Impact of rainfall variability on soybean yields in Southern Brazil

As the third soybean-producer state in Brazil, Rio Grande do Sul (RS) presents a known year-on-year unevenness for soybean production, mainly due to water availability. This study aimed to assess the weather effects, with special focus on rainfall during 25 soybean growing seasons and 11 producing regions around the State. Sites were divided into three Clusters according to soybean yield and the effect of El Niño Southern Oscillation (ENSO) was considered in association with soil water balance. Neutral ENSO phases occurred in 32% of the years, while El Niño and La Ninã occurring in 36% and 32% of the years, respectively. Seasons under El Niño normally present higher accumulated rainfall, whereas those under La Niña present a reduction. Data from neutral-year sites of Clusters B and C seems to be more disturbed. No season had statistical difference of rainfall among Clusters under Neutral conditions. In addition, thermal gradient in RS from October to January benefited sites of Cluster A. Interaction of soils with higher water-storage capacity and cooler temperature reduces the water consumption by soybeans, causing lower values of water deficiency. A boundary function relating soybean yield and rainfall displays the limit of 800 mm for significant yield increments, and such amounts of rainfall were only achieved in El Niño seasons. The combined effect of rainfall and soil type on soybean yield, represented by the actual soybean yields-water deficit relationship, led to water propitiate from -3.7 to -15.2 kg mm -1 ha -1 . Decision-making on public policies and investments on the soybean industry can be supported from our results, either to better planning the investments on the soybean farming systems depending on the ENSO phase predictions, either to reduce the production risks in the region inherent to local weather.

According to Brazilian National Supply Company (Conab, 2019), the average on-farm soybean yields of the last ten crop seasons ranged from 1. 55  In such region soybean yield is heavily controlled by farming systems technology but it is also well-known that the irregular distribution of rainfall during the soybean growing season presents itself as a limiting factor to the potential yield of the crop (Berlato & Fontana, 1999; Sentelhas et al., 2015; Zanon et al., 2016).
Several studies have demonstrated the teleconnections between ENSO and anomalies in the seasonal rainfall patterns in subtropical southeastern South America (Grimm, 2004;Grimm and Tedeschi, 2009;Tedeschi et al., 2015), influencing the spring-summer harvest in RS, especially soybean crop. ENSO is a coupled ocean-atmosphere phenomenon characterized by sea surface temperature anomalies in the equatorial Pacific Ocean (Philander, 1983). Its warm phase called El Niño is associated to some positive precipitation anomalies observed in RS, while the ENSO cold phase is called La Niña, which triggers negative anomalies in the state (Grimm, 2000 Certainly ENSO's influence is geographically different within the state (Gelcer et al., 2013), depending directly on the ENSO intensity/type (Grimm, 2000;Grimm, 2004), and such spatial variability makes the overall impacts of ENSO on soybean yields still uncertain in such region (Berlato & Fontana, 1999;Matzenauer et al. 2018).
To better understand the relationship between climatic variability and agricultural productivity (Berlato and Fontana, 2003;Arsego et al. 2018) and provide some support, analyzes based on drought or seasonal water indexes (Gelcer et  To our knowledge, the relationship between weather variability to soybean yields in the region is not yet well defined. To fill the gap in the knowledge about the effect of ENSO and weather variability on soybean yield in RS State, and provide some forecast insights for decisionmakers and growers, we analyzed a long-term weather database containing data from 11 weather stations together with the ENSO phases. Specific objectives were: (i) to split the effects of soil and weather on crop yields based on crop modeling simulations and actual yield data (ii) to assess the soybean potential yields and yield gaps for different rainfall levels and regions and (iii) to evaluate soybean crop yield loss due to water deficit in each region of RS.  Rainfall missing data were replaced by data from the closest weather station. Air temperature missing data was replaced from linear relationships between the values from nearby stations. Validation methods were applied to identify erroneous data from weather sensors measurements (Estévez et al. 2011). The analyses were based on three classes of consistency tests: range test, step test and internal consistency test (Table 1). The quality of the weather data was examined searching for outliers when compared to other years and to those observed in neighboring stations. By visual inspection, we did not find any outliers after the statistical tests showed in Table   1 were applied to the datasets. Therefore, based on climatic regions of the State of Rio Grande do Sul as defined (Maluf & Caiaffo 2001)  Typical soil of each site (Figure 1) was defined based on references, describing soil taxonomy, profile features, granulometry and soil density ( Table 2 and Table 3

