A systems approach in research offers to examine the bio-physical constraints and decision-making of farmers exposed to climate variability. In this project, a systems perspective was achieved by combining computer-based simulation modelling, farmer surveys, and field experimentation to explore current and potential agronomic management practices crucial to smallholder maize (Zea mays L.) farmers to manage climate risks in the semi-arid regions of Ethiopia. The study aimed at investigating a suit of management options to identify opportunities that can improve crop productivity while reducing the production risk in smallholder maize-based cropping systems in the Central Rift Valley (CRV) Ethiopia. To establish better insights into farmers' perceptions of, and management responses to climate variability, farmer surveys or rapid rural appraisal (RRAs) were conducted. The RRAs were conducted in three villages from two districts (Bosset and Adamitulu Jido-Kombolcha (AJK) in the CRV region of Ethiopia. Information collected from the interviews of 60 farmers and two focus group discussions in the study area were used to acquire baseline information of how farmers in the CRV perceive climate variability, particularly rainfall variability, and how their understanding of climate variability translates into farm management decisions and actions. During RRAs, assessments were made regarding farmers' perceptions of the local climate variability, along with how farmers' observation and knowledge of the seasonal climate affect their agronomic decisions. Generally, farmers gave similar criteria to describe seasonal climatic conditions and to distinguish seasons as 'good', 'average' or 'bad' indicating a shared experience. Farmers' perceptions of seasonal climate variability and risk were mainly related to seasonal rainfall parameters in regards to crop growth and yield. Furthermore, in most cases, farmers' ratings of season 'types' were in agreement with the official classification published by the National Meteorological Service of Ethiopia. Of the rainfall characteristics, total amount of seasonal rainfall was rated less critical than variations in the timing of rainfall onset and dry spells during the growing season. The historical pattern, local weather observations, and other indicators allowed farmers to form expectations of what the rainfall conditions are likely to be in the season ahead. Many of the farmers agronomic decisions are based on the actual and expected seasonal rainfall, however, not all farmers respond in the same way. Farmers indicated that rainfall indicators are particularly important as many of the key management decisions (i.e., sowing date, cultivar choice, the portion of land allocated to maize and other crop species) are flexible according to the timing of the onset of the seasonal rains. Historically, farmers used to sow their late maturing maize if rain started early in the Belg season (March/April‚Äö-May), however, many farmers stated that they had noticed that the onset and distribution of early seasonal rainfall had become less reliable and more variable from the 1990s onwards. Farmers explained Belg season as unreliable due to post-sowing dry spells of varying length that can risk their crop to fail and they often need to re-sow. Still, around 30% of the respondents at Bosset and 60% at AJK opted to sow a late-maturing cultivar if Belg rain did occur, while the remaining 60% of farmers would wait until June if rain established well in the Kiremt season (June‚Äö-September). In this study, less than 30% of the respondents applied mineral nitrogen (N) fertiliser, at sub-optimal rates, while 70% did not apply N fertiliser at all. Of the 70% of respondents who did not apply N fertiliser, nearly 40% of the respondents assumed that their fields were sufficiently fertile or non-responsive at all and there would be no yield advantage from applications of commercial N fertiliser. In 2012, a maize field experiment was conducted season at Melkassa, in the CRV, to obtain a comprehensive quality data set suitable for modelling purposes and to evaluate the responses of two locally adapted maize cultivars to contrasting sowing dates and N fertiliser application rates. Data included daily weather, crop properties (phenology, growth pattern, plant N concentration, grain and biomass yield of the locally adapted and medium-maturing maize cultivar, Melkassa-2), soil water and N characteristics and crop management details, along with the initial conditions of the soil profile (soil water and mineral N content and surface residue). These data were used to parameterise the Agricultural Production Systems sIMulator (APSIM) for one of dominant soil type representative of 'good cropping conditions' in the region. The parameterised model was evaluated against independent data from six maize experiments conducted between 2006 and 2012 at Melkassa. The model was evaluated by comparing the simulated and observed