Peer-reviewed articles published from 1996 to 2020 were accessed from Google Scholar, ScienceDirect, ResearchGate, and Francis and Taylor databases. We used the search string (maize* OR corn* OR wheat* OR sorghum* OR teff* OR barley*) AND (nitrogen* OR nitrogen fertilizer* OR nitrogen use efficiency*) AND (yield*) AND (Ethiopia*). For a study to qualify in this meta-analysis, the following selection criteria had to be met:

· the study was conducted in the field, not pot or greenhouse experiments

· each treatment had a minimum of three replications

· the study reported grain yield and/or NUE

· the experiment was conducted in Ethiopia, and

· experimental and control treatments were applied to the same agricultural site and system.

Results presented in graphs were extracted using GetData Digitizer 2.26 software (http://getdata-graph-digitizer.com/). Overall, 82 studies that met the aforementioned criteria qualified for the final database (see Table1 and TableS1 in supplementary information). Finally, the experimental sites of the studies included in this meta-analysis were plotted using ArcMap 10.4 (ESRI, 2018) (Figure 1).

Relevant variables were dissected and included in our database to assess the magnitude of yield response (Table 2). The major explanatory variables such as mean annual precipitation (MAP), mean annual temperature (MAT), N-application rate, soil texture, pH, initial soil organic carbon (SOC), and available phosphorous (AP), were categorized into different groups/levels to evaluate their effects on

n: number of observations; treatment yield: +N yields; minimum, maximum and the average treatment yield was expressed in kg∙ha⁻¹.

yield under N-fertilizer application. The annual temperature ranged from 11.6˚C to 33.5˚C, whereas MAP ranged from 249.2 to 1800 mm. The percentage clay content was used to categorize the soil textural classes if it was not directly indicated in the study according to [31]. The soils with a clay content below 20%, between 20% - 32%, and >32% were categorized under sandy, loamy and clayey soils, respectively. The percentage of soil texture included in this study was 75.38%, 21.61%, and 3% of clay, loam, and sandy soils, respectively. If the study has reported soil properties of different soil layers (depth), only the topmost layer (0 - 20 cm) was considered. The AP (mg∙kg⁻¹) and SOC (g∙kg⁻¹) were categorized as extremely low (≤6), low (6 - 12), moderate (12 - 18) and high (>18), respectively. The NUE, agronomic use efficiency (AE) (the increase in grain yield per unit of fertilizer N applied), and PFP (the ratio of treatment yield to N-inputs) were estimated accordingly.

The percentage yield response, the yield obtained from the application of nitrogen fertilizer over control, was estimated as indicated in Equation (1). In order to determine the robustness of the study, sensitivity analysis of the response ratio (RR) i.e., the ratio between the yield on the nitrogen applied plot and control plot, was performed, using the standard procedure to evaluate the overall effect sizes as given in Equation (2). The RR distribution of cereal yield response under N-application is not a typical normal distribution (p < 0.05) (Figure 2) and therefore, we have used a non-parametric Kruskal-Wallis one-way analysis of variance on ranks to evaluate the differences within two sub-groups of each indicator as indicated in [32].

% Y R = ( Y t − Y c Y c ) * 100 (1)

ln ( R Y ) = ln ( Y t Y c ) (2)

where YR, Yt, and Yc are yield responses, the grain yield of N-applied and N-omitted (control) yield, respectively.

The meta-analysis was performed in SPSS statistic version 22 and all plots

were designed in Sigma-Plot version 12.5 software. The mean effect size and bias-corrected and accelerated/BCA i.e., the 95% confidence interval (CI) for each categorical variable, were generated by bootstrapping at 4999 iterations in SPSS software. Differences between treatments (N-applied) and controls (N-omitted) were considered as significant (p < 0.05) if the 95% CI did not cross the line of zero effect (either a significant increase or decrease) and non-significant if it crosses the line of zero effect.