Study area and weather and soil conditions
where fc is the soil moisture at field capacity (kg kg -1 ); pwp is the soil moisture at permanent wilting point (kg kg -1 ); Clay, Silt and Sand content (kg kg -1 ); SD is the soil density (kg m -3 ); and AW is the maximum available water (mm).
From a cluster analysis performed by Melo et al. (2004) with the Ward method, using the Euclidean distance and data from 210 municipalities of soybean production in RS, the 11 sites ( Figure 1) were divided into three classes according to soybean yield (Mg ha -1 ), soybean production (tons) and percentage of soybean cultivated area data (ratio between soybean area and total area of the municipality) as: Ahigh yield, Bmedium yield and C -low yield ( Table 2). For each production region, soils are described in terms of texture and hydraulic parameters in Table 3.

Boundary function to soybean yield gap analyses
To where x is the seasonal water supply (rainfall, mm) and a, b and c are parameters of the equation model.
Regarding the 11 sites, a previous treatment of the soybean yield data had to be done. Given improvements in genetic and fertilizer applications over the 25 years, which on average caused a yearly increase of 52.9 kg ha -1 in RS (CONAB, 2018), it was necessary to statistically detrend the time-series of crop yield to remove these factors, and isolate the role of weather. So, soybean yield was detrended using linear regression (Goldblum, 2009).

Soybean water balance
Furthermore, as a complementary analysis to define the role of water on soybean crop, relationships between soybean yield and water deficit were fitted, therefore integrating soil and plant features and weather data considering the El Niño, La Niña or Neutral weather condition.
Through this methodology, each unit of water deficit was assessed for different production clusters.
To account for the water deficit, water balances (BH) were calculated using the concept of Thornthwaite & Mather (1955). The maximum available water (AW) reflected the root system depth simulation (RSD), thus considering the maximum availability of water [AWr (%)] for each soybean development sub period (Table 4).
where Qo/2.45 is the extraterrestrial solar radiation (mm day -1 ); Tmax is the maximum air temperature (°C); Tmin is the minimum air temperature (°C); Tavg is the average air temperature.
where DAE means days after plant emergency.
Mean comparison statistical analyses were performed using the Tukey test at 5% of error probability.

