Crop Upgrading Strategies and Modelling for Rainfed Cereals in a Semi-Arid Climate—A Review

Spatiotemporal rainfall variability and low soil fertility are the primary crop production challenges facing poor farmers in semi-arid environments. However, there are few solutions for addressing these challenges. The literature provides several crop upgrading strategies (UPS) for improving crop yields, and biophysical models are used to simulate these strategies. However, the suitability of UPS is limited by systemization of their areas of application and the need to cope with the challenges faced by poor farmers. In this study, we reviewed 187 papers from peer-reviewed journals, conferences and reports that discuss UPS suitable for cereals and biophysical models used to assist in the selection of UPS in semi-arid areas. We found that four UPS were the most suitable, namely tied ridges, microdose fertilization, varying sowing dates, and field scattering. The DSSAT, APSIM and AquaCrop models adequately simulate these UPS. This work provides a systemization of crop UPS and models in semi-arid areas that can be applied by scientists and planners.


Introduction
Food security refers to physical and economic access to sufficient, safe, and nutritious food that meets people's dietary requirements and food preferences for an active and healthy life [1,2]. Despite efforts from 1990 to 2012 to reduce the number of people suffering from hunger, the hungry population in Sub-Saharan African (SSA) countries was reduced by only 5%. Millions of people in this region are still not able to meet food demand for their families due to land degradation, land use pressures, and climate change [3]. Other causes include population growth, poverty, weak institutions and infrastructure, high dependency on rainfed agriculture, unequal global trade relationships, and poor soil fertility [4][5][6]. There are also constraints related to the production value chain in general, food access [5,7] due to geographical locations and associated eating behaviors [8], socio-economic status [9], technology adoption [10,11], and sustainable production [12][13][14].
A total of 57% of the population in East African countries (Tanzania, Kenya, Uganda, Rwanda and Burundi) experienced food insecurity during 2015-2016, which is only a slight decrease from

Materials and Methods
We collected papers from different literature repositories that describe the roles of upgrading strategies in the production of cereals ( Table 1). The search keywords that were used and the examples of combinations are presented in Table 1. The papers were screened in succession starting from a global context, then in arid and semi-arid climate countries and finally to the topic of this study. A total of 187 papers were relevant to this review from more than 1000 papers. The papers covered broadly the roles of UPS for cereal production in arid and semi-arid regions. In addition, the papers described the usefulness of modelling UPS for enhancing cereal production under low and highly variable rainfall and poor soil fertility conditions. The other conditions for the inclusion of papers in this review were the following: i.
Peer-reviewed articles on crop UPS and modelling in English published in 1990 and after. The UPS that were included were categorized as rainwater harvesting, soil moisture conservation, means of water application and productivity, nutrient addition to the soil, soil conservation, drought coping measures and measurements of rainfall variability. ii.
Reports with qualitative or quantitative empirical findings and perceptions and views discussed in relation to UPS and modelling. iii.
Conference proceedings papers related to UPS and modelling.
All the UPS identified in the selected papers were ranked according to importance, affordability, possibility, and effectiveness ( Table 2). The authors arbitrarily defined these criteria in terms of UPS adequacy to address the rainfall variability and soil fertility conditions faced by farmers in semi-arid areas.
The constraints categories for the UPS were scored by six experts, ranging from 1 (highest constraining potential), 2 (medium constraining potential) to 3 (lowest constraining potential). These experts were familiar with UPS and were capable of making informed judgements about them. Using the experts' results, the constraint with highest scores was regarded as having the least constraining potential. The authors of this study then scored the UPS using the scores given by the experts. The UPS with the highest efficiency in overcoming the stated constraint in each criterion as judged by expert scores was given a score of 3, the medium efficient was given a score of 2 and the least efficient was given a score of 1. If half or more of the six experts gave a score of 3 to the constraint (i.e., expert one score = 3, expert two score = 3, expert three score = 1, expert 4 score = 3, expert 5 score = 2, expert 6 score = 1) then we picked 3 as a score. If the scores tied, for instance, three experts gave a score of 3 and two experts gave a score of 2, then we picked the highest score, which is 3. Similarly, if two experts each gave similar scores we picked the highest score of the three given (i.e., two experts gave 1, the other two gave a score of 2 and the other gave a score of 3, then we picked 3). For example, for the criterion "importance" the constraint social acceptance was given a score of 1 because four experts gave a score of 1, for constraint "labor" the score was 2 because three out of six experts gave a score of 2 and for constraint "sustainability" the score was 2 since four experts gave a score of 2. In the next step, all constraints of a given criterion were averaged, producing suitability scores ranging from 1 to 3, with 3 being the highest suitability, while 1 means low suitability. Finally, all criteria scores for a given UPS were averaged to form a final score. Once the four UPS with the highest overall average scores according to the criteria in Table 2 were selected, we searched for literature that described the biophysical models able to simulate them. The main criterion for model selection was that the reported models must have been used to adequately simulate the selected UPS. The models that were the most robust in simulating these UPS were further described and analyzed. Depending on the number of the UPS simulated by the biophysical models, we ranked the models from those that simulated all the four selected UPS to those which simulated at least one UPS. We described three of the most frequently used models pertaining to the four selected crop UPS. We used the FAO database (FAOSTAT) (http://www.fao.org/faostat/en/#data, accessed on 7 July 2017) to extract information on cereals' global harvested area and production. We then computed the relationship between the model coverage of the four UPS by counting the number of papers from our collection under each cereal and the global harvested area of the cereals to determine whether there was any association. In addition, we analyzed the relationship between the global harvested area of different cereals and the usage of the four UPS.

