cool SSTs weaken the monsoon flow, leading to below‐normal rainfall.

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

So, has an SST‐to‐rainfall teleconnection been stationary (i.e. the sign and magnitude of relationship has been maintained for many decades), and will remain stationary into the future?

​

After so many investigations over a number of African regions, including the Sahel and Guinea Coast, the answer is NO!

​

The results appear to be that the fluctuations in the strength of SST–rainfall teleconnections are in fact more likely to have arisen from sampling effects rather than from genuine multidecadal variations of the climate system.

​

That is, the predictive relationship between SST and rainfall has probably been more or less stationary.

​

What’s next……………………….?

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Types of Forecasts:

​

Statistical Forecasts

• Involve studying past historical data in order to deduce a forecast for the future. Examples are, Analogues, Linear Regression, Canonical Correlation Analysis (CCA), Principal Component Regression, Discriminant Analysis, and Non-stationary Methods.

​

​

Dynamical Forecasts

• The use of models to simulate the atmosphere by replicating the physics (see Chapter 10 for more details).

​

​

Combined Forecasts

• A situation where forecasts of different types are combined in order to minimize the errors.

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Analogues:

​

To deduce a forecast, the forecaster scans the historical record for cases when similar precursor events occurred and assesses what happens to the predicted event in these ‘analogue’ cases.

​

The forecaster then assumes the same will happen in the ‘current’ case, the event one is trying to predict.

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Linear regression:

​

Has the advantage that use is made of all the historical data available rather than a few analogue years.

​

Regression is used to predict the value of a single ‘dependent’ variable from one or more independent variables.

​

Linear regression works by constructing a mathematical model of the form

Y a1 = X1 + a2X2 + ……………..+ anXn + C

​

This equation is also referred to as a regression model.

​

The regression model is constructed using historical data, also referred to as training data.

​

In order to make a forecast, the appropriate recent predictor values are substituted into the equation.

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Canonical Correlation Analysis (CCA):

​

• In CCA, the variables are ‘fields’ containing many values.

​

• CCA identifies pairs of patterns in a predicted and a predictor field that are highly correlated in time.

​

• Calculations of empirical orthogonal functions (EOFs) of both the predictor and predicted fields are done.

​

• The logic is that if the predictor field resembles a CCA field, then the corresponding predicted field will resemble its partner field.

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Figure 8.5: Example pair of CCA patterns and temporal scores representing the association of West African rainfall and pre-season SST.

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Principal Component Regression (PCR):

​

PCR is a variant of linear regression where the independent variables are principal components (PCs).

​

It is a ‘halfway house’ between linear regression and CCA where the predictors are ‘fields’ represented by PCs or EOFs, and predictands are a single ‘dependent’ variable.

​

Here, EOFs are used to reduce dimensionality in the predictor field, reducing many independent variables to a more manageable number of representative EOFs and associated PCs.

​

PCR has the added advantage that PC predictors are by definition ‘orthogonal’, removing the problem of collinearity in predictors that can adversely affect linear regression.

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Discriminant Analysis:

​

The product of discriminant analysis is not a point value as in regression, nor a field of values, as with CCA, but, rather, probabilities of two or more events.

​

Probabilities are predicted for a fixed number of events that are mutually exclusive and cover the full range of possibility. An example could be categorising a season’s rainfall into three equiprobable tercile categories.

​

Discriminant analysis determines the ‘true’ probability of an event, given a forecast of that event, by reference to the retrospective performance of the forecast system.

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Figure 8.6a: Twenty-one year centred moving correlation windows (green line, right y-axis) between the expansion coefficients U corresponding to predictor field (blue bars, left y-axis) and V corresponding to predictand field (red line, left y-axis) obtained from the leading mode of co-variability from MCA analysis between both anomalous fields considering monthly lag 3 (AMJ) for predictor. Coloured triangles indicate significant correlation under a Monte-Carlo test at 90 %.

Non-stationarity Methods:

Highlights the need for study of the stationarity of the relationship between the predictor and predictand fields.

​

Non‐stationary models, such as S4CAST, help to seek and understand the physical mechanisms behind the modulations on different timescales (e.g. interannual, multidecadal).

