Nutrient supply affects the yield stability of major European crops—a 50 year study

The Dikopshof long-term fertilization experiment was established in 1904 in Western Germany near Cologne. The soil type is a Chromic Luvisol (Rueda-Ayala et al 2018). It was developed from a >100 cm loess layer over a sandy-gravelly, highly permeable pleistocene middle terrace of the Rhine river. The arable layer has a loamy silt texture with a thickness of around 35 cm. The Atlantic climate with mild winters and summer has mean annual temperature of 10.3 °C and a mean annual precipitation of 693 mm (1955–2008, source: German Meteorological Service) (Zhao et al 2015). During the observation period a 5 year crop rotation typical for the region was maintained with (in the following order) Persian clover (Trifolium resupinatum L.), potato (Solanum tuberosum L.), sugar beet (Beta vulgaris), winter wheat (Triticum aestivum L.) and winter rye (Secale cereale L.). Each crop was grown in 24 fertilizer treatments (Rueda-Ayala et al 2018). The experimental design was not randomized. For each crop × treatment combination there is only one replicate (one observation) in each year. The five crops were grown in parallel, resulting in a total number of 120 plots, with a plot size of 18.5 × 15 m (supplementary figure S1 (available online at https://stacks.iop.org/ERL/16/014003/mmedia)). After harvest, potato leaves and stubbles of cereals were left standing until the beginning of October, when they were incorporated into the soil during tillage, whereas sugar beet leaves and clover cuttings were removed. Crops were grown under conventional management practice. Growth regulators, herbicides, and crop protection (fungicides and pesticides) were applied starting from late 1950s and early 1960s, respectively. In 1964 the ploughing depth was increased from 22 to 30 cm due to agricultural mechanization.

Yield data of potato and sugar beet (fresh weight), winter wheat and winter rye (14% moisture content) for the period 1955–2008 are reported as t ha^–1 (figure 1, supplementary table S1). We analysed data from 6 out of 24 treatments where nutrient supply was kept constant for all crops since 1953. The six selected treatments were fertilized with synthetic (mineral) fertilizers only. Plots had either received all major macronutrients (complete fertilization), did not receive nitrogen (–N), phosphorus (–P), potassium (–K) or calcium (–Ca) or had not received any fertilizer (–NPKCa) (table 1). Data from 1975, 1998 and 1999 were excluded from the analysis due to a technical failure in harvest within one treatment (1998) and exceptional sowing of spring cereals (1975, 1999) instead of the typical winter-cereals sown in fall of the previous years.

Zoom In Zoom Out Reset image size

For obtaining yield anomalies, data trends, which are considered as long-term changes in yield levels due to the combined effect of improved technology, agronomic management, and breeding, have to be removed. To account for curvilinear trends and non-stationarity we applied the locally weighted polynomial regression function to derive yield trends (T) (figure 1). Yields (Y) were normalized by the corresponding trend value (T), representing the expected yield value in each year (index t) to obtain relative yields (Y_r) (Lu et al 2017):

$$\begin{equation}{Y_r} = \frac{{{Y_t}}}{{{T_t}}}\end{equation} \tag{ 1 }$$

By subtracting 1 from each data point resulting anomalies indicate the relative deviation of yields from the expected annual yield in each year (with the latter being described by the trend).

$$\begin{equation}RYA = {Y_r} - 1\end{equation} \tag{ 2 }$$

Resulting relative yield anomalies (RYA) allow for a direct comparison of data obtained for different crops and within different treatments (figure 2).

