Can soil bunds increase the production of rain-fed lowland rice in south eastern Tanzania?

Transcription

1 agricultural water management 89 (2007) available at journal homepage: Can soil bunds increase the production of rain-fed lowland rice in south eastern Tanzania? D. Raes a, *, E.M. Kafiriti b, J. Wellens c, J. Deckers a, A. Maertens d, S. Mugogo b, S. Dondeyne a, K. Descheemaeker a a K.U. Leuven University, Faculty of Bioscience Engineering, Division of Soil and Water Management, Celestijnenlaan 200 E, B-3001 Leuven, Belgium b Naliendele Agricultural Research Institute, P.O. Box 509, Mtwara, Tanzania c DRAHRH-HB (Regional Board of Hydraulics and Agriculture of the Upper-Basins), 01 BP 3526, Bobo-Dioulasso, Burkina Faso d Department of Applied Economics and Management, Warren Hall, Cornell University, Ithaca, NY 14853, United States article info Article history: Accepted 18 January 2007 Published on line 13 March 2007 Keywords: Soil water balance model Yield response to water Rainwater harvesting Erratic rainfall abstract Rain-fed lowland rice is by far the most common production system in south eastern Tanzania. Rice is typically cultivated in river valleys and plains on diverse soil types although heavy soil types are preferred as they can retain moisture for a longer period. To assess the effects of soil bunds on the production of rain-fed lowland rice, the crop was cultivated in bunded and non-bunded farmers plots under the common agronomic practices in the region, in three successive seasons on Grumic Calcic Vertisols (Pellic). For the three seasons and for the two plot types, crop transpiration was simulated with the BUDGET soil water balance model by using the observed weather data, soil and crop parameters. Comparison between the observed yields and the simulated crop transpiration yielded an exponential relationship with a determination factor of 0.87 and an RMSE of 0.15 tonnes ha 1. With the validated soil water balance model crop yields that can be expected in bunded and non-bunded fields were subsequently simulated for wet, normal and dry years and various environmental conditions. Yield comparison shows that soil bunds can appreciably increase the production of rain-fed lowland rice in south eastern Tanzania in three quarters of the years (wet and normal years) when the soil profile is slow draining (K SAT equal to or less than 10 mm day 1 ). In normal years a minimum yield increase of 30% may be expected on those soil types. In wet years and when the soil hardly drains (drainage class of 0 5 mm day 1 ), the yield may even double. In dry years the yield increase will be most of the time less than 10% except for plots with a percolation rate of 0 5 mm day 1. # 2007 Elsevier B.V. All rights reserved. 1. Introduction South eastern Tanzania borders with the Indian Ocean to the east and Mozambique in the south. In the coastal area, the reference crop evapotranspiration (ETo) is fairly constant throughout the year ranging from 3.5 mm day 1 in the dry months to 5.5 mm day 1 at the start of the rainy season. The rainfall pattern is uni-modal (Fig. 1). December April are wet months and the period sets the length of the cropping season (Griffiths, 1972). Rice is next to maize a popular staple food crop in south eastern Tanzania and ranks third in importance for food * Corresponding author. Tel.: ; fax: address: dirk.raes@biw.kuleuven.be (D. Raes) /$ see front matter # 2007 Elsevier B.V. All rights reserved. doi: /j.agwat

2 230 agricultural water management 89 (2007) Fig. 1 Average 10-day reference evapotranspiration (ETo, line) and 10-day rainfall (bars) for various probabilities of exceedance for Mtwara in the coastal area of south eastern Tanzania. security after maize and cassava (Ministry of Agriculture, 1992). About 94% of the crop is cultivated on smallholdings of about ha. Rain-fed lowland rice is by far the most common production system. In the lowlands rice is typically cultivated in river valleys and plains on diverse soil types although heavy soil types are preferred as they can retain moisture for a longer period. Due to increasing demand for rice consumption, an average of 10 25% of the total consumption is imported annually to cover the shortfall (Kafiriti, 2004). One of the main problems associated with rain-fed lowland rice in south eastern Tanzania is the erratic rainfall pattern (De Pauw, 1989). The onset, cessation and length of the rainy season, as well the occurrence of long dry spells during the growing season, are extremely variable (Kafiriti et al., 2001). When rainfall is low in a particular decade, it covers only about