Roles of Crop Upgrading Strategies for Cereal Production
We identified 13 UPS that are useful for cereal crop production (Table 3). Table 3. Reported upgrading strategies used for enhancing and securing cereal crop production.

Mulching
China [44,49,50], Zambia [51], Zimbabwe [52] Enhance infiltration rate of rainfall and reduces evaporation of moisture from soil Availability of mulch to cover large fields

Crop rotation, and intercropping
Ohio [53], China [54], Turkey [55], Spain [56], Iran [57], India [58,59], Burkina Faso [60], Zimbabwe [61] Crop rotation enhances soil structure and hence reduces runoff and soil erosion Enough land is needed to allow the rotation of the same amount of land for cultivation Reservoir tillage or Pit cultivation Tanzania [62], Zimbabwe [63,64] Potential for improved soil water availability and crop productivity. Additionally, reservoir tillage of sandy loam is effective for the infield harvesting of high-intensity rainfall Care should be given to harvest rainwater so as not to destruct the water balance of the catchment, especially groundwater storage Micro-dose fertilization Iran [65], China [66][67][68], Nepal [69] Micro-dose fertilization improves nutrient availability to the soil and grain yield and provides income for poor communities Limited by availability of sufficient soil moisture Crop substitution Iran [70], Canada [71] Substitution of maize with pearl millet was found as the best option in future climates for the production of fodder Traditional and cultural practice challenges Varying planting/sowing dates India [72,73], Brazil [32], Nigeria [74], Burkina Faso [75], Tanzania [74], South Africa [74,76], Zimbabwe [77], Egypt [78], Hungary [79], Syria [80], Lebanon [81], Nepal [69], Iran [82] Reduces the effects of yield loss due to temporal rainfall variability Timing of the start of planting dates Use of groundwater Oman [83], Iran [84], China [85][86][87] Adds to the fresh water that can be made available for agriculture Requires energy to lift water from below ground to the fields Spatial plot distribution Niger [88], [29], Greece [89], Kazakhstan [90], Benin [91] Reduces the risk of complete production failure Requires ownership of spatially distinct land

