3.1. Simulated Smallholder Maize Production in Tanzania
Maize production in smallholder farms in Tanzania is characterized by low productivity, large year-to-year variability, and dependency on rainfall [35
]. Through a national-scale grid-based simulation for the period of 1984–2013, we attempted to reproduce these characteristics. The simulation was conducted with common management strategies currently practiced by smallholder famers, i.e., low fertilizer application, middle maturity cultivar, and rainfall-dependent sowing. We kept these settings static and universal for regions within a given simulation due to during insufficient information of the nation’s heterogeneous production areas.
Simulated historical maize yield averaged across the country over the period (1995–2008) was 1.28 t/ha, similar to the reported figure of 1.21 kg/ha (Figure 1
). This yield similarity between the simulation and reported confirms the dominant and prevailing low-investment practices in Tanzania. Simulated yield variability denoted as Coefficient of Variability (CV) was 8.1%, lower than the reported value of 16.2%. This indicates climate variation explains roughly 50% of reported yield variability for maize in Tanzania. A value that significantly higher than its global mean of 39% [36
], suggests the high dependence of Tanzania’s maize production on weather. However, simulation-based estimates tended to overestimate reported national yield for most years, likely due to lack of consideration of year-to-year changes in areas planted with maize and the lack of representation in the model of other biophysical stresses, such as extreme temperature, pests, and diseases.
3.2. Importance of Rainfall on Tanzania’s Maize Production
Precipitation is important for Tanzania’s smallholder maize production for two reasons. First, sowing is highly dependent on rainfall onset, which is highly variable [35
] and delayed under recent climatic warming trends in the region [6
]. Second, production is highly dependent on seasonal rainfall and thus, increasingly threatened by rainfall variability and deficit during the growing season [13
Crop model simulations revealed substantial variability of sowing date due to changes in annual rainfall patterns. First, sowing date averaged across areas planted with maize in 2000 [31
] demonstrated a large inter-annual variation by up to 27 days (Figure 2
a). Second, substantial number of locations (grid cells) exhibited delay of the sowing, likely due to climate change, indicated by a moving of sowing day from a bimodal distribution at the first decade to a bell shape with peak around the 20th day from the beginning of sowing window (Figure 2
Of all the 7494 simulation grid cells, 48.8% experienced delayed sowing from the first to second decade, with a mean delay by 3.5 days (Figure 3
a), and 39.1% grids experienced delay from the second to third decades, with a mean lateness of 6.2 days largely concentrated in the northeastern part of the country (Figure 3
b). This is confirmed by farmers’ perceptions that delays in rainfall patterns are now a common occurrence: “We have supposed to plant in February and March but the seeds that been sown have not got rainfall until now (mid-March)” [6
]. However, sowing date fluctuates in most areas with a large part of the country not showing detectable changing trends in sowing date during the time period considered here. This is likely caused by periodic precipitation patterns and our sowing window is limited to the fixed time window in the source data set.
Even a minor delay of sowing poses large detrimental effects on maize production, as it substantially increases drought risk, affecting crop grain filling. We found negative relationships between days of delay and maize yield at both national and regional scales. A day of sowing delay could lead to 0.8% decrease of the maize yield across the country (Figure 4
a). For example, the maximum delay scenario (27 days) from 2002 to 2003 equates to a 23.5% reduction in national maize yield in Tanzania, seriously undermining the country’s food supply.
The simulations also confirmed the negative effects of insufficient rainfall on Tanzania’s maize production. Simulated national yield with current management was positively associated to growing-season rainfall, with R2
= 0.432 (Figure 4
b). This suggests that over 40% of the yield variability can be explained by growing-season precipitation. This estimate is higher than previous reports of over 20% of maize yield variability in Tanzania explained by precipitation [35
]. This is likely because our estimate takes into account the effects of both precipitation variability and its impact on sowing date. Based on the relationship between national maize yield and growing-season precipitation, we estimated a one-hundred-year drought like 1985 could damage the national maize production by 9.7%, similar to estimated production loss at the global level caused by drought [38
Furthermore, late sowing showed different effects across regions due to differences in precipitation patterns and crop responses. Maize production in southern Tanzania was more susceptible to yield damage by late sowing, reflected by their significant and negative correlation between late sowing and yield (p
< 0.05) (Figure S1a
). In these areas, if the sowing was postponed by 5 days due to late rainfall, up to 10% of yield would be lost (Figure S1b
Unlike the effect of late sowing, insufficient rainfall has more widespread effects on Tanzania’s maize production. Rowhani et al. analyzed the relationship between maize yield and precipitation from 1992 to 2005, concluding that there was a significant relationship between maize yield and seasonal mean precipitation, with an increase in precipitation favoring yield [35
]. Our simulation confirmed this finding. The simulated annual yield was significantly and positively correlated to growing-season precipitation in over 70% grids (Table 2
), implying that the majority of the country has been affected by insufficient rainfall. Furthermore, estimated yield loss due to deficit of rainfall was substantial in many areas, with larger production damages on smallholder farmers. For example, with a 100 year insufficient rainfall event, the estimated yield loss averaged across the villages participating the PPP was 29.3% (indicated by red dots in Figure 5
), pushing the production of smallholder farmers in these areas down to the subsistence line, i.e., less than 1 t/ha [5
]. In addition, this susceptibility to insufficient rainfall emerged beyond current maize area. Simulation over all grids indicated the susceptibility was prevailing over most of its arable land, except the western districts, e.g., Kigoma, Katavi, and Rukwa (Figure S2a
). Estimated yield loss due to deficit of rainfall was even larger in southeast, central, and northeast areas, where a large portion of the land was not under maize cultivation (Figure S2b
). With periodic occurrence and increasing frequency of insufficient rainfall, this prevalence of yield damages with insufficient rainfall indicates the high and widespread drought risks for smallholder farmers in Tanzania, constraining their expansion of maize cultivation areas through conventional land clearing practices.