Overall, the application of nitrogen fertilizer significantly (p<0.05) lifted the average yield of all the crops to 3775.8 kg∙ha⁻¹ (64.8%) when compared to the control treatments which had 2593.3 kg∙ha⁻¹ (Figure 3). The overall yield response for all the crops was 1180.9 kg∙ha⁻¹, with wheat having 1545 kg∙ha⁻¹; maize, 1454 kg∙ha⁻¹; barley, 1155 kg∙ha⁻¹; sorghum, 499.8 kg∙ha⁻¹, and teff, 421.4 kg∙ha⁻¹ higher than their controls (Figure 4(a)). This clearly indicates that wheat responded more (96.5%) to N-application followed by barley, 84.36%; teff, 50.48%; maize, (40.7%); and sorghum, 23% (Figure 4(b)).

The AE_N and PFP_N to nitrogen fertilization were presented in Figure 5(a) & Figure 5(b). Overall, AE_N and PFP_N were significantly influenced by N-application.

The average AE_N and PFP_N values across all studies were 18.2 and 71.81 kg∙kg⁻¹, respectively. Also, AE_N and PFP_N were declined with the increase of the N fertilizer. At N rates of <30, 30 - 60, 60 - 100, and >100 kg∙ha⁻¹, the AE_N values were 23.42 kg∙kg⁻¹, 19.21 kg∙kg⁻¹, 16.43 kg∙kg⁻¹, and 14.35 kg∙kg⁻¹, while PFP_N were 139.5 kg∙kg⁻¹, 74.14 kg∙kg⁻¹, 51.93 kg∙kg⁻¹, and 32.24 kg∙kg⁻¹, respectively. Moreover, AE_N and PFP_N varied with crop type. The highest AE_N and PFP_N were realized in maize (23.8 and 104.13 kg∙kg⁻¹) followed by barley (22.65 and 76.9 kg∙kg⁻¹), and wheat (19.96 and 65.51 kg∙kg⁻¹), respectively, while the lowest AE_N and PFP_N were recorded in sorghum (9.46 and 63.44 kg∙kg⁻¹) followed by teff (6.36 and 26.92 kg∙kg⁻¹), respectively.

Mean annual temperature, MAP, nitrogen application rate, soil texture, soil pH, SOC, and AP had a significant impact on cereal yield. The yield of five cereal crops could linearly and significantly increase with N-application rate and was maximum at >100 kg∙ha⁻¹ (123.9%) and lowest (28.9%) where <30 kg∙ha⁻¹ of N was applied (Figure 6(a)). The lower temperatures of <16˚C resulted in a 71.8% yield increase, while a 52.4% yield increase was realized at temperatures above 22˚C (Figure 6(b)). Also, N application increased crop yields by 80.2% and 79.9% with a MAP of <700 mm and >1000 mm, respectively (Figure 6(c)).

The result revealed that the cereal yield response due to N-application was considerably higher at the pH range of 6 - 7 (77.2%) than at pH of <6 (58%) and >7 (58.3%) (Figure 7(a)). The highest yield response was observed in clay soil (75.5%) followed by loam soil (72.2%), while lowest in sandy soil (57.1%) (Figure 7(b)). Soil AP concentration resulted in minimum and maximum yield responses of 101.6% and 29.6%, respectively (Figure 7(c)). Moreover, N-application significantly increased yield by 92.4% where the initial SOC concentration was low, and by 50% where the initial SOC was extremely low (Figure 7(d)).

In the present meta-analysis study, N-application significantly increased grain yield in maize, teff, wheat, barley and sorghum by 64.8% overall (Figure 3 and Figure 4). The increase of crop yield under N application is due to the critical importance of nitrogen fertilizer as a macronutrient in the agricultural production system and its potential to augment the low nitrogen levels in soil [33]. It is also related to the high importance of N fertilizer in enhancing plant leaf area, and photosynthesis efficiency; which resultantly enhance harvest index, plant dry matter, and crop yield [2] [22]. Rational nitrogen fertilization on agricultural fields has paramount importance to improve soil and crop productivity, while overdose leads to deterioration of soil quality [29] [34]. In line with the present findings, several studies reported the positive response of crop yield to nitrogen application [2] [31] [35] [36]. Abera et al. [2] reported that maize crop fertilized with half (55 kg∙ha⁻¹) and full recommended (110 kg∙ha⁻¹) of N-fertilizer in Ethiopia has a grain yield advantage of 18% to 209% and 18% to 254% over the control, respectively. Also, a meta-analysis study conducted in Zimbabwe revealed a positive and significant yield response of 33.7% under N-application over the control [33].

