Effect of drought on wheat production in Poland between 1961 and 2019
Assigned to Associate Editor Paulo Teodoro.
Abstract
The impact of drought on wheat (Triticum aestivum L.) production is shown, using an example data set of almost 60 yr from six climate-specific regions in Poland. Drought was measured using the standardized precipitation index (SPI) and the hydro-thermal coefficient of Selyaninov (HTC). Yield trends were estimated by Bayesian linear regression over two periods, 1961–1991 and 1992–2019, identified by a changepoint detection method. Bayesian inference is used as it allows the estimation of a credible interval of regression coefficients instead of point estimates and asymptotic confidence intervals, thus comparisons between regression coefficients are more meaningful. We detected an increase in yield in both time periods and in all regions. The average winter wheat yield increased by 97% in the first period and by 35% in the second (19.8–39.1 dt ha−1 and 32.9–44.5 dt ha−1, respectively). Spring wheat yield increased by 96% in the first period and by 42% in the second (16.8–37.9 and 22.9–32.5 dt ha−1, respectively). Yield losses in drought years were estimated using the paired t test to compare mean difference between real yields and yields estimated from regression lines for nondrought years. The highest yield losses due to drought were in regions I (–19.3% spring wheats, –6.3% winter ones) and III (–16.1% spring and –8.3% winter wheats) over the 1992–2019 period.
Abbreviations
-
- COBORU
-
- Research Centre for Cultivar Testing
-
- GUS
-
- Polish Central Statistical Office
-
- HTC
-
- hydro-thermal coefficient of Selyaninov
-
- MCMC
-
- Markov Chain Monte Carlo
-
- SPI
-
- standardized precipitation index.
1 INTRODUCTION
Crop yield is affected by tolerance to environmental stresses (both biotic and abiotic) which is reflected in the market value of the crop (Hayward et al., 1993; Ward et al., 2022). In 2017, the seed market in agricultural crops was valued at about US$500 million in Poland, of which the wheat (Triticum aestivum L.) seed market accounts for more than one-third (Oleksiak, 2019). Of the abiotic stresses, periodic water shortage is the main factor limiting plant production (Schittenhelm et al., 2019). In central Europe the primary causes of agricultural drought include high temperatures, groundwater shortages, and changes in the character of precipitation from moderate rains and drizzle to sustained rainfall, with large amounts of water falling in a short period (Lobell et al., 2011). Based on data between 1936 and 1997, a linear upward trend has been observed in the number of days with heavy rains (about 4% per decade), but it has been accompanied by a decrease in total precipitation (1% per decade) in eastern Europe (the European part of Russia). At the same time, the prediction for Germany is a 6.5% increase in heavy rains per degree Celsius of average rise in temperature (Groisman et al., 2005). In Poland, the average annual temperature is expected to rise by about 3.5 °C by the year 2100 (EEA, 2004; Kozyra & Górski, 2008). Further, the predominance of light soils, characterized by limited water retention capacity in the arable layer, has always been a significant problem exacerbating periodic droughts, which in Poland affects about 60% of the arable land area (Krasowicz at al. et al., 2009). According to FAO (2003), the country is threatened by drought when annual water use exceeds 20% of natural supply. In Poland, it is estimated that more than a fifth of deep groundwater resources is used up annually, which is thus beyond the ecological safety limit of 20%. By this criterion, the threat of drought is greater than that in Poland only in Bulgaria, Spain, Germany, and North Macedonia. It is impossible to counteract yield loss or even a significant yield decline caused by extreme climate events. Nonetheless, plants respond to moderate environmental stresses by activating adaptive mechanisms and attempt acclimatization, which stabilizes yield (Feller et al., 2017). The selection of better yielding genotypes over several years and in multiple environments indirectly extends the acclimatization capacity of plants and can stabilize yields under moderate environmental stresses (Brancourt-Hulmelá et al., 2003).
Agricultural drought is monitored in Poland by the Institute of Soil Science and Plant Cultivation-IUNG (Doroszewski et al., 2012). Almost every year, there are several days without rain, leading to drought in 5–6-yr cycles and severe droughts in 10–11-yr cycles (Doroszewski et al., 2012; Łabędzki & Bąk, 2014). Since wheat land area is close to 40% of cereal crops sown annually, its drought tolerance is economically highly significant. The share in the crop structure is continuing to increase due to the favorable agronomic features of winter wheat cultivars: higher yields resulting from an extended vegetative season, starting in the autumn of the previous year, making it more likely to avoid spring droughts. Spring cultivars are sown most often due to agrotechnical reasons: either when autumn sowing is not possible or in years where winter crops experience significant losses due to adverse conditions in the winter (Rybka & Nita, 2015; Senapati et al., 2018; Simova et al., 2016).