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Figure 8.7 Maps corresponding to the leading mode of co‐variability and monthly lag 3 (April–June) for the predictor. (a) Panels refer to the whole period; (b) panels refer to the SC period; (c) panels refer to no‐significant correlation period. Left panels show the leading mode for the predictor field (squared region) obtained from projecting the expansion coefficient U into the global domain (°C per standard deviation). Central panels show the leading mode for the predictand field (squared region) obtained from projecting the expansion coefficient U into the local domain (mm day−1 per standard deviation). Right panels show correlation skill‐score obtained by Pearson correlation coefficients between observed and simulated maps showing a maximum correlation value of 0.23 (EP), 0.56 (SC) and no skill (NSC). Coloured regions indicate significant values under a Monte‐Carlo test at 90%. Source: B Rodríguez Fonseca and R Suarez (Universidad Complutense de Madrid, Spain).

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Representation of Predictors (SST EOF Predictors):

​

This approach uses Empirical Orthogonal Functions (EOFs). EOFs represent patterns of variance in the data, and thus indicate how SST anomalies evolve independent of any teleconnection

​

• EOF2 is a representation of the ENSO cycle that has some influence on West African rainfall.

​

• The most important pattern for African rainfall forecasting is EOF3. This shows a contrast in sign between Northern and Southern Hemispheres and is considered to represent the IHC pattern.

​

• The identification of EOF3 was perhaps the most important result of the Met Office research to develop seasonal predictions for West Africa in the 1980s and 1990s.

​

• It is still widely used as the key predictor of West African rainfall by the African Center of Meteorological Application for Development (ACMAD), PREvision Saisonairre en Afrique del l’Ouest (PRESAO) and others.

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Figure 8.8: Three global SST EOF patterns used for forecasting West African rainfall. The final panel is EOF 2 of the “varimax rotated” EOFs. (Crown Copyright Met Office).

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Representation of Predictands:

​

A forecast system requires a variable or set of variables to predict.

​

Station data have traditionally been the fundamental building block of seasonal rainfall records used for this purpose.

​

Averaging data for a group of stations in a climatologically homogeneous zone can be problematic in many ways.

​

Therefore, a number of databases have now been created using satellite data alone or satellite data blended with observations.

​

However, there are still large uncertainties in observed data. As observations are required to produce statistical forecasts, observational uncertainty can adversely affect statistical forecast systems.

​

To illustrate this uncertainty four example maps of observed precipitation anomalies for the same period, June–August (JJA) 2011, are presented below:

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Figure 8.9: Four estimates of rainfall over Africa for June-July-August 2011.

(a) TRMM anomaly for JJA 2011 (relative to 1998-2011 climatology)

The satellite based TRMM (http://trmm.

gsfc.nasa.gov/) plot suggested it was mostly wet except near the coast.

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Figure 8.9: Four estimates of rainfall over Africa for June-July-August 2011.

(b) FEWS ARC2 anomaly for JJA 2011 (relative to 1983-2011 climatology)

CPC‐FEWS ARC2, a blended satellite /

gauge dataset, has a near equal coverage of wet and dry.

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Figure 8.9: Four estimates of rainfall over Africa for June-July-August 2011.

(c) CAMS-OPI June-July-August 2011 anomaly from 1979-2000. Reproduced by permission of International Research Institute for Climate and Society, (iridl.ideo.columbia.edu).

The CAMS‐OPI map

(another satellite outgoing longwave radiation‐based

database) suggests it was mostly dry.

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Figure 8.9: Four estimates of rainfall over Africa for June-July-August 2011.

(d) NCDC precipitation anomalies for JJA 2011 as % 1961-1990 climatology (from NOAA National Centers for Environmental Information, State of the Climate: Global Analysis for August 2011, published online September 2011)

NCDC (http://www.ncdc.

noaa.gov), based on gauge data, does not show a strong

signal either way.

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Representation of Rainfall Using Box Zones:

​

Box zones have been used to define regions of homogeneous climate that can be predicted as one variable. The three regions in Figure 8.2 (slide 5) are an example for JAS rainfall.

​

The homogeneity of the regions shown in Figure 8.2 can be seen by viewing the leading eigenvectors/EOF of gridded rainfall datasets (Figure 8.10; next slide).

​

The first EOF of both datasets shows a bimodal structure with a divide at about 10°N. The area to the north of 10°N coincides quite well with the Sahel and Soudan regions.