We found the overall AE_N and PFP_N of 18.2 and 71.81 kg∙kg⁻¹, respectively, which also largely varied with N rate and crop type. However, our results show an overall inverse relation between NUE and N-application rate (Figure 5(a) and Figure 5(b)). This indicates N applied at low rates is efficiently utilized by the crop for biomass accumulation while continuous application of N beyond the optimum rate results in reduced N use efficiency and yield determinant factors other than N becomes more limiting when the crop approaches its maximum yield potential [27] [37]. Plants cannot absorb nutrients applied in excess due to their absorption mechanisms becoming oversaturated. Under these conditions, there exists a high chance of unabsorbed N loss to the environment through different mechanisms such as volatilization and leaching. The variation of NUE across the studied crops could be related to genotypic variation, environment, soil indigenous nutrient supply, management methods and nutrient application rate [11] [23] [38].

The variability in crop response, soil fertility differences, and climatic conditions and other factors makes the management and improvement of NUE more difficult. More importantly, understanding such variabilities is highly essential to design area-specific nutrient management practices. Several approaches have been developed to improve NUE in agriculture such as integrated soil fertility management [38], nutrient stewardship [39] [40], use of organic fertilizer [41], root-zone nutrient management [42], integrated soil-crop system management (ISSM) [43], using nutrient use efficient cultivars [44], precision farming [45] and so on. Overall, rationalization of the fertilizer use and adoption of an integral management approach based on an inclusive understanding of the yield and NUE limiting factors is an important step to achieve high crop productivity and high NUE [2] [25].

The present study revealed that the cereal yield response to N fertilizer was largely positive and significant, but the magnitude of yield response varied based on N-supply rate, MAP, MAT, soil texture, pH, initial SOC, and AP (Figure 6 and Figure 7). The observed linear relation between N-rate and grain yield (Figure 6(a)) emphasizes the need to increase yields by applying high rates of nitrogen than the average current N-application rate in Ethiopia. Gotosa et al. [33] found a linear relation of maize yield with nitrogen input rate, and the highest yield response was revealed at N-rate > 100 kg∙ha⁻¹, which was in line with the present findings.

Temperature is an important yield determining factor and its deviation from the optimum due to climate change or other factors negatively impacts crop productivity. A study conducted in China revealed that a unit increase of climate warming resulted in a reduction of maize yield by 2.6% [46]. However, the magnitude of change in yield varies depending on location, season, and soil inherent fertility status [46], which was concordant with our finding (Figure 6(b)). MAP is among the major factors that affect crop production, particularly in rain-fed regions [47], where there is a scarcity or lack of irrigation systems. For example, in Ethiopia, despite the smallholder accounts for over 95% of the total maize area and production, the irrigated areas account for only 1% of the total [17]. The distribution and pattern of rainfall considerably influence the grain yield and NUE [11] [48]. However, the observed increase-decrease-increase yield response trends in MAP (Figure 6(c)) might be related to the variation in water requirements of each crop.

The soil pH is a major indicator that plays a significant role in the availability of soil nutrients and also affects plant nutrient uptake and use efficiency. The higher yield response in soils with pH values ranging from 6-7 (Figure 7(a)) is implying the profound effect of soil pH on soil nutrient availability. In this range, most soil nutrients are optimally available to plants. Chen et al. [49] found a higher yield response in the near-neutral pH range (6.6 to 7.3). Regarding the soil texture, higher yield response was observed in clay soils followed by loamy soils (Figure 7(b)). The reason is the potential of clay and loamy soils to hold water and sequester organic carbon. Whereas, the lower yield response of sandy soil is due to its high leaching potential, low clay content, poor water retaining potential, and poor OM content [33] [50]. The highest yield response in low SOC (6 - 12 g∙kg⁻¹) and AP content (6 - 12 mg∙kg⁻¹) could be due to the potential of nitrogen fertilizer to improve infertile soils and resultantly crop productivity (Figure 7(c) and Figure 7(d)). The supply of soil nutrients in already fertile soils slightly enhances or even declines the grain yield. Another study revealed a higher yield increase under P-fertilizer application when the soil phosphorous is low [51] and under controlled-release nitrogen application when the SOC content is low [52]. Therefore, it was observed that the N-application in Ethiopia significantly increases crop yield, although the magnitude is affected by several explanatory factors and crop type.