Physiological use efficiency (PUE), recovery fraction of applied nutrients and indigenous soil nutrient supply form the basis of site-specific fertilizer recommendations. To derive these parameters, and understand their variability, as well as yield responses and fertilizer use profitability, nutrient omission trials (NOTs) were conducted in farmers’ fields across different agro-ecologies in Bako (n = 37), Central Rift Valley (CRV) (n = 66) and Jimma (n = 44) regions of Ethiopia in the main crop growing seasons of 2015 and 2016. The treatments used in the NOTs were control, PK, NK, NP, NPK and NPKSM, where SM refers to secondary and micro nutrients, and applied levels of N, P and K, were 120, 40 and 40 kg /ha, respectively. The results showed that the average yields of the control treatment were 4.5, 3.1 and 2.9 t/ha in Bako, CRV and Jimma, whereas the average yields for the NPK treatment were 8.3, 4.9 and 7.9 t/ha in the respective regions. Nitrogen was limiting grain yield in all the three regions, whereas P limited yield only in CRV and Jimma. The average N agronomic efficiencies in Bako, CRV and Jimma were 25.7, 13.3 and 35.5 kg grain kg⁻¹ of applied N, respectively, under NPK fertilizer use. With the levels of fertilizer used in the NOTs, NK, NP and NPK treatments were profitable in Bako and Jimma, whereas PK was not. None of the fertilizer treatments were profitable in CRV. Soils in Bako and Jimma supplied more N and K but less P than the soils in CRV. The PUE at maximum accumulation, median and maximum dilution of N were 27, 54 and 80 kg grain kg^-1 N, while for P, the values were estimated to be 194, 350 and 505 kg grain kg^-1 P, and for K they were 16, 52 and 87 kg grain kg⁻¹ K. The estimated average N, P and K recovery fractions were 0.29, 0.05 and 0.06, respectively, in Bako, 0.22, 0.10 and 0.15 in CRV, and 0.38, 0.10 and 0.01 in Jimma. While these average parameter values are relevant, in particular agronomic use efficiencies and recovery fractions showed large variability and, moreover, averages were lower than what is deemed feasible with good agronomy. We discuss the variability in the derived parameters, the relation with yield levels, soil nutrient supply and rainfall, and conclude that caution is needed when deriving fertilizer recommendations from parameters obtained in on-farm experiments. Using single estimated average values is not sufficient: variability in these parameters and sub-optimum values need to be explained first, and derived insight should be used when developing site-specific fertilizer recommendations.

生理利用效率（PUE），施用养分的回收率和本地土壤养分供应构成了特定地点施肥建议的基础。为了推导这些参数并了解其变异性以及产量响应和肥料使用的获利能力，在中部纵谷（CRV）的巴科（n = 37）的不同农业生态中的农民田间进行了养分去除试验（NOT）。 ）（n = 66）和埃塞俄比亚的吉马（n = 44）地区在2015年和2016年的主要农作物生长季节。NOTS中使用的处理方法是对照，PK，NK，NP，NPK和NPKSM，其中SM指的是次要和微量养分以及氮，磷和钾的施用量分别为120、40和40千克/公顷。结果表明，Bako，CRV和Jimma的对照处理的平均产量分别为4.5、3.1和2.9 t / ha。NPK处理的平均产量在各个地区分别为8.3、4.9和7.9吨/公顷。氮限制了这三个地区的谷物产量，而磷仅限制了CRV和吉马的产量。Bako，CRV和Jimma的平均农艺效率为25.7、13.3和35.5 kg谷物kg在氮磷钾肥料的使用下，施用的氮分别为^-1。根据NOTS中所使用的肥料水平，Bako和Jimma的NK，NP和NPK处理有利可图，而PK则无济于事。在CRV中，所有肥料处理均无利可图。与CRV中的土壤相比，Bako和Jimma中的土壤提供了更多的N和K，但提供了较少的P。N的最大累积量，中值和最大稀释度下的PUE分别为27、54和80 kg谷物kg ^-1 N，而P的值分别为194、350和505 kg谷物kg ^-1 P，K它们分别是16、52和87公斤谷物公斤^-1K。估计的平均N，P和K回收率在Bako中分别为0.29、0.05和0.06，在CRV中分别为0.22、0.10和0.15，在Jimma中为0.38、0.10和0.01。尽管这些平均参数值是相关的，但特别是农艺学的使用效率和回收率显示出较大的变异性，此外，平均值低于良好农艺学认为的可行性。我们讨论了衍生参数的可变性，与产量水平，土壤养分供应和降雨的关系，并得出结论，从农场实验获得的参数推导出肥料推荐时需要谨慎。仅使用单个估计平均值是不够的：首先需要解释这些参数和次优值的差异，