Optimizing fertilization schemes to narrow the maize yield gap in smallholder farming systems in southern Ethiopia

Abstract

Maize (Zea mays L.) is an important food crop in Ethiopia, but productivity is low mainly due to low soil fertility and suboptimal fertilization. Therefore, this study aims to determine the yield, nutrient use efficiency and economic feasibility of maize production under various fertilizer applications and test the suitability of the Quantitative Evaluation of the Fertility in Tropical Soils (QUEFTS) model for predicting maize yield response to fertilization in Sidama region, southern Ethiopia. On-farm experiments were conducted at six sites (Site 1–6) of Sidama region, southern Ethiopia during the 2019 growing season. The experiments were laid out in a randomized complete block design (RCBD) with three replications. The experiment was a nutrient omission trial with seven treatments: control, two full NPK treatments and four nutrient omission treatments with contrasting N and P rates. Omitting N resulted in 5–28 % yield loss and omitting P resulted in 4–44 % yield loss compared to the lower rate of full NPK treatment across all study sites. Whereas omitting K resulted in 21 % yield loss only at sites 2 and 3. An increase in maize yield was mainly associated with an increase in both nutrient uptake and nutrient use efficiency of NPK. The results showed the need to revise blanket recommendations since the highest mean grain yields and net economic returns with acceptable marginal rates of return were obtained with NPK application or higher levels of NP (N2P2 treatment). In addition, the present QUEFTS model validation study revealed the good fit between QUEFTS model predicted maize grain yields (6.3 t ha⁻¹) to the average actual yields (7.4 t ha⁻¹) and this was also confirmed by small average values of RMSE = 1.5 t ha⁻¹ and PBIAS = 6.9 %. Thus, the model can be a promising option for development of site specific fertilizer recommendations under smallholder farming systems in the region.

Abbreviations

QUEFTS

Quantitative Evaluation of the Fertility in Tropical Soils model

RCBD

A randomized complete block design

1. Introduction

Smallholder farming systems in Ethiopia are generally characterized by low productivity due to limited adoption of improved technologies (fertilizers, quality seeds, agrochemicals and mechanization), low crop diversification (dominantly mono-copping), and poor tillage practices, which result in considerable soil degradation [1,2]. In Ethiopia, degradation of arable land due to inherent low soil fertility, sub-optimal fertilizer use (negative soil nutrient balances) and soil erosion are of great concern. These problems have been accelerated due to population increase, farmers’ low adaptive capacity of improved technologies and climate change [3,2].

Maize (Zea mays L.) is one of the strategic commodities over the coming ten years (2021–2030) in Ethiopia as part of the national food security strategy [4]. The crop is the first in productivity and total production among all cereals and second in area coverage next to teff [Eragrostis teff (Zucc.) Trotter]. Nearly 9 million smallholder farmers in Ethiopia are cultivating maize, mostly for human consumption and as a source of income [4]. Maize is a nutrient demanding crop, with high uptake rates of nitrogen (N), phosphorus (P) and potassium (K) [5]. The national average yield is 3.5 t ha⁻¹, which is far below its agronomic potential (Yp = 15.8 t ha⁻¹) (www.yieldgap.org). Particularly insufficient fertilization is causing low yields [6].

In Ethiopia, fertilizer use for maize is generally far below actual crop demand. Fertilizers are applied as blanket recommendations (64 kg N and 20 kg P ha⁻¹), with no adjustment to diversified agro-ecologies, limiting crop productivity and impacting soil quality in terms of soil nutrient depletion and low organic matter turnover [7,8]. However, studies revealed that optimum N and P application rates varied for different maize growing locations [9,10], showing that using blanket fertilizer recommendations can no longer be an appropriate practice. Other studies confirm that ignoring important soil nutrients, other than N and P, in any crop production system in the country could result in significant yield losses, at least in specific locations [11,12]. For instance, K fertilizer was not applied in soil fertility management schemes due to a misconception that Ethiopian soils are rich in K [13]. Continuous cultivation of crops without K fertilizer [14], substantial removal of straw from the field [15], and introduction of high yielding crop varieties [16], can result in a considerable drain of soil K reserves, thereby increasing the responsiveness of crop yields to K fertilization [17,14,18,19].

In this context, it is imperative to assess maize yield response to NP and balanced NPK application to improve nutrient use efficiency, minimize soil fertility degradation and sustain intensification of crop productivity in southern Ethiopia. Raising fertilizer use in the region requires, in part, better fertilizer recommendations, which, in turn, will require more accurate assessments of likely yield responses in different locations and management contexts [20]. It is thus important to develop site-specific nutrient management options for Ethiopia, where management practices, soil types and fertility status, climatic conditions and other factors governing yield response to nutrients are considerably heterogeneous [21,22,23].

Nutrient omission trials are useful to assess which nutrients are limiting to formulate site-specific fertilizer management [22,24,25]. However, conducting repeated field experiments to determine yield responses to NPK fertilizers is an expensive approach [26]. Complementary, empirical models such as the Quantitative Evaluation of the Fertility in Tropical Soils (QUEFTS) model has the potential to identify yield limiting nutrient on the basis of limited soil data [27] which saves money and time. Furthermore, the QUEFTS model can be used as a tool to predict yields, income or net returns for different combinations of fertilizer application rates for site specific fertilizer recommendation [27,28]. The model has been applied to different locations with a range of soil types and crops in tropical countries including Ethiopia [29,30,31,32,77]. However, it has rarely been tested in Sidama region of southern Ethiopia. It is for the first time the QUEFTS model was tested using soil samples collected from the study sites and validated using grain samples harvested in southern Ethiopia as it is important to test the suitability of the model first to make site specific fertilizer recommendation for the sites.