There is no direct standardized field selection method to assess drought tolerance even though automatic mass phenotyping systems based on spectral indices are being developed (Araus et al., 2018). However, multi-environmental trials and post-registration cultivar evaluation results in a gradual increase in crop yields even in seasons affected by drought (Senapati et al., 2018; Simova et al., 2016; Studnicki et al., 2019). Further, it is expected that continued selection aimed at high yield potential will suffice as a means to counter the expected increase in droughts in Germany for the foreseeable future (Schittenhelm et al., 2019). Until the beginning of the 1990s, in Poland, there was a gradual intensification of technology, mineral fertilization, and the use of chemical pesticides. During the period of economic transformation in the country and the process of market liberalization, livestock production was reduced, the level of mineral fertilization with phosphorus and potassium declined significantly, and the crop structure was simplified (FAO, 1992). The value of fore-crops for wheat deteriorated as the share of cereals increased in farming practices. Again, an increase in the level of nitrogen fertilization and the intensity of chemical protection, measured by the number of treatments can be seen starting from 1992. The share of fertilization and plant protection in the increase of yield was estimated at 24 and 14%, respectively (Nalborczyk, 1997). The aim of our study was to examine the impact of drought on wheat yield in Poland over a period of almost 60 yr. The analysis is carried out for six climate-specific regions, with varying amounts of drought. The following methodological approach was employed as there are differences in climate, soil quality as well as in the level and dynamics of technology changes between regions.
Core Ideas
- Wheat production is limited due to spring droughts.
- Yield loss in spring wheat cultivars is more pronounced compared to that in winter ones.
- Yield-based selection generally results in improvement of wheat drought tolerance.
The yield trend lines, which can be an important asset in estimating a possible loss due to drought, were estimated using Bayesian linear regression in two separate periods identified by changepoint detection (Killick et al., 2016). Bayesian inference offers the advantage of estimating (simulating by Markov Chain Monte Carlo [MCMC]) a credible interval of regression coefficients instead of point estimates and asymptotic confidence intervals, which makes comparisons between regression coefficients more meaningful. Additionally, MCMC methods are applicable even when it is not possible to directly draw samples from a distribution, they work for complicated distributions in high-dimensional spaces, even when we do not know where the regions of high probability are, they are relatively easy to implement, and they are also reliable. Moreover, Bayesian methods are based on a single tool which is the Bayes theorem, making them simpler overall (Bolstad, 2007). Using these regression lines, we predicted the mean yield values in years of drought along with yield values in nondrought years, and we looked for differences between real and predicted yields, to calculate the magnitude of the yield lost due to drought.
2 MATERIALS AND METHODS
2.1 Yield data
The mean yields of winter and spring wheat are published annually by the Polish Central Statistical Office (GUS) (https://bdl.stat.gov.pl/BDL/start), and so being publicly available, we used the values for the years 1961–2019. GUS calculates wheat yield in the regions as the quotient of production to the cultivation area. The data were collected annually at the Plant Breeding and Acclimatization Institute (http://www.ihar.pl/index_en.php), Poland, from printed statistical yearbooks for 1961–2004 and for remaining years, electronically (GUS, 2020). Data were collected separately for six regions, according to the Research Centre for Cultivar Testing (Coboru) (http://www.coboru.gov.pl/English/index_eng.aspx), in line with administrative divisions and climate differences between regions (central Europe, Supplemental Figure S1 in the Supplemental Materials file). The regions are latitudinal, with two regions alongside each other. The northern pair along the Baltic Sea spread over the Zachodniopomorskie, Pomorskie, and part of Wielkopolska voivodships (region I) and the Warminsko-Mazurskie and Podlaskie voivodships (region II) containing Lakelands of Pomorze (I) and Warmia and Mazury (II). Region III includes portions of the Wielkopolska and Kujawsko-Pomorskie voivodships, whereas region IV is spread across the Łódzkie, Mazowieckie, and Lubelskie voivodships. Regions V and VI are situated along the southern border regions of Poland, across the Sudetes (V) and Carpathians (VI). Cereal crops, including wheat, are not irrigated in Poland, so the wheat is cultivated using water available naturally in the soil.
2.2 Outlier detection
Outliers were detected using Grubbs’ test in a univariate dataset and is defined as the following hypothesis test:
: there are no outliers in the data set
: there is an outlier in the data set
2.3 Detection of changepoints in yield dataset
In statistical analysis, changepoint detection is used to identify time points where the probability distribution of a time series has a significant change. This change can be a change in the mean, variance, correlation, spectral density, or trend. In this work, we wanted to estimate yield trends, and so we examined changepoints on the yield trend (regression with time as an explanatory variable). The algorithm used to compute changepoints in regression is based on a previously published dynamic programming algorithm (Killick et al., 2016; Zeileis et al., 2002). The idea behind this algorithm is to compute a triangular residual sum of squares matrix, which gives the residual sum of squares for a segment starting at observation i and ending at observation i′. Following this, a decision is made about the optimal number of changepoints and which these changepoints are, based on an information criterion (Bayesian Information Criterion: BIC; see Supplemental Material). We used the “Envcpt” (R package version 1.1.3), “strucchange” (R package version 1.5.2) and “changepoint” (R package version 2.2.2) packages in R for changepoint detection (Killick et al., 2016; Killick et al., 2021; Zeileis et al., 2002).
2.4 Bayesian linear regression model
The trends of the yield data for winter and spring cultivars in the six different regions are estimated using Bayesian linear regression. The advantage of Bayesian linear regression against classical linear regression (as referred to in the Introduction) is that it yields the whole range of trend values (posterior distribution – credible interval) rather than a point estimate and asymptotic confidence intervals as in classical linear regression using the maximum likelihood method (Wasserman, 2004).