​

The second modes both have strong positive weights for a region between about 7.5°W and 7.5°E and south of 10°N, and opposite sign elsewhere. The region with strong weights coincides quite well with the Guinea Coast region. The locations of the regions are determined by the position of the zone of maximum rainfall in JAS.

​

These homogeneous regions are season dependent.

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Figure 8.10: First two EOFs of (a) FEWS ARC2 for JAS (1983-2012), (b) NCEP PREC/L for JAS (1983-2012) (Crown Copyright Met Office).

(a)

(b)

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Assessment of Statistical Methods

Assessing statistical methods can be problematic, as statistical forecast systems represent apparent relationships in the historical data. If one tests a statistical system on the same data that were used to produce it, the result will suggest there is skill.

​

A solution to this is to test a statistical forecast system on independent data; that is, data that were not used to produce the forecasts.

​

Hence, cross‐validation or jack‐knifing technique.

• In its simplest form, if one has N years of data to create a forecasting system, one removes the first year from the analysis and makes a prediction equation using the (N 1) remaining years. That prediction equation is then tested on the year that was excluded (the first year). The process is then repeated in turn for all N years. Thus, N equations are produced, each equation created using a slightly different set of (N 1) years and used to make a prediction for each of the N years.

​

Using cross‐validation windows >1 is recommended when the data are highly auto-correlated (meaning that there is ‘memory’ in the data from year to year), or when skill is low and removal of ‘spike’ years is likely to have an excessive influence on the prediction equation.

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Figure 8.11a): Cross-validated forecasts of Sahel rainfall plotted against observations. Source: (Crown Copyright Met Office).

Cross‐validated assessments of predictions of rainfall using EOF3. The correlation between cross‐validated forecasts of JAS Sahel rainfall using March–April EOF3 and observations is r 0.52. The correlation rises to r 0.61 if we use concurrent (JAS) SST to predict rainfall.

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Figure 8.11b): Cross-validated forecasts of Sahel rainfall plotted against observations. Source: (Crown Copyright Met Office).

Cross‐validated assessments of predictions of rainfall using rotated EOF2. Forecasts from the three predictors are a slight improvement on using EOF2 alone, if concurrent JAS SST is used ( r 0.70 ), but the skill of the three predictors using March–April SST ( r 0.53 ) is similar to using EOF3 alone ( r 0.52 ).

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Performance over 1996–2009

Whilst the EOF‐based predictors show skill over the 111‐year period (Figure 8.11), unfortunately it is a different story when forecasts are assessed over the 14‐year period 1996–2009 (next slide).

​

Skill from March–April is reduced to near or below zero and skill from JAS from all three predictors is 0.25 or lower, well short of the 5% significance level.

​

In particular, the March–April forecasts failed completely to predict the relatively extreme dry and wet years between 1996 and 1999, but a weak upward trend from 2000 to 2009 has been correctly predicted by the March–April forecast.

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Figure 8.12: Sahel forecasts for JAS 1996-2009 verified using five different datasets. Source: (Crown Copyright Met Office).

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Dynamical Forecasts

The use of models to simulate the atmosphere by replicating the physics.

​

Dynamical models have been used for weather forecasting for over half a century, but seasonal dynamical models have only been widely available for about 15 years, since sufficiently powerful computers have become available.

​

Dynamical models are generally run in groups of repeated runs called ensembles.

​

Ensembles are run to allow for uncertainty in initial conditions and sometimes to allow for uncertainty in model formulation.

​

Ensembles can also be used to produce probability forecasts.

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Figure 8.13: Dynamical predictions from the Met Office GloSea4 system for the three regions shown in Figure 8.2. (Crown Copyright Met Office).

Hindcasts for the 1996–2009 seasons are plotted as red crosses against corresponding observations represented by X symbols.

​

The correlation score in the title provides a measure of performance.

​

The Guinea Coast correlation is somewhat lower at 0.282 and disappointing. A possible explanation for this low skill, related to poor model performance in predicting tropical Atlantic SST

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Correction of Systematic Errors in Models

​

Models have limitations due to:

• understanding and representation of some processes in the climate system
• coarse resolution of the climate model
• representation and consideration of small‐scale processes (turbulence, convection),
• lack of knowledge of the initial state of the forecasting system
• truncation error,
• computational limitations (storage or processors), etc.