In addition, several studies undertaken to examine yield response to fertilization rarely considered the importance of economic analysis prior to making fertilizer recommendations for smallholder farming systems. Although the increase in yield is important, the costs associated with enhancing yields and how the recommendations and their associated risks may affect farmers’ profits needs to be considered [33,34,35]. The Marginal Rate of Return (MRR) can be used to determine whether it is profitable for a farmer to invest in fertilizer application [36]. A MRR is acceptable if the return from each extra unit invested outweigh the cost of the extra unit invested. A MRR greater or equal to 100 % is used as a baseline for recommendations [37].

To develop strategies for improved nutrient management and optimize fertilizer recommendations in specific regions, there is a need to understand the nutrient status of the soil, the magnitude of crop response to fertilization and the nutrient use efficiencies in specific locations. The objectives of this study were to: (1) determine yield, nutrient use efficiency and MRR for maize in response to the application of different fertilizer combinations, and (2) test the suitability of QUEFTS for predicting maize yield response to fertilization in the Sidama region of southern Ethiopia.

2. Materials and methods

2.1. Study locations

The field experiments were conducted at six sites (Sites 1–6) in three districts (Loka Abaya, Boricha, and Hawassa Zuria) of the Sidama region, southern Ethiopia (Fig. 1), during the 2019 growing season. There was 1426 mm rainfall in Boricha, 1186 mm in Hawassa Zuria and 1402 mm in Loka Abaya during the growing season (from March to October). Similarly, the mean monthly maximum and minimum temperatures were 24 and 22 °C, 20 and 19 °C, and 21 and 19 °C, respectively for the three districts. The soils at all study sites have been under extensive management with a history of minimal agricultural input use prior to the establishment of the experiment. Furthermore, the sites were cultivated with maize, at least for five years prior to the establishment of the field experiment. The dominant soil type at the study districts are chromic luvisols and eutric vertisols at Boricha (FDREPCC, 2008), Cambisols at the Loka Abaya (Nigussie et al., 2021) and Andosols at Hawassa (Kebede et al., 2019).

Fig. 1.

Map of study districts and sites, Loka Abaya district (Sites 1 and 2); Boricha district (Sites 3 and 4); Hawassa Zuria district (Sites 5 and 6).

The soils of the study sites were generally clay loam in texture and had a bulk density ranging from 0.93 to 1.16 g cm³ (Table 1). The soils were moderately acidic in pH-H₂O [38] ranging from 5.62 at Boricha (Sites 3 and 4) to 6.13 in the other locations (Hawassa Zuria and Loka Abaya) (Table 1). The soil organic carbon (SOC) contents at the sites were in an optimum range from 1.5 to 2.9 % [38]. Total soil N (% TN) at Sites 1, 3, 4, 5 and 6 ranged from very low to low, while Site 2 was high in TN [Table 1; 38]. The soils of the study sites had generally very low plant available soil P (Av.P), very high exchangeable potassium (Kex) and high cation exchange capacity (CEC) [Table 1; 20]. Other nutrient contents of the soils at the experimental sites are depicted in Table 1.

Table 1.

Soil properties in the surface soil layer (0–20 cm) of six on-farm experimental sites in Loka Abaya, Boricha and Hawassa Zuria districts of the Sidama region, southern Ethiopia.

Soil parameters	Districts
	Lokabaya		Boricha		Hawassa Zuria
	Site 1	Site 2	Site 3	Site 4	Site 5	Site 6
BD (g cm−3)	1.12	1.16	0.99	0.93	1.03	1.01
Sand (%)	30	20	34	26	34	38
Clay (%)	32	42	38	38	28	26
TC	CL	C	CL	CL	CL	CL
pH	6.1	6.0	5.6	5.6	6.1	6.2
OC (mg kg−1)	20 × 103	23 × 103	23 × 103	29 × 103	16 × 103	15 × 103
TN (mg kg−1)	1 × 103	8 × 103	0.4 × 103	1.3 × 103	0.4 × 103	0.3 × 103
Av.P (mg kg−1)	2.7	2.9	6.1	4.0	4.1	5.3
Kex(cmol kg−1)	0.8	1.7	1.4	1.4	2.4	2.0
CEC (cmol kg−1)	25	41	21	19	38	39
Ca (mg kg−1)	2169	1737	2551	2412	1723	1768
Mg (mg kg−1)	221	194	235	219	192	198
Fe (mg kg−1)	246	296	184	175	154	145
Mn (mg kg−1)	93	109	85	89	52	54
Cu (mg kg−1)	3.8	3.6	5.4	7.5	12.5	14.3
Zn (mg kg−1)	0.002	0.1	11.4	7.4	1.4	0.6

Note. BD, bulk density; TC, textural class; CL, clay loam; C: clay; OC, soil organic carbon, TN, total nitrogen, CEC, cation exchange capacity, Av.P, available phosphorus, Kex, exchangeable potassium.

