Nat Hazards (2012) 64:1187–1207 DOI 10.1007/s11069-012-0286-2 ORIGINAL PAPER
An innovative tailored seasonal rainfall forecasting production in Zimbabwe Desmond Manatsa • Leonard Unganai • Christopher Gadzirai Swadhin K. Behera
•
Received: 18 April 2011 / Accepted: 6 July 2012 / Published online: 18 July 2012 Ó Springer Science+Business Media B.V. 2012
Abstract Farmers’ adaptation to climate change over southern Africa may become an elusive concept if adequate attention is not rendered to the most important adaptive tool, the regional seasonal forecasting system. Uptake of the convectional seasonal rainfall forecasts issued through the southern African regional climate outlook forum process in Zimbabwe is very low, most probably due to an inherent poor forecast skill and inadequate lead time. Zimbabwe’s recurrent droughts are never in forecast, and the bias towards near normal conditions is almost perpetual. Consequently, the forecasts are poorly valued by the farmers as benefits accrued from these forecasts are minimal. The dissemination process is also very complicated, resulting in the late and distorted reception. The probabilistic nature of the forecast renders it difficult to interpret by the farmers, hence the need to review the whole system. An innovative approach to a regional seasonal forecasting system developed through a participatory process so as to offer a practically possible remedial option is described in this paper. The main added advantage over the convectional forecast is that the new forecast system carries with it, predominantly binary forecast information desperately needed by local farmers—whether a drought will occur in a given season. Hence, the tailored forecast is easier for farmers to understand and act on
D. Manatsa (&) Geography Department, Bindura University of Science, Bindura, Zimbabwe e-mail:
[email protected] D. Manatsa International Center for Theoretical Physics (ICTP), 34151 Trieste, Italy D. Manatsa S. K. Behera Department of Ocean Technology, Policy, and Environment, University of Tokyo, Tokyo, Japan D. Manatsa S. K. Behera Research Institute for Global Change/JAMSTEC, Yokohama, Japan L. Unganai Environmental Management Agency, Ministry of Environment, Harare, Zimbabwe C. Gadzirai Agriculture Department, Bindura University of Science, Bindura, Zimbabwe
123
1188
Nat Hazards (2012) 64:1187–1207
compared to the conventional method of using tercile probabilities. It does not only provide a better forecasting skill, but gives additional indications of the intra-seasonal distribution of the rainfall including onsets, cessations, wet spell and dry spell locations for specific terciles. The lead time is more than 3 months, which is adequate for the farmers to prepare their land well before the onset of the rains. Its simplicity renders it relatively easy to use, with model inputs only requiring the states of El Nino Southern Oscillation (ENSO) and Indian Ocean Dipole (IOD) climate modes. The developed forecast system could be one way to enhance management of risks and opportunities in rain-fed agriculture among small-holder farmers not only in Zimbabwe but also throughout the SADC region where the impact of ENSO and/or IOD on a desired station rainfall is significant. Keywords Tailored seasonal rainfall forecast Zimbabwe farmers Maize yield Indian Ocean Dipole El Nino Southern Oscillation
1 Introduction Crop production in Zimbabwe is predominantly rain-fed and seasonal rainfall is highly variable, making crop failure due to rainfall extremes common (Martin et al. 2000). Changing climate and exposure to natural climatic hazards, such as droughts, floods and tropical storms, has resulted in increasingly devastating impacts on rural livelihoods (Eriksen et al. 2008; Fussel 2007). Farmers might considerably reduce their exposure to risks associated with climate variability through use of climatic information (e.g., Ziervogel et al. 2010). At the same time, it can be argued that improved climate information and enhanced application can represent a new wave of agricultural technology for small-scale farmers (Phillips et al. 2001; Ziervogel 2004). Improvement of the advance warning systems of rainfall extreme events should be of high priority, especially considering the resource poor farmers, who are the most vulnerable group of the southern African region. A well-functioning extended forecasting system allows farmers to plan the cropping calendar according to the risks and opportunities that deficient, surplus, early, timely or late seasonal rains entail (e.g., Phillips et al. 2001). The potential for producers to benefit from seasonal forecasts depends on factors including the flexibility and willingness to adapt farming operations to the forecast, the timing and accuracy of the forecast, and the effectiveness of the communication process (Unganai and Kogan 1998). Convincing demonstrations of forecast value are therefore desirable to support assimilation into practice. Moreover, by responding well to the current climate variability, farmers automatically enhance their adaptation strategy to the changing climate as a whole (Vogel 2000). Thus, farmers increasingly require tailored climate assessments, seasonal forecasts and prediction to manage climate risks. This group lacks experience in making use of scientific and technical knowledge (Patt 2001) and faces a diverse set of decisions that vary from year to year. Explicit integration of climate forecast products into agricultural operations of farmers in Zimbabwe is currently largely non-existent, although climate forecasts are issued annually through the southern African regional climate outlook forum (SARCOF) process. SARCOF is a regional seasonal weather outlook prediction and application practice adopted by the southern African development community (SADC) that produces a consensus forecast for the sub-region. This study seeks to explore ways to improve the application of climate information among small-holder farmers through the analysis of climatic information. In this way, a tailored seasonal climate forecast system could be one way to enhance management of risks and opportunities in rain-fed agriculture among small-holder farmers not only in Zimbabwe but also elsewhere where the SARCOF forecast application is constrained.