The yield loss for years with drought was estimated using paired t test for examining the mean of the differences between yield for drought years and estimations by Bayesian linear regression in drought years using the nondrought regression line. More specifically, we investigated the difference between yield in drought years and yield estimated if these years were nondrought. For example, for a drought year we compare the value recorded and the estimated value from the solid black regression line which represents the yield trend line in nondrought years. The same applies for all regions and drought years.
2.5 Meteorological data
We used publicly available precipitation and temperature data, validated and published by the Institute of Meteorology and Water Management-National Research Institute (IMGW-PIB), Poland, between 1961 and 2019 for the months: April, May, and June. For each region, which includes two or three provinces, the data from at least three meteorological stations in each province, with a full data-set were analyzed as sum of monthly precipitations and mean daily temperatures (IMGW, 2019). The meteorological stations were chosen from areas where the intensity of wheat cultivation was the highest, based on the agricultural census in 2010 (Głębocki, 2014).
2.6 Calculation of drought indices
Drought indices enable quite easy identification of critical periods using long time weather data, which is a widely accepted methodological simplification that facilitates drought determination (Gąsiorek & Musiał, 2015; Głębocki, 2014; Keyantash, 2021; Singleton, 2012). We have used two drought indices: Standardized precipitation index (SPI) and hydro-thermal coefficient of Selyaninov (HTC) (mm/°C) (Edwards & McKee, 1997; Łabędzki & Bąk, 2014; Tokarczyk & Szalińska, 2014).
where S is the sum of monthly precipitation; is the normalized sum of precipitation;is the average value of normalized sums of precipitation; and is the standard deviation of normalized precipitation sums. Rainfall was normalized to change the gamma distribution- typical for rainfall data – to a normal distribution. Usually, the cube root function is used, and this was followed by checking whether the transformed data has a normal distribution. The SPI was calculated separately for: April, May, and June, as well as for the April–June period. Drought was classified as: (SPI > −0,5) no drought (ND); (−0.5 ≥ SPI > −1.0) moderate dry (MD); (−1.0 ≥ SPI > −1.5) dry (D); (−1.5 ≥ SPI > −2.0) very dry (VD); and (−2.0≥ SPI) extremely dry (ED).
The HTC was calculated for the May–June period, but not for April when the average daily temperatures did not exceed 10 °C, which is necessary for reliable identification of drought using this coefficient. A five-point drought scale was used: (HTC > 1.3) no drought (ND); (1.3 ≥ HTC > 1.0) moderate dry (MD); (1.0 ≥ HTC > 0.7) dry (D); (0.7 ≥ HTC > 0.4) very dry (VD); and (0.4 ≥ HTC > 0.0) extremely dry (ED).
In years indicated as dry, crop loss was determined as the difference between the value estimated based on the trend line for years that were not defined as dry and the yield value in years with spring drought.
3 RESULTS
3.1 Regional differences in wheat production
The area of Poland (between geographical coordinates: 54o 50′ N and 49o 00′ N latitude, and between 14o07’ E and 24o08’ E longitude) is varied, so the impact of yield-based selection on wheat drought tolerance was analyzed following the six-member regionalization accepted by the Research Centre for Cultivar Testing (Coboru). The share of any particular region in the annual national wheat production and yield was between 10 and 25%, based on data from 2015 to 2017 (Figure 1). We chose these years as being illustrative, recent years .

From Figure 1 it is clear that winter wheat is the predominant form, the share of spring wheat does not exceed 7% in terms of either yield or land area. Regions III, IV, and V together produce about 70% of the wheat harvest. In the recent past, the highest wheat yields (>50 dt ha−1) comes from regions I and V. The yield in region III is 45–50 dt ha−1, in regions II and IV it is 40–45 dt ha−1, and it is under 35–40 dt ha−1, in region VI. In region V, both wheat cultivation area and yields make up about 50% of the share of all cereal crops. Similarly, in region I, wheat constitutes about 40% of the sown area and 50% of yield of all cereals, while in region VI it is about 40% of both yield and land area. Neither region I nor VI is a significant wheat producer. About 25% of the area sown with cereals in regions II–IV is given over to wheat and the wheat share of the cereal production is also similar (Figure 1). Also, as supplemental material (Supplemental Figure S4a and S4b), principal component analysis (PCA) biplots combined with K-means clustering are given and show that regions I, III, V and II, IV, VI are clustered together as these regions have similar yield for both winter and spring wheat (I, III, V higher yield in general). With regard to time, the average winter wheat yield increased from 19.8 to 39.1 dt ha−1 (an increase of 97%) in the 1961–1991 time period and from 32.9 to 44.5 dt ha−1 (an increase of 35%) in the 1992–2019 time period. The yield increases for spring wheat were 16.8–32.9 dt ha−1 (an increase of 95%) and 22.9–32.5 dt ha−1 (an increase of 42%). In the region with the highest wheat production the yield increase for winter forms between 1961 and 1991 was 82% (23.1–42.1 dt ha−1), and 39% (36.9–51.4 dt ha−1) between 1992 and 2019. Spring wheat yields increased by 79% (20.9–37.5 dt ha−1), and 32% (26.7–35.3 dt ha−1), respectively.