2.2. Experimental design and treatments

The field experiments were laid out in a randomized complete block design (RCBD) with seven treatments and three replications with a plot size of 3 × 4 m, leaving 0.5 and 1 m spaces between the plots and blocks, respectively. Treatments were: a control without fertilizer application, four nutrient omission treatments (two levels of NP (N1P1, N2P2), N1K, and P1K), and two full NPK treatments (N1P1K and N2P2K) (Table 2). Urea [CO (NH₂)_2, 46 % N], Triple supper phosphate [Ca (H₂PO₄)₂·H₂O, 20 % P] and potassium chloride [KCl, 52 % K] were used to fertilize the plots based on treatments.

Table 2.

Treatments and fertilizer application rates used in the field experiment.

	Treatment	N applied	P applied	K applied
		____ kg ha−1 _____
1	Control	0	0	0
2	P1K	0	20	33
3	N1K	46	0	33
4	N1P1	46	20	0
5	N2P2	69	30	0
6	N1P1K	46	20	33
7	N2P2K	69	30	33

2.3. Soil sampling and analysis

Before planting, a composite soil sample from the surface soil (0–20 cm) was collected at each site by combining 20 randomly taken samples per site from the experimental fields. The composite soil samples were air-dried and passed through a 2 mm sieve for analyses of soil particle size distribution using the Hydrometer method [39], soil pH in 1:2.5 soil:water ratio using a digital pH meter, CEC using the ammonium acetate method [40], A_v.P using the Olsen method [41], K_ex using the NH₄OAc method [42], and available micro-nutrients (Fe, Mn, Cu, Zn) following the DTPA method [43]. The soil samples were further passed through a 0.5 mm sieve for determination of SOC by wet oxidation method [44] and total N by Kjeldahl digestion (Bremmer and Mulvaney, 1996). Additionally, undisturbed surface soil samples (0–20 cm depth) were taken from each experimental site for measuring bulk density as described by Ref. [45].

2.4. Agronomic management

We used the recommended high yielding maize variety (Zea mays L., cv. BH-540) as test crop. Maize seeds were sown early March 2019 using a spacing of 75 cm between rows and 25 cm between plants. Two seeds per hole were hand sown and the stands were thinned to 1 plant per hole at about 2 weeks after emergence. The P and K fertilizers were applied all at once at planting whereas the urea was split-applied (half at planting and the remaining half at flowering) using band application method. All management practices including weeding, were done uniformly for all treatments. The weeding was undertaken three times during the growing season at all sites as recommended for the area. The incidence of American army worm (Mythimna unipuncta) on maize, particularly at Sites 3 and 4 were controlled by the application of Cypermethrin 1 %. Maize biomass and grain yields were determined at physiological maturity from plants in the three central three rows from a net plot area of 7 m². At harvest in late September 2019, aboveground biomass and grain yield was measured after sun drying until constant weight was attained in greenhouse shade, and weighed using a digital balance. Hundred seed weight was determined from random grain samples in each plot. Grain moisture content (MC) was measured using a grain moisture tester (Dickey-John Multigrain) and the final grain yield and hundred seed weight were adjusted to a MC of 12 %.

2.5. Plant tissue sampling and analysis

Maize grain and stalk samples were separately sampled at physiological maturity from each plot at Sites 1, 3 and 5. Plant tissue samples were collected from each plot, dried and analyzed separately for N, P and K concentration. Nitrogen was determined by the Kjeldahl method [46]. Concentration of P and K were determined by ash and HCl dissolving methods by inductively coupled plasma atomic emission spectroscopy (ICP-AES) [47].

2.6. Data analysis

Grain and stalk nutrient uptake were calculated by multiplying N concentrations in the tissues with the respective stalk and grain yields ha⁻¹. Total nutrient uptake was calculated as the sum of grain and stalk uptake and was expressed in kg ha⁻¹ of the nutrients. The empirical formulas (Equations (1), (2))) were used to determine agronomic efficiency (AE) and apparent recovery efficiency (ARE) [48].

Agronomic efficiency (AE), the economic production obtained per unit of nutrient applied, is calculated as:

$${AE}_i=\left({GY}_f^i-{GY}_i^o\right)/I_i$$ (1)

Where AE_i is the agronomic efficiency for nutrient i (N, P, K); GYⁱ_f is the grain yield with addition of fertilizer contains nutrient i; GY⁰ is the grain yield in the control or unfertilized treatment; I is the input of nutrient i in kg ha⁻¹.

Apparent crop recovery efficiency (RE) of applied nutrient (kg increase in nutrient uptake per kg nutrient applied) is computed as:

$${RE}_i=\left(U_f^i-U_0^i\right)/I_i$$ (2)

Where RE_i is the recovery efficiency for each nutrient i (N, P, K); Uⁱ_f is the uptake of nutrient i with the addition of fertilizer contains nutrient i; $U_0^i$ is the uptake of nutrient i in the control or unfertilized treatment; I_i is the input of nutrient i in kg ha⁻¹.