Requires labor and is not easily mechanized
Reduced tillage and zero tillage India [106,107], Ethiopia [108], Mexico [109,110], Canada [111], Pakistan [112][113][114] It may be applied when onset is uncertain and for conserving soil moisture and nutrients Reduced yield due to a less favorable crop growth environment Determining the balance between the strengths and limitations of the UPS is important because these will respectively encourage and discourage farmers during the adoption stages. For instance, in the study on the suitability of on-farm ponds (irrigation), Roost et al. [37] had to analyze soil characteristics in order to understand water dynamics, of which deep percolation was a concern. This implied that even if on-farm ponds are suitable for improving yields, they may work better when soils have a robust ability to retain the stored water. In addition, land availability that can be reserved for ponds within farmers' fields will be a challenge as most poor farmers' landholdings in semi-arid areas have small plot sizes [3]. Thus, sustainability of these practices becomes a challenge since it largely depends on the type of soil. Ponds primary purpose is to supplement a soil moisture deficit through irrigation. Other operational costs such as pumping or intensive labor for water lifting may be needed. Under high rainfall variability water storage for supplemental irrigation is important; the main challenge remains to be affordability and possibility.
Similar concerns arise when we examine the use of groundwater as an alternative source of water for supplemental irrigation [84]. This UPS, despite being important and effective, also faces challenges of affordability and possibility as it requires an initial investment for boreholes or a large amount of labor for hand dug wells. Additionally, during irrigation, the operational costs of lifting water via pumping are untenable for most poor farmers. Raising water for irrigation by hand is unattractive due to drudgery work involved. In general, this makes the UPS difficult to implement and labor intensive, and it may also face the challenge of availability of groundwater, hence jeopardizing its adoption possibilities.
Mulching helps to prolong soil moisture by protecting it from excessive evaporation [115], and in that regard is reported as an important strategy by many authors (Table 3). It is socially acceptable, sustainable, environmentally friendly, not constrained by land availability, and helps improving microbial activities in the soil, and productivity. Mulch also is well suited with zero tillage. However, when rainfall is minimal mulch may obstruct rainfall from falling directly on the soil. Under high rainfall variability, mulch works better when combined with a field scattering strategy and it may work with irrigation to improve water use efficiency [49]. The main challenge is mulch availability, and the labor involved in laying down the mulch. Some authors did not find significant advantage for yields when mulch was used [51], and farmers are reluctant to practice mulching [52]. Other UPS such as crop rotation, crop substitution, reservoir/pit tillage, and irrigation using surface water, all have their strengths and weaknesses (Table 3). In our analysis, we found four UPS to be the most suitable for boosting cereal production in semi-arid areas.
In Table 4, the experts' scores for each constraint and criterion related to individual UPS are presented. The results indicate the amplitude of scores but also show that this procedure can identify the most suitable UPS for the target region. For example, irrigation is among the most important and effective UPS to improve crop performance; however, considering the context in which this assessment is done, it is very costly and requires extensive knowledge, thus, scoring very low in affordability and possibility. When considering the four UPS with higher overall scores (Figure 1), the strategies were shift planting/sowing dates, contour strips and tied ridges, spatial plot distribution and micro-fertilization.

Modelling Cereal Crops UPS for Enhancing Production
Many models have successfully simulated the four selected crop UPS and have suggested methods to tackle the production challenges of cereal crops in semi-arid areas. Examples of these models include DSSAT, APSIM, AquaCrop, HYDRUS-2D, EPICSEAR, MSM and others (Table 5). Table 5. Different models reported for simulating UPS in semi-arid areas.