3.2 Regional differences in climate
Climatic differences are readily identifiable by comparing rainfalls and mean temperature for the primary growing season: the months of April, May, and June. Most of the rainfall was in June, in the South (regions V and VI). It was about 20% more than in the North (regions I and II). In contrast, rainfall was the lowest in April, again being nearly 40% more in the South (V–VI) than in northern and central regions (I–IV) with an average total of 33 mm rain (Table 1).
Rainfall | Temperature | ||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|
Region/month | Max. | Min. | V | D | Max. | Min. | V | D | |||
mm | °C | ||||||||||
I | |||||||||||
April | 33.9 | 149.2 | 0.3 | 0.67 | –8.9 | 7.0 | 12.3 | 3.7 | 0.23 | 2.0 | |
May | 51.0 | 142.7 | 3.9 | 0.51 | –1.4 | 12.0 | 16.8 | 7.6 | 0.15 | 1.7 | |
June | 63.0 | 258.5 | 0.0 | 0.58 | 4.8 | 15.5 | 21.5 | 12.3 | 0.09 | 0.8 | |
II | |||||||||||
April | 37.2 | 99.1 | 1.6 | 0.56 | –7.3 | 7.1 | 12.5 | 3.6 | 0.23 | 1.9 | |
May | 56.9 | 186.4 | 5.3 | 0.46 | 3.8 | 12.9 | 18.5 | 8.0 | 0.13 | 1.5 | |
June | 72.1 | 227.1 | 1.0 | 0.49 | –7.2 | 16.1 | 27.5 | 13.0 | 0.11 | 0.6 | |
III | |||||||||||
April | 32.3 | 131.8 | 0.2 | 0.60 | –9.0 | 8.3 | 13.7 | 0.0 | 0.20 | 2.1 | |
May | 52.3 | 159.3 | 4.0 | 0.52 | –4.9 | 13.6 | 18.0 | 9.2 | 0.12 | 1.7 | |
June | 61.0 | 298.6 | 2.2 | 0.59 | –13.1 | 16.8 | 29.1 | 13.7 | 0.09 | 0.9 | |
IV | |||||||||||
April | 38.0 | 132.6 | 0.0 | 0.57 | –5.6 | 7.9 | 13.8 | 2.5 | 0.21 | 1.8 | |
May | 61.9 | 239.9 | 8.0 | 0.52 | 7.4 | 13.3 | 17.3 | 5.4 | 0.12 | 1.5 | |
June | 70.9 | 198.9 | 8.0 | 0.50 | –9.5 | 16.5 | 22.4 | 10.2 | 0.09 | 0.9 | |
V | |||||||||||
April | 43.3 | 165.2 | 0.0 | 0.58 | –8.0 | 8.5 | 14.4 | 4.0 | 0.20 | 1.7 | |
May | 74.7 | 511.5 | 4.0 | 0.58 | –2.9 | 13.4 | 17.8 | 8.1 | 0.12 | 1.6 | |
June | 82.5 | 350.2 | 7.3 | 0.52 | –11.5 | 16.6 | 22.7 | 12.3 | 0.09 | 1.3 | |
VI | |||||||||||
April | 47.8 | 156.9 | 0.2 | 0.58 | –4.4 | 8.2 | 14.7 | 2.4 | 0.22 | 1.3 | |
May | 78.0 | 302.4 | 13.4 | 0.50 | 13.5 | 13.3 | 17.5 | 7.9 | 0.13 | 1.4 | |
June | 88.3 | 255.1 | 3.7 | 0.50 | –13.4 | 16.5 | 22.6 | 12.1 | 0.10 | 1.3 |
- Note. D, the difference between averages calculated for 20 recent years (2000–2019) and 20 yr at the beginning of the analyzed data set (1961–1980); Max., maximal; Min., minimal; V, coefficient of variation of the regional average value; , average value.
The coefficient of variation of total precipitation for all the months and regions was high – between 46 and 67%. In May, rainfall in the East (regions II, IV, and VI) and the southwestern region V, was higher than in the West (regions I and III). The largest decrease in rainfall (>20 mm) for April–June was in the West (regions III and V), unlike the temperature, which increased on average by about 1.5 °C. Average temperatures ranged from 7 to 8 °C in April to 16 °C in June. The coefficient of variation for mean temperatures for all months and regions was lower than for rainfall (9–23%; Table 1).
Based on precipitation and temperature data, the SPI and the HTC were calculated (Table 2; Supplemental Tables S1 and S2), as a widely accepted methodological simplification that facilitates identification of drought periods (Gąsiorek & Musiał, 2015; Głębocki, 2014; Keyantash, 2021; Singleton, 2012).