​

These lead to errors that grow as the model steps forward through time.

​

For some models, systematic errors are obvious and easy to correct.

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Figure 8.14: Example of systematic mean bias error between Model and Observation. Source: Contributing Author, Ndiaye, O (Senegal Met Service)

It is clear that the GCM is able to capture well the rainfall variability but present systematic positive bias on the mean.

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

A simple comparison between model and observation statistics at a specific location allows us to identify the systematic difference between the two, and then to implement a correction.

​

Correction can be applied either in the mean or in the variance, or in both, as in the equation;

​

​

​

​

​

where X is the raw value,

X is the model forecast mean,

Xobs is the corresponding observed mean,

σXobs the observed standard deviation and

σX the standard deviation of the model forecasts.

​

​

​

​

​

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Other biases might not be as obvious as the systematic difference because model and observation are not collocated at the same grid point, when there is a shift or misrepresentation in space.

​

In such cases, decomposition or filtering of the GCM outputs into modes of variability is necessary to ‘see’ the captured signal.

​

Usually, EOF analysis is used. Tippet and Giannini (2006), using signal‐to‐noise analysis, identified the most reproducible GCM patterns of African summer precipitation related to the SST forcing.

​

They used EOF analysis, rather than the ensemble mean, to identify the most predictable components of the GCM and they were able to increase GCM skill over the Sahel.

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Figure 8.15 a) Spatial pattern of EOF1 of the 925 hPa GCM zonal wind used for the MOS and b) Sahel rainfall skill from various AGCMs forced with observed SST, for rainfall (shaded bars) and after applying a MOS (open bars) over 1968-2001 (from Contributing Author Ndiaye et al., 2009; 2011).

Ndiaye et al. (2009, 2011) identified the low‐level wind over the Atlantic basin as a good predictor of Sahelian rainfall. They used the EOF1 (Figure 8.15a), which captures 33% of the total wind variance, to correct poor GCM rainfall skill. They showed that GCMs in general were able to represent wind better than rainfall over this region.

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Figure 8.15 a) Spatial pattern of EOF1 of the 925 hPa GCM zonal wind used for the MOS and b) Sahel rainfall skill from various AGCMs forced with observed SST, for rainfall (shaded bars) and after applying a MOS (open bars) over 1968-2001 (from Contributing Author Ndiaye et al., 2009; 2011).

They were able to recover most of the rainfall variability (Figure 8.15b) through a model output statistics (MOS) approach.

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Need to Improve Simulation of South Atlantic SST

​

The performance of the seasonal forecast for the Guinea Coast region is, usually, substantially worse than that of the forecasts for the Sahel and Soudan zones.

​

One reason for this is that models poorly represent SST in the Atlantic.

​

The Guinea Coast has a much stronger association with the tropical South Atlantic SST than the Soudan and Sahel regions.

​

One of the worst failures of GloSea4 in predicting Guinea Coast rainfall was 2004, in which all the forecast members were too dry. Example of this is illustrated in the next three slides.

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Figure 8.16: Observed SST anomalies in the South Atlantic from (a) April 2004. Source NCEP plots, http://www.emc.ncep.noaa.gov/research/cmb/sst_analysis/#_sstplots/.

One of the clearest features of this SST anomaly map is the colder than average SST in the tropical South Atlantic, particularly in the Gulf of Guinea. Such cold anomalies are associated with drier than average conditions near the Guinea Coast.

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Figure 8.16: Observed SST anomalies in the South Atlantic from (a) April 2004. Source NCEP plots, http://www.emc.ncep.noaa.gov/research/cmb/sst_analysis/#_sstplots/

However, come the height of the wet season in August, the anomalies in the South Atlantic are very different: the colder than average anomalies have all but disappeared and the majority of the tropical South Atlantic is now warm. These anomalies are consistent with the near‐average season that was observed on the Guinea Coast.

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Figure 8.16: (c) Forecast SST anomalies from GloSea4 model for August 2004. Source (c): (Crown Copyright Met Office).

Meanwhile, GloSea4 model predicted colder than average SST anomalies to persist in the Gulf of Guinea

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

An underlying cause of this problem may be in the process of developing dynamical models for seasonal forecasting.