The economic feasibility of fertilizer use was assessed for every site by estimating net benefit, which was calculated by subtracting the labor and fertilizer costs from gross return. Gross return was calculated by multiplying the yield adjusted downward by 25 % [49] with market price. The price of urea, TSP and KCl fertilizers were 1000, 1654 and 1500 ETB per 100 kg, respectively in 2019 (One Ethiopian Birr (ETB) = 0.034 US$ in 2019). The local market price of maize in the experimental year was 10 ETB kg⁻¹ and the labor costs for land preparation, sowing, fertilizer applications, weeding and harvesting was 50 ETB person⁻¹ day⁻¹ and for these, the same cost considered for all the treatments except for the control or non-fertilized treatment that the cost of fertilizer application was not included. The labor cost needed for maize cultivation per hectare for each treatment was estimated based on the approximate cost incurred in the farmers’ field. The dominance analysis procedure was used to select potentially profitable treatments. The method comprised ranking of treatments in order of ascending total variable cost (TVC), from the lowest to the highest [49]. This helps to eliminate those fertilization treatments costing more but producing a lower net benefit than the next lowest cost treatment. The selected and rejected treatments by using this technique were referred to as undominated and dominated treatments, respectively. For undominated treatments, a percentage marginal rate of return (% MRR) was calculated. The ratio of monetary value of extra maize yield to the fertilizer costs (MRR) was then used for judging whether or not a given type and amount of fertilizer is profitable for maize production in the study area.

For determination of fertilization effect on maize yield responses, AE and RE were evaluated using General Linear Models (GLM) procedure of the Statistical Analysis System (SAS) version 9.2 [50]. The data from each site was separately analyzed as error variances were heterogeneous [51]. Mean comparison for cropping systems effect on response variables were done by the Fisher's LSD test at 0.05. Pearson's correlation coefficients were computed to determine the nature of relationships between grain yields of common bean and maize and other agronomic parameters (traits).

2.7. Model evaluation

The QUEFTS model (Janssen, 1990) was evaluated using field experimental data collected during 2019 season from the three experimental sites (Sites 1, 3 and 5). The Excel version of QUEFTS model was run with the following input parameters (i) soil properties (i.e. soil pH, SOC, K_ex, Av.P): from the soil analysis results shown in Table 1, (ii) 7 different NPK fertilization treatments, (iii) default QUEFTS model maize nutrient recovery values of 0.5 for N and K, and 0.1 for P fertilizer and (iv) using default QUEFTS maize internal efficiency values (maximum IE borderline values (i.e. 70, 600, 120 kg above ground biomass per kg N, P and K, respectively) and minimum IE borderline values (i.e. 30, 200, and 30 kg above ground biomass per kg N, P and K, respectively [27]. Model outputs were used to predict maize grain yields [27].

To test the accuracy of the QUEFTS model, predicted values were plotted against measured values. Further, three statistical tests including root mean square error (RMSE) (Eq. (3)) and percent bias (PBIAS) (Eq. (4)) were used to evaluate model accuracy.

$$RMSE=\frac{\sqrt{{\sum }_{i=1}^n{\left(O_i-P_i\right)}^2}}{\mathrm{n}}$$ (3)

Where P_i is the predicted value, O_i is the observed value, and n is the number of observations. The RMSE is an error index where lower value indicates better model performance [52].

$$PBIAS=\left[\frac{\sum _{i=1}^n\left(O_i-P_i\right)*100}{\sum _{i=1}^n{O}_i}\right]$$ (4)

Where $P_i$ is the predicted value, $O_i$ is the observed value.

The optimal value of PBIAS is 0, with low-magnitude values indicating accurate model simulation. Positive values indicate model underestimation bias, and negative values indicate model overestimation bias [53].

3. Results

3.1. Effect of NPK fertilization on maize grain yields

Irrespective of the rate, NP and NPK fertilizer applications resulted in significantly higher maize grain yields compared to the unfertilized control and NK and PK fertilization (P < 0.05). Indeed, maize yield response to the applied fertilizers varied across the experimental sites (Fig. 2). The highest grain yield was obtained from N2P2 at Sites 1, 4 and 5, whereas the highest grain yields were obtained from N2P2K at sites 2 and 6 and from N1P1K at Site 3. Important is to note that the grain yield from the treatments that included both N and P fertilization (N1P1, N1P1K, N2P2, and N2P2K) were comparable at most sites. There was no difference in maize grain yields between N1K and P1K treated plots at all sites (P > 0.05), except at Site 3, where the yield response to P1K application was significantly higher than that of N1K application (P < 0.05).

Fig. 2.

Grain yield responses of maize to different fertilizer combinations at sites 1 to 6 and average response of all six sites; Error bars at sites 1 to 6 indicate standard deviation (n = 3 replications); Error bars in sites average indicate standard deviation (n = 6 sites); Different small letters denote significant differences between treatments at each site at P < 0.05.

The increase in fertilization rate from N1P1 to N2P2 resulted in an increase in grain yield of maize except at Site 6, although the increase was only significant in Sites 3 and 4 (P < 0.05) (Fig. 2). Across sites, significantly highest average grain yields were obtained from the application of N2P2, N1P1K and N2P2K (Fig. 2, average). Whereas, significantly lowest average grain yield with fertilizer application was obtained with the application of N1K and P1K and for the control treatment (Fig. 2, average). Among the locations (irrespective of fertilization), highest average maize yield (7.3 t ha⁻¹) was obtained at Hawassa Zuria district (Sites 5 and 6) followed by at Loka Abaya (Sites 1 and 2) (7.0 t ha⁻¹) and Boricha district (6.9 t ha⁻¹) (Sites 3 and 4).