half of the 10-day ETo (Fig. 1). Hence crop water stress and yield depression occurs often during the rainy season. Even in years with abundant rainfall, average rice yields in farmers fields are only 2 tonnes ha 1 as a result of dry spells and the absence of fertilizers applications. In normal years on average 1 tonnes ha 1 rain-fed rice is produced, and in dry years the rice yield drops to 0.5 tonnes ha 1 or farmers might even fail to harvest (URT, 2000). Kafiriti et al. (2003) reported average rice yields of 3.2 tonnes ha 1 in smallholdings when fertilizers and irrigation are applied. This is still about 50% of the 6 tonnes ha 1 that can be obtained under fully controlled conditions in experimental stations. To reduce crop water stress during the growing season, there have been calls by extension services to promote onfarm rainwater harvesting techniques. Such techniques consist of retaining surface runoff rainwater within the field thereby altering the soil water status within the root zone (Lal, 1995). In rice fields the technique consists of building soil bunds around the fields to store excess rainwater. As bunds around the fields limit surface runoff and therefore raise the efficiency of the rainfall, one would expect a positive effect on the yield of rice. A Benefit Cost analysis (Senkondo et al., 2004) indicates that rice production with rainwater harvesting is profitable. By considering the magnitude and frequency of the rainfall storms in the region, Maertens (1999) found that 0.15 m is the theoretical optimal height for the soil bunds. However given the roughness of the soil surface, the Gilgay micro relief of the land and the risk of damage by excess rainfall, a height of 0.3 m is recommended for plots of maximum 0.10 ha. Although the extension service has started to promote bunds in the region, the practice is not yet adopted by most of the farmers because of lack of awareness of the importance of bunds in relative flat areas, specific guidelines for where to apply it and the labour involved in the construction of the bunds. No information is yet available for which environmental conditions soil bunds might be effective. To understand the profitability of soil bunds on the production of rain-fed lowland rice in south eastern Tanzania an experiment was set up in the smallholdings in the region. Rice was produced in bunded and non-bunded farmers plots in three successive seasons. In this paper the results are analyzed and used to calibrate and validate a simulation model. Rice yields that can be expected in bunded and nonbunded fields for various environmental conditions were subsequently simulated with the calibrated model. The results are currently used to formulate guidelines for farmers. 2. Material and methods 2.1. Classification of years Long series of daily rainfall were collected for three locations in the region (Table 1). To classify the years in wet, normal and dry years, the total rainfall during the five wet months (December April) was analyzed with the RAINBOW software (Raes et al., 1996, 2006b). Although series of wet years alternates with series of dry years in cycles of about 10-years, the homogeneity test revealed that the analysed periods can be considered as more or less homogeneous. The statistical tests in RAINBOW revealed that the total rainfall for Lindi and Masasi and the square root of the total rainfall for Mtwara during the five wet months are normal distributed. By means of a frequency analysis the total rainfall that can be expected with various probabilities of exceedance in the wet months were estimated (Table 2) Observations in farmers fields In the lowland close to Mkwaya ( E, S, 40 masl), rain-fed rice is cultivated for over 20 years in a large number of small plots of m 2. In the growing seasons 2002/2003, 2004/2005 and 2005/2006 about 30 farmers scattered over the cultivated area were selected. Each farmer was requested to cultivate rice in a bunded and a non-bunded plot by respecting identical agronomic practice. The average bund height was 0.30 m. Water from the non-bunded plots, was drained to plots which did not belong to the experimental set-up. Land preparation with a hand hoe started typically in October. Ploughing in of crop residues and weeds was done in wet fields. When the fields are still too dry to plough, the residues