​

A key metric used in development of the models is how well they simulate ENSO events in the Pacific.

​

In contrast, less attention is paid to the Atlantic.

​

However, results like this show a need to correctly model the Atlantic too; a challenge for developers of future models.

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Figure 8.17: Long-lead predictions of West Africa climate zone rainfall anomalies for July-August-September from preceding November using two models. (Crown Copyright Met Office)

A typical example of Predictability at long lead times. This is nearly always likely to be harder than at shorter lead times.

​

SST anomalies evolve slowly enough to make short‐lead seasonal forecasts viable, but over longer periods the SSTs may change significantly, and thus it may be harder to detect a predictive signal.

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Combined Forecasts

Forecast users just want one forecast indicating what the forthcoming season is going to be like.

​

Combining forecasts of different types has the added advantage in that errors can be minimized.

​

The logic behind the use of multiple models is related to the uncertainty in their representation of the ocean and atmosphere.

​

Therefore, it is a good idea to have a collection of models, each with different errors that will ultimately cancel each other out.

​

It is important, however, that all the models used have good skill and pick up seasonal predictability

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Dynamical forecasts can be divided into three components:

​

Forecast = direct signal + indirect + signal noise

​

Direct signal is when the model directly predicts the event; for example, the mean of precipitation output from the model predicts an anomalously wet or dry season e.g. raw ensemble probabilities.

​

Indirect signal is when the model fails to predict the event, but does predict climate phenomena related to the event. For example, the model may fail to predict rainfall in a particular location, but it may predict El Nino events that are known to be related to the rainfall, e.g. CCA-calibrated forecasts.

​

The remaining component is noise. This includes atmospheric variability that is unpredictable due to chaos or due to the model not being good enough.

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

In order to produce a combined forecast one needs to know the ratio of direct signal to indirect signal, in order to determine how much weight to give each forecast.

​

This is not a straightforward question and can only be estimated using a relatively small number of hindcasts.

​

This has been done for West Africa seasonal precipitation forecasts using the relative operating characteristic (ROC) skill measure.

​

ROC scores are discussed on the Met Office web site:

http://www.metoffice.gov.uk/research/areas/seasonal‐to‐decadal/gpcoutlooks/user‐guide/interpret‐roc

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Since the ratio of direct to indirect signal is unknown, the forecast system was tested giving various weights to the CCA‐calibrated forecasts (representing the indirect signal) and the raw ensemble forecasts.

​

Combinations using 0, 50, 70 and 100% weights to the CCA‐calibrated predictions were tested and the ROC skill is presented in Figure on slide 56.

​

Blue in the Figure represents areas where there is no skill. Highest skill is indicated by orange and red.

​

Overall, the 70% weighting has the highest skill on average ( (60.2 56.6)/2 58.4%).

​

Therefore, the 70% weighting is regarded as the best guess, as this combination had the highest average ROC skill overall.

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Figure 8.18: Assessments of combined forecasts from five GPC models giving various weights to the calibrated and uncalibrated components. (Crown Copyright Met Office)

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Figure 8.19: Combined forecasts from five GPC models giving various weights to the calibrated and uncalibrated components. (Crown Copyright Met Office).

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Figure 8.20: (a) Volta Basin region

(Crown Copyright Met Office)

The Volta Basin covers Ghana and extends north covering more than half of Burkina Faso and also parts of Benin, Mali, Togo and Ivory Coast; thus, it bisects the Guinea Coast, Soudan and Sahel zones.

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Figure 8.20: (b) Annual cycle of rainfall and inflow .(Crown Copyright Met Office)

Lake Volta outflow occurs predominantly between June and November and peaks in September each year, following the West African wet season.

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Figure 8.20: (c) Hindcasts for 1996-2009 and forecast for 2012. (Crown Copyright Met Office)

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Prediction of Meningococcal Meningitis (MCM) in the West Africa Dry Season

​

MCM is a contagious, infectious disease caused by the bacteria Neisseria meningitis. MCM epidemics occur worldwide, but the highest incidence is observed in the ‘meningitis belt’ of sub‐Saharan Africa, stretching from Senegal to Ethiopia.

​

The severity of the MCM epidemics varies from year to year, with the number of cases in the region varying from 25 000 to 250 000.