Zoom In Zoom Out Reset image size

Yield stability was assessed using (a) the interquartile range of the RYA, (b) the standard deviation of the RYA, (c) first order lower partial moments (LPM) and (d) the relative frequency of negative, normal and positive yield anomalies. LPM are frequently used in agricultural economics (Antle 2010) to describe the shortfall risk below a specific threshold (k). We computed the first order LPM which describes the conditional expected value of shortfalls (accounting for the probability and magnitude of shortfalls). As a threshold we considered negative RYA deviating >15% from the expected (trend) yield value (k = −0.15). Results were affected by the value selected for k (with values ranging from −0.10 to −0.30 having been tested), but the overall pattern did not change. We further computed relative frequencies of normal years (anomalies within ±15% from the expected yields value) and positive/negative yield anomalies (>15% above/below the expected yield value, respectively). Stability measures were calculated for the data from the complete, the first (period 1: 1955–1981) and the second half (period 2: 1982–2008) of the study period. To further assess changes in yield anomalies and meteorological data over time we tested for the presence of monotonic trends using the non-parametric Mann Kendall test (Mann 1945). For RYA the absolute (non-negative) data values were used. Significant trends (p ≤ 0.05) were quantified using the non-parametric Sen's slope estimator (Sen 1968).

Meteorological data were obtained from the German Meteorological Service (http://www.dwd.de) and interpolated to a 1 × 1 km grid. The reference evapotranspiration (ET0) was calculated using the Penman–Monteith equation. For further details on interpolation and calculation approaches we here refer to (Zhao et al 2015). Daily data were aggregated to annual means and sums. The water balance (WB) was calculated as difference between annual ET0 and precipitation sums (in mm). While it was beyond the scope of the study to quantify the contribution of meteorological variability to yield stability, data were used as complementary information.

The association between RYA across treatments is considered as an indicator for the impact of nutrient supply on the response of crop yield anomalies to interannual weather variability. Strong associations indicate that the effect of annual weather conditions dominated over the effect of differentiated nutrient supply. In contrast, a lower level of association suggests that additional effects, such as nutrient supply, affected crop yield responses. Of particular interest for our study are changes in the number of simultaneous positive and negative yield anomalies among treatments over time. Associations were therefore quantified using the non-parametric Goodman and Kruskal's gamma coefficient (γ) which is based on the normalized difference between discordant and concordant data pairs (Goodman and Kruskal 1954). The γ coefficient describes the difference between the probability of concordant observations and the probability of discordant observations. Thus, negative values indicate a negative and positive values a positive association. Perfect positive and negative association are given at γ = +1 and γ = −1, while a value of 0 reflects no association at all.

All analyses were performed using the R statistical software version 4.0.2 (R Core Team 2020).

For winter wheat the interquartile range of the RYA was highest if grown on unfertilized plots (figure 3(a)). For potato and sugar beet and for rye it was highest on soils that haven not received K or N fertilizer, respectively (figures 3(c) and (d)). For a given treatment, the interquartile range of potato yield anomalies systematically exceeded those of all other crops. Winter rye yield anomalies were least affected by differences in nutrient supply (figure 3(b)).

Zoom In Zoom Out Reset image size

The standard deviation of potato and sugar beet yield anomalies exceeded that of cereals across treatments (figure 4(a)). The stability of rye was rather similar across treatments with slightly lower stability if grown on unfertilized soils or those where N supply has been omitted (figure 4(a)). Wheat, potato and sugar beet yields were least stable on zero fertilization and under K omission. Ca omission had a low effect on yield stability.

Zoom In Zoom Out Reset image size

The conditional expected value of yield shortfalls (with k = −0.15) for wheat and sugar beet (figure 4(b)) agreed with the standard deviation of the RYA (figure 4(a)): it was highest under no fertilization, followed by K omission. For rye it was highest in unfertilized plots and those where Ca was not supplied. For potato crops conditional expected values of shortfalls were rather similar across treatments. They were on average higher than for the two winter cereals.

All crops, except rye, showed a higher relative frequency of negative yield anomalies if grown under long-term omission of K or under no fertilization (supplementary figure S2). Correspondingly, the frequency of normal years (−0.15 > RYA < 0.15) was lower under these treatments and highest under complete fertilization or if only Ca had been omitted. For wheat and sugar beet frequencies of positive were similar to those of negative anomalies (supplementary figure S2). This implies that an increase in normal yield anomalies (higher stability) was related to a decrease in both, positive and negative anomalies.