Results and discussion
Air temperature data was not analyzed for associations with El Niño, La Niña or neutral events (Table 5, Figure 2). A simple and comparative analysis presents the thermal nuances among the sites in order to identify its influence on soybean evapotranspiration and consequently on the water balance results and crop yield.
From October to January while soybean crops are predominantly in vegetative development stages, besides the thermal differences among sites basically due to the relief, the Rio Grande do Sul also presents a temporal distinction increase of air temperature from spring to summer among sites from low to high altitude (Table 5 and Figure 2). Cluster A sites (Ibirubá, Júlio de Castilhos, Lagoa Vermelha and Passo Fundo), the highest altitude and colder soybean crop areas in RS also show less thermal increase. So, these areas seems to be not thermally suitable for soybean, considering the optimum temperature range for soybean growth is between 20 and 30°C (Silva et al., 2015).
However, there is an apparent inconsistency since these are the most soybean yieldness sites of RS.   In the northern region of RS, where most of the soybean production area of the state is, such as Cluster A, there are predominantly clayey soils, with deep horizons A and B with at least 0.8m, resulting in maximum water availability ranging from 102 to 206mm (Table 3). In sites of Clusters B, the soils are clayey and deeper too, excepted Iraí ( Table 3).
Sites of clusters A and B differ mainly by altitude and air temperature ( Table 2). As previous set ( Table 5, Figure 2) sites in A (higher altitude) have lower temperature than sites in B, an important issue regarding to soybean evapotranspiration, which is minimized, reducing occurrences of water deficiency and thus favoring production (Pilau et al., 2018).
Sites of cluster C are in less clayey and also shallower soils areas, with lower water holding capacity (Table 3). In addition, they represent the areas of low altitude and higher temperatures Regarding the 25 seasons considered in our study ( Results (Table 6)   It is important to note that although the average rainfall is lower during La Nina (Table 7), there were several El Niño and neutral seasons in which rainfall was lower than those thresholds.
Regarding cluster A 599,0mm (La Niña average, see  From mean rainfall data of each cluster (Table 7) it is defined that under neutral condition total rainfall (within soybean production period) is higher in Cluster A than in B and C (statistically not defined), which under influence of El Niño have much more close values. Therefore, as already emphasized B and C sites seem to be more benefited by the phenomenon (Figure 3) Even so, water deficit is lower on Cluster A sites, basically due to soil (water retention) (Table 3) and air temperature (more crop suitable) ( Figure 2). Under La Niña events, Clusters B and C sites again appear to be more vulnerable to changes, in those cases with negative signals because the total rainfall in the soybean production time remains below those from Cluster A (Table 7).
Our results agreed with Grimm et al. (2000) who point out Southern Brazil, specially Rio Grande do Sul, as the region with the strongest signal in the El Niño event in Southern South America (Table 6; Table 7). Based on average rainfall values it can be observed that Rio Grande do Sul has higher rainfall at El Niño events and less rainfall at La Niña phase (Table 7) for all clusters.
Despite the differences it is essential to highlight the statistical equality between data of the neutrality condition and La Niña.
Although average data (Table 7)   Average precipitation data reveal the difference between the hot and cold phases of ENSO (Table 7). However, the variability observed among years under the same weather condition characterizes the importance of the intensity of the phenomenon ( Table 6). The relationship between cumulative rainfall and Tavg for each Cluster shows distinct influence of the phenomenon on local weather due to its intensity ( Figure 3). Linear coefficients indicate mean rainfall for Neutral weather of 702 mm for Cluster A, 676 mm for Cluster B and 648 mm for Cluster C, all below 800 mm required to maximize soybean yield based on the full attendance of water requirement (Figure 4).
The weather condition already unfavorable to soybeans in neutral years becomes even worse under the influence of La Niña (Figure 3). Angular coefficients of the linear adjusted graphic ( Figure 3) distinguished them among Clusters. Cluster A sites confirm to be the least influenced by ENSO, due to higher stability of water availability for the soybean crop. On the other hand, Cluster C sites (158mm °C -1 ) are the most disturbed, characterizing sites with higher climate risk of yield loss (MAPA 2018). In cluster sites A the interaction between soils with greater water storage capacity (Table 3) and colder air temperature ( Figure 2) results in lower values of soybean water deficiency (Table 7). So, in most soybean crop years, even with total rainfall close to Clusters B and C sites especially at neutral condition (Table 7), Cluster A seems to be more suitable and least risk places for the production of soybeans in RS (Figure 3).
Besides soil type (due to maximum water availability -AW) ( Table 3), an important issue related to the increase/decrease rainfall relative to the ΔTavg (Figure 3) is the water drainage capacity (not shown). This last issue is extremely important in Cluster C areas, such as Bagé, Encruzilhada do Sul and Santa Maria, in which soils show limited drainage with superficial water table. As these sites are highly influenced by the positive phase of ENOS (El Niño) (Figure 3), often with intense rainfall above normal (Table 5), these areas are more susceptible to damages not only due to water deficiency, but also caused by flooding (Zanon et al., 2015).  Based on the 800 mm found out as the threshold (Figure 4), it can be inferred that 80.8% of site-years cases were water limited (for Neutral 88% and La Niña 94%) ( Table 6 and Figure 4). The results for El Niño influence (61%) ( Table 6) highlights the point that, even in the face of a largescale phenomenon, generally considered positive for soybean production because it brings punctual increases in rainfall, weather can often imposes restrictions for soybean crop (Table 6; Figure 4) as described by Cirino et al. (2015). Additionally, Calviño and Sadras (1999) already indicated that water availability was limiting on farm yield in 54% of the years in the Argentinean Pampas and also for RS soybean areas, suggesting the profitability of this cropping system can be substantially enhanced with practices and cultivars aimed at increasing available water and water-use efficiency.
As established by Purcell and Specht (2004) and Nóia Junior and Sentelhas (2019) the availability of water to the plant depends not only on the amount and temporal distribution of rainfall and its disturbances caused by phenomena such as ENSO (Figure 3), but indisputably on soil type -water storage capacity (Table 3), as well as crop growth stage and variation in available energy -solar radiation and temperature. All of these show natural variability even in small tracts of land. Considering these variables, water balance can make water deficiency available as an alternative index to be correlated with soybean grain yield ( Figure 5).
Through individual analysis of each site, the combined effect of rainfall + soil on soybean yield can be seen in Figure 5, in which linear adjustments between soybean yield and water deficit can support water valuing from the angular coefficient (a) and project the attainable soybean yield (municipality) performed from the linear one (b). Although in Cluster A water deficit was lower than 200 mm ( Figure 5), sites less influenced by ENSO phenomena ( Figure 3) and with high soil water retention (Table 3)  It is clearly the combination of weather predisposed by ENSO phenomenon (Figure 3), coupled with soils less suitable for soybean cultivation (Table 3), that make the sites less productive and therefore less costly in relation to water deficiency ( Figure 5), leading Cluster B to an average loss of -7.4 kg mm -1 ha -1 and Cluster C -3.7 kg mm -1 ha -1 .
From the identification of the harvests corresponding to Neutral, La Niña or El Niño weather conditions ( Figure 5), it is El Niño that normally leads to significant yield gains, especially in extreme events by increasing rainfall (Figure 3) and thus reaching the limit to maximize yield ( Figure 4). The reason for the positive yield response of the rainfed soybean is that, there is an increase in rainfall compared to Neutral, which is already limiting to meet the crop water needs, and La Nina years, in which there is an even worse condition for soybean in terms of water availability.
Results such as those obtained by us help to understand the relationship between interanual climate and soybean, although as sad to Letson and McCullought (2001) they were not able to attribute the relationship to ENSO or to say that it is economically important. Figure 5. Relationship between soybean yield and water deficit (mm) for different ENSO phases (Neutral, El Niño and La Niña) for Clusters A, B and C.