UPS Models and Authors Country Crops
Contour strip and tied ridges ACRU [116], SWAT [117], AquaCrop model [118], HYDRUS-2D Software [119], APSIM [120] Zimbabwe In general, the simulations of UPS for semi-arid climates are reported by only a few papers compared to the multitude of papers available on modelling/simulation. We found that 30 out of the 187 papers contained in this review specifically covered the modelling of at least one of the four selected UPS in semi-arid areas. One strong reason for such a shortage is the limitation in the availability of, and access to, quality data [100,143,144]. Maize was covered by 51% of the papers reviewed. The other cereal crops that were covered were wheat (23%), rice (10%), barley (8%), sorghum (5%) and pearl millet (3%), (Figure 2). We found a strong positive correlation (r) between the global harvested area and the frequency of crop coverage by the models in the papers that simulated the four crop UPS in semi-arid regions ( Figure 2). Thus, the cereals with large global harvest areas not only possess a good chance to be cultivated with UPS but also attract more modelling studies to improve their production. This is evident because the three major cereal crops (wheat, rice, and maize), which occupy more than 80% of the global area of harvest for cereals, show a high frequency of usage of models and UPS (Figure 2) compared to millet sorghum and barley. We also found a relationship between the productivity of global cereals ( Table 6) and their frequency of model coverage for the UPS in the reviewed papers, which may best be represented by a quadratic model (Figure 3).   Crops such as pearl millet and sorghum, which are among the major staple food crops in semi-arid areas, especially in Africa, have gained less modelling attention. This lack of attention could be due to the disadvantage related to global harvest area, as explained by the relationship in Figure 2. This disadvantage is also due to the population that consumes these crops which is small when compared to other cereals. The relationship can be associated with the analysis of White [145]. However, these crops can be of major importance locally as they represent large potential for reducing food insecurity and poverty in semi-arid areas.
From our review, we selected three models to describe and discuss; these are DSSAT, APSIM and AquaCrop. These models cover the four UPS adequately. In summary, in this paper's framework, the DSSAT model dominated the studies, followed by APSIM and AquaCrop. The reason for the dominance of these models can be understood further through their descriptions and discussion, which is also presented in this paper.

Discussion
The choice of UPS depends on the involvement of stakeholders and their knowledge of labor limitations, financial constraints, restrictive governmental policies, environmental and climatic conditions, social and cultural conditions and related food security risks [20]. Thus, UPS are important to sustainable crop production and food security. In our review, all 13 UPS were found to be suitable for improving production; however, they all had constraints in their adoption. In Tables 2 and 4 we have identified the constraints, assigned a score as supported by the literature and ranked the UPS according to the average scores of the constraints. The scores of the constraints were given by experts as a result of the understanding found in the literature describing the ability of the UPS to overcome these constraints. We have picked the four highest ranked UPS and described them below.

Contour Ridges and Tied Ridges
"A ridge is a long, narrow, elevated strip of land or any raised strip or band. A ridge is called a contour ridge when it is created along the upslope furrow to accommodate runoff from a catchment strip between the ridges that are 15 to 20 cm high" [146]. "These ridges may be 1.5 to 10.0 m apart depending on the microcatchment system, as the catchment is a function of the distance between ridges. The precise distance should be calculated according to the expected rainfall and the soil texture of the region" [146]. When a ridge is crossed by the ridge of another perpendicular earthen band, it is called a tied ridge (Figure 4). "Small earthen ties are made within the furrows at 4 to 5 m intervals to prevent the lateral flow of water. The main objective of a tied ridge system is to collect local water runoff and store it within the soil profile in the vicinity of the plant roots" [146].
Tied ridges prolong the retention of soil moisture and enhance nutrient uptake by crops [101,104,147]. Tied ridges provide a better crop growth environment [102,148,149] in areas where the temporal rainfall variability is a challenge. Evidence of higher cereal yields and increased rain water use efficiency [27,101], and even better income [92,104], have been reported. Tied ridges can be improved to be more effective, for instance by combining tied ridges with microdosage techniques [147]. This UPS is found to be the most useful in reducing the acuteness of temporal rainfall variability. Araya and Stroosnijder [104] report that tied ridging and mulching can increase soil water availability in the root zone by more than 13% and crop grain yield (barley) by at least 44% during below average rainfall years. Other evidence shows that tied ridges are beneficial in drought-prone areas. Arid and semi-arid areas can benefit from enhanced production, up to five times greater, of sorghum yields grown in loamy soils [103].
Care should be taken when farming using tied ridges, and it is advised to carefully open the ridges when excess water is expected to cause waterlogging [104]. Thus, when rainfall is well distributed during the cropping season, tied ridging is not a necessary tillage alternative for cereals such as maize on flat land. However, when tied ridges are used in areas with modest slopes, they markedly increase soil water reserves at 0.15 to 0.60 m depths after maize harvest [102]. This strategy is limited by the difficulties related to mechanization [150] and excess rainfall. The limitations necessitate "an assessment of the major agro-meteorological challenges, existing tillage, hoeing and associated land management practices" [150].