3.2. Effects of NPK fertilization on nutrient uptake

The effects of fertilizer treatments on nutrient uptake and nutrient use efficiencies were computed only for Sites 1, 3 and 5 (one site from each study district), as nutrient content of the maize grain and stalks were only measured for these particular sites. Application of N significantly increased the above ground N uptake on average by 15, 16 and 25 % at Sites 1, 3 and 5, respectively compared to the none fertilized (control) treatment (Fig. 3). Combined application of N with P (NP) and PK (NPK) significantly increased total N uptake in above ground biomass more than combined application of N with K (NK) and P with K (PK). Total N (stover and grain) uptake in maize increased with increasing rate of NP (from N1P1 to N2P2) and NPK (N1P1K to N2P2K) at most sites, although this was not always significant.

Fig. 3.

Maize total nitrogen (A) and phosphorus (B) uptake at Site 1, Site 3 and Site 5; Error bars indicate standard deviation (n = 3). Different small letters denote significant difference between treatments at each site at p < 0.05.

P application also increased above ground P uptake by 13 %, 28 % and 3 % at Sites 1, 3 and 5, respectively. Combined application of P with N (NP) increased total P uptake in above ground biomass more than combined application of P with K (P1K) (Fig. 3; Fig. S1). However, this effect is only significant at Sites 1 and 2 with combined application of N2P2. However, application of K had no significant effect (P > 0.05) on K uptake in all the three sites (Fig. S1).

3.3. Relationship between maize yields and nutrient uptake

Maize grain yield was positively and significantly correlated with total N uptake at all the three sites (Sites 1, 3 and 5), with total P uptake at Site 3 and K uptake at Site 5 (P < 0.001) (Table 3). The majority of the maize yield versus nutrient uptake values examined in this study were within the standard internal nutrient use efficiency (IE) borderlines (Fig. 4).

Table 3.

Correlation coefficients of maize grain and stover yield, correlated with NPK uptake in grain, stover and above ground biomass (sum of grain and stover).

Nutrient Uptake	Grain yield			Stover yield
	Site 1	Site 3	Site 5	Site 1	Site 3	Site 5
GN	0.93***	0.92**	0.93**	0.62**	0.49*	0.65**
GP	0.82***	0.85**	−0.03	0.55***	0.49*	0.2
GK	0.37*	0.71**	0.82***	0.07	0.17	0.75**
STN	0.24	0.22	0.27	0.63***	0.84***	0.54*
STP	−0.35*	0.13	0.51*	−0.21	0.2	0.7**
STK	−0.02	0.27	0.5*	0.25	0.62*	0.35
TN	0.74**	0.81***	0.90**	0.78***	0.78***	0.81**
TP	0.02	0.73***	0.33	0.04	0.53*	0.29
TK	0.04	0.38	0.65**	0.24	0.59**	0.5*

Note. GN, grain N uptake; GP, grain P uptake; GK, grain K uptake; STN, stover N uptake; STP, stover P uptake; STK, stover K uptake; TN, total N uptake; TP, total P uptake; TK, total K uptake; the symbols *, **, and *** are used to show significance at the 0.05, 0.01, and 0.001 p-values.

Fig. 4.

Relationship between above ground biomass yields of maize and NPK uptakes at Sites 1, 3 and 5. For each nutrient, the lower and upper boundary lines represent the maximum accumulation (a) and the maximum dilution (d) of that nutrient in the plant.

3.4. Effect of fertilization on nutrient use efficiency

3.4.1. Agronomic efficiency

Agronomic efficiency (AE) (kg grain kg⁻¹ of applied nutrients) of maize was significantly affected by nutrient application across study sites (Table 4). The AE of N (AE_N) ranged from 6.2 kg grain kg⁻¹ of applied N (N1K treatment) to 66.2 kg grain kg⁻¹ applied N (N2P2K treatment) across study sites. Significantly highest AE_N was obtained either from NP or NPK fertilizer application, whereas the lowest AE was obtained for the N1K treatment at most sites. AE_N did not significantly improve with increasing rate of N in both treatment combinations (NP and NPK) in all sites.

Table 4.

Agronomic efficiency (AE) of maize in response to different fertilizer treatments in Sidama region, southern Ethiopia.

Treatments	AEN (kg grain kg−1 applied N)			AEP (kg grain kg−1 applied P)
	Site 1	Site 3	Site 5	Site 1	Site 3	Site 5
N1P1	36.7a	32.2c	42.9ba	90.0c	73.9a	98.8a
N2P2	29.8ba	45.3d	38.6b	104.4cb	60.3ba	88.8a
N1K	19.1bc	10.9e	6.2c	–	–	–
P1K	–	–	–	113.6b	37.1bc	19.3b
N1P1K	22.3bc	74.1a	44.6ba	170.5a	51.4bac	102.5a
N2P2K	18.2c	66.2b	56.8a	101.5cb	27.9c	87.1a
Mean	25.2	45.74	37.8	166	50.9	79.3
CV (%)	22.7	8.6	19.6	10.1	25.5	20.9
LSD	10.8	6.7	24.07	22.1	24.0	31.3

Note. Within columns, means followed by different letter(s) are significantly different according to LSD (P < 0.05).