Conclusions
In this paper we explore the role of weather, ENSO and soils on soybean yield in Southern phase, indicate sites of Clusters B and C as more severely disturbed by the ENSO phenomenon.
Relationship between soybean yield and rainfall for each soybean production year (1991 to 2017) and sites pointed the 800-mm rainfall as needed to maximize soybean yield. Differences between soybean attainable (Yw) and average yields of sites quantitatively establish the productive gap of the soybean in RS, and highlighting the variability among producing regions in terms of investment, technology and cropping systems. An inverse relationship between soybean water deficiency and yield reinforces the lower quality of the soil for soybean production and the negative effects even more pronounced in years of Neutral and La Niña phenomenon and its rainfall well below what is required to assure high yield levels. We identified the Cluster A as benefited by the thermal regime positively affecting soybean growth, development and water use. Decision-making on public policies and investments on the soybean industry can be supported from our results, either to reduce temporal production variability in the region and risks inherent to local weather.

Declarations
Funding: not applicable

Conflicts of interest/Competing interests:
The authors declare that they have no conflict of interest.

Availability of data and material:
The weather data that support the findings of this study are openly available in INMET and NOAA websites. Soil data are available in the literature cited. All material is described in the manuscript. More information can be shared upon requested.

Code availability: not applicable
Author's contributions: FGPconceptualization, data collection and organization, data analysis, writing and editing; FRMdata analysis, writing and editing; DAVGdata analysis and editing; GAD -data analysis and editing.  Vertical thermal gradient (°C 100m-1) from October to May among all weather stations.

Figure 3
Relationship between mean cumulative rainfall (mm) and mean temperature deviation of the Central Equatorial Paci c Ocean Surface (°C) (OND to MAM seasons).

Figure 4
Soybean yield plotted against seasonal rainfall, boundary layer for attainable yield (Yw) and soybean yield data from cultivars trials.

Figure 5
Relationship between soybean yield and water de cit (mm) for different ENSO phases (Neutral, El Niño and La Niña) for Clusters A, B and C.