https://africanswift.org/ - PowerPoint Presentation

1. https://africanswift.org/Sub-seasonal (S2S) Monsoon Onset forecasting over West Africa EGU General Assembly 2020Elisabeth Thompson, Caroline Wainwright, Linda Hirons, Felipe Marques de Andrade, and Steven Woolnoughe.m.thompson@reading.ac.uk GCRF African SWIFT - Science for Weather Information and Forecasting Techniques1

2. GCRF African SWIFT - Science for Weather Information and Forecasting Techniqueshttps://africanswift.org/S2S and Monsoon Onset ForecastingSkilful onset forecasts are highly sought after in West Africa and provide essential information for many sectors, including: - agriculture - disease prevalence - energy provision With research on the sub-seasonal timescales bridging the gap between short-range and seasonal weather forecasts, sub-seasonal forecasts may provide useful information in the period preceding monsoon onset. As part of the Sub-seasonal work package of the GCRF African-SWIFT Project, this study examines the challenges of operational monsoon onset forecasting. Specifically, it compares two methods of monsoon onset determination over Ghana as well as the impact of geographical scale. Motivation 2Research Questions Does geographical scale impact the usefulness of the onset determination method?How does model bias in the forecast rainfall impact the onset definition?Is monsoon onset predictable on an S2S timescale?What onset determination methods are useful of S2S onset forecasts? Does scale impact which onset determination is most useful?

3. GCRF African SWIFT - Science for Weather Information and Forecasting Techniqueshttps://africanswift.org/DataECMWF S2S Model Version CY45R1 Forecast Length: 0-46 daysHorizontal Resolution: 16km ≤day 15; 32km > day 15Thursday initialisations from the 31st January 2019 Forecast Ensemble Size: 51Hindcast Ensemble Size: 11Hindcast Length 1999-2018TAMSAT version 3 daily satellite estimates from 1999 to 2019 (Maidment et al., 2014; Tarnavsky et al. 2014) 3Figure 1: Map of the four agro-ecological zones and Ghana Meteorological Agency (GMet) stations, sourced from Amekudzi et al. (2015)

4. GCRF African SWIFT - Science for Weather Information and Forecasting Techniqueshttps://africanswift.org/Data Challenges: Model BiasScaling of the forecast and hindcast data to account for the model bias was conducted prior to onset determination [shown as the difference between the hindcast and TAMSAT observation climatologies in Figure 1].4This study scaled the model data using the following method:Calculate the TAMSAT daily climatology for each initialisation date (t to t+42)Calculate the daily hindcast average for each initialisation date (t to t+42)Calculate the bias by dividing the tamsat daily climatology by the ecmwf hindcast climatologyCorrect the forecast values for each initialisation date by multiplying by the bias calculated in step 3. Figure 1: 2019 ECMWF forecast, initialised on the 22nd February, with Model and Observation climatology data indicating the presence of a bias in the model.

5. GCRF African SWIFT - Science for Weather Information and Forecasting Techniqueshttps://africanswift.org/Data Challenges: Resolution Change Day 15 to Day 16Negative rainfall values and inconsistencies caused by the model resolution change from day 15 to day 16, were corrected after the calculation of daily total rainfall values.Figure 1: Accumulated rainfall (mm) values by day from initialisation dateFigure 2: Number of negative rainfall values after the resolution change for 29th April 2019 forecast initialization5

6. GCRF African SWIFT - Science for Weather Information and Forecasting Techniqueshttps://africanswift.org/1) Anomalous Accumulation Technique (Dunning et al., 2016)Defining Monsoon Onset Monsoon onset occurs on the day of the cumulative daily rainfall anomaly minimum[first purple dot].Red line: Climatological daily mean rainfall for each day of the year Blue line: Climatological daily mean rainfall anomalyGreen Line: climatological cumulative daily mean rainfall anomaly for 9.5∘N, 14.5∘W from GPCP averaged over 1997–2014. Magenta dots: Start and End of the climatological water season.6