The AE of P (AE_P) ranged from 27.9 kg grain kg⁻¹ applied P (N2P2K, Site 3) to 170.5 kg grain kg⁻¹ of applied P (N1P1K, Site 1) across study sites (Table 4). Significantly highest AE_P was obtained from combined applications N1P1K at Site 1 and both rate of NP at Sites 3 and 5. In addition, equally significantly highest AEp was obtained with NP application at Site 5, whereas significantly lowest AE_P values were obtained from N1P1 at Site 1, from N2P2K at Site 3 and from P1K at Site 5. AE_P decreased with increasing levels of P in both treatment combinations (NP and NPK) at most sites, although this was not always significant.

3.4.2. Effect of fertilization on apparent nutrient recovery

The apparent recovery fraction of N (RE_N) varied between treatments at all sites (P < 0.05) except at Site 5 (Table 5). The RE_N ranged from 0.3 (for N1K treatment, Site 3) to 0.9 (for N1P1, N1P1K, Site 3) across study sites. The lowest RE_N was obtained from N1K treatment at all sites and equally lowest RE_N values were obtained from N2P2K and N1P1 at Site 5. On the other hand, the highest RE_N across sites was obtained for treatments that included the application of both N and P. For Site 1, N1P1, N2P2 and N2P2K had the highest RE_N, for Sites 3 and 5, N1P1K had the highest RE_N. Irrespective of treatments, the same higher average RE_N (0.68 kg N uptake kg⁻¹ of N applied) was obtained at Sites 1 and 3, followed by Site 5 (0.48 kg N recovered kg⁻¹ of N applied).

Table 5.

Effects of fertilizer treatments on apparent nutrient recovery of nitrogen (N) and phosphorus (P) in the Sidama region, southern Ethiopia.

Treatments	REN (kg N taken up kg−1 of applied N)			REP (kg P taken up kg−1 of applied P)
	Site 1	Site 3	Site 5	Site 1	Site 3	Site 5
N1P1	0.8a	0.9a	0.4a	0.3a	0.3a	0.1a
N2P2	0.8a	0.6ca	0.5a	0.2a	0.2a	0.2a
N1K	0.4c	0.3c	0.4a	–	–	–
P1K	–	–	–	0.2a	0.2a	0.1a
N1P1K	0.6b	0.9a	0.7a	0.2a	0.2a	0.2a
N2P2K	0.8a	0.7ba	0.4a	0.2a	0.2a	0.2a
Mean	0.68	0.68	0.48	0.22	0.22	0.16
LSD (%)	0.1	0.3	ns	ns	ns	ns
CV	8.1	20.9	30.8	52.1	23.9	47.7

Note. Within columns, means followed by different letter(s) are significantly different according to LSD (0.05).

The apparent recovery fraction of P (RE_P) did not differ between treatments at all sites (Table 5). The RE_P ranged from 0.1 kg of P uptake kg⁻¹ P applied to 0.3 kg of P uptake kg⁻¹ P applied. Unlike RE_N, the RE_P neither increased nor decreased at all sites.

3.4.3. Relationships among soil properties, grain yield and nutrient use efficiency

The correlation among nutrient use efficiency and grain yield varied with sites (Table S1). Grain yield significantly (P < 0.05) and positively correlated with N recovery and agronomic nitrogen use efficiency only at Site 3 but not in the other two sites. Whereas, grain yield significantly and positively correlated with P recovery and P agronomic use efficiency at Sites 1 and 5 but not at site 3. We also tested for correlations between soil properties, grain yield and use efficiencies, but most were found non-significant (Table S2).

3.5. Economic feasibility of fertilizer use

Irrespective of treatments, the net benefit from fertilized treatments were 5–39 % higher than non-fertilized control at all sites (Fig. 5). Relatively, the highest net benefits with acceptable MRRs were obtained with the application of N2P2 at Sites 3, 4 and 5; with the application of N1P1 at Sites 1 and 6, and N2P2K at Site 2 (Fig. 5). The combined application of K with N (N1K treatment) and NP (N2P2K) were profitable with acceptable MRRs at most sites but was not with P (P1K) except at Site 3 where P1K application resulted in a higher economic return with acceptable MRRs (Table S3).

Fig. 5.

Net economic return of different fertilizer treatments for maize production in the six experimental sites; line = average of six sites.

3.6. Actual versus QUEFTS predicted yields

QUEFTS predicted grain yields showed a reasonable agreement with measured actual yields (average RMSE = 1.5 t ha⁻¹ (Fig. 6a). The model showed a small underestimation bias in four out of six sites (PBIAS = 14 %), where the average QUEFTS predicted yield was 0.7 t ha⁻¹ lower than the average measured yield (Fig. S2). In two out of six sites, QUEFTS showed a small overestimation bias (PBIAS = −8.5 %) where the average QUEFTS predicted yield was 0.6 t ha⁻¹ higher than the average measured yields (See details for each site and treatment in Fig. S2). Generally, the agreement between actual and QUEFTS simulated yields was better for the NP and NPK treatments than for the control, N1K and P1K treatments at most sites (Fig. 6b). The performance of the model in estimating maize yield response for the fertilization treatments varied with sites. Based on RMSE values, relatively, the performance of the model in the study sites ranked as Site 5 < 4 < 1 < 6 < 2 = 3 (Fig. 6; Fig. S2).

Fig. 6.