Micro-Fertilization
After water availability, soil nutritional status is the greatest constraint to food production across the arid and semi-arid SSA countries [105,151,152]. When soil moisture is adequate for plant growth, the crop will reach its maximum growth potential if nutrient supply is adequate. Several authors have reported a challenge of poor soil fertility that farmers in semi-arid areas are facing [153][154][155]. The problem is acute for poor farmers in semi-arid Sub-Saharan Africa countries [151,156,157]. The precision-farming technique referred to as "microdosing" has been developed by experts [158][159][160]. It "involves the application of small quantities of fertilizer with the seed at the time of planting or as top dressing three to four weeks after emergence" [161]. This method enhances the efficiency of fertilizer use when compared to the broadcasting application method. Micro-fertilization helps poor farmers increase the returns on small initial investments [160,162].
In agreement with Bielders and Gerard [162], farmers have to develop the capacity to apply recommended amounts of fertilizer over time. The best option could be the application of manure or organic fertilizers. However, manure availability can be a challenge for farmers who are not livestock keepers. In such conditions, microdose fertilization remains an important option to these farmers. So far, microdosing has been introduced in Zimbabwe, Mozambique, and South Africa, which are in the southern part of the African continent [151,157]. Farmers apply fertilizer in small quantities (micro-fertilization) to cover large areas of their farms, and this method has helped them increase crop yields [163]. The results from Niger and Mali show a yearly average of 20 kg P 2 O 5 ha −1 and 30 kg N ha −1 increased stover yields by 400 to 1500 kg ha −1 , and the stover yield was not further increased with higher application rates of 40 kg P 2 O 5 ha −1 and 60 kg N ha −1 [164]. Similar reports of millet yield success have been reported in Niger [162]. Other promising results from initial on-farm trials in Zimbabwe show increases in yield by 30-100% when microdoses, as little as 10 kgN ha −1 [151] were applied.
Thus, micro-fertilization UPS are recommended in low-yielding plots, as microdosing in plots with minimum yield bears a non-negligible financial risk [151]. Twomlow et al. [151] reported that micro-fertilization was an effective strategy to counteract the disadvantages of late sowing, and it better suited as a means of famine mitigation [151]. They provided yield data that smallholder farmers could increase their yields by 30-100% through application of micro-fertilization as little as 10 kg Nitrogen ha −1 [151]. However, these experiences from West Africa have shown that the adoption of micro-fertilization technology requires supportive and complementary institutional innovation [151]. The major constraints to the adoption of this strategy are the initial access to fertilizer, access to credit, insufficient flows of information and training to farmers, and inappropriate policies [151]. Of much interest is what additional benefits can be obtained by combining microdose fertilization with other UPS, such as varying planting dates, tied ridges and field scattering.