​

Children under 15 are particularly affected. Mortality rates average around 10%, with 10–20% of survivors affected by serious neurological repercussions.

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

MCM epidemics tend to occur during the latter part of the dry season (February–April) and usually end with the onset of rains , and have been associated with wind meridional component.

​

Statistical analyses show that the winter climate can explain 25% of the interannual MGM variance in Niger (albeit with lower skill in Burkina Faso).

​

Overall, the skill of the statistical models is encouraging, and development of similar forecasts for other parts of the Sahel is recommended.

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Applying Seasonal Forecasts to Assist Agriculture in Senegal

​

A pilot project was set up by the Senegal National Meteorological Agency to communicate climate information to farmers in the Kaffrine district of Senegal.

​

The project involved building a bridge of trust between the climate scientists producing the forecasts and the farmers as users;

• this included understanding how the farmers currently use indigenous knowledge to predict climate and weather events related to their needs,
• getting the farmers to share their needs for climate information and to identify what new information could be useful to them,
• and investigating ways of communicating forecast information to the farmers.

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Initially, 33 farmers were involved in the project, but it has been recently expanded to involve up to 1000 farmers.

​

Farmers were divided into groups, some gender specific, to say what climate information they needed and answer the question ‘If you knew the future climate, what would you do differently?’

• Foreknowledge of the rainy season onset date was considered important as it guided what and when to plant.
• There was also interest in the length of the wet season and the onset data for next year’s season.

​

Ways of communicating were also discussed.

• Use of SMS texting
• radio broadcasts, but it was pointed out that electricity supplies are often unreliable.
• Personal contact was considered better.
• Women identified social gatherings (e.g. naming ceremony) as a good medium, as people would be in attendance for other (cultural or social) reasons.

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

International Seasonal Forecast web sites:

Sources of Information

Figure 8.21: The 12 Lead Centres. Source: www.wmolc.org/

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

World Meteorological Organization Lead Centre Multimodel Ensemble web site

https://www.wmolc.org/

This web site was set up as a single location where seasonal forecasts from all reliable sources could be viewed together in a common format and, therefore, compared and combined. The seasonal forecasts presented here are all produced by World Meteorological Organization (WMO)‐approved lead centres.

​

Forecasts are presented of means of precipitation, surface temperature, surface pressure and 500 mb height for the next three calendar months. Forecasts are updated around the 20th of the month.

​

A UserID and password are required to view the forecasts, which can be obtained by accessing the site and clicking on ‘sign up’.

​

Long‐range Forecast Verifications

http://www.bom.gov.au/wmo/lrfvs/

This site contains skill assessments of the lead centre models using the ROC and root‐mean‐square error skill measures.

​

​

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

UK Met Office web site

Met Office forecasts produced using the GloSea system of 3‐month mean precipitation, surface temperature, surface pressure and 500 mb height out to 6 months are available here for free public access.

​

ECMWF/EUROSIP web site

http://www.ecmwf.int/en/forecasts/documentation-andsupport/long-range/seasonal-forecast-documentation/eurosip-user-guide/multi-model

This site contains forecast output from the EUROSIP models (Met Office GloSea4, Meteo‐France ARPEGE, NCEP Climate Forecast System (CFS) as well as ECMWF).

​

Included are forecasts of surface temperature, precipitation, 500 mb height and 850 mb temperature for 3‐month periods out to 6 months ahead. Forecasts for the tropics (30°N–30°S including West Africa) are publicly available.

​

​

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

NOAA/NCEP web site

http://www.cpc.ncep.noaa.gov/products/people/wwang/cfsv2fcst

Forecasts of 1 and 3‐month averages of surface temperature, precipitation, 200 mb height and 700 mb height from the CFS model out to 9 months ahead are publicly available here in the form of global maps.

​

International Research Institute for Climate and Society Seasonal Forecast web site

http://iri.columbia.edu/climate/forecast/net_asmt/

This site provides forecasts of 3‐month mean temperature and precipitation up to 6 months ahead, which are presented as maps for regions, including Africa and for the globe.

​

The IRI products are based on a combination of dynamical and statistical models and differ from those discussed earlier in that human forecaster intervention is involved in producing the forecasts.