123
Nat Hazards (2012) 64:1187–1207
1189
Fig. 1 Location of the study area in the south-east of Zimbabwe
The innovativeness of this study arises from the fact that it provides original options to improve the local climate information delivery and application systems at the district level and also to improve the farmers’ confidence in the use of the seasonal forecasts. The strategy employed is unique as it develops a region-specific nonparametric seasonal rainfall forecasting model that uses ENSO and Indian Ocean dipole (IOD) phases as the main inputs. It also proposes a different institutional mechanism for the generation, flow, translation, communication and application of seasonal rainfall forecast. Lastly, it provides a hybrid application framework for the seasonal rainfall forecast. Under this novel rainfall forecast dispensation, the local climate early warning system can accomplish its intended mandate as the most important adaptive strategy to climate change (Acher et al. 2007) for the local rural communities. 1.1 Study area The geographical area of study is Chiredzi district (Fig. 1), which is in the south-west of Zimbabwe. The district was chosen as a pilot site due to the high vulnerability of its smallholder subsistence farmers. The prevailing climate of the region is semi arid, characterized by low rainfall which varies widely in the temporal and spatial domain both intraseasonally and inter-seasonally. The relatively high temperatures and erratic rainfall pattern limit agricultural production of the region to dryland crops and cattle rearing. These unreliable climatic conditions could be contributing factor to the difficulties experienced in developing dependable climate forecasts for the end users.
2 Data and methods 2.1 Data Climate data for the area, including perceptions on the current seasonal forecast and the use of indigenous knowledge, were collected from the grey literature. Other data sources were focus group discussions, workshops and key informant interviews. The sea surface temperature data for potential predictors ENSO and IOD were obtained from Hadley Centre SST and JAMSTEC data set websites //www.esrl.noaa.gov and //www.jamstec.go.jp//iod, respectively. Daily rainfall data, covering the period 1966–2009, to represent the predictant
123
1190
Nat Hazards (2012) 64:1187–1207
for the district were point data obtained from Buffalo Range research station in Chiredzi. The seasonal forecasts for the period 1997/8–2006/7 were obtained from the SARCOF statements, provided by the Drought Monitoring Centre in Botswana at their website www.sadc.int/dmc/SARCOF/. Annual maize yield data for the period 1980–2007 for Chiredzi district were obtained from the national central statistical office (CSO). 2.2 Methods Anomalies for the annual rainfall and maize yield time series were standardized, using their respective standard deviations to obtain the corresponding indices. The regression analysis was then used to find the extent of the relationship between rainfall and maize yield indices. Correlation analysis was used to find the relationship between the indices of the predictors and the predictant. A combination of participatory and statistical approaches was also used to come up with a nonparametric seasonal rainfall forecasting system that also incorporated the indigenous knowledge systems in some of the response options. However, an important aspect of the design methodology used for developing a tailored climate applications framework for Chiredzi district was a strong interaction with Extension Services and farmers. It ensured that the information provided in the system is relevant for user needs and that the language and formats used are appropriate. While a number of activities did not necessarily require interactions with end users, such as the development and validation of the seasonal forecast model, the design of the applications framework was based on an intense interaction with end users for testing and validation of the findings. 2.2.1 Verification method Verification of the Chiredzi district seasonal rainfall forecast with the highest confidence tercile (above, near and below normal) category was done using 3 9 3 contingency tables. This approach was adopted because most farmers interviewed who were able to understand the probabilistic forecast from the Met Office, indicated that they usually take the forecast in a deterministic way. Thus, we also considered the general outcome of the season to materialize according to the tercile with the highest confidence. The forecast and observed categories are simply classified in a table of 3 rows and 3 columns (Table 2a, b). There is a row for each observed category, and a column for each forecast tercile category (above, near and below normal). For each year since the season 1997/98, 1 is added to the grid element of the contingency table for each event, according to the intersection of the forecast category and the observed category. Many scores can be calculated from the contingency tables. For example, the per cent correct is the sum of the diagonal elements over the total elements (multiplied by 100), and it is used as the standard score for the long-range forecasts. A detailed explanation of each score can be found in the literature from the World Meteorological Organization (Stanski et al. 1989). 3 Results and discussion 3.1 Deficiencies within the current forecast system Before the development of a tailored seasonal rainfall forecast system for Chiredzi district, it was necessary to study the prevailing maize yield—rainfall relationship so as to estimate the effectiveness of the current rainfall forecast system. Table 1 shows correlation coefficients between average maize yield and total rainfall for the first and second parts of the
123
Nat Hazards (2012) 64:1187–1207
1191
Table 1 Correlation coefficient of Chiredzi maize yield with OND, JFM and seasonal rainfall total
Correlation coefficient with maize yield
OND rainfall
JFM rainfall
Seasonal total rainfall
0.35
0.27
0.39
Values in bold indicate confidence above the 99 % level
rainfall season that stretches from October to December and January to March, respectively, and the whole period of the rainfall season that covers October through March over the period 1980–2007. It can be deduced from the table that the seasonal rainfall totals of all the three mentioned periods are able to explain statistically significant fractions (above the 95 % level) of the observed maize yield variation. This clearly indicates the significant dependence of the Chiredzi maize yield on rainfall. However, the maize yield variance explained by the rainfall totals of less than 16 % demonstrates that there are some deficiencies inherent in the effectiveness of the current seasonal rainfall forecasts. The corresponding bars of the Chiredzi rainfall index, alongside those of the yield, depicted in Fig. 2, bring out clearly these anomalies. An ideal situation of an effective forecasting system in a rain-fed agricultural system should have rainfall that explains more than half of the yield variance (e.g., Phillips et al. 2001) as there are bound to be more frequent coincidences of high (low) yields with good (poor) rainfall. But, here, we observe from Fig. 2 that the negative rainfall indices are not necessarily followed by negative maize yield indices (e.g., 1982/1983, 1986/1987 and 1989/1990), and the positive rainfall indices are also not essentially accompanied by above average maize yields (e.g.,