Region | ||||||
---|---|---|---|---|---|---|
Years | I | II | III | IV | V | VI |
1961 | D | SPI | ||||
1962 | ||||||
1963 | D | D | HTC | D | D | |
1964 | D | D | HTC | D | D | |
1965 | ||||||
1966 | D | D | ||||
1967 | ||||||
1968 | HTC | |||||
1969 | HTC | HTC | ||||
1970 | HTC | |||||
1971 | ||||||
1972 | ||||||
1973 | HTC | |||||
1974 | SPI | |||||
1975 | D | D | ||||
1976 | D | SPI | D | D | SPI | SPI |
1977 | HTC | HTC | D | |||
1978 | D | D | SPI | |||
1979 | D | D | D | D | HTC | |
1980 | ||||||
1981 | HTC | SPI | ||||
1982 | HTC | D | ||||
1983 | HTC | HTC | HTC | D | SPI | |
1984 | ||||||
1985 | ||||||
1986 | HTC | D | ||||
1987 | ||||||
1988 | HTC | D | D | |||
1989 | D | D | ||||
1990 | HTC | SPI | D | HTC | ||
1991 | ||||||
1992 | D | D | D | D | D | SPI |
1993 | HTC | D | HTC | D | D | SPI |
1994 | D | HTC | HTC | HTC | ||
1995 | ||||||
1996 | HTC | |||||
1997 | ||||||
1998 | HTC | |||||
1999 | ||||||
2000 | D | D | D | D | D | |
2001 | D | HTC | ||||
2002 | D | HTC | ||||
2003 | D | HTC | D | D | D | D |
2004 | HTC | SPI | ||||
2005 | D | HTC | ||||
2006 | HTC | D | HTC | D | ||
2007 | HTC | SPI | SPI | |||
2008 | HTC | D | D | D | D | D |
2009 | ||||||
2010 | ||||||
2011 | D | HTC | D | HTC | SPI | |
2012 | HTC | |||||
2013 | ||||||
2014 | ||||||
2015 | D | D | D | D | D | SPI |
2016 | HTC | HTC | HTC | D | D | |
2017 | HTC | HTC | HTC | |||
2018 | D | D | D | D | D | D |
2019 | D | HTC | D | D | D | |
∑ | 26 | 23 | 31 | 22 | 18 | 20 |
1961–1991 | 14 | 10 | 13 | 8 | 4 | 10 |
1992–2019 | 12 | 13 | 18 | 14 | 14 | 10 |
- Note. D, drought detected by both indices; SPI or HTC for droughts detected by only one of the indices.
The highest number of droughts was recorded in regions I and III in the West. Droughts in region I accounted for about 50% of the years and in region III for about 60% of the years (Table 2). The highest temperature was also detected there (Table 1), which showed that the drought was detected exclusively by HTC index in regions: I and II – eight times and in region III – 18 times, while the total number of droughts were 26, 23, and 31, respectively (Table 2).
A PCA biplot (Supplemental Figure S4c) combined with K-means clustering based on weather data (temperature, rainfall) shows that region I is alone (Region I is next to the Baltic) and V, VI are clustered together, which makes sense, since they are in a higher altitude than the others. The remaining three regions (II, III, IV) of European Plateau are clustered together.
3.3 Yield trend lines
Before estimating yield trend lines, the yield data set for each region was inspected for outliers using Grubbs’ test and for uniform regression by a changepoint detection method (Killick et al., 2016). The year 1991 was detected as a significant changepoint, dividing the time under study into two time periods (1961–1991 and 1992–2019). It is important here to clarify that this division into two time periods is purely statistical, and is based on the algorithm described in a previous section for identifying changepoints in regression (Killick et al., 2016; Zeileis et al., 2002).
It would hence be wrong to apply a single regression model (or more than two models) for the entire duration under study as there are no other changepoints identified by the statistical method described in Materials and Methods (see also the Supplemental Material file). Moreover, a search for outliers in the yield data found only one outlier, the year 2015 in the western region I. Therefore, the influence of spring drought was analyzed for data divided into two time periods by comparing yield trend lines in dry seasons (Table 2) to the yield trend lines for periods without droughts (Figure 2).

All the trends were estimated using Bayesian linear regression. Increasing tendencies in the yielding of wheat were found in all regions of Poland. In all cases, the yields showed a growing trend, except in region III for spring wheat after 1991. The yield trend lines for winter wheat in nondrought years usually showed a higher estimated regression coefficient (trend) on average compared to drought years (Figure 2).
In the northern and central regions of the country we observed much greater differences in the yields and their growth dynamics in the second period, and they were dependent on the occurrence of spring droughts. The growth dynamics of winter wheat yields in dry years in regions particularly threatened by drought, that is, in regions I, II, and III, were lower by 18, 15, and 35 kg per year, respectively. Analogous drops in the yield growth dynamics for spring wheat amounted to 13, 2, and 16 kg per year. At the same time, it should be noted that in region V, the impact of periodic droughts on the yield is not statistically significant.
Nevertheless, yield difference is statistically significant at the 10% confidence interval (p value = .092) only for the more recent time period (after the changepoint 1991) in region II and at the 5% confidence interval (p value = .041) in region III, which recorded the highest number of drought events on average (18 vs. 13) in the entire country (Tables 2 and 3). None of the other comparisons (drought vs. nondrought) showed any significant differences in the trend before 1991 for winter and spring wheat (see Supplemental Figures S2 and S4).