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

African Center of Meteorological Application for Development web site

http://www.acmad.net/

ACMAD is in a demonstration phase, preparing to become a WMO designated Regional Climate Centre for Africa with a long‐range forecasting mandate for West Africa.

​

​

​

​

​

​

These lead centres are also known as GPCs.

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Consensus Forecasts

Having described several sources of seasonal forecast information for West Africa. How should one use this information?

​

After all, most users require just one forecast.

​

The RCOF process was set up in the late 1990s to help resolve this problem.

​

RCOFs are events where forecasters can get together to produce a consensus

forecast and disseminate it to users.

​

The consensus forecast is produced by a conference of experts from

(1) West African National Met Services,

(2) ACMAD,

(3) overseas institutes such as the UK Met Office and Meteo‐France.

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

To produce the consensus forecast the following information is used:

​

1) Calibrated (systematic errors corrected) and downscaled GPC model dynamical forecasts for climatically homogeneous regions in individual countries.

​

2) Forecasts for these same zones using pre‐season observed SST produced using PCR and simple multiple linear regression.

​

3) SSTs and precipitation forecast output from the 12 GPC models as posted on the lead centre website (www.wmolc.org(www.wmolc.org) and individual GPC websites (e.g. www.ecmwf.int).

​

4) The IRI seasonal forecast from http://iri.columbia.edu/climate/forecast/net_asmt/.

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

5) ACMAD’s Regional Climate Centre Demonstration phase’s consolidated long‐range forecast product for Africa. These forecasts are produced by a subjective analysis of observed sea temperatures and trends. Rainfall predictions are derived from the sea temperatures based on known teleconnections and from investigating ‘analogue’ years with similar observed sea temperature patterns and trends to the current year, particularly in the tropics. The forecasts are crosschecked against the GPC forecasts.

​

6) Expert interpretation of recent SST and subsurface sea temperature anomalies and their relation to teleconnections (including recent values of Global EOF3).

​

Forecast maps like the those on the next two slides is drawn by hand based on the aforementioned information.

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Figure 8.22: Consensus forecast for 2012 season issued by PRESAO15. Figure supplied by Contributing Author, Andre Kamga, ACMAD (http://www.acmad.net/new/)

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Figure 8.23: JAS 2012 forecast for Burkina Faso consistent with neighbouring countries and the regional consensus of May 2012. Figure supplied by Contributing Author, Andre Kamga, ACMAD (http://www.acmad.net/new/)

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Software and Tools

The following tools are available to develop forecasting system

​

The Climate Predictability Tool (CPT)

http://iri.columbia.edu/our‐expertise/climate/tools/cpt/

The CPT is a powerful, purpose‐built and easy to use facility for developing seasonal forecast models and forecasts. It is free and easily downloadable from the link given.

​

International Research Institute for Climate and Society Data Library

http://iridl.ldeo.columbia.edu/

A large selection of data is available here that can be used directly in CPT.

​

KNMI Climate Explorer

http://climexp.knmi.nl/start.cgi?id=someone@somewhere

The KNMI Climate Explorer is a web application to analyse climate data statistically.

​

NOAA Earth System Research Laboratory Interactive web site

http://www.esrl.noaa.gov/psd/data/correlation/

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Other web sites Including Forecast Applications

​

FEWS NET

www.fews.net

FEWS NET produce high‐resolution maps showing food security risk based on available seasonal forecasting information and precursor conditions. Maps are usually for the next 3 months.

​

AGRHYMET, Niger

www.agrhymet.ne

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Hints and Recommendations

• Statistical forecasts are becoming obsolete for most seasonal forecast applications in West Africa.
• This is due to the weakness of statistical methods, a consequence of poor quality observations, climate change and changing land use, and to the relative success of dynamical methods.

​

However, there may be some localised applications where statistical methods may still be of use. This may correspond to applications where dynamical model resolution is too coarse to provide a reasonable prediction. For example, for years during which the Mediterranean SSTs and/or northern Africa near‐surface land temperature influence significantly Sahel rainfall (e.g. 2012), statistical tools are more useful because very few dynamical models have Mediterranean SST/Sahel rainfall mechanisms well represented.

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman

Thank You!

​

Merci be coup!

Meteorology of Tropical West Africa: The Forecasters’ Handbook

Chapter 8: Seasonal Forecasting – Lead author: Andrew Coleman