3 agricultural water management 89 (2007) Table 1 Location of the climatic stations, the average annual reference evapotranspiration (ETo) and rainfall with indication of the number of years (N r ) with complete daily rainfall records for the wet months (December April) Station Location Latitude Longitude Altitude (m a.s.l.) ETo (mm) Rain (mm) Period N r Lindi Coastal S E Mtwara Coastal S E Masasi Inland S E Table 2 Expected total rainfall in the five wet months (December April) for various probabilities of exceedance Station Expected total rainfall (mm) in the five wet months (December April) 10% a (1 b ) 20% a (2 b ) 30% a (3 b ) 40% a (4 b ) 50% a (5 b ) 60% a (6 b ) 70% a (7 b ) 80% a (8 b ) 90% a (9 b ) Lindi Mtwara Masasi Type year Wet Normal Dry a Probability of exceedance in percentage. b Probability of exceedance in years out of 10. and weeds were burned and ploughing was only done superficially. The sowing was largely determined by the onset of rains, and varied from early December in 2004 to early January in The major rain-fed rice variety grown in the area is Supa with a growing cycle of 5 months. No fertilizers were applied since farmers believe that their soil is fertile enough. The main period of harvest coincides with the start of the dry season and ranged from the end of May in 2003 and 2005 to the end of June in 2006 depending on the sowing date. Farmers harvested the crop by cutting the plants at the base and then threshed in the field. The average size of the plots and the observed yields are presented in Table 3. The high standard deviation of the yield is typical for farmers plots and results from differences in location and bund heights, soil percolation rates, maintenance of bunds during the season and agronomic practices (date of sowing and weeding). A paired sample t-test revealed that observed yields in bunded plots were significantly higher than yields in unbunded plots in two out of 3 years (Table 3). Rainfall was recorded daily in Mkwaya. When compared with the long term rainfall of the December April period for Lindi (located 20 km North), the 2002/2003 season was dry (probability of exceedance of 84%), 2004/2005 was normal (probability of exceedance of 51%), and 2005/2006 was extremely wet (probability of exceedance of less than 1%). The monthly weather data for the three seasons is presented in Table 4. The mean monthly reference evapotranspiration (ETo) in Table 4 was calculated with mean monthly climatic data of the location as available in the New-LocClim software (FAO, 2005). In the absence of air humidity, radiation and wind speed, ETo was estimated with the FAO Penman Monteith equation with the minimum (T min ) and maximum (T max ) air temperature according to the calculation procedures outlined by Allen et al. (1998), whereby dew point temperature is estimated from T min and solar radiation from the difference between T max and T min. The calculation procedures were validated for the area by Kafiriti et al. (2001). The soils in the lowland are dark heavy cracking soils, classified as Grumic Calcic Vertisols (Pellic) (Kafiriti, 2004; IUSS Working Group WRB, 2006). The soil water contents as reported by Dondeyne and Kafiriti (2000) are 54 vol.% at saturation (u SAT ), 47 vol.% at field capacity (u FC ) and 32 vol.% at wilting point (u WP ). The corresponding total available water (TAW) is 150 mm per meter soil depth. During periods of heavy rainfall, excess rainwater could be stored for several days between the bunds. Graduated rulers were installed in the bunded plots in the growing seasons 2004/2005 and 2005/2006. The average recorded decline of ponded water during periods of no rainfall was 14 mm day 1 (standard deviation of 5 mm day 1 ). By considering an average rice evapotranspiration of 5 mm day 1, an average percolation rate through the saturated soil (K SAT ) of 9 mm day 1 can be considered Simulations Simulations were run with the soil water balance model BUDGET (Raes et al., 2006a). BUDGET is composed of a set of validated subroutines describing the various processes Table 3 The number of farmers (N r ) involved in the successive trials, the average plot area, the growing period, the average observed rice yields of bunded and non-bunded plots (standard deviation reported between parentheses), and the p-value of a paired sample t-test for the three seasons Season N r Plot area (m 2 ) Growing period Rice yield (Mg ha 1 ) p Bunds No bunds Bunds No bunds 2002/ (138) 212 (222) January May (1.094) (0.293) / (179) 214 (121) December May (0.656) (0.638) / (184) 204 (120) January June (1.588) (1.102) 0.001