1961–1991 | 1992–2019 | |||||
---|---|---|---|---|---|---|
Region | Average real yield | No. of spring droughts | Yield changes | Average real yield | No. of spring droughts | Yield changes |
dt ha–1 | dt ha–1 (%) | dt ha–1 | dt ha–1 (%) | |||
Winter wheat | ||||||
I | 31.9 | 14 | 3.26 (11.0)** | 45.2 | 12 | –3.34 (−6.3)* |
II | 27.3 | 10 | 0.97 (1.7) | 37.0 | 13 | –2.07 (−4.8)* |
III | 31.6 | 13 | –1.46 (−4.3)* | 41.1 | 18 | –4.06 (−8.3)** |
IV | 26.4 | 8 | –0.23 (−1.2) | 35.2 | 14 | –0.67 (−1.8) |
V | 31.5 | 4 | 0.36 (1.2) | 45.4 | 14 | –0.24 (0.0) |
VI | 25.0 | 10 | 0.02 (0.2) | 32.2 | 10 | 1.59 (5.1) |
Spring wheat | ||||||
I | 27.2 | 14 | 0.96 (3.9) | 33.6 | 12 | –6.79 (−19.3)** |
II | 24.2 | 10 | –0.48 (−3.5) | 30.2 | 13 | –2.83 (−8.9)* |
III | 27.1 | 13 | –2.13 (−7.6)* | 32.5 | 18 | –5.82 (−16.1)** |
IV | 24.5 | 8 | –1.14 (−5.2) | 30.3 | 14 | –2.63 (−8,4)* |
V | 27.9 | 4 | –1.09 (−3.2) | 35.7 | 14 | –1.43 (−3.8) |
VI | 21.9 | 10 | –0.25 (−1.1) | 29.8 | 10 | 1.94 (6.6) |
- Note. The statistical significance was tested using a paired t test.
- *p value < .05. **p value < .01.
To avoid any misunderstanding, this does not mean that there is no yield loss in drought years (Table 3). Table 3 presents the changes in winter and spring wheat yield in drought years, against predicted yields, based on the yield trend (regression) lines for years without any drought (as presented in Figure 2). In the period until 1991, winter wheat yields showed a significant loss in region III (−1.46 dt ha−1). In contrast, a significant yield gain was calculated (3.26 dt ha−1) in region I. A yield loss in response to drought was found for spring wheat in all regions except region I, during 1961–1991 and for region VI during 1992–2019. The loss is statistically significant in both periods especially in region III. In the period after 1991, droughts led to significant yield loss for winter wheat grown in the western regions (I and III) and the northeastern part of region II. Spring wheat yields were affected significantly in all regions, except regions V and VI, and were particularly severe in regions I and III, where the yield loss is estimated at −6.79 dt ha−1 drought year−1 and −5.82 dt ha−1 drought year−1, respectively.
All regions, but region III after 1991 for spring wheat, showed an increase in the average wheat yield in dry periods; the only exception where the trend coefficient was not statistically significant (zero is included in the 95% credible interval of the posterior distribution of the coefficient). That interval is marked in red in Figure 3 (spring wheat, region III). Figure 3 shows the comparison between densities of the estimated regression coefficient values (trends) for all regions after 1991. The densities before 1991 showed no significant differences between trends of wheat yields in nondrought and drought years (see Supplemental Figure S2).

We also saw differences between yield growth in regions as revealed by the density (posterior distribution) of the regression coefficient (yield trend) values for drought years, estimated by Bayesian linear regression using MCMC, for winter wheat yields. The yield trend lines for regions II–IV, were characterized by lower slope factors than in regions V and I. The slope values differed in regions I and VI, where the overall national share of wheat production is marginal – the slope for region I was significantly higher than for region VI (see Figure 4). The statistical significance in Figure 4 was examined with a z test (see Equation 1) with a p value threshold equal to 0.00333 (0.05 divided by 15 – the number of pairwise comparisons of the areas). When multiple comparisons are made, it is necessary that p value correction keeps the type I error rate (false positive rate) to less than or equal to the specified p value cutoff which is usually .05. The simplest way to adjust the p values is to use the Bonferroni correction method which divides the threshold by the number of tests (Jafari & Ansari-Pour, 2019). No significant differences for the yield trend between regions were found for spring wheat (see Supplemental Figure S2).

4 DISCUSSION
4.1 Wheat production and the threat of drought
Winter wheat is the predominant cereal species cultivated in Poland, even though it is the most demanding among cereal crops in terms of soil quality. The proportion of land used for wheat cultivation has steadily increased from 20% in the 1960s, to 25% in the 1980s, to the current 39%. At the same time, the average yield has gradually increased: from 2.14 in the 1960s and 2.86 in the 1980s, up to 4.46 dt ha−1 at present (Arseniuk et al., 2003; GUS, 2020). Yield reduction due to prolonged periods of drought, affecting stem elongation up to post-anthesis, is higher than that due to shorter droughts (e.g., only during stem elongation), as the pattern of assimilate allocation is different (Tatar et al., 2015). Yield reduction in response to high temperatures during flowering and seed set is the result of accelerated leaf senescence that leads to lower flower number, and consequently to reduced grain number (Asseng et al., 2011; Maccaferri et al., 2011). However, we here analyze data using drought indices as described in the Materials and Methods section (SPI and HTC), and this simplification does not allow us to examine details such as drought duration (Gąsiorek & Musiał, 2015; Głębocki, 2014; Keyantash, 2021; Singleton, 2012). The HTC identifies a larger number of drought periods than the SPI index for most regions in temperate climate (Gąsiorek & Musiał, 2015), and this is consistent also with our data (Tables 2 and 3). According to predictions made using the Sirius model (Rothamsted Research), heat tolerance would be the limiting factor for yield in central Europe (Stratonovitch & Semenov, 2015). The HTC therefore seems to be more appropriate for drought detection in such scenarios, which is clearly seen in the western region III, where higher number of droughts was detected by HTC than SPI. In general, the total rainfall shows a diminishing trend over the period examined here, with the increasing maximal daily precipitation (data not shown), due to convectional rainfall as a result of rising temperatures, which finally manifest as an increasing number of droughts (Easterling et al., 2000; Sillmann & Roeckner, 2008; Trenberth et al., 2003).