Fig. 2 Yearly comparisons of maize yield and rainfall indices for Chiredzi district for the period 1980–2007. Respective linear trends are shown in the insert
123
1192
Nat Hazards (2012) 64:1187–1207
1999/2000, 2000/2001, 2003/2004 and 2004/2005). The observation of poor maize yields in above normal rainfall years could be a reflection of poor within season rainfall distribution when a correct rainfall total has been given or failure of farmers to take advantage of the good rains due to poor forecasts. We cannot rule out other constraints which affected the farmers during this time like the national economic downturn. However, in the presence of a reliable seasonal rainfall forecast system, it is expected that farmers will maximize opportunities associated with favourable rains and minimize losses in the event of poor rains, thus increasing the yield rainfall association. Since 1997, farmers have also been complaining about misleading seasonal rainfall predictions. If the farmers have been acting according to the forecasts, then the deterioration of this product is apparent in Fig. 2, where the seasonal rainfall and maize yield mismatches increased from 1997. For example, rainfall events prior to 1997 of marginally deficit rainfall total, which translated into exceptionally good harvests like 1989 and 1990, could hardly be associated with similar yields afterwards. Later episodes of abundant rainfall, like that of 1999, 2003 and 2004, respectively, were not taken advantage of in order to produce correspondingly increased yield. Coincidentally, 1997 is the year when SARCOF coordinated seasonal forecasts was initiated in southern Africa. Unfortunately, this noble regional process started with a glaring miss, when a presumed devastating drought that was in forecast regionally due to the successful projections of a major El Nino by the International Research Institute for Climate Prediction (IRI) during the season (Mason et al. 1999), never materialized (Manatsa et al. 2007, 2008). IRI is the institute that plays a major part in the SARCOF process, and hence, its general circulation model output contributes significantly to the regional seasonal forecasts. Linear trends inserted for the rainfall and yield also suggest a general falling ability of the local farmers to take advantage of the seasonal rainfall, especially after 1997. The yield trend is declining significantly above the 95 % confidence level, but under (slightly) increasing rainfall trend. On a national scale, Manatsa et al. (2011) examined the temporal trend in the maize yield for poor resourced famers and also noted similar post-1996 years with abundant rainfall that are not matched by increased maize yield but display significantly suppressed production instead. This could be a clear testimony to the gradual ineffectiveness of the local seasonal forecast system. It is fortunate that in rain-fed farming systems, effective application of seasonal rainfall forecasts has the potential to reverse this negative direction of the crop yield trend so as to match or even surpass that of the rainfall (Acher et al. 2007; Phillips et al. 2001). Ultimately, the deductions from Table 1 and Fig. 2 warrant an investigation of the Met Office seasonal rainfall forecast process itself, for possible explanations of the perceived ineffectiveness and strongly supports the need to produce a better ‘tailored seasonal forecast system’. 3.2 Analysis of the SARCOF seasonal rainfall forecast system The current format of seasonal rainfall forecasts based on the SARCOF procedure started to be developed and issued in 1997. This technique is based primarily on ENSO related information and downscaling of General Circulation Models (Klopper 1999). The seasonal forecasts which are expressed in terciles of any two consecutive combinations of above normal (AN), near normal (N), and below normal (BN), are issued in September. These forecasts provide probabilistic information on future climate of time scales of three to 6 months, hence covering the first and second half of the rainfall season, OND and JFM, respectively. An analysis of the probabilistic seasonal rainfall forecast performance since 1997 reveals that a forecast of below normal rainfall as the highest probability to indicate the
123
Nat Hazards (2012) 64:1187–1207
1193
Fig. 3 Temporal evolution the Met Office tercile seasonal forecasts from 1997 to 2008 for OND (a) and JFM (b). The amount of rainfall in mm can be read from the right vertical axis
123
1194
Nat Hazards (2012) 64:1187–1207
Table 2 (a) OND and (b) JFM: 3 9 3 contingency tables of Met Office forecast from 1997/1998 to 2008/2009
OND Forecasts
Total
Below normal
Near normal
Above normal
Below normal
0
0
0
Near normal
4
2
3
9
Above normal
1
1
0
2
5
3
3
11
(a) OND observations
Total
JFM Forecasts
0
Total
Below normal
Near normal
Above normal
Below normal
0
0
0
Near normal
3
4
2
9
Above normal
0
1
1
2
3
5
3
11
(b) JFM observations
Total
0
strongest confidence in a drought occurring has never been issued (Fig. 3a, b). In the JFM period, the near normal rainfall category with the highest probability of 40 % and above has been forecast in 10 of the 12 years. On the other hand, the below normal tercile, with a probability of 25 % or less, has been forecasted on nine of the eleven occasions, while the above normal tercile is appearing in only two occasions. For OND, it is surprising that the near normal tercile has been forecasted consecutively since 2001, and the above normal tercile has never been forecasted since 2000 and worse still, below normal rainfall has never been forecasted, ever since the commencement of the SARCOF process, about 14 years ago. Thus, it appears that SARCOF has a tendency of avoiding forecasting droughts, despite the fact that several drought years, including extreme droughts of the century, have occurred in Chiredzi district during this period (Fig. 3a, b). In the next section, we investigate the current forecast value to the intended users, the small-scale farmers whose crops are dominantly rain-fed. 3.3 SARCOF forecast value to small-scale farmers During data collection, Chiredzi district farmers expressed great concern regarding the current seasonal rainfall forecast and were very sceptical of the value of these forecasts. Despite the seasonal forecasts being issued yearly by SARCOF through the Met Office for the past 14 years, the extent of uptake in Chiredzi district remains limited, with very low rate (17 %) of utilization among local small-holder farmers. This general display of unpopularity of the seasonal forecast with the local farmers could be emanating from the forecast evaluation results depicted in Table 2. The forecast skills are not only very low, but also heavily biased towards normal conditions. Very high confidence of the forecast is persistently in the near normal range, but this confidence is never attributed to drought forecasts. As a result, the above normal rainfall events are rarely forecasted and droughts are never in forecast. The forecast is also too general, in both spatial and temporal resolution, hence neither specific to an area nor specific to a particular application. The high climate variability over