Varying Sowing/Planting Dates as a Measure to Counteract Rainfall Uncertainties
Varying planting dates increases the efficiency of the use of available water or rainfall and it has the potential to reduce the risk of temporal rainfall variability [29]. Under current climate change and variability, the onset of seasonal rainfall is uncertain for most years [76,[165][166][167]. Therefore, farmers cannot rely on the planting dates they previously used. To reduce the risk of total production failure, farmers are advised to vary planting dates to capture the best planting date for the season. This method guarantees the probability of at least one planting date secures a harvest. In this case, the timing of the start of planting dates is crucial and very challenging [74].
In addition, a challenge related to the affordability remains, as poor farmers have a limited capacity to purchase enough seeds to cover several planting dates. A study in Southern Africa established optimum planting dates for six locations. These dates were from the 5 to 27 December, 26 December to 7 January, and 4 to 7 December for Malawi, Mozambique, and Zimbabwe, respectively [76]. Sacks et al. [168] discovered that maize planting dates might vary more widely in tropical regions. The findings reflected that sowing time can be optimized for stable rainfed maize yields, but the effect of sowing dates on rainfed maize yield is modified by both soil depth and soil fertility.
In general, varying sowing/planting dates increases the chance of crop production under high temporal rainfall variability; for some cereal crops, this strategy has significantly influenced seed germination, plant height, tillering, total biomass, and overall grain yield and water productivity under rainfed conditions. Such cereals include wheat [168,169], rice [170,171], maize [172,173] and sorghum [174][175][176]. The weakness of the method is related to the difficulties in controlling the harvest loss for all dates combined. Thus, this strategy provides an opportunity for farmers to have a wide range of planting date options that can help them secure a possible optimum harvest for the season. The timing of planting should be adjusted based on the forecast of the onset of seasonal rainfall to harness the maximum potential of available seasonal rainfall.

Spatial Plot Distribution as a Measure to Combat Spatial Rainfall Variability
Rainfall can vary over a spatial distance as small as 800 m [29]. Efforts to protect crop yields from the impacts of climate change should, therefore, consider spatial rainfall variations [177]. Spatial plot distribution (cultivating fields at different locations within a few kilometers) reduces the risk of complete production failure when there is less rainfall and high spatial variability in the area; however, fields cannot avoid production losses altogether [29,30]. The spatial plot distribution strategy has been successfully used in Niger [29] and Nigeria [30]. Spatial plot distribution is limited by land and financial resources of farmers to be able to farm fields in different locations. The strategy is useful in areas that are not constrained by land resources and have low rainfall with high spatial variability [29,30].

Roles of Models for Cereal UPS
The models are good at exploring different questions that arise from the limitations of UPS at the field level and extrapolating the results to a large scale that allows for more informed decision making [178]. Modelling and simulation studies help in making more sensible decisions for the current climatic status and the future production of cereals. They provide an alert for the measures to take in case food security is at risk. The models should subsequently be improved as they have several limitations, as described below.