For winter wheat the standard deviation of yield anomalies as well as the conditional expected value of shortfalls mostly indicated an increase or no changes in yield stability during the observation period (figures 4(a) and (b)). Yields in plots that did not receive K fertilizer (standard deviation and LPM) and the plots without P fertilizer (only LPM) were more stable during the second half (period 2) of the observation period (figure 4). The lower risk of yield failures was supported by an increased frequency of normal yield anomalies and a decreased frequency of negative yield anomalies (supplementary figure S2). Correspondingly, on the K omission plot we estimated a significant negative trend of absolute RYA values (slope: −0.002, p-value: 0.014).

For rye grown under N omission (–NPKCa and –N treatments), yield stability was slightly decreased in the second period (figure 4(a)). This is linked to a slight decrease in normal anomalies (supplementary figure S2). However, the expected value of shortfalls under Ca omission and no fertilization was decreased.

If grown in plots having received all nutrients, potato yield stability increased over time. This is indicated by a decreasing trend in RYA values (slope: −0.003, p-value 0.012), a lower standard deviation of RYA (figure 4(a)) and a higher frequency of normal yield anomalies (supplementary figure S2). However, both, frequencies of negative and positive yield anomalies were decreased. On Ca or K omission plots the frequency of normal years in period 2 was increased as well. However, this was rather linked to a decrease in positive yield anomalies (supplementary figure S2) and an increase in the standard deviation (figure 4(a)). If unfertilized, potato yield stability was higher during period 1. On such plots the frequency of both positive and negative yield anomalies were increased during the second half of the observation period (supplementary figure S2).

For sugar beet the standard deviation of RYA was higher during the second period across treatments except for complete fertilization and K omission where changes were small (figure 4(a)). The frequency of negative yield anomalies and normal years was increased and decreased in period 2, respectively, for almost all treatments (supplementary figure S2). The frequency of positive yield anomalies remained nearly unchanged. If unfertilized, sugar beet showed a significant upward trend of RYA values (slope: 0.007, p-value: 0.019).

Patterns of the standard deviation of RYA are similar to those of the coefficient of variation of absolute yield values (supplementary figure S3). Thus, the overall effects of differentiated nutrient supply are independent of whether absolute or relative yield anomalies are considered..

The association of yield anomalies across treatments was higher for cereals than for row crops (figure 5). At least 37% (for wheat) and 40% (rye) of the observation years yield anomalies were related to each other (concordant observations: positive or negative yield anomalies in both treatments). Higher proportions of ranked pairs in agreement indicate a more similar response of treatments to weather conditions across years. This suggests that weather-induced anomalies of cereals were less affected by nutrient omission (treatment) than those of row crops. For wheat and sugar beet grown under complete fertilization associations were lowest with the −NPKCa, followed by the −K treatment (figures 5(a) and (d)). For completely fertilized potato relations were lowest with the −K (figure 5(c)), followed by the −NPKCa treatment while for rye they were lowest with −NPKCa and −N omission plots (figure 5(b)). Thus, except for rye, the omission of K in particular led to a decrease in the association with anomalies observed under complete fertilization.

Zoom In Zoom Out Reset image size

Across treatments, yield stability was highest for winter rye, followed by winter wheat and sugar beet and was lowest for potato. The higher long-term stability of cereal yields compared with that of root crops agrees with results from other long-term experiments in Europe (Chloupek et al 2004). Especially winter cereals benefit from their long growing season. If soil water is replenished due to sufficient winter precipitation, early leaf and root growth allows them to use water more efficiently than spring-sown crops (Gan et al 2000). For spring-sown crops, such as potato and sugar beet, timing is more critical. Periods with unfavourable weather conditions and lower tuber growth cannot be compensated. This is especially true for potato crops, which have a shallow root system. The root density of sugar beet crops in deeper layers exceeds that of potato plants (Willigen and Noordwijk 1987). Thus, they are less vulnerable to seasonal fluctuations in growing conditions and their yields are more stable. In this context it is important to point out that cereal yields were reported as dry matter while for row crops yields were only available as fresh matter. This might have amplified differences in yield stability between crops. The higher stability of cereals was accompanied by a higher fraction of concordant data pairs across treatments – compared with those of row crops (figure 5). These stronger associations indicate a similar response of crops to annual weather conditions (positive or negative anomalies were observed across treatments). Thus, for cereals differences in nutrient supply were less critical for leading to either positive or negative anomalies than for row crops.