4 232 agricultural water management 89 (2007) Table 4 Mean monthly reference evapotranspiration (ETo), and monthly rainfall for Mkwaya for the three seasons Month ETo (mm) Rainfall Season 2002/2003 Season 2004/2005 Season 2005/2006 mm Events mm Events mm Events December January February March April May June Total rainfall: December April 587 mm 789 mm 1338 mm involved in water extraction by plant roots and soil water movement in the absence of a water table. The water stored in the root zone is determined on a daily basis by keeping track of incoming and outgoing water fluxes at its boundary. By considering the reference crop evapotranspiration (ETo), the soil evaporation rate and crop transpiration rate (T c ) of a wellwatered soil is calculated with the help of the dual crop coefficient procedure as presented by Allen et al. (1998). When the soil water content in the root zone drops below a threshold value, water stress occurs and the actual transpiration rate (T a ) drops below T c. Water stress will occur when a certain amount of the plant extractable water is depleted from the root zone. This amount, the so-called readily available soil water (RAW), is expressed as a fraction ( p) of TAW (total available soil water, i.e. the amount of water available in the root zone between field capacity u FC and wilting point u WP ). Given the simulated soil water content in the root zone, the actual crop transpiration (T a ), relative crop transpiration (T a /T c ) and the corresponding crop water stress (1 T a /T c ) is determined for each day of the growing period. In the simulations, the observed daily rainfalls in each of the seasons were used as rainfall data. Given the minimal variation in evapotranspiration between years and in the season, the simulations were run with the mean monthly ETo (Table 4). As a result of the incorporation of organic matter and land preparation at the beginning of the season, rainwater can be stored between the roughness of the soil surface and in the topsoil before it slowly infiltrates in the soil profile. In BUDGET this was described by putting a lighter clay layer (u SAT =52vol.%; u FC =44vol.%; u WP =23- vol.%) of 0.1 m on top of the heavy clay soil. This extra layer represents the soil roughness and the more sandy characteristics of the top layer of the soil profile. By considering a K SAT of 35 mm day 1 for this top layer, rainfall could be stored in the topsoil when the soil was not too wet by previous rainfall. In the simulations, plots with and without bunds were considered. In the plots without bunds runoff was simulated by means of the Curve Number (CN) method developed by the US Soil Conservation Service (United States Department of Agriculture, 1964; Rallison, 1980). A default CN value of 80 was considered (the selected CN value is adjusted by the model during the simulation run to the varying wetness of the topsoil). In fields with bunds, runoff is only considered when the surface storage exceeds the bund height of 0.30 m. Rainwater stored between bunds will infiltrate on the successive days until all stored water is percolated or lost by evaporation. A single sowing day for each of the season was considered in the simulation runs and was selected as the day when successive large showers would have wetted sufficiently the top soil. Since farmers do not sow before December, the first of December was considered as the initial search date. The obtained sowing dates (1 January 2003, 10 December 2005, and 10 January 2006) were typical for the three seasons and observed in the plots. As a result of rains before sowing it was assumed that the top soil (0.1 m) at sowing was soaked (u = 52 vol.%) and the soil water content in the subsoil ( m) was at 50% of TAW (u = 39.5 vol.%). Next to the crop coefficients for rice (Table 5) as reported by FAO (Allen et al., 1998), the model requires an effective rooting depth (Z r ) and a depletion factor ( p) for the simulation of the soil water balance of the cropped soil. The effective rooting depth refers to the soil depth in which most of the roots are concentrated. A Z r of only 0.3 m was considered and was confirmed by sampling of the plots. Doorenbos and Kassam (1979) report that when the water content of the soil decreases to 70 80% of the saturated value, rice yields begin to decline. This corresponds with a p value of about 0.2 in the considered soil type ( p = 0.11 in the top 0.1 m, and 0.27 in the lower part of the root zone). Given the simulated mean water stress throughout the season (1 T a /T c ), the resulting depression in yield is estimated in BUDGET by means of the functional model presented by Doorenbos and Kassam (1979), describing the relation between seasonal water stress and the corresponding expected yield: 1 Y a Y m ¼ K y 1 T a T c Table 5 Crop coefficient (K c ) for rice (source Allen et al., 1998) Crop growth stage Length (days) K c 1. Initial stage (dry soil surface) 1.10 (wet soil surface) 2. Crop development 30...! Mid-season stage Late season stage (1)