123
Nat Hazards (2012) 64:1187–1207
1195
short distances inherent in the region, even within villages, reduces the utility of forecast information, especially when provided at the current, national scale. Since the seasonal rainfall forecast is expressed only through probabilities of the OND and JFM totals, the forecast inherits an uncertain start and end of the season. Consequently, the length of the season is very vaguely implied in the forecast, if defined at all. This also explains the absence of the definition of other intra-seasonal variations in the forecast information, such as the distribution of both wet and dry spells, including their spatial distribution. This is despite the fact that these were cited as crucial information in the planning process by the local farmers during the initial survey of the study. Since Table 2 indicates that many of the above normal rainfall forecasts have been incorrect, this situation could have resulted in both massive financial and material losses by the small-scale farmers. This stems from the actions that a farmer might have adopted (significant investments in inputs and area planted), following several wet forecasts that turned out to be misses. Also, compounding the problem is the fact that below normal rainfall events have never been in forecast, yet several droughts have occurred in Chiredzi district. Thus, the intended benefits from these official forecasts drop considerably, as the small-scale farmers are hit unexpectedly by the extreme seasonal rainfall events. High rainfall is not taken advantage of, and the farmers are also ravaged by the droughts. Since the current forecasts expose the poor farmers to the full wrath of extreme rainfall events due to unmitigated impacts (e.g., Roncoli et al. 2001), some innovative ways should be sought that tailor the seasonal rainfall forecast in such a way to reduce the farmers’ vulnerability. Farmers’ concerns also centre on the lead time period that is less than a month, which was considered rather inadequate to make seasonal decisions and plans. The probabilistic nature of the forecast itself proved to be incomprehensible and often too difficult for the farmers to interpret, let alone apply. Hence, the interpretation was deterministic, with only the tercile with the highest percentage value considered. At the same time, the forecasts principally referred to the ‘meteorological’ component, neglecting the agricultural part required by the farmers. There were no effective dissemination systems in place, and the communication channels chosen were not easily accessible (e.g., radio, TV and newspapers) to the poor farmers. Most farmers neither own a radio, nor own a TV set, and the majority have problems in timely accessing the newspaper, let alone being able to read it (e.g., Phillips et al. 2001). In the following section, we do a simple verification exercise to the forecast, according to the WMO technical report (Stanski et al. 1989), so that some approximated quantifiable values can be attached to the forecast. 3.4 Verification of the seasonal rainfall forecast The verification process, whose results are summarized in Table 3, provides a very poor rating to the forecast. It reveals that the skill of the forecast for both OND and JFM cannot provide an economic value to the farmer (skill less than 50 %). The success rate that can be derived from the forecast use, as measured by the critical success index is either zero or close to it, for all the forecasted rainfall events. It is reasonably reliable only for the near normal forecasts, with index values of 0.7 and 0.8 for OND and JFM, respectively, but virtually unreliable (OND and JFM index values of zero) for the rest of the tercile forecasts. Furthermore, it is indicated in Table 3 that the forecast is worse than both chance (random) and persistence forecast for OND, but slightly better than both for JFM with a Heidke skill score that is negative and positive, respectively. It strongly over forecasts (positive bias) near normal events rainfall, but virtually does not forecast below normal events. On the other hand, the forecast cannot detect below normal rainfall events, weakly
123
1196
Nat Hazards (2012) 64:1187–1207
Table 3 Verification statistics according to the WMO technical report (Stanski et al. 1989) Statistic
OND
JFM
Forecast skill (percentage correct)
18 %
45 %
Heidke skill score 1 (chance)
-0.25
0.21
Heidke skill score 2 (persistence)
-0.09
0.13
Bias below normal event
No bias
No bias
Bias near normal event
2.0
3.3
Bias above normal event
1.5
1.5
Probability of detection of below normal event
None
None
Probability of detection of near normal event
0.22
0.4
Probability of detection of above normal event
0.0
0.5
False alarm ratio of below normal forecasts
1.0
1.0
False alarm ratio of near normal forecasts
0.3
0.2
False alarm ratio of above normal forecasts
1
0.7
Reliability of below normal forecasts
0.0
0.0
Reliability of near normal forecasts
0.7
0.8
Reliability of above normal forecasts
0.0
0.3
Critical success index for below normal forecasts
0.0
0.0
Critical success index for near normal forecasts
0.2
0.4
Critical success index for above normal forecasts
0.0
0.3
detects near normal categories and can hardly project above normal events for JFM, while at the same time, it is not capable of identifying above normal rainfall events for the OND period. Although the forecast picks out most of the near normal rainfall, it detects well less than a third truths about near normal rainfall for both sub-periods. Lastly, it misleads the user completely about all the below normal rainfall events of both sub-seasons and all above normal OND events. This state of affairs calls for immediate remedial action. In the following section, we provide an innovative seasonal forecast system for the region and propose its application framework. 3.5 The use of ENSO and IOD phases as predictors Traditionally, the Met Office has used the ENSO phases in its seasonal forecasts as several authors believe that this climate mode has the dominant influence on the regional rainfall (e.g., Reason and Jagadheesha 2005; Jury et al. 2002). But it is apparent from the analysis and evaluation of the SARCOF’s seasonal forecast, which the seasonal rainfall for Chiredzi district is extremely difficult to predict. In developing a tailored forecasting system for the district, the complementary use of the Indian Ocean Dipole and ENSO phases can greatly facilitate the prediction of Chiredzi district seasonal rainfall. The motivation comes from the work by Manatsa et al. (2008) and Manatsa and Mukwada (2012), which noted that the IOD is a better predictor of Zimbabwe seasonal rainfall both in terms of lead time and forecasting skill. In the development of the forecast model, we first establish the relationship between ENSO and IOD phases, with Chiredzi seasonal rainfall terciles. The following Tables (4, 5, 6, 7, 8) show the coincidences of the different phases of ENSO and IOD with Chiredzi district seasonal rainfall terciles. The analysis of the relationship between El Nino and/or positive IOD and the observed rainfall terciles presented in Table 4 reveals that the occurrences of at