(A). QUEFTS predicted yield (in t ha⁻¹) against actual yield (in t ha⁻¹). The different symbol colours and shapes indicate different sites. The solid line indicates the 1:1 (A). Across sites average of actual yield (in t ha⁻¹) against QUEFTS yield (in t ha⁻¹) for each treatment.

4. Discussion

4.1. Maize yield response to fertilization

The results showed large variation in maize yield response to fertilization treatments among the study sites. This shows a substantial effect of soil and other biophysical variability on maize yield. Similarly, other studies have reported variable maize yield responses to fertilization in different soils in Ethiopia [54,9]. Maize yields of the N omission (P1K) and P omission (N1K) treatments were mostly lower than that of NP and NPK treated plots, showing that both N and P are the most limiting nutrients in the soils of the study sites. Maize yields of the N omission plots (P1K) were lower than that of the P omission (N1K) at Sites 1, 2, 4 and 6, whereas it was higher at Sites 3 and 5, showing that N is the most limiting nutrient in the former sites, while P is most limiting in the latter sites. This justifies the need for site specific fertilizer management in smallholder farming systems for sustainable maize production. Similarly, Giller et al. [55] and Vanlauwe et al. [56] noted the need for modifying soil fertility management through balanced crop nutrition that takes into account site specific deficiencies in nutrients to sustainably increase crop productivity in SSA including Ethiopia. Omitting K application resulted in 21 % yield reduction particularly at Sites 2 and 3, suggesting the importance of considering K application in the future, as farming without K application cannot be sustainable, especially when N, P application rates are increased which lead to increased K demand [57]. Crop response to K application has also been documented in some parts of Ethiopia where the soils were previously considered to have adequate K prior to agricultural intensification [58,59].

Across sites, treatments with the higher N and P application rates (N2P2 and N2P2K) resulted in the highest yields closely followed by the treatments with the low N and P application rates (N1P1 and N1P1K). The high yields under these treatments were possibly due to the increased N and P rates which were also identified as the most yield limiting nutrients in the soils. In agreement with our results, other researchers indicated a yield advantage with increasing N and P rates above the blanket fertilizer recommendation rates in Ethiopia (46 N kg ha⁻¹ and 20 P kg ha⁻¹) [60,61].

The variation in yield responses to the fertilization treatments across the study sites may be attributed to the variation in soil characteristics or soil fertility, climate and management. The relatively higher yield responses to fertilizer application compared with the control treatment at Sites 5 and 6 are expected to be due to their corresponding low soil organic carbon and total nitrogen contents. Njoroge et al. [57] also reported the variability in crop yield responses to fertilizer application within and across sites in SSA which are often attributed to differences in inherent soil fertility and management. This suggests that site specific optimization of nutrient inputs and production costs are necessary for enhancing crop yields and sustaining production for subsistence farmers (77, 78). However, unbalanced nutrient supply might result in low plant nutrient uptake and utilization, and can cause low yields and depletion of soil fertility [62,54].

4.2. Fertilizer application, nutrient uptake and use efficiencies

In the present study, NP or NPK fertilization enhanced N, P and K uptakes as a result of increased nutrient availability in the soil with mineral fertilizer application. Moreover, enhanced root growth could have resulted in further increased nutrient uptake from deeper soil layers. The lowest stover and grain N, P and K uptakes were recorded from the control treatment followed by either P1K or N1K at most sites, showing N and P (NP) or balanced fertilization (NPK) is critically important in determining maize yields and the level of nutrient uptake (Fig. 3; Fig. S1). A close correlation between nutrient uptake and maize yields has also been reported previously [63,64,65]. Others have reported the complicated interactions among N, P and K in various crops. Osborne et al. [66] noted that P and K uptakes either decreased or were unaffected with increasing N application. Our results revealed that P and K uptakes were higher when applied with N, as evidenced by greater P and K accumulation in NPK and NP than in PK treated plots, which clearly indicates the synergistic effect of N on P and K uptakes. This seems to indicate that with increased uptake of N, the other nutrients also start to be limiting and the plant will take up more of the other nutrients as well. This would increase P and K uptakes since the plant requires balanced nutrition for optimal growth. In addition, with higher N uptake, better growth of the plant is achieved resulting in more uptake of P and K. Further, the supply of nitrogen enhances the growth of small roots and root hairs, which in turn facilitates the high absorbing capacity per unit of dry weight [67]. Across sites, the average aboveground N and P uptakes (118 kg N ha⁻¹ and 42 kg P ha⁻¹) were more or less equivalent or higher than the average total N and P supplied (119.3 kg N ha⁻¹ and 39 kg P ha⁻¹) including N and P applied as fertilizer (46 kg N ha⁻¹ and 20 kg P ha⁻¹) and N and P supplied from the soil as estimated from N and P uptake from the control treatment. Hence, the fertilizer levels applied in this study may not be sufficient for sustainable agriculture in the long run in the soils with continuous soil nutrient depletion. As a percentage of total uptake, K was removed more than any other nutrient followed by N and P, respectively, suggesting that soils could be rapidly depleted of K. Previous studies in sub-Saharan African countries have also shown negative nutrient balances for N, P and K contributing to declining soil fertility in small scale farming systems [68,69,70].