5 agricultural water management 89 (2007) where Y a /Y m is the relative yield, and (1 Y a /Y m ) the relative yield decrease. The response of yield to water stress for a given environment, is quantified through the yield response factor K y. Since crop yield is proportional to the relative crop transpiration, the observed yields in the bunded and non-bunded plots in Mkaway were used to calibrate the BUDGET model by selecting an appropriate seasonal K y coefficient. With the calibrated model, rice yields that can be expected in bunded and non-bunded farmers fields in the lowlands on the above described Grumic Calcic Vertisols (Pellic) were subsequently simulated for various environmental conditions: for four different percolation rates that can be expected in the heavy clay soil (K SAT is (a) 2.5 mm day 1, (b) 7.5 mm day 1, (c) 12.5 mm day 1, and (d) 17.5 mm day 1 ); for two locations in south eastern Tanzania. Of the analysed stations Lindi and Mtwara were selected. Although both stations are in the coastal zone, Lindi is representative for a station some 15 km inland as a result of the shape of the coastline, while Mtwara is a typical coastal station. The difference in location is reflected by their differences in ETo, which is higher in Lindi, and rainfall which is higher in Mtwara (Table 1); for the five (Mtwara) to six (Lindi) dry years, the six to eight normal years and the five to six wet years of the series (Table 1) for which daily rainfall data were available. For each of the years and for each of the locations, the sowing date (first day of simulation run) was determined according to criteria presented by Sivakumar (1988) and Stewart (1988). Once the cumulative rainfall (after 1 December) within a 10-day period is equal to or greater than 30 mm, rice was assumed to be sown. Wellens (2000) validated the criterion for the case of Mtwara and Masasi. As described above, an initial soil wetness of 52 vol.% in the top 0.1 m of the soil profile and of 39.5 vol.% in the remaining part of the root zone was assumed at the start of each simulation run. Fig. 2 Observed yields with indication of the standard deviation vs. simulated relative transpiration for rain-fed lowland rice cultivated in Mkwaya in the growing season 2002/2003, 2004/2005, and 2005/2006 in bunded (circle) and non-bunded (diamonds) plots. A linear and an exponential trend line are plotted. A good agreement (Fig. 2) is obtained between observed yields (Y a ) and the simulated mean seasonal relative transpiration (T a /T c ). The linear trend line (Y a = (T a /T c ) ) describing the relationship between crop transpiration and yield has a determination factor (R 2 ) of The slope of the line corresponds with a seasonal yield response factor (K y )of 1.7 (Eq. (1)) and refers to a very sensitive crop to water stress as can be expected for rice. The maximum yield (Y m ) that can be expected when no water stress occurs is according to the linear relationship 2.3 tonnes ha 1. At a relative transpiration of 41%, yield can no longer be obtained. The linear trend line confirms the observations in the region and responds to the relationship between transpiration and yield described by Doorenbos and Kassam (1979). Although with the ANOVA test on the residuals no significant difference is found between the linear and the exponential trend line (R 2 of 0.875), the RMSE drops from 0.25 to 0.15 tonnes ha 1 when the exponential trend line (Y a = exp(4.2225(t a /T c ))) is considered. The maximum yield of 3.1 tonnes ha 1 obtained with the exponential relationship is in correspondence with the 3.2 tonnes ha 1 reported by Kafiriti et al. (2003) in irrigated paddy fields in the region. Average yields that can be expected as simulated with the BUDGET model for various environmental conditions are plotted in Figs. 3 and 4. Due to periods of dry spells, seasonal soil-water deficit occurred in almost all simulation runs. The average relative transpiration ranged from 95% (in Mtwara, for wet years, in bunded fields, with soil drainage class a) to 58% (in Lindi, for dry years, in non-bunded fields, for all soil drainage classes). Since the exponential trend line better 3. Results and discussion Fig. 3 Expected average yield in bunded (light bars) and non-bunded (dark bars) plots of rain-fed lowland rice in Mtwara in farmers fields in wet, normal and dry years for different soil drainage classes (K SAT of 2.5 mm day S1 (a), 7.5 mm day S1 (b), 12.5 mm day S1 (c) and 17.5 mm day S1 (d)).