Highest yield stability was observed in plots receiving all nutrients (+NPKCa). Under no fertilization standard deviations of the RYA were two to three times higher than for complete fertilization plots (except for winter rye). Omitting N (for rye) or K (for all other crops) supply was the key factor for decreasing yield stability. These patterns were independent from data processing and also visible in the absolute yield data (supplementary figure S3). Adequate N supply that matches the demand for a crop rotation in an integrated system (Spiertz 2010) has also been reported as a critical factor for yield stabilization in other studies (Berzsenyi et al 2000, Varvel 2000, Hao et al 2007, Macholdt et al 2019). It promotes plant growth, rooting depth and density and thus, modulates soil characteristics and enhances plant resilience to adverse growing conditions. K supply mainly affects the water use efficiency of crops and is especially critical under growing conditions with frequent occurrence of drought stress periods (Grzebisz et al 2013, Zörb et al 2014). Low yield stability on K omission plots therefore suggests that the amount of crop available water was one of the main factors affecting yield stability at our site. Findings highlight the critical role of K supply for a sustainable crop production under rainfed conditions. They further suggest investigating the relation between water availability and yield stability at our site. While the mean annual WB (indicated in figure 2) cannot capture complex interactions at the field scale, the use of process-based models might help understanding moisture effects related to yield stability. Low yield stability on unfertilized plots might be related to sparse vegetation cover and therefore high soil evaporation losses. In addition, due to a poorer soil quality, they are more susceptible to waterlogging than those plots receiving more nutrients. Nutrient supply further affected the association strength between treatment-specific yield anomalies (figure 5). Anomalies observed under complete fertilization were least strongly related to those of unfertilized treatments. Thus, long-term nutrient omission strongly decreased the probability that crops respond similarly to weather conditions than crops grown under complete fertilization (positive or negative yield anomaly in both treatments). The omission of K was the most (for potato) or second most (for winter wheat and sugar beet) important factor for reducing associations. Since yields were most stable in the +NPKCa treatment these results provide further evidence on the critical role of K supply for stabilizing yields under changing weather conditions at our site.

Yield anomalies are largely controlled by changes in meteorological conditions (Moore and Lobell 2015, Ray et al 2015). In the 1980s a major regime shift has been observed leading to changes in meteorological conditions over Europe, such as increased temperatures (Reid et al 2016). In accordance with this, air temperature, radiation and ET0 sums significantly increased over time at our site (supplementary table S2). Trends suggest that under rain-fed cultivation conditions, as in this study, water availability had decreased over time. Correspondingly, the number of years with a negative WB (based on the ET0) was higher in the second half of the observation period (22 in period 2 versus 14 in period 1) and the number of years with a positive WB lower (12 in period 1 versus 3 in period 2) (figure 2). Sugar beet yields are largely controlled by water supply (Hoffmann and Kenter 2018). Thus, increased evapotranspiration rates and a decrease in soil water availability during the last decades might have contributed to the lower stability (despite a prolonged growing season). This was reflected in higher standard deviation of sugar beet yield anomalies (in N omission and unfertilized plots, figure 4(a)) and higher probability of sugar beet yield failure (N and Ca omission in figure 4(b)) in the second half of the observation period. We further observed a strong decrease in the association of yield anomalies among treatments in period 2 (supplementary figure S4(d)). Thus, the long-term omission of nutrients strongly reduced the probability, that plants grown in the different treatment plots respond similar to changing weather conditions (positive or negative yield anomalies across treatments). However, with full fertilizer application, sugar beet yield anomalies tended to be low and were rather constant over time (figures 3(d) and 4). When receiving all nutrients, the stability of potato yields was higher after 1981 (lower frequencies of positive and negative anomalies in supplementary figure S2 and the negative Sen's slope). However, it tended to be lower if plants were grown on plots where K was omitted (slightly increasing standard deviation in period 2 in figure 4 and lower frequency of normal years in period 2 in supplementary figure S2). Considering the importance of K fertilization under dry field conditions (Grzebisz et al 2013, Zörb et al 2014), the destabilizing effect of omitting K on potato yields in period 2 was thus probably related to a decrease in water availability over time. As for sugar beet, the fraction of concordant data pairs (positive or negative yield anomalies in treatment data pairs) was lower in period 2 (supplementary figure S4(c)). Thus, fertilization did not only affect potato yield anomalies in single years (figure 2) but also led to a contrasting response of crop yields to meteorological fluctuations over longer-term time scales. Contrasting with the findings for potato, the stability of wheat yields in K omission plots increased over time (negative trend of yield anomalies). The development of short-straw cultivars (being less susceptible to lodging) and the progress in agronomic management (e.g. plant protection and growth regulators) might have led to a lower dependency of wheat yield anomalies on K supply under adverse weather conditions. The lower dependency consequently led to stronger associations between anomalies under complete nutrient supply with those of the −K treatment during period 2 (1982−2008) (supplementary figure S4(a)).

On a longer-term time scale crop yields are influenced by changes in agricultural management (e.g. technology, fertilizer management and properties, management dates, crop protection, seed treatment) and crop varieties. These factors could not be kept constant over time and affected yield stability. Application of chemical crop protection strongly increased from the early 1960s to the 1990s. Research trials such as our site were subject to continuous observations. Nevertheless, chemical crop protection probably contributed significantly to a lower risk of yield loss. This is especially true for treatments receiving higher amounts of nutrients which are characterized by denser canopies with microclimates favourable for fungal infection and the spread of diseases (Tivoli et al 2013). Thus, crop protection might have contributed to the increase in yield stability of treatments receiving more nutrients over time (i.e. for potato in figure 4 and indicated by the negative trend in RYA). At our site, the cumulative effect of nutrient omission affected soil properties, such as pH-value and content of soil organic carbon. The water holding capacity of all treatments might have benefitted from an increase in the ploughing depth in the mid-1960s. Further, crop rotational effects (i.e. with respect to the nutrient availability after harvesting the preceding crop and the nitrogen fixation of clover) must be considered. We did not account for cultivar-specific traits, such as phenology (Rezaei et al 2018) and root characteristics (Li et al 2016). These traits might affect crop responses to fertilization (Addy et al 2020) and their adaptation capacity to variable environmental conditions (Blum et al 1989, De Vita et al 2010). Exemplarily, drought stress tolerance of potato cultivars was researched much later than that of cereals (Monneveux et al 2013). The selection for drought tolerance was less critical in countries where water shortage has not been an issue in the past or where potato is frequently irrigated, such as Germany. Therefore, more detailed studies are required to confirm and explain the patterns and trends in yield anomalies that we observed.

It was beyond the scope of this study to attempt a detailed mechanistic explanation for the yield stability observed at the Dikopshof long-term fertilization experiment. The interaction between genotype, management, and environment creates a high level of complexity (Lobell and Gourdji 2012) and crop yield responses to weather conditions strongly vary with crop phenological stage. Due to the high number of confounding factors statistical relations between weather and yield data and yield anomalies cannot be used for assessing causal relationships (Siebert et al 2017). Process-based crop models that allow for varying one input factor at a time, while keeping all others fixed, are required for separating weather effects from those of other factors. However, pronounced differences in yield stability and association between treatments illustrate that stability was strongly affected by the supply of nutrients, in especial by the omission of nitrogen and potassium (research question (1)). These effects differed between crops and were strongest for row crops and lowest for winter rye (research question (2)). Stability changed over time, but the statistical significance and the direction of trends differed between crop × treatment combinations and cannot be generalized (research question (3)). Our understanding of management effects on past and future yield stability will benefit from investigations across similar long-term experiments.