Omission of P (i.e. N1K treatment) extraordinarily reduced AE_N in all study sites, suggesting that it is required to fertilize the crop not only with N, but also with P to achieve higher yields over the long-term. Particularly combined application of N and P is needed as the study sites are shown to be deficient in these nutrients. Most of the AE_N values in this study except a few, were in the range of 30–90 kg grain kg⁻¹ N, confirming earlier findings by Dobermann (17). AE_P of the present study range between 19.3 and 170.5 kg grain kg⁻¹ of P, which is higher than the maximum AE_P values (50–52 kg grain kg⁻¹ of P) in other Africa countries [71]. However, they were in line with the maximum AE_P (100.8 kg grain kg⁻¹ of P) reported for high rainfall area in Ethiopia [21].

The average RE_N is 0.61 kg N taken up kg⁻¹ of N applied, which is close to the estimated ‘global’ average of 0.55 kg N taken up kg⁻¹ of N applied [72]. Additionally, Dobermann et al. [48] noted an increase in N recovery from 50 to 80 % within well-managed systems at low levels of N use. A study from Ethiopia also reported highest RE_N of 87.4 % for maize [59]. Our results revealed that RE_N was higher for NPK and NP treatments compared with NK and PK treatments. Unlike RE_N, RE_P was not affected by fertilization treatment at most sites. The RE_P obtained in this study ranging from 3 to 30 %, is mostly in line with the reported values of 10–30 % [73]. The difference in maize NPK uptake, agronomic and recovery efficiencies for the same fertilization treatments among the study sites seem to be attributable to differences in soil characteristics, nutrient supply, management and agro-ecological factors [74,29].

4.3. Economic feasibility of fertilization

Smallholder farmers are mostly financially constrained and thus require high net returns to justify their use of fertilizer for crops [75]. Our results revealed the net benefit from fertilizer treatments was higher than the control treatment at all sites. This clearly indicates that the low soil fertility and low fertilizer application rates are the problem and not the return on investment. High net returns were also reported for fertilizer use in Ethiopia [76] and other sub-Saharan Africa countries such as Kenya [77], Burkina Faso [78] and Malawi [76]. Generally, the highest net benefits with acceptable MRRs (>100 %) were obtained with the application of NP followed by NPK and NK applications (Fig. 5 and Table S3). From this, it seems that the combined application of N and P causes the positive yield response and better net benefits.

4.4. QUEFTS model and site-specific fertilization

Average measured actual maize grain yield was 7.4 t ha⁻¹ while average QUEFTS predicted yield was 6.3 t ha⁻¹. Generally, the model fit was good as the average actual yields were not too far off from the QUEFTS predicted values at most sites. This was confirmed by average RSME (1.5 t ha⁻¹) and average PBIAS (6.97 %). For instance, the calculated RMSE for some sites in this study is comparable with a RMSE (1.1 t ha⁻¹) reported by Shehu et al. [32] and Wijayanto and Prastyanto [79]. The performance of the QUEFTS model is particularly quite good for the treatments that include N and P fertilization. However, for the control, N1K and P1K treatments, the modelled yields were mostly lower than the measured values. Hence, it seems that the model predicts a lower soil N and P supply than observed in reality. These results seem to suggest that further efforts are required to parameterize and recalibrate the QUEFTS model for better prediction of maize yield response to fertilization and thus to come up with better site specific fertilizer recommendation in smallholder farming systems in the region.

The variation in QUEFTS model performance across sites might be due to the discrepancy in the model assumption that there are no yield limiting conditions other than NPK nutrients [80]. This might not be the case in these sites where varying weather conditions within a growing season might have resulted in drought stress at certain crop growth stages. Furthermore, there could be some imperfections in management such as insufficient weeding and insect pest attacks. Similarly, several studies noted the variation in model predicted and actual yields with varying conditions and imperfect management such as late weeding, drought, animal or insect attacks [20,81,32].

5. Conclusions

The results showed that average grain yields of N2P2, N2P2K, N1P1K treatments were significantly higher than those of P1K, N1K and control, suggesting that application of both N and P is strongly recommended to significantly increase the grain yield of maize in southern Ethiopia and the application of either N or P is not effective to increase the yield significantly even though K is applied. The results also showed the need to revise blanket fertilizer recommendations (N1P1) as the response to nutrient application varied across sites and better yield response obtained with higher NP rate and combined application of NP with K at most sites. Across study sites, more N, P and K were taken up than applied to the soil, hence fertilizer rates should be increased for long-term sustainable production as the soils are depleted of the nutrients. The present QUEFTS model validation study revealed that that there is a good correlation between the QUEFTS predicted and actual yields at most sites implying that the QUEFTS model can be a promising option for development of site specific fertilizer recommendations under smallholder farming systems in southern Ethiopia.

Disclosure statement

The authors report there are no competing interests to declare.

Data availability statement

The data that support the finding of this study are available from the corresponding author upon reasonable request.

CRediT authorship contribution statement

Tigist Yimer: Writing – original draft, Investigation, Funding acquisition, Formal analysis, Data curation, Conceptualization. Girma Abera: Writing – review & editing, Visualization, Validation, Supervision, Project administration. Sheleme Beyene: Writing – review & editing, Supervision. Arie Pieter Paulus Ravensbergen: Writing – review & editing, Conceptualization. Amrachu Ukato: Investigation, Data curation. Frank Rasche: Writing – review & editing, Supervision.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

The following is the Supplementary data to this article: