Maize (Zea mays L.) is a major crop; currently, more than 1 × 109 t of grain is produced annually around the world. For the last two to three decades, global maize production has grown faster than the area planted, and hence the total yield has grown rapidly1. The USA grows the most maize and harvests about one-third of the world’s total. Its maize yield has increased six-fold from the early 20th century to date2. It is conceivable that the turning point of improved yields came when transgenic hybrid (TH) cultivars with introgressed insect resistance and herbicide tolerance started being commercially grown in the mid-1990s. During the late 19th century3, only open-pollinated (OP) cultivars were planted, yielding poorly before the introduction of inorganic fertilisers. The adoption of hybrids increased in association with the development of fertiliser-responsive conventional hybrid (CH) cultivars and the increased use of fertilisers in the 1930s.

The world is increasingly concerned with the effects of climate change on agriculture, particularly in drylands, which is projected to make society vulnerable to climate variability in the coming decades. South Africa, a water-scarce country, is the leading maize-producing country in Sub-Saharan Africa but is at high risk of climate hazards like drought, frost, and floods. Since the crop is grown mostly under rainfed conditions, its production is highly affected by drought, which during the last four decades has occurred in the 1982–83, 1983–84, 1991–92, 1994–95, 2006–07, 2014–15, and 2015–16 cropping seasons4. There is little direct evidence that cultivar improvement is effective for further increasing yields at the national scale, and its relationship to fertiliser application is yet unclear. Here, by cointegrating time-series variables to determine the relationship of yield with fertiliser application in association with cultivar adoption, we show how the yield of dryland maize was historically increased by cultivar improvement and agronomic management under climate risk adaptation in South Africa (Fig. 1). This is particularly relevant to the Maize Triangle—comprising North West, Free State, Gauteng, and Mpumalanga provinces—which accounts for the majority of maize production.

Fig. 1: The Maize Triangle on the highveld of South Africa.
figure 1

The western area is classified as semi-arid and the eastern area as temperate. Maize planting typically begins in October in the east and in December in the west, although the timing depends on the start of rainy season. Air temperature drops to below the base temperature of maize (10 °C) in April.

Full size image

The growth of national-scale yield (YN, t/ha) was faster during the CH-to-TH transition (1998–99 to 2016–17), increasing from 2.57 t/ha to 6.40 t/ha over 19 years, during which TH cultivars were being adopted, than during the OP-to-CH transition (1949–50 to 1979–80), increasing from 0.96 t/ha to 2.49 t/ha over 31 years, during which CH cultivars were being adopted (Fig. 2). This was because the adoption of TH technology was faster than that of CH technology and coincided with the growth of NPK fertiliser application rates (FNPK, kg/ha); moreover, there was a nearly threefold faster increase in national total production (PN) than in national total planted area (AN) from the OP phase (1922–23 to 1948–49; PN = 1.8 ± 0.5 × 106 t; AN = 2.4 ± 0.4 × 106 ha) to the CH phase (1980–81 to 1997–98; PN = 8.0 ± 3.0 × 106 t; AN = 3.7 ± 0.4 × 106 ha) and an increase in PN with decreasing AN from the CH phase to the TH phase (2017–18 to 2021–22; PN = 14.2 ± 2.2 × 106 t; AN = 2.5 ± 0.2 × 106 ha) (Supplementary Fig. 1). Our model of cointegration with regime shifts fits the time series of the dependent variable YN and the independent variable FNPK well (P < 0.01), since the first-order differences of the non-stationary time series are stationary (Supplementary Fig. 2). It depicts the long-term trend with five regime shifts: 1962–63 (first), 1981–82 (second), 1991–92 (third), 2002–03 (fourth) and 2012–13 (fifth). The first shift occurs at ~40% of CH cultivar adoption, when YN variation starts to increase along with rapidly decreasing fertiliser use efficiency (FUE = 103·YN/FNPK, kg/kg). The second and third shifts are identified, respectively, as the beginning and end of yield stagnation with high variation due to rainfall variability (Supplementary Fig. 3)5 in the CH phase, during which CH cultivars were planted in almost all production areas. Similarly, the model with five level shifts fits the time series of YN and FNPK well (P < 0.01), but the first two shifts occur at approximately 5% and 90% of CH cultivar adoption in 1953–54 and 1976–77, respectively. The third shift occurs during the worst drought on record. The fourth shift occurs at 7% of TH cultivar adoption, when YN starts to increase again along with slowly increasing FUE, and the fifth shift occurs at 85% of TH cultivar adoption, when YN fluctuates widely owing to rainfall variability (Supplementary Fig. 3)5, but starts to stagnate again with no change in FUE. These shifts indicate that once breakthrough cultivars are largely adopted, no further efficiency of fertiliser application and improvement of yield can be expected. In contrast, the yield trend in the USA—a region characterised by relatively high rainfall—has shown a steady increase since the introduction of hybrid cultivars, with no evident shifts associated with subsequent cultivar adoption3. Consequently, the findings of this study may be specific to the South African context.