7. GCRF African SWIFT - Science for Weather Information and Forecasting Techniqueshttps://africanswift.org/2) Operational Rainfall Threshold Technique (Ghana Meteorological Agency)Defining Monsoon Onset Monsoon onset occurs after 1st February on the third day of three consecutive days over which the cumulative rainfall is greater than 20mm, and after which there are no successive periods of 10 days or more with less than 1mm of rainfall. [e.g. the 5th March below]7Note 1: the previous anomalous accumulation onset method here would give an onset date of the 12th AprilFigure 1: Daily rainfall totals (blue bars) and the cumulative rainfall anomaly (green line) from the 1st March to the 30th April. Note 2: If it hadn’t rained on the 15th March, the 10 day dry spell would have meant onset occurred on the 15th April.

8. GCRF African SWIFT - Science for Weather Information and Forecasting Techniqueshttps://africanswift.org/Differences in Monsoon Onset Definitions8Figure 1 shows the difference between the accumulation and GMet onset determination techniques for each year. The accumulation method tends to forecast for later onset dates. Note: Model biases mean that the 20mm GMet rainfall threshold is not exceeded as often as observations. Additionally, the model bias and spatial scale challenge for forecasting is that the lower spatial resolution and associated regional averaging in the model results in not as many days with less than 1mm of rainfall, which means there are not as many dry spells as in the observations, impacting the GMet onset determination, by forecasting earlier onset dates. What onset determination methods are useful of S2S onset forecasts? Figure 1: Boxplots of the TAMSAT climatology for all years, the TAMSAT onset date for each year [blue/orange circle], as well as the hindcast ensemble mean and spread for each year, from 1999 to 2018, split by the Accumulation and the GMet onset determination techniques.Monsoon Onset Dates for Accra

9. https://africanswift.org/Geographical Scale ComparisonMonsoon Onset Dates for Forecast initialised 14th February 2019, split by onset determination method, and the TAMSAT onset climatologyFigure 3: Accra versus Coast Region Figure 2: Kumasi versus Forest Region 9GCRF African SWIFT - Science for Weather Information and Forecasting TechniquesThe geographical scale or resolution of the data can make a big difference to the determination of onset dates. Figure 1 shows how the observations at a grid point compared to a larger region can give very different monsoon onset dates [27th March compared to 7th March, respectively, for the anomalous accumulation technique].Figure 1: 2005 TAMSAT estimates for Kumasi versus the Forest Region Figures 2 and 3 show the ensemble spread in forecast onset dates, as well as reiterating both the difference in mean onset with geographical scale from city to regions, and the impact of applying different onset determination methods, within the ECMWF model. The accumulated method tends to find later onset dates, as do the larger spatial areas.Days of the Year30405060708090100110Gmet Method Accum. Method TAMSAT KumasiKumasiTAMSAT ForestForestTAMSAT KumasiKumasiTAMSAT ForestForestDays of the Year406080100120Gmet Method Accum. Method TAMSAT AccraAccraTAMSAT CoastCoastTAMSAT AccraAccraTAMSAT CoastCoastHow does the scale we’re measuring over impact the definition?

10. GCRF African SWIFT - Science for Weather Information and Forecasting Techniqueshttps://africanswift.org/Geographical Mean Onset ComparisonIs monsoon onset predictable on an S2S timescale?What onset determination methods are useful of S2S onset forecasts?10These two figures show geographical maps of mean monsoon onset for the TAMSAT observations and the ECMWF hindcast for 2000, using the anomalous accumulation and operational threshold techniques. Although there are locations where the two techniques show similar onset dates, specifically for the hindcast onset, these also show the variation in onset date caused by onset determination method.Figure 2: Onset Hindcast for 2000[initialised 21st March 2019]Figure 1: TAMSAT Onset Satellite Estimate for 2000TAMSAT Onset dates using the Accumulation method for 200014°N12°N10°N8°N6°N4°N5°W2.5°W2.5°E0°5°W2.5°W2.5°E0°TAMSAT Onset dates using the GMet method for 2000Mean Onset hindcast for 2000 using the GMet method from initialisation date 21st March 20195°W2.5°W2.5°E0°5°W2.5°W2.5°E0°Mean Onset hindcast for 2000 using the Accumulation method from initialisation date 21st March 201914°N12°N10°N8°N6°N4°N