123
Nat Hazards (2012) 64:1187–1207
1197
Table 4 The coincidences of the warm phases of ENSO and/or IOD with the rainfall terciles IOD phase
ENSO phase
OND tercile
JFM tercile
1967/1968
Positive IOD
Neutral
BN
BN
1968/1969
Negative IOD
El Nino
NN
NN
1969/1970
Positive IOD
El Nino
AN
BN
1972/1973
Positive IOD
El Nino
BN
BN
1976/1977
Positive IOD
El Nino
NN
AN
1977/1978
Positive IOD
El Nino
AN
AN
1979/1980
Negative IOD
El Nino
AN
NN
1982/1983
Positive IOD
El Nino
NN
BN
1986/1987
Positive IOD
El Nino
NN
BN
1987/1988
Positive IOD
El Nino
AN
BN
1991/1992
Positive IOD
El Nino
BN
BN
1994/1995
Positive IOD
El Nino
BN
NN
1997/1998
Positive IOD
El Nino
BN
NN
0.38 %
54 %
Probability of drought
Table 5 Rainfall tercile coincidences of negative IOD
IOD
ENSO
OND
JFM
1968/1969
Negative IOD
El Nino
NN
NN
1970/1971
Negative IOD
La Nina
NN
BN
1971/1972
Negative IOD
La Nina
NN
AN
1974/1975
Negative IOD
La Nina
NN
NN
1975/1976
Negative IOD
La Nina
NN
AN
1979/1980
Negative IOD
El Nino
AN
NN
1980/1981
Negative IOD
Neutral
NN
AN
1983/1984
Negative IOD
La Nina
AN
NN
1988/1989
Negative IOD
Neutral
NN
BN
1989/1990
Negative IOD
Neutral
NN
AN
1995/1996
Negative IOD
Neutral
BN
NN
1996/1997
Negative IOD
La Nina
NN
AN
92 %
83 %
Probability of non drought
Table 6 Rainfall tercile coincidences of neutral IOD and La Nina
IOD
ENSO
OND
JFM
1983/1984
Neutral
La Nina
BN
NN
1988/1989
Neutral
La Nina
NN
BN
1995/1996
Neutral
La Nina
BN
AN
1999/2000
Neutral
La Nina
BN
AN
75 %
25 %
Probability of drought
least El Nino and/or positive IOD events phases are not strongly connected to below normal rainfall (38 % for OND and 54 % for JFM, respectively), but are better linked compared to climatology (more than 33.3 %). All the three worst (extreme) droughts during the period
123
1198 Table 7 Rainfall terciles for cooperating ENSO and IOD neutral events
Nat Hazards (2012) 64:1187–1207
IOD
ENSO
OND
JFM
1966/1967
Neutral
Neutral
BN
AN
1978/1979
Neutral
Neutral
AN
AN
1981/1982
Neutral
Neutral
BN
NN
1985/1986
Neutral
Neutral
NN
NN
1990/1991
Neutral
Neutral
NN
NN
1993/1994
Neutral
Neutral
AN
AN
33.3 %
0.0 %
Probability of drought
Table 8 Events in the period for the forecast model validation
IOD
ENSO
OND
JFM
2000/01
Neutral
La Nina
NN
NN
2001/02
Negative IOD
Neutral
AN
BN
2002/03
Positive IOD
El Nino
NN
AN
2003/04
Negative IOD
Neutral
AN
NN
2004/05
Neutral
El Nino
BN
BN
2005/06
Neutral
La Nina
BN
NN
2006/07
Positive IOD
El Nino
BN
NN
1967–2000 (1967/1968, 1972/1973, 1991/1992) occurred at least during the positive IOD events, and both sub-seasons of these years had below normal rainfall. However, it is apparent that the extreme droughts did not occur in the presence of either La Nina or negative IOD events. In addition, we note that the only events when both sub-seasons did not have droughts are when there was a negative IOD event and that above normal rainfall is rare during the warm phases of either ENSO or IOD (23 % for OND and 15 % JFM). The coincidences of negative IOD (irrespective of ENSO state), with the rainfall terciles shown in Table 5, illustrate that there is a relatively high chance that no drought occurs when the season has a negative IOD event (92 % for OND and 83 % for JFM, respectively). At the same time, the presence of the negative IOD seems to suppress completely the occurrence of back to back droughts for both OND and JFM. Although the sample in Table 6, which indicates the coincidences of La Nina (without negative IOD), with the rainfall terciles is small, it is most likely that OND will have drought (75 %), but with JFM having less chances of developing a drought (25 %). However, there are no chances of a simultaneous drought for OND and JFM, and no probability of an extreme drought. When the conditions in both the equatorial Pacific and Indian Ocean are not anomalous (absence of both ENSO and IOD extreme events), Table 7 indicates the absence of drought during JFM but relatively less chance during OND (33.3 %). At the same time, the chances of a drought for both OND and JFM including an extreme drought are absent in these normal conditions. 3.6 The development of a seasonal rainfall forecasting model From the above observations that are based on the period 1967/1968–1999/2000, the following model to forecast the Chiredzi OND and JFM season can be constructed for known ENSO and IOD phases during May/June. It has to be noted that these concurrent
123
Nat Hazards (2012) 64:1187–1207
1199
Fig. 4 Chiredzi district seasonal rainfall forecast model for predetermined ENSO and IOD phases. The ENSO and IOD phases can be determined with some acceptable degrees of accuracy in either May or June
ENSO and IOD phases used above can be determined with some acceptable degrees of accuracy in either May or June. In Fig. 4, we illustrate the nonparametric model that uses ENSO and IOD phases as inputs and produces a two tercile confidence forecast as outputs. In this model, all the possible combinations of the ENSO and IOD phases are taken into consideration, and the most likely two rainfall terciles to emanate from the resulting physical atmospheric conditions thus created. For validation purposes of the model, we used the events during period 2000/2001–2006/ 2007 (Table 8) that did not participate in the forecast model construction. For 2000/2001 and 2005/2006, the events correspond to row number two in the model (Fig. 4), and the description suits both observed rainfall. The same applies to the seasons 2001/2002 and 2003/2004, which correspond to row number one in the model, and has a description that also suits their observed rainfall. However, there is a miss in the season 2002/2003, which corresponds to row number two in the model. Here, the opposite of the expected occurred. The period 2004/2005 corresponds to row number four in the model, and there is a hit with the observed rainfall, which shows that it was an extreme drought. Having a single miss in the validation period of seven seasons is a sure sign that the new model has an improved skill and can work. 3.7 The applications framework of the forecast model The realization of the social benefits from the emerging rainfall forecasting model capabilities is derived from the forecast application, rather the forecasts themselves (Acher
123
1200
Nat Hazards (2012) 64:1187–1207
et al. 2007). It is fortunate that while the Chiredzi rural farming community confidence in the national weather service was exceptionally low (17 %), confidence in Department of Agriculture and Extension Services (Agritex) extension officers appeared relatively high (93 %). It was quite apparent that these farmers worked closely with the extension workers and strongly believed that extension agents were reliable. This trust is even more cemented by the fact that Agritex officers live in the area and are believed to be successful local farmers. They speak the local languages and have integrated themselves within the community. Given this background, it is most probable that confidence in the new forecasting model will increase and become a significant factor in the small-scale farm management practices, if consistently disseminated through extension agents. The rainfall prediction in the study is based on a nonparametric forecasting model that uses two terciles as its output. Thus, the application framework consists of how this model is used in conjunction with the Chiredzi rainfall tercile characteristics, preferably by Agritex extension agents, to come up with usable information for OND and JFM by farmers. Fig. 5 gives the background and information on how the model is used. We have noted during the development of the forecasting model that certain terciles of sub-seasonal rainfall are more likely to manifest under a defined combination of ENSO and IOD events. It is therefore expected that these defined event combinations of the climatic modes should impose certain common intra-seasonal characteristics in the years in which they strongly influence. Thus, these average characteristics can easily be determined by aggregating all finer rainfall details of the OND and JFM seasons for the three terciles. It is then assumed that the expected temporal distribution of the rainfall within each tercile can be detected through the expression of rainfall producing/suppression mechanisms as imposed by any defined combination of the two major tropically based climate modes over the sub-region (e.g., Manatsa et al. 2008; Manatsa and Mukwada 2012). In Fig. 6a–c, we present the intra-seasonal character of Chiredzi seasonal rainfall as shown for the 3-day averages, for all the years when the rainfall was above normal (a), near
Fig. 5 The application framework for the Chiredzi tailored seasonal rainfall Forecast
123
Nat Hazards (2012) 64:1187–1207
1201
Fig. 6 Temporal distribution of 3-day rainfall totals for a above normal; b near normal and c below normal rainfall years
123
1202
Nat Hazards (2012) 64:1187–1207
normal (b) and below normal (c). Using these figures, it is possible to derive the multiple characteristics for each of the three rainfall terciles. For example, it is possible to approximate the amount of expected rainfall, the projected intra-seasonal distribution, onset and cessation dates, the location and intensity of mid-season dry spell, the expected intensity of rainfall, and the general locations of wet spells. For example, in the above normal rainfall season (Fig. 6a), we note that OND has an average of 292 mm and JFM has 395 mm with the above normal season, averaging 687 mm, and consistent rainfall starting on day 45 (mid-November). However, the first week of December (day 63–69) is relatively dry, and a mid-season dry spell is located in January with less than 20 mm of rainfall total in the first and third decades of the month. It becomes relatively dry from the second week of March (less than 20 mm), thus marking the general termination of the season. On the other hand, the OND sub-season for a near normal season has an average of 188 mm, and JFM has 291 mm, with the total season having 479 mm. We note that consistent rainfall starts during the third week of November and terminates at the end of February. In this particular tercile, there is no distinct dry spell within the season, and the rainfall intensity is moderate, with the rainfall rarely exceeding a 25 mm total in 3 days. Lastly, the below normal rainfall season has the OND season having an average of 105 mm, and JFM has 142 mm, with the season having a total of 347 mm. In this season of notable rainfall deficit, the consistent rainfall starts late during the second week of December and ends during the first week of March, thus making the season length the shortest of the terciles. However, the tercile properties provided are not exhaustive, other relevant distinctiveness of each tercile can also be derived through careful scrutiny of the given graphs. With expected tercile rainfall amount and distribution characteristics readily available, it is possible to strategize so as to maximize opportunities and minimize climate risks, since farm management decisions are framed on the basis of the length of growing season and the quality of rainfall season. At the same time, in any given year, the outcome of the onset of the rains can assume any of the following three definitions, normal, early and late. In a similar fashion, seasonal rainfall outcomes can be near normal, below normal, above normal and extremely wet or dry. It also follows that the possible drought outcomes include, early season, mid-season, terminal, seasonal and extreme drought. Since almost all these important rainfall events within the season are generally captured in the tercile information provided in Fig. 6a–c, by anticipating these inter- and intra-seasonal events in any given season, a farmer has an opportunity to maximize opportunities or minimize risks associated with the ‘weather within the climate’ for that season, depending on strategies deployed. For example, when at least a negative IOD is anticipated, the farm preparation should go in line with averaged seasonal rainfall characteristics depicted in Fig. 6b and c, but with more of near normal expectations for both OND and JFM. Neither an extreme drought nor drought that affects both OND and JFM should be anticipated. Staggered planting dates should take into consideration onsets and cessation dates in both figures. Average seasoned crops may be grown. However, choice of plant varieties should take into consideration the location of the mid-season dry spells so that they do not coincide with the most critical moisture demand stages of plant growth like tussling. Weeding can be done in conjunction with the more reliable 10-day forecasts issued by the Met Office (e.g., Klopper 1999). The rainfall should be enough for good pastures and drinking water for livestock; hence, restocking could be considered. The whole season should be generally good. In the event that the IOD is in a neutral state but in the presence of La Nina, then it is most likely that the first part of season could be dry but with a better performing second half. Late planting should be considered for a bigger portion of the field after taking into consideration the rainfall amount and distribution depicted in Fig. 6c. Supplementary
123
Nat Hazards (2012) 64:1187–1207
1203