Description of DSSAT-Models
The "decision support system for agrotechnology transfer" (DSSAT) encompasses 16 different crop models for different evaluation purposes [179]. The models in DSSAT integrate the effects of soil, crop phenotype, management options and weather. A DSSAT model has been adopted in a wide range of countries and continents (Table 5), which indicates its strength in accommodating a wide range of climatic and agricultural conditions. For instance, one simulation study found clear signs of decline in the future production of maize and rice in India [73]. Additionally, they determined that shifting planting dates would have a positive impact on production. Rezaei et al. [70] used the DSSAT model and found that maize crops could be substituted by pearl millet to counter yield reduction in the future, while other studies have suggested that a change of cultivar would be an option to sustain food production in the future [32,180].
In this survey, we found that the models have satisfactorily explained farmers' attitudes to planting/sowing time, microdosage under nutrient management, and scattered field or spatial field dispersion. In most of the papers, data used for calibration were daily weather records, measured physical, chemical, and morphological characteristics of the soils as model inputs, and crop genetic coefficients that are specific to each cultivar of interest. In a study by Jagtap and Abamu [127], DSSAT model used farmers' planting dates to simulate maize growth and found that "early planting together with manure from cattle allowed to graze on crop residues increased yield response to fertilizer N" [127]. The results also showed that changing the planting date by 30 days (more or less) had minimal effect on obtainable maize grain yields in production systems characterized by high N inputs (120-150 kg/ha) [127].
DSSAT offers nutrient management functions, where fertilizer application can be varied from optimum to minimum applications (micro-fertilization). Nutrient management has been simulated extensively and reported by many authors. For example, Gerardeaux et al. [129] used the DSSAT 4.5 cropping system model (DSSAT-CSM) to explain the impact of various climate changes on rice productivity in Madagascar using two tillage strategies and two fertilization rates: low and high nitrogen. They used a locally cultivated rice cultivar to calibrate and validate the model, with a 6-year experimental data set. Nitrogen was found to be a major constraint while zero tillage demonstrated no advantage in addressing impacts of climate change. Another study by Jeong et al. [130] using DSSAT found that adjusting the split N fertilizer application rate was enough to attain a satisfactory yield of rice, providing an additional 10-20 kg/ha. Other studies also done on wheat [126], pearl millet [131] and maize [32,125] demonstrated the potential of using DSSAT in simulating fertilizer microdosage.
DSSAT has also been reported to simulate field scattering as an UPS for improving cereal yields in semi-arid areas. In a study by Jalota and Vashisht [73], they found that "in the future, the magnitude of climate change and variability would vary with agro-climatic zone" as well as the yields of maize and rice [73]. A simulation of wheat at different locations using a DSSAT model [140] also recorded different yields, which were a result of different agro-climatic zones within India. Hurtado et al. [138] also reported that the DSSAT was able to simulate corn yields related to field spatial variability. The authors used data of "maximum and minimum temperatures, rainfall and solar radiation; soil data in the 0-27, 27-45, 45-68, 68-80 and 80-100 cm layers to each experimental plot, management information of corn crop and genetic parameters of the corn hybrid" [138]. The authors reported that the DSSAT was able to simulate yields as related to the spatial variables of the soil attribute base saturation.
Although DSSAT models are reported to adequately simulate planting dates, micro-fertilization and spatial plot distribution UPS, we found no evidence that DSSAT models were used to simulate tied ridges strategies. This is a gap that crop modelers may be required to fill as this UPS is among the four prioritized UPS found to be effective in reducing the adverse effects of rainfall variability and in conserving soil fertility and moisture during crop growth in semi-arid areas. Another summary of the strengths and weaknesses of DSSAT models is presented by Brilli et al. [181]. The limitations reported include overestimation or underestimation of soil organic carbon and soil nitrogen.

