By Emily Black
A new early warning and decision support system: TAMSAT-ALERT
For subsistence farmers in Africa, decisions on what variety of crop to grow, when to plant, and when to apply fertilizer are of life and death importance. The new TAMSAT* Agricultural Early Warning System (TAMSAT-ALERT) combines multiple streams of environmental data into probabilistic assessments of the risk faced by farmers when making these decisions. The assessments can be issued directly to farmers to support day-to-day decision making, or provided via drought warning bulletins from regional meteorological and agricultural organisations.
TAMSAT-ALERT risk assessments can be based on any metric that can be generated from meteorological data. Our work has so far focused on meteorological and agricultural drought – encapsulated respectively by deficit in cumulative rainfall and soil moisture.
Figure 1: Predicted cumulative rainfall anomaly for a location in Kenya for the 2000—2001 DJF rainy season. The forecast was carried out on 10th Decemeber 2000.
Figure 2: Predicted soil moisture anomaly (left) and Water resource Satisfaction Index (right) for the 2017 October-December rainy season in Kenya. This prediction was made on 1st December, 2017.
TAMSAT-ALERT complements existing systems because it is sufficiently lightweight to be run on a standard PC, using the computing facilities available at African meteorological and hydrological services . Example outputs are shown in Figure 1 and 2. The system can be implemented both for detailed analysis at the individual community scale (Figure 1) and for regional/national level assessments (Figure 2). The national forecast shown in Figure 2, for example, took only ten minutes to generate. All the code for TAMSAT-ALERT is freely available on GitHub , and TAMSAT is working closely with African organisations to build their capacity to use the system.
Figure 3: A screen shot of the TAMSAT-ALERT web interface
For less expert users, we are developing a web interface, which will provide a limited set of output plots for any point in Africa (Figure 3).
Figure 4: TAMSAT-ALERT example output. The ‘true’ historical (thick blue lines) and projected soil moisture in northern Ghana (thin red lines), left, for 2011 (top) and 2003 (bottom) and the corresponding evolving drought probability (right), where the vertical black lines represent the time steps of the three left plots. The drought probability is probability that the projected soil moisture for the growing season will be in the lowest quartile of climatology (from Brown et al, 2017)
The science behind TAMSAT-ALERT
Agricultural outcomes are affected by weather over an extended period, ranging from days to months. For example, low yield is caused by soil moisture deficit at critical periods during a ~ three month growing season; germination failure is caused by low rainfall during the two weeks after planting. The risk assessment on a given day, therefore, needs to take into account weather in the past and the future. In TAMSAT-ALERT, weather in the past is taken from observations, and an ensemble of future weather is derived from the climatology. Meteorological forecast information is integrated into the assessments by weighting the ensemble using probabilistic output from a numerical forecast model. This process is illustrated by Figure 4 and by the example above.
An example…
A prediction of soil moisture deficit for the 2018 March-May growing season carried out on 1st April, will be created by driving a land surface model, such as the UK land surface model, JULES, with observations from 1st March – 1st April 2018 spliced together with data a 2nd April – 30th May climatology. If the climatology is from 1983-2012, the ensemble will have 30 members. Ensemble member 1 will be yield output from the model driven with historical observations for 1st March – 1st April 2018, spliced with historical observations from 2nd April – 30th May 1983. The second member will use 2nd April – 30th May 1984 for the future period, and so on.
The result will be 30 possible predictions of soil moisture – a frequency distribution.
The next step is to integrate meteorological forecasts. In this year (2018), we have a prediction that the probability of regional MAM rainfall being in the lowest tercile is 60%, the middle tercile is 30% and the upper tercile is 10%. For example, if in 1983, regional rainfall was in the lowest tercile, the 1983 ensemble member is weighted by 0.6; if in 1984, rainfall is weighted by 0.1, and so on. Probabilistic risk assessments are then derived by analysing this weighted frequency distribution.
In essence, the system quantitatively addresses the question: ‘Given the state of the land surface, the climatology and the meteorological/climate forecast, what is the likelihood of some adverse food production event over the coming cropping period?’ As such, TAMSAT-ALERT is an ‘impacts-based’ forecast system, providing information aligned with the needs of operational food security risk assessments.
TAMSAT-ALERT is run at the scale of the input meteorological observational data. It thus implicitly downscales and bias corrects regional seasonal forecast data. The system has recently been extended to run in gridded mode. Pilot projects confirm that the system is sufficiently lightweight to run in African agrometeorological agencies.
*TAMSAT stands for Tropical Applications of Meteorology using SATellite data and ground-based observations.
References:
Asfaw, D., Black, E., Brown, M., Nicklin, K.J., Otu-Larbi, F., Pinnington, E., Challinor, A., Maidment, R. and Quaife, T., 2018. TAMSAT-ALERT v1: A new framework for agricultural decision support. Geoscientific Model Development, 11(6), pp.2353-2371. DOI: 10.5194/gmd-11-2353-2018
Brown, M., Black, E., Asfaw, D. and Otu‐Larbi, F., 2017. Monitoring drought in Ghana using TAMSAT‐ALERT: a new decision support system. Weather, 72(7), pp.201-205. DOI: 10.1002/wea.3033