feeding including zhombe plant (indigenous cattle feed used during drought) use should be ready for the first part of the season to enable cattle enough energy to plough in case of possible late onset of the rains that replenish the pastures. No preparation for a wholesale drought should be done as extreme droughts, and droughts that affect both OND and JFM are not expected. In the case that both the conditions in the tropical Pacific and Indian Ocean are normal, that is, a situation devoid of ENSO and IOD extremes, the second part of the season is bound to be good with higher chances of above normal rainfall. However, the first part may have below normal rainfall. Neither extreme droughts nor droughts that affect both JFM and OND are expected. Farm preparation should be for deficit OND, but in anticipation of a far better JFM conditions. The remaining combinations of ENSO and IOD phases seem not to be conducive for triggering favourable rainfall conditions. The rainfall during these events is very erratic. A near normal to above normal season is rare, but possible. These are the occasions when all the extreme droughts and all the situations with drought for both OND and JFM during the study period occurred. This calls for minimum investment in rain-fed farming. Farm preparations should be done mostly according to the rainfall distribution, illustrated in Fig. 6c. Here, even drought resistant varieties of maize may fail with this type of rainfall distribution and amount (e.g., Phillips et al. 2001), hence the need to consider growing small grains over the much bigger portion of the fields. Destocking should be considered and make preparation for supplementary feeding. Generally, we prescribe conservative farming action, which encourages risk-averse behaviour in farmers and so that even if in error, they will only forfeit minimal benefits (e.g., Orlove et al. 2004; O’Brien and Vogel 2003). A higher than expected rainfall will still sustain normal crop growth for the drought resistant crops and is even bound to produce unexpectedly higher yields. Thus, farm preparations should be for a drought but with some bias towards a near normal season. These are just guiding principles derived directly from the forecast model, but some modifications can be done by the extension officers in accordance with the local area and indigenous knowledge, to provide more relevant information for farmers’ planning purposes. 3.8 Advantages of the tailored forecast over the Met Office forecast It is apparent from the analysis of the current official seasonal forecast that it inherits a considerable risk of misinformation on the coming season’s climate. The high risk ramifications of applying these rainfall forecasts are evident as already been illustrated in the Table 3. On the other hand, it is clear that the newly developed forecasting system has several advantages over the official rainfall forecast. The new forecast system has a longer lead time as the anticipated ENSO and IOD phases can be determined with some degree of confidence in May/June, 3 months before the release of the official forecast from the Met Office in August/September. Its simplicity is so overwhelming that a site-specific forecast model can easily be replicated countrywide using station data. The trained extension officer only needs to know the states of the IOD and ENSO phases, in order to use the developed site-specific model and advise the farmer accordingly, on different options of applying the forecast. The new forecast eliminates the unnecessary persistent nightmare of deciphering the probabilities that characterize the convectional forecast, before pondering on which farming decision to make. It also has the capacity to forecast droughts with higher confidence, rather than subjecting the small-scale farmers to unreasonable levels of operational risk by ignoring the drought forecasts as in the official forecasts. The communication hurdles, that dominated the convectional forecast system dissemination process, were overcome as illustrated by the flow chart depicted in Fig. 7. The
123
1204
Nat Hazards (2012) 64:1187–1207
Fig. 7 Proposed Institutional mechanism for the climate information management in Chiredzi district
department of Agritex simply provides its extension officers with predicted ENSO and IOD events, from global climate prediction centres on the Internet websites (//www.esrl.noaa.gov and //www.jamstec.jp, respectively). These Agritex officers then generate the seasonal rainfall forecast from the input of ENSO and IOD phases into the ‘Two tercile forecast nonparametric model’. They further employ the generated forecast together with the tercile rainfall information explained above, and other information including indigenous knowledge systems of the area to adapt and transform the forecast into an applicable format for the smallscale farmers at ward level.
4 Conclusions and recommendations The possibility of developing a tailored seasonal climate forecasts to Zimbabwe smallholder farmers’ needs has been investigated. The primary aim of this study was to tailor ENSO-based seasonal climate forecasts to small-holder farmers’ needs by cooperating the IOD events. This process culminated from the realization that the seasonal rainfall
123
Nat Hazards (2012) 64:1187–1207
1205
forecasts issued by the Met Office Department through the SARCOF process have limited application value to the Chiredzi farmers. The local farmers are largely interested in a binary seasonal climate forecast with a longer lead time that provides information of whether a drought will occur in a given season. This is a major departure from the official tercile forecasts which have an inherent poor forecast skill and inadequate lead time. Devastating recurrent droughts recorded throughout Zimbabwe in recent decades probably due to climate change have never been in forecast, and the high confidence in near normal conditions is almost perpetual. Consequently, benefits from this forecast drop considerably, as farmers are bound to be hit by all the recurrent droughts unexpectedly with unmitigated impacts, and the opportunity of benefiting from correct above normal rainfall forecasts is substantially reduced. The dissemination process is very complicated and long, resulting in the intended user receiving the forecast information late, and in most cases distorted. In addition, the probabilistic nature of the forecast renders it difficult to interpret by the farmers. In response, this study developed a new product with an adaptation opportunity for the small-scale farmers. The new tailored forecast product does not only provide better forecasting skill of the seasonal total, but gives indications of the intra-seasonal distribution of the rainfall including onsets, cessations, and wet and dry spell locations for specific terciles that are in forecast. The lead time gives ample time for the farmer to prepare well before the onset of the rains. Its simplicity in structure is remarkable and hence renders it relatively easy to use. The participatory processes used in this research enabled the tailored forecast tool to be developed and applied in ways useful to farmers. Hence, the developed tailored seasonal climate