phenology, grain and biomass yields of maize cv. Melkassa-2 across a range of production situations at Melkassa. Generally, evaluation of the parameterised model against independent data showed that it was able to predict key crop responses including crop phenology, grain yield and biomass production as evidenced by different statistical indices for the goodness of fit between the simulated and observed values. The results demonstrated that APSIM-Maize is reliable and suitable for scenario analyses of maize production systems in semi-arid environments of Ethiopia. Subsequently, the APSIM-Maize model was configured to run long-term simulation experiments to explore the maize yield response to agronomic factors, which farmers who participated in the RRAs had identified as being important in managing climate risks. In the long-term simulations, a combination of varying sowing window, cultivar type and N fertiliser rates were considered to represent local management practices of typical farmers. In addition, agronomic recommendations of research and extension services were simulated along with other agronomic management measures. Simulations of maize yield were run for each year of the available historical weather records from weather stations nearby to the study villages (i.e., 34 years ranging from 1982 to 2015 at Adamitulu and 39 years ranging from 1977 to 2015 at Melkassa). For the sowing windows, early, normal and late sowing dates were considered. Cultivar choices included early-, medium- and late-maturing maize cultivars and three rates of N fertiliser were applied: 0 kg N ha\\(^{-1}\\) (N0), 25 kg N ha\\(^{-1}\\) (N25), and 50 kg N ha\\(^{-1}\\) (N50). Altogether, there were 54 simulation scenarios to analyse for both Adamitulu and Melkassa. The production risk associated with each combination of agronomic factors were estimated thereby creating best management options that farmers may possibly consider in the future when making decisions related to maize production under their local environment, which is characterised by highly variable and uncertain climate. Early sowing (March/April‚Äö-May) ensured a sowing opportunity in more years compared to normal or late sowing, however, the likelihood of complete crop failure was greatest for early sowing (10%), due to a false start of rain or a risk of post-sowing dry spells, with risk decreasing as sowing was delayed from a normal (5%) to late window (<5%) during the Kiremt season. For late sowing, crop failure was unlikely, except for the late-maturing cultivar at Adamitulu where crop failure was ~15% more likely. For the early sowing, the late-maturing cultivar out-yielded the earlier cultivars at all levels of cumulative probability in 90% of the years. For the normal sowing, there was at least 88% likelihood of yield gain from selecting late-maturing cultivars compared to earlier cultivars irrespective of the N rate applied. At Adamitulu, the yield advantage of the late-maturing cultivar was greater if sown early (1 March‚Äö-30 May) instead of later (early- to mid-June or mid- to end-June). At Melkassa, the yield gain was greater if the late-maturing cultivar was sown during the normal (1‚Äö-15 June) and late (16‚Äö-30 June) sowing window rather than early (1 April‚Äö-30 May). For both locations, the long-term median yield of the late cultivar was greater than the early or normal cultivar, especially in high to average yielding years. In contrast, selecting an early cultivar reduced median yield. Irrespective of sowing time, there was at least an 85% likelihood of a yield loss from using an early cultivar than the medium and the late cultivars. However, for the late sowing at Adamitulu, the likelihood of yield penalties was only 65% when using an early or medium cultivar instead of the late one. Application of N fertiliser produced greater yields compared to unfertilised maize in at least 85% of the years regardless of the sowing window and cultivar type. With application of N50, there was a 65% likelihood that the yield gain would be more than the maximum yield that could ever be achieved with the application of N25. Averaged across locations, application of fertiliser could result in increases in the long-term median yields of 77% at N25 and 133% at N50 (2.7 and 3.5 t ha\\(^{-1}\\) vs. 1.7 t ha\\(^{-1}\\)) compared to the baseline N0. There were large shifts in cumulative distribution functions towards greater yields with application of either N25 or N50 compared to N0, although to varying degrees depending on the sowing time and the cultivar type. For a late cultivar sown at early and normal sowing windows, and for a medium cultivar sown late, the long-term simulations showed that application N25 could increase yield in more than 95% of the seasons without affecting the inter-seasonal variations in yield (as indicated by CV%) compared to N0. On the other hand, the locally recommended rate of N50 reduced maize yields in as much as 20% of the seasons comp...