11. https://africanswift.org/2019 Ensemble Mean Onset Comparison by forecast initialisationIs monsoon onset predictable on an operational S2S timescale?11GCRF African SWIFT - Science for Weather Information and Forecasting TechniquesIn order for S2S onset forecasts to be useful, the appropriate determination method and resolution need to be carefully considered, as well as forecast timing . 14°N12°N10°N8°N6°N4°N5°W2.5°W2.5°E0°Onset forecast for 2019 using the Accumulation method, initialisation date 20190207 Onset forecast for 2019 using the GMet method, initialisation date 20190207 Onset forecast for 2019 using the Accumulation method, initialisation date 20190131 Onset forecast for 2019 using the GMet method, initialisation date 20190131 14°N12°N10°N8°N6°N4°N5°W2.5°W2.5°E0°TAMSAT Onset dates using the Accumulation method for 2019TAMSAT Onset dates using the GMet method for 201914°N12°N10°N8°N6°N4°N5°W2.5°W2.5°E0°Onset forecast for 2019 using the Accumulation method, initialisation date 20190221 Onset forecast for 2019 using the GMet method, initialisation date 20190221 Onset forecast for 2019 using the Accumulation method, initialisation date 20190214 Onset forecast for 2019 using the GMet method, initialisation date 20190214

12. GCRF African SWIFT - Science for Weather Information and Forecasting Techniqueshttps://africanswift.org/Summary12Model biases need to be corrected prior to use for onset forecasting, with forecasts scaled to the observation data.Geographical scale and resolution of the data can impact the determination of onset dates. The accumulated method tends to find later onset dates, as do the larger spatial areasLarge variations occur in onset dates when comparing different onset determination methods.Timing of the onset forecast, in association with good observational data is important for operational distribution and usefulnessThe accumulated technique is more useful for onset forecasting due to the model bias and scaling issues, which results in a reduction of the dry spells in the model and has a greater impact on the Gmet definition.

13. GCRF African SWIFT - Science for Weather Information and Forecasting Techniqueshttps://africanswift.org/Further work is ongoing, including: probabilistic onset forecasts [Are deterministic or probabilistic forecasts more useful when predicting onset?]examination of the wider impact and appropriate solutions for the ECMWF day 16 resolution changewider comparisons of the difference in geographical scale,analysis of quantitative differences and trends in onset determination method; and skill analysis [What skill metrics are useful for forecasting monsoon onset?]Next Steps13

14. 3GCRF African SWIFT - Science for Weather Information and Forecasting TechniquesThank you! Elisabeth Thompsone.m.thompson@reading.ac.uk 14References:Amekudzi, L.K., Yamba, E.I., Preko, K., Asare, E.O., Aryee, J., Baidu, M. and Codjoe, S.N.A. (2015) Variabilities in Rainfall Onset, Cessation and Length of Rainy Season for the Various Agro-Ecological Zones of Ghana. Journal of Climate, 3. pp. 416-434.Dunning, C. M., Black, E. C. L. and Allan, R. P. (2016) The onset and cessation of seasonal rainfall over Africa. Journal of Geophysical Research: Atmospheres, 121 (19). pp. 11405-11424.Maidment, R., D. Grimes, R.P.Allan, E. Tarnavsky, M. Stringer, T. Hewison, R. Roebeling and E. Black (2014) The 30 year TAMSAT African Rainfall Climatology And Time series (TARCAT) data set Journal of Geophysical Research DOI: 10.1002/2014JD021927Tarnavsky, E., D. Grimes, R. Maidment, E. Black, R. Allan, M. Stringer, R. Chadwick, F. Kayitakire (2014) Extension of the TAMSAT Satellite-based Rainfall Monitoring over Africa and from 1983 to present Journal of Applied Meteorology and Climate DOI 10.1175/JAMC-D-14-0016.1