6 234 agricultural water management 89 (2007) yield gap between non-bunded and bunded fields drops to 15% for the c drainage class (K SAT of 0 15 mm day 1 ). In dry years the yield increase will be most of the time less than 10% except for plots with an a drainage class. Acknowledgements Fig. 4 Expected average yield in bunded (light bars) and non-bunded (dark bars) plots of rain-fed lowland rice in Lindi in farmers fields in wet, normal and dry years for different soil drainage classes (K SAT of 2.5 mm day S1 (a), 7.5 mm day S1 (b), 12.5 mm day S1 (c) and 17.5 mm day S1 (d)). describes the yield differences in bunded and non-bunded plots as long as the relative transpiration is not too small, the equation was used to convert the simulated mean seasonal relative transpiration to crop yield. The average coefficient of variation (standard deviation/mean) between the yield simulations of the 5 8 years for each type of year was 31% for Mtwara and 23% for Lindi. As can be observed from the graphs, the simulated yields were higher at the coast (Mtwara) than in the interior (Lindi) as a result of differences in the seasonal rainfall amount, the interval between showers, and the evapotranspiration rate. Thanks to the use of soil bunds the yield doubled in slow draining plots in the wet years. The yield difference reduces to 68% in Mtwara and 21% in Lindi in the dry years. The yield gap between non-bunded and bunded plots almost disappears when the plots are permeable (class d) and when the years are dry. 4. Conclusions From the field trials and the simulation results it is clear that, even in wet years, soil water deficits commonly occur during the rainy season in the lowlands in south eastern Tanzania. Improvement of rainwater management is therefore a strategy that should be promoted. Soil bunds which will keep the excess rainwater during several successive days on the fields help to reduce the soil water stress in the rain-fed rice production schemes, with higher yields as a result. The simulation results show that with rice produced on the heavy clay soils in the plains with an a or b drainage class (K SAT of 0 5 and 5 10 mm day 1 ), an appreciably increase in the production of rain-fed lowland rice in the region can be obtained in 75% of the years (wet and normal years). The yields might even be doubled in wet years for soils with an a drainage class. In normal years a minimum yield increase of 30% can still be expected on soils with a b drainage class. The The enthusiasm of the farmers in the participatory trials in Mkwaya was greatly appreciated. We would also like to thank Dr. S.H. Shomari, the Zonal Director for the Naliendele Agricultural Research Institute (NARI) in south eastern Tanzania, for strongly supporting the research. The active contribution of Naliendele ARI and the much appreciated financial contribution of the Flemish Inter-University Council of Belgium (VLIR South Initiative project ZEIN2004ZVVT24) allowed us to conduct the research. references Allen, R., Pereira, L.S., Raes, D., Smith, M., Crop evapotranspiration guidelines for computing crop water requirements. FAO Irrigation and Drainage Paper No. 56. Rome, Italy, 300 pp. De Pauw, E., Models of the climatic growing period: an inquiry into concepts, analysis and applications in regional agricultural planning. Doctoral dissertation. R.U. Gent, Gent, Belgium. Dondeyne, S., Kafiriti, E.L., Annual report no. 1. Irrigation research project. Ministry of Agricultural and Co-operatives. Naliendele Agricultural Research