3.3. Better Management Practices Increase National Productivity
All management strategies considered here substantially increased national maize yield (Figure 6
), with a potential two to five times increase of national production depending on the strategy. Given 2013 production of 5.3 million tons of maize, a basic investment on seeds and chemical fertilizer could potentially increase national production up to 25 million t/year without expansion of land under cultivation, and can largely secure the import demand imposed by other African regions, such as eastern and southern Africa. Same as controlled experiments conducted in Africa, such as ref [39
], strategies with later maturity, drought resistant seed, and highest nutrient input exhibited the largest potential to increase national production (five times), whilst strategies of early maturity maize with less fertilizer generated lowest production growth (1.8 times). Spatial variation in yield increases under different management strategies, demonstrating different yield responses in specific weather and soil conditions. The spatial variation in yield responses was larger with higher-yielding strategies (Figures S3 and S4
), suggesting varying investment returns for smallholder famers among regions. For example, northern Tanzania exhibited larger yield increase than the south, particularly northwestern Tanzania (Figure S3
). Districts Mara in the north and Kigoma in the westernmost part are likely to have the highest yield return of over 7 t/ha under strategy s (long-season and drought resistance maize and 120 N input) (Figure S3
Higher yields tend to increase yield ranges over the 30 years due to periodic abiotic shocks, such as extreme weather, but estimated yield variability does not increase with all strategies in our simulation. Changes in yield variability are reflected by simulated yield range (difference between maximum and minimum yield), standard deviation (SD), and coefficient of variability (CV) (Figure 6
). All strategies with long and middle maturity maize demonstrated increased range and SD on national maize yield. For example, the range of national maize yield for strategy j (long maturity with 120 kg N) was 2416 kg/ha, 7 times the range for the yield with current management (355 kg/ha). However, because strategies with larger yield range and SD usually experienced higher mean yield, larger range and SD are not always translated to greater yield variability in statistics. For instance, the CVs of most improved strategies (3.4–9.0%) are slightly higher or even lower than the CV with current management (7.1%), suggesting that adoption of better management practices does not increase yield variability. Moreover, strategies with short maturity varieties all experienced reduced range and SD, reducing the CV by half with current management, which demonstrates the potential of the strategies to fulfill the dual goal to increase crop yield and reduce variability.
3.4. Better Management Practices Decrease the Prevalence of Poor Harvest
All improved management strategies demonstrated big potentials to mitigate the prevalence and occurrence of poor harvests for smallholder farmers. This was indicated by the substantial shrinkage of areas exposed to periodic poor harvest under adverse weather (Table 2
and Figure S5
). Of 7494 simulation grid cells, 80% (5844) experienced poor harvest at least once over the 30 years under current management (strategy a). The poor harvest areas are distributed over the majority of the country except a small portion in the west (e.g., the districts Tabora and Katavi), implying widespread harvest failure for smallholder farmers. Improving seeds and adding fertilizers substantially decreased the prevalence of poor harvest, reducing the areas to 23%, 44%, and 24%, respectively, under management groups with short (strategies b–d), middle (strategies e–g), and long season varieties (strategies h–j). Meanwhile, improved managements substantially reduced the occurrence rate of poor harvest, from 1 in 3 years under current management to 1 in 10 years under most improved strategies (Table 2
Although characterized by low mean yields over the years, strategies with short-season varieties showed the highest potential to decrease the areas experiencing poor harvest, concentrated in northeastern, central, and southeastern Tanzania. These reductions are significant for smallholder farmers, meaning more stable production while less investment in seeds under variable climate. These results are consistent with previous understanding that adopting short-season varieties is a preferred economical option for small holder farmers to escape bad weathers such as drought or heat waves, and stabilize their incomes [13
]. However, we found similar potential of the high-yielding varieties (e.g., long-season) in alleviating the prevalence of poor harvest, but with a stronger ability to boost farmers’ incomes. For example, strategies h–j (with long-season varieties) reduced the poor harvest areas by similar magnitudes as strategies b–d (short-season varieties), whilst doubled the mean and minimum yields of the strategies b–d (Table 2
and Figure 7
). Our result suggests that investment in higher-yields varieties is not only able to secure smallholder farmers’ food supply and incomes, but also enhances their capacities to cope with production shocks caused by adverse weather.