Fig. 2
figure 2

Time series of national-scale maize yield (YN), NPK fertiliser application rate (FNPK), fertiliser use efficiency (FUE) and cultivar adoption (CA) (CH conventional hybrid, TH transgenic hybrid) during the 1922–23 to 2021–22 crop seasons in the Maize Triangle.

Full size image

A farm survey on TH cultivar adoption in South Africa highlights better yield performance in wet years than in dry years6, implying that the yield advantage of TH cultivars is limited under rainfed conditions in drier areas. As South Africa’s maize is produced mainly in such areas—namely, Free State and North West provinces (Supplementary Fig. 4)—the recent yield improvement at the national scale can be attributed to TH cultivar adoption but is confounded by the increased use of fertiliser. Thus, fertiliser is a primary yield determinant along with cultivar adoption in the Maize Triangle. Farmers have coped with the interannual variability of the dryland climate by adjusting the fertiliser rates needed for given cultivars each season and have improved the yield by more than sevenfold over the last seven decades. In addition, planting has been moving to rainier areas; the proportion of planted areas has increased in Free State and decreased in the North West (Fig. 1, Supplementary Fig. 4). In the coming decades, climatic zones in the maize-producing areas of South Africa are projected to become drier7. Interestingly, increasing CO2 in the atmosphere has little effect on maize yield in the absence of water stress, but it mitigates yield reduction under water stress8. This favourable effect of rising atmospheric CO2 may help reduce drought risk. However, extreme droughts such as those experienced in the past can exceed the limits of the farmers’ coping practices in the Maize Triangle. Accordingly, the farmers’ decision-making on the fertiliser rates needed to attain target yields for given environments will be made harder by increasing drought risk under the future climate, indicating the need for new cultivars for coping with drought stress.

Towards dryland farming, future cultivar improvement will have to depend on cutting-edge biotechnology such as genome editing9, and substantial effort in maize breeding will be needed to adapt to climate change in droughty environments. All in all, the effectiveness of cultivar adoption in dryland farming, which has benefited from up-to-date breeding technology, is exemplified nowhere better than in the Maize Triangle.

Methods

Data on the national total area of maize planted (AN, ha) and commercial production (PN, t) in the 100 crop seasons from 1922–23 to 2021–22 were obtained from the Crop Estimates Committee of South Africa (https://www.sagis.org.za/cec_reports.html). The national-scale yield (YN, t/ha) was calculated as PN per unit of AN. The first CH cultivar was introduced into South Africa in 1949–50, and the adoption of CH cultivars as a share of seed sales increased over the following three decades: from 6% in 1954–55, to 29% in 1960–61, to 64% in 1965–66 and to 98% in 1979–8010. This upward trend followed a roughly sigmoidal curve. CH cultivars dominated until the introduction of insect-resistant TH cultivars in 1998–99, herbicide-tolerant TH cultivars in 2003–04, and pyramided cultivars in 2007–086. Adoption of TH cultivars increased from 1.3% of the planted area in 1999–2000 to 93.8% in 2016–17. Data on the TH planted area were compiled from reported data11,12,13,14. Data on the national average application rate of NPK fertilisers (FNPK, kg/ha) to maize from 1945–46 to 2021–22 were compiled from reported data15,16,17 and the Fertiliser Society of Southern Africa (https://www.fertasa.co.za).

The following models cointegrate time series (harvest year t = 1–n) of the dependent variable YN (yt) and the independent variable FNPK (xt) with k structural shifts (i = 1–k):

$$\begin{array}{cc}{y}_{t}={\alpha }_{0}+\mathop{\sum }\limits_{i=1}^{k}{\alpha }_{i}{D}_{i,t}+{\beta }_{0}{x}_{t}+{u}_{t} & ({\rm{level\; s}}{\rm{h}}{\rm{ifts}})\end{array}$$

and

$$\begin{array}{cc}{y}_{t}={\alpha }_{0}+\mathop{\sum }\limits_{i=1}^{k}{\alpha }_{i}{D}_{i,t}+{\beta }_{0}{x}_{t}+\mathop{\sum }\limits_{i=1}^{k}{\beta }_{i}{x}_{t}{D}_{i,t}+{u}_{t} & ({\rm{regime\; s}}{\rm{h}}{\rm{ifts}})\end{array}$$

where α0 is the intercept before the shifts, αi is the intercept at the time of the ith shift, β0 is the cointegrating slope before the shifts, βi is the slope at the time of the ith shift, ut is the error, and Di,t is a dummy variable that is equal to 0 if t ≤ [nτi] or 1 if t > [nτi], where the brackets denote the integer part and the unknown parameter τi, which is an element of (0, 1), denotes the timing of the ith shift18,19. Cointegration model analysis was performed in GAUSS v. 23 software with the TSPDLIB package20.