Description of APSIM Model
"The Agricultural Production Systems Simulator (APSIM)" [182]simulates several systems and addresses the interactions among plants, animals, soil, climate, and management. The model allows for the analysis of a whole-farm system, including crop and pasture sequences and rotations, and livestock [182]. Maize was found to be a dominant crop simulated by the APSIM model followed by wheat and sorghum. For example, Ngugi et al. [120] adopted the APSIM model for a scenario analysis of the effect of tied ridges on maize yield in two study sites in Kenya and found different responses to tied ridges and fertilizer application. In general, when combined, tied ridges and fertilizer application boosted the maize yield significantly [120]. This provides evidence for the concept of combining different UPS to counteract the effect of rainfall variability and poor soil fertility. In this study, the combined UPS increased yields of maize grain between 507-3370 kg/ha [120].
In Zimbabwe, the APSIM model was used to explore the "risk associated with N fertilizer use by smallholder farmers and management strategies to minimize that risk" [121]. They used long term data from Nitrogen fertilizer maize trials for calibration and validation of the model before using it in their exploration. In their results, they found moderate fertilizer application rates and early sowing will give economic benefits to farmers [121,122]. Kisaka et al. [122] found similar results in Kenya and recommended that "for subsistence farming, low-cost recommendations are geared towards some 'guaranteed' yield stability each cropping season" [122]. Ahmed et al. [123] also calibrated and validated the APSIM model and used it to predict nitrogen use efficiency of wheat under rain-fed conditions for the Pothwar region of Pakistan. The simulation revealed the potential of the APSIM model to provide fertilizer recommendation rates that may achieve agricultural production and economic benefits, a finding that is in agreement with [142].
In the study by Ngugi et al. [120], the authors reported the potential of the APSIM model to inform the planting dates UPS. They found that timing of planting was key in improving yields. The results are in agreement with Heng et al. [133] who reported that "early sowing is important for achieving high yields by avoiding terminal water deficit" [133]. The algorithm for simulating planting dates in the model provides opportunities to determine optimum planting dates, which can help farmers to avoid total crop failure. The model, in addition, could "simulate maize yields at four locations with 8 years of reliable weather records" [142], which highlights the potential of the model in determining the effect of spatial plot distribution UPS. Under high spatial rainfall variability, the model can help in predetermining the potential positions of fields where farmers can expect to obtain optimum yields.
Some of the weaknesses of the model have been summarized by [181], which include underestimation of soil nitrogen and soil organic component, overestimation of annual leaching, and underestimation/overestimation of N 2 O emissions. In this regard, the APSIM model still requires periodic improvements so that it can be used with more certainty. In addition, [181] provides a detailed summary and review of the strengths and weaknesses of the APSIM model.

Description of the AquaCrop Model
"The AquaCrop model simulates major herbaceous crops as a function of water consumption under rainfed, supplemental, deficit, and full irrigation conditions" [183]. It also simulates evapotranspiration as separately soil evaporation and crop transpiration. The model also simulates the biomass accumulated each day using crop specific parameters that are normalized for reference evapotranspiration [183][184][185], which makes the model useful to in different climates [132,186,187]. The model has been successfully used to simulate different cereal UPS in semi-arid areas. We found six papers that discussed planting dates for maize [77,184], barley [136,186], winter wheat [135], and varying nitrogen levels (microdose fertilization) for maize [132]. We found the model to have the potential for extrapolating successful cereal UPS in arid and semi-arid areas. Expansion of the usage of the model for simulating tied ridges and modifying the model to accommodate spatial field dynamics will be of importance for its robust usage.
Most of the weaknesses of the AquaCrop model are mentioned on the FAO website for land and water (http://www.fao.org/aquacrop/overview/limitations/en/). Our overview finds that the weaknesses resemble the limitations of DSSAT and APSIM. These limitations include the following: (i) AquaCrop simulates a single growth cycle only; (ii) the model predicts crop yields at a single point in its simulations [183] and assumes that the field is uniform without spatial differences in crop development, transpiration, soil characteristics or management [183]; and (iii) the AquaCrop model considers only incoming vertical (irrigation capillary and rainfall rise) and outgoing (transpiration, deep percolation and evaporation) water fluxes in its simulations [183].

Conclusions
Four UPS were identified as the most promising to improve rainfed cereal production for poor farmers in semi-arid areas. These UPS are tied ridges, microdose fertilization, varying sowing/planting dates and field scattering. These strategies are suitable for reducing the risks of cereal production failure under low rainfall, high spatiotemporal variability, and poor soil fertility conditions.
Among the different biophysical models, three of them-namely, DSSAT, APSIM and AquaCrop-have more frequently studied the four selected UPS. We found that both the strengths and the weaknesses of these models were comparable. In general, the models have performance weaknesses when temporal and spatial variations in soil moisture, soil temperature and aeration are considered. In addition, they do not include algorithms for each UPS or for soil organic content and they consider only vertical water fluxes. The frequency of the use of models is an important proxy for their robustness. This assessment also identified the main points necessary to improve the selected UPS's. Despite their limitations, we encourage researchers and decision makers to use these models since they have been instrumental in research and decision-making.