We have focused on the fact of steady yield increase even in years of drought, and not on the reasons for yield decline. Independently, in a recent experiment, we analyzed changes in water saturation deficit in field-grown wheat, during the hot and dry summers of 2017 and 2018 (Spyroglou et al., 2021), and we found that the Belissa, Bonanza, and Frisky winter wheat cultivars from the Polish crop-seed market were tolerant to drought. Belissa is a Polish cultivar, selected by pedigree breeding, based on monitoring phenotypic traits (habit, health, quality traits, overwintering) to the F6 generation, followed by an assessment of features related to the yield potential and other economically valuable properties (Matysik et al., 2007). The estimated genetic gain in wheat yield is about 1% per year at the global scale (Reynolds et al., 2012). The genetic gain in Poland is similar (Krzymuski et al., 1993). Oleksiak (2003) estimated that the contribution of breeding programs in years 1986–2001 was about 41–47 kg ha−1 per year (depending on method of estimation used) with a cultivar share 42% of the yield increase, and its significance tends to grow over the course of time. Rudnicki and Piekarczyk (2018) estimated a cultivar impact of 37%, Nalborczyk (1997) of 52% on the increase of winter wheat yields between 1970 and 2016, and Harasim and Matyka (2009, 2020) estimated the agrotechnology impact on yield increase to be 39% (fertilization: 25% and plant protection: 14%). Generally, fertilization affects the yield stability at the global scale (Ahrends et al., 2021). In France, the share of breeding in winter wheat yield increases was calculated as 30–50%, depending on the agronomic treatment (Brancourt-Hulmel et al., 2003). Higher growth rate before and during anthesis and better grain filling are the main traits associated with genetic gains in winter wheat grain yield (based on experiments in the United Kingdom; Shearman et al., 2005).
4.2 Differences between agricultural regions
The territory of Poland is geographically diverse, and the cultivar evaluation regions are in line with these differences and with administrative divisions of the country. Since the availability of land with high-quality soil is limited, wheat is increasingly grown on less optimal soils, where water shortages occur ever more frequently due to climate change (Rosenzweig et al., 2014). Water retention in the arable layer following heavy rain is quite poor – especially when poorly cultivated – and the drought is consequently more severe (Rybka & Nita, 2015; Skłodowski & Bielska, 2009). Factors of the agroecosystem influencing drought occurrence include landform, hydrological conditions, the quality and water retention capacity of the soil and the degree of forest cover. A further agroecological contributor is human population and the consequent degree of urbanization and industrialization, the extent of the road network, as well as demarcated and protected natural and recreational areas (Kondracki, 1998). These features were taken into account in the designation of COBORU evaluation regions in which crop varieties are assessed annually.
The western regions (I, III, V) have a humid Atlantic climate characterized by warmer winters and slightly cooler summers compared to the eastern ones (II, IV, VI) that are more influenced by continental climate (cooler winters and warmer summers). Central regions (III and IV) located on the Great European Plain, are the major wheat producers, with a lower forest cover (ca. 20%), a smaller number of lakes, and are more urbanized. These regions comprise 9.1 million ha of agricultural land (3.7 million ha in region III and 5.4 million ha in region IV) mainly of medium quality. Given such a geography, these central regions are also more prone to droughts. Most northern regions (I and II) lie along the Baltic seashore, and include glacial lakes, they are highly forested (>30%) with 54% of land under cultivation, and only 15% high-quality arable land (I–III) (Kondracki, 1998). Further, the climate in region II is the harshest, so the average yields and wheat production share are the lowest. Coupled with the fact that regions I and II have only slightly fewer drought years than regions IV and V, much greater yield loss due to drought are noticed there. On the other hand, the most significant difference in winter wheat yields between optimal and drought conditions were detected in regions II and III, in region II due to geography and in region III due to the highest number of droughts accompanied by high temperatures. Regions V and VI are mountainous, with foothills for the most part, with an average forest cover of 30%. Region V is a significant wheat producer, excluding the highly industrialized central portion. 2.4 million ha of arable land more than 40% of which is high-quality, combined with a mild climate result in the highest wheat yields and production share (Kondracki, 1998). The impact of periodic droughts on the yield is statistically insignificant in region V, because soils have the highest retention capacity and the region is characterized by high-quality soil cultivation systems (Krasowicz et al., 2009). In region VI, wheat production is minor because of the climate, soil, forest cover, and mountainous topography (Kondracki, 1998). In general, because soils of wheat complex constitute only 28% on average in all the areas combined, a portion of the better rye complex lands are also being used for wheat cultivation (Skłodowski & Bielska, 2009). That accounts for why regions III and IV produce more wheat than what can result purely from soil quality.