Institute, Mtwara, Tanzania, 87 pp. Doorenbos, J., Kassam, A.H Yield response to water. FAO Irrigation and Drainage Paper No. 33. FAO, Rome, Italy, 193 pp. FAO New_LocClim software local climate estimator. Environment and Natural Resources Working Paper No. 20. Environment and Natural Resources Service, FAO, Rome, Italy (CD-ROM). Griffiths, J.F., In: Morgan, W.T.W. (Ed.), Climate East Africa: Its People and Resources. Oxford University Press, London, UK. IUSS Working Group WRB, World Reference Base for soil resources. World Soil Resources Report No. 103, FAO, Rome, Italy. Kafiriti, E.M Integrating conventional and participatory research: experiences from trials with rice farmers in south eastern Tanzania. PhD Dissertation. Dissertationes de Agricultura No. 595, Fac. of Agric. Sciences, K.U. Leuven University, Belgium, 137 pp. Kafiriti, E.M., Dondeyne, S., Msomba, S., Deckers, J.A., Raes, D., Variation in agronomic characteristics of irrigated rice varieties: lessons from participatory trials in south eastern Tanzania. J. Food Agric. Environ. 1 (2), Kafiriti, E.M., Wellens, J., Dondeyne, S Annual report no. 2. Irrigation research project. Ministry of Agricultural and Cooperatives. Naliendele Agricultural Research Institute, Mtwara, Tanzania. 108 pp. Lal, R., Tillage systems in the tropics, management options and sustainability implications. FAO Soil Bulletin No. 71, FAO, Rome, Italy. Maertens, A., Intensivering van de rijstteelt in zuidoost Tanzania. Eindverhandeling bio-ingenieur. K.U. Leuven University, Leuven, Belgium, 123 pp.

7 agricultural water management 89 (2007) Ministry of Agriculture, Diagnostic Survey Report of West Lindi/East Nachingwea/North East Masasi. Farming Systems Research/Socio-economics Unit, ARI Naliendele, July Raes, D., Geerts, S., Kipkorir, E., Wellens, J., Sahli, A., 2006a. Simulation of yield decline as a result of water stress with a robust soil water balance model. Agric. Water Manage. 81, Raes, D., Mallants, D., Song, Z., RAINBOW a software package for analysing hydrologic data. In: Blain, W.R. (Ed.), Hydraulic Engineering Software VI. Computational Mechanics Publications, Southampton, Boston, pp Raes, D., Willems, P., Gbaguidi, F., 2006b. RAINBOW a software package for analyzing data and testing the homogeneity of historical data sets. In: Proceedings of the Fourth International Workshop on Sustainable management of marginal drylands, Islamabad, Pakistan, January, pp Rallison, R.E., Origin and evolution of the SCS runoff equation. In: Symposium on Watershed Management, ASCE, New York, NY, pp Senkondo, E.M.M., Msangi, A.S.K., Xavery, P., Lazaro, E.A., Hatibu, N., Profitability of rainwater harvesting for agricultural production in selected semi-arid areas of Tanzania. J. Appl. Irrig. Sci. 39, Sivakumar, M.V.K., Predicting rainy season potential from the onset of rains in Southern Sahelian and Sudanian climatic zones of West Africa. Agric. For. Meteorol. 42, Stewart, J.L., Response Farming in Rainfed Agriculture. WHARF Foundation Press, Davis, USA. United States Department of Agriculture, Estimation of direct runoff from storm rainfall. National Engineering Handbook, Washington DC, USA. Section 4 Hydrology, Chapter 4, pp URT, Integrated agricultural survey 1997/98. Rural Tanzania mainland Statistics Unit Ministry of agriculture and co-operatives and National bureau of statistics. Dar-Es- Salaam, Tanzania. Wellens, J., BudGIS A tool for yield forecasting during the ongoing growing season MSc Dissertation IUPWARE. K.U. Leuven University, Leuven, Belgium, 82 pp.