forecast could be one way to enhance management of risks and opportunities in rain-fed agriculture among small-holder farmers not only in Zimbabwe but also throughout the SADC region where the impact of both ENSO and IOD is significant. The model inputs, which are the states of the ENSO and IOD events, are simply derived from the internet and communicated directly to the extension workers. The model output normally discounts the possibility of occurrence of a third tercile; hence, it is more precise than the convectional forecast which gives probabilities to all three terciles. This becomes the main added advantage over the convectional forecast as the tailored forecast now carries with it, predominantly binary forecast information desperately needed by local farmers—whether a drought will occur in a given season. This makes the tailored forecast easier for farmers to understand and act on compared to the tercile probabilities. For example, neither extreme droughts nor other forms of droughts that affect both the OND and JFM can be anticipated in the presence of either a negative IOD event and/or La Nina or when both the tropical Pacific and Indian Ocean sea surface temperatures are neutral. It is recommended that, since the Agritex extension officers are the ones best positioned to effectively communicate with the farmers, this seasonal rainfall forecast model can be availed to them, so that they can generate the forecast themselves by simply using the ENSO and IOD phases as inputs. The model outputs are then used in conjunction with the rainfall tercile characteristic charts, to transform the forecast into applicable farming information for the Chiredzi farmers. In this way, the communication chain is shortened and principally managed by one department, with the application options coming directly from the forecast generators, who are readily available and accessible to every farmer. Although the novel approach demonstrated in this study seems to be effective, it is errant to assume an identical procedure conducted for other regions will entail improved benefits. However, given the current success (despite the limited model validation and testing), exploration is warranted. Some locations that are susceptible to high climate variability where low forecast skill is apparent are ripe for investigation. However, the
123
1206
Nat Hazards (2012) 64:1187–1207
forecast model presented here is generally grounded on stationary climate principles. Multi-decadal effects or climate change that are paramount to the current study’s bigger national project of ‘Coping with Drought and Climate Change Project’ are not explicitly addressed here. On the other hand, while tailoring the seasonal forecasts this way is clearly beneficial, incentives to improve the convectional SARCOF forecast emanating from the Met Office Department to potentially draw even greater returns are undeniable. Acknowledgments Funding for this research has been provided through UNDP sponsored ‘Coping with Drought and Climate Change Project’ in Zimbabwe. Advice from the two reviewers was very instrumental in shaping the publishable state of the paper. The author participated in the SEI-ISDR-UNU Writeshop in February 2011 and acknowledges the valuable support, especially from Professor G. Kranjac-Berisavljevic. Bindura University is also thanked for partial funding and providing research facilities.
References Acher E, Mukhala E, Walker S, Dilley M, Masamvu K (2007) Sustaining agricultural production and food security in southern Africa: an improved role for climate prediction? Clim Change 83(3):287–300 Eriksen S, O’Brein K, Losentrater L (2008) Climate change in eastern and southern Africa: impacts, vulnerability and adaptation. Report, Global Environmental Change and Human Security, p 2 Fussel HM (2007) Adaptation planning for climate change: concepts, assessments approaches, and key lessons. Sustain Sci 2(2):265–275 Jury MR, Enfield DB, Melice JL (2002) Tropical monsoons around Africa: stability of ENSO associations and links with continental rainfall. J Geophys Res C10(15):1–17 Klopper E (1999) The use of seasonal forecasts in South Africa during 1997–98 rainy season. Water SA 25:311–316 Manatsa D, Mukwada G (2012) Rainfall mechanisms for the dominant rainfall mode over Zimbabwe relative to ENSO and/or IODZM. Sci World J. doi:10.1100/2012/926310 Manatsa D, Chingombe W, Matsikwa H, Matarira CH (2007) The superior influence of the Darwin sea levelpressure anomalies over ENSO as a simple drought predictor for southern Africa. Theor Appl Climatol. doi:10.1007/s00704-007-0315-3 Manatsa D, Chingombe H, Matarira CH (2008) The impact of the positive Indian Ocean dipole on Zimbabwe droughts. Int J Climatol. doi:10.1002/joc.1695 Manatsa D, Nyakudya IW, Mukwada G, Matsikwa H (2011) Maize yield forecasting for Zimbabwe farming sectors using satellite rainfall estimates. Nat Hazards. doi:10.1007/s11069-011-9765-0 Martin RV, Washington R, Downing TE (2000) Seasonal maize forecasting for South Africa and Zimbabwe derived from an agroclimatological model. J. Appl Meterol 39:1473–1479 Mason JM, Goddard L, Graham NE, Yulaeva E, Sun L, Arkin PA (1999) The IRI seasonal climate prediction system and the 1997/98 El Nino event. Bull Am Meteorol Soc 80(9):1853–1873 O’Brien K, Vogel CH (2003) Coping with climate variability: user responses to climate forecasts in southern Africa. Ashagte, Burlington Orlove BS, Broad K, Pretty AM (2004) Factors that influence the use of climate forecasts. Bull Am Meteorol Soc 85:1–9 Patt A (2001) Understanding uncertainty: forecasting seasonal climate for farmers in Zimbabwe. Risk decis Policy 6(2):1–15 Phillips JG, Makaudze E, Unganai L (2001) Current and potential use of climate forecasts for resource poor farmers in Zimbwe. In: Rosenzweig C (ed) Impacts of El Nino and climate variability in agriculture. American Society of Agronomy, Special Publication 63, Madison Reason CJC, Jagadheesha D (2005) A model investigation of recent ENSO impacts over southern Africa. Meteorol Atmos Phy 89:181 Roncoli C, Ingram K, Kirshen P (2001) The costs and risks of coping with drought: livelihood impacts and farmers responses in Burkina Faso. Clim Res 19(2):119–132 Stanski HR, Wilson LJ, Burrows WR (1989) Survey of common verification methods in meteorology. World meteorological organisation techical report. 8, WMO/TD 358, p 114 Unganai LS, Kogan FN (1998) Drought monitoring and corn yield estimation in southern Africa from AVHRR data. Remote Sens Environ 63:219–232 Vogel C (2000) Usable science and: an assessment of long-term seasonal forecast among farmers in rural areas of South Africa. S Afr Geogr J 82:107–116
123
Nat Hazards (2012) 64:1187–1207
1207
Ziervogel G (2004) Targeting seasonal forecasts for integration into household level decisions: the case of smallholder farmers of Lesotho. Geogr J 170(1):6–21 Ziervogel G, Johnston P, Matthew M, Mukheibir P (2010) Using climate information for supporting climate change adaptation in water resource management in South Africa. Clim Chan. doi:10.1007/s10584009-9771-3
123