An increase in fertilizer application from 60 to 120 kg/ha led to a small increase of maize yields. This was indicated by small changes in yield distributions between strategies within the same maturity group varieties (Figure 7
). Only strategies with highest-yield maize (long-season) exhibited moderate increases with more fertilizers, suggesting increased efficiency of fertilizer with higher-yields seeds. Fertilizer increase also had limited effect on the prevalence and occurrence of poor harvest (Table 2
and Figure S6
). We consider our simulation likely overestimated the roles of fertilizers because of the simplification in the simulation regarding fertilizer application. For example, we neglected fertilizers of phosphorus and potassium due to lack of detailed information on soil nutrient content and assumed they were unlimited in the soils. This might overestimate the effects of nitrogen given the widespread high P depletion and unbalanced application among N, P, K in Africa [4
]. Although fertilizer is currently the core strategy for securing the food production in many African countries [41
], our result suggests that applying higher mineral fertilizer may not be the best option for smallholder farmers in terms of its cost-benefit ratio.
Although gene-modified drought resistant maize is increasingly adopted in Africa [13
], our results suggest a relatively weak role of drought resistant maize in increasing yields, i.e., maize varieties with and without drought resistance have similar yield distributions (Figure 7
). Plant drought resistance involves a series of mechanisms, including drought avoidance by adjusting morphological structure (such as root depth) or growth rates, and drought tolerance through stomatal movement and photosynthesis. However, we only considered the adjustment of root depth and root density in the soil profile as a means to increase the water extraction capacity of the crop, which largely under predicts the capacity of drought tolerance. Despite the underestimation, some management strategies explicitly reveal the great potentials of drought resistant crops in mitigating the production risks for smallholder farmers, particularly the strategies with middle maturity varieties and low to middle fertilizer application rates. For example, there were 3318 grids exhibiting poor harvests (<1 t/ha) during years of adverse weather under strategy f (middle maturity varieties and 90 kg/ha N inputs), while this number went down to 2457 for simulations with drought resistant varieties (Table 2
). This suggests that drought resistance could be an efficient option to cope with weather shocks, mitigating the risks of yield failure for smallholder farmers.
3.5. Drought-Related Risks under Improved Management Practices
As drought is one of the major climatic disasters in Africa, we further estimated the yield loss caused by drought related shocks-late sowing and insufficient growing-season rainfall (Table 2
and Figures S7 and S8
). Not surprisingly, strategies with short maturity maize exhibited largest capacity to deal with delayed sowing and insufficient growing-season precipitation. With a 5 day delay of the sowing, national maize yield dropped by around 3% with strategies b–d (short with 60, 90, and 120 N fertilizers, respectively), which was half of the strategies e–g with middle maturity maize, and 70% of the strategies h–j with long-season varieties. Adopting drought resistant maize further diminished yield loss to 2% (strategies k–m). Similarly, short-season maize demonstrated the highest potential to cope with insufficient growing-season rainfall. Compared to yield loss of 10.7% for strategies e–g with middle-season maize and 13.7% with long-season varieties, strategies b–d with short-season varieties experienced only 2.5% yield loss under a one-hundred-year insufficient rainfall event. Adopting drought resistant maize could neutralize or even reverse the yield losses, pushing the yield responses to zero or positive (strategies k–m). This shows the potential of strategies with short maturity and drought resistant maize to reduce drought risks and stabilize national food production.
The decreased drought risks under strategies with short maturity and drought resistance maize are significant for smallholder famers. First, smaller areas are susceptible to late sowing and insufficient rainfall with strategies using short maturity maize than those using middle or long maize varieties. For example, only 8% of the simulation grid cells had significant and decreased maize yields due to delayed sowing under strategies b–d and k–m (using short maturity varieties), far less than the strategies using middle or long maturity varieties (i.e., 20%). Only a few maize area in districts of central, northeastern, and southern Tanzania exhibited detectable yield loss to insufficient rainfall with short maturity and drought resistance maize, which is half of the area under current and improved strategies with middle or long varieties (Figure S8
). Second, strategies with short and drought resistance crops demonstrated substantial potentials to offset yield loss caused by drought related shocks. For instance, estimated yield losses 31%~42% for a one-hundred-year insufficient rainfall event under strategies e–j and n–s could be largely decreased to 10% with strategies k–m by adopting drought resistant and short maturity varieties.
Although most effective strategies for coping with drought related shocks are these with short-season varieties, their long-term benefits are smaller than those of high yielding strategies with middle maturity varieties for instance. This is indicated by total land productivities in the long term such as the mean yield over 30 years, and capacity to cope with poor harvest such as occurrence of yields less than 1 t/ha. Adoption of the improved management strategies depends on a range of factors, including short versus long-term costs and benefits, shocks such as climatic disasters and their magnitudes and occurrence rates, yield responses to drought risks and their spatial variations, etc.—all of which are worth further investigation.