4.3 Yield trend lines
Before estimating yield trend lines, the yield data set for each region was inspected for outliers. We have no explanation as to why the 2015 yield in region I was an outlier.
The changepoint in 1991 is related to political and economic changes in eastern Europe, when farm sector economic reforms began in Poland. The FAO (1992) webpage includes comments concerning the rate of increase in input prices and overall consumer prices to seven or eight times 1989 levels in comparison with only fourfold increase in farmgate prices. Severely lower incomes forced drastic savings, which impacted investment in agricultural production and resulted in yield reduction and a reversal of yield trends.
The yield trend lines show a steady yield growth even in years of drought. The current upper limit for wheat yield using advanced agronomic techniques is about 10 T ha–1 (Mackay et al., 2011), which can lead to yield stagnation, or even to its decline, as seen in Great Britain (Brisson et al., 2010; Stratonovitch & Semenov, 2015). In Poland, where the share of extensive agriculture is still significant, wheat yields have not yet plateaued since the average yields in wheat production are lower than 5 T ha–1 (Studnicki et al., 2019). So there seems to be still room for the yield potential to be improved. In Poland the threshold criterion for increasing the yield potential even in years of drought, yield and disease resistance at least as good as the reference variety, is a sufficient criterion in the forseable future, as it was in the case of Germany (Schittenhelm et al., 2019). Yield trend lines showed the most significant difference in winter wheat yields between optimal and drought conditions in regions II and III.
Yield reduction is to be expected in areas with relatively poor soils and where weather fluctuations are most drastic. Agrotechnical treatments limiting soil water loss and increasing plant drought tolerance thus become increasingly important. For temperate regions, where drought in one season can be followed by excessive snows/rains leading to flooding like that seen in April–May 2010 (Bocianowski et al., 2019; NASA, 2010), highly adaptible cultivars are preferred. Selection of such cultivars is usually done by parallel testing in locations with light soils and frequent droughts using a yield-oriented approach, which results in a gradual increase in crop yields, even in years affected by drought (Romagosa & Fox, 1993; Studnicki et al., 2018) – this observation is also confirmed by data presented in this paper.
Making progress in increasing drought tolerance is difficult because of the complexity of this trait being determined by many environmentally controlled genes (Tarawneh et al., 2019). Therefore, the breeding of drought-tolerant cultivars requires the integration of various knowledge systems and methodologies (Araus et al., 2018; Cendrero-Mateo et al., 2017).
5 CONCLUSIONS
- The linear increase in the wheat yield rate (both long-term averages and yields in years of drought) varies in different regions of Poland;
- Drought most often occurs in region III and causes the greatest drops in yields there;
- Winter wheat cultivars are better suited for cultivation in regions threatened by droughts, their yield declines are smaller than those for spring cultivars;
- Selection using a yield-oriented approach results in a gradual increase in crop yields, even in years affected by drought;
- Breeding of drought-tolerant cultivars should counteract the threat of yield losses in years of drought.
ACKNOWLEDGMENTS
Tadeusz Oleksiak and Ioannis Spyroglou are co-first authors. The authors would like to express their thanks to Dr. Zygmunt Nita, Plant Breeding Strzelce, Ltd. Co., Poland, for inspiration and discussion. Plant Sciences Core Facility of CEITEC Masaryk University is acknowledged for technical support. This work was supported by: (1) The National Centre for Research and Development (NCBiR) Poland, as a part of the R&D Activity of the Intelligent Development Operational Program 2014–2020 co-financed by the European Regional Development Funds; grant number: POIR-01.01.01-00-0782/16-00; (2) Ministry of Agriculture and Rural Development-Poland subvention in Area 3: Breeding and seed production of arable crops, Task 3.5: Separation of cultivated plant forms with increased resistance to periodic water shortages; (3) European Regional Development Fund-Project “SINGING PLANT” (No. CZ.02.1.01/0.0/0.0/16_026/0008446) which received a financial contribution from the Ministry of Education, Youths and Sports of the Czech Republic in the form of special support through the National Programme for Sustainability II funds.
AUTHOR CONTRIBUTIONS
Tadeusz Oleksiak: Data curation; Investigation; Methodology; Validation; Writing – review & editing. Ioannis Spyroglou: Formal analysis; Methodology; Writing – original draft; Writing – review & editing. Przemysław Matysik: Funding acquisition; Project administration; Resources; Validation; Writing – review & editing. Markéta Pernisová: Funding acquisition; Project administration; Writing – review & editing. Krystyna Rybka: Conceptualization; Investigation; Validation; Writing – original draft; Writing – review & editing.
CONFLICT OF INTEREST
The authors have no conflict of interest to declare.
Open Research
DATA AVAILABILITY STATEMENT
All data used were publicly available: the wheat yield data published by Polish Central Statistical Office (GUS) and COBORU; climatic data (precipitation and temperature) published by the Institute of Meteorology and Water Management-National Research Institute (IMGW-PIB).