Does Climate Change Influence Russian Agriculture? Evidence from Panel Data Analysis

Abstract

Agriculture is one of the economic sectors primarily affected by climate change. This impact is very uneven, especially for countries with large territories. This paper examines the contribution of climate change to the improvement in agricultural productivity in Russia over the past two decades. Several ensembles of fixed effects regressions on yields and gross harvests of grain, fruits, and berries, potato, and vegetables were evaluated for a sample of 77 Russian regions over the 2002–2019 period. In contrast to similar studies of the climate impact on Russian agriculture, we considered a larger set of variables, including both Russian and global climate trends, technological factors, and producer prices. Russian weather trends such as winter softening and increase in summer heat have a significant but opposite effect on yields. An interesting finding is a significant and mostly positive influence of global climatic variables, such as the CO₂ concentration, El Niño and La Niña events on both harvests and yields. Although technological factors are the main drivers of growth in Russian agricultural performance over the past 20 years, we found a strong positive effect on yield and gross harvest only for mineral fertilizers. The influence of the other variables is mixed, which is mainly due to data quality and aggregation errors.

1. Introduction

1.1. Climate Change and Performance of the Russian Agricultural Sector over Past Years

The impact of climate change on agriculture is controversial. The past two decades have witnessed both the highest global temperatures and the greatest number of natural disasters. Increasingly frequent extreme weather events are one of the main reasons for the rising number of undernourished and food insecure people [1]. Lobell and Field [2] showed that climate change observed in 1981–2002 caused a decrease in global yields of wheat, maize and barley, but not for soybeans. However, estimated climate-driven losses of global agriculture over this period were overcompensated by agricultural technological advances and by fertilization effects of increased CO₂ levels.

In addition, the impact of climate on agriculture is uneven across the world. Increased drought will reduce agricultural production in Africa, the Middle East and South and Southeast Asia. On the contrary, it is expected that higher latitude regions such as Russia will benefit from climate change by extending the vegetation period [3]. At this point, there is much evidence to support this assumption for comparable countries. Due to global warming, an increase in grain yields is expected to occur in Ireland [4], Finland [5,6], the USA [7,8] and Canada [9,10].

However, climate change in Russia can affect crop yields in several dimensions. The main negative effect is an increase in the number of abnormal adverse weather events, especially droughts. Severe and extensive droughts can cause a 40–50% reduction in gross grain yields in major grain-producing regions [11]. Another risk factor is the increased negative impact of pests and crop pathogens. Climate warming and thawing of permafrost lead to the expansion of pest habitats and population growth. Increases in the annual sum of active temperatures of the surface air layer caused the boundaries of the Italian locust distribution to shift significantly northward in 1996–2015 compared to 1956–1976 [12]. Estimates of the potential distribution of the Colorado Potato Beetle under further climate warming in the 21st century indicate that its habitat will expand substantially in the northern, northeastern, and eastern directions [13].

Nevertheless, the Federal Service for Hydrometeorology and Environmental Monitoring of Russia (Roshydromet) identifies three main effects of climate warming that could at least partially lead to positive effects on agriculture: (1) an increase in the heat availability of crops and the duration of the growing season; (2) an increase in winter air temperatures, determining the conditions for overwintering of crops; (3) changes in moisture conditions, which are caused by an increase in precipitation in the cold season and a decrease in precipitation in the warm season [11].

The 1975–2008 period revealed a climate-driven increase in the yield of grain and leguminous crops on the entire territory, except for the regions of the Central federal district (18 sub-federal Russian territorial entities located around Moscow, the capital of Russia). The greatest positive effect was observed for winter grain crops. In some regions of the North Caucasus and the Volga region, the increase peaked at 10–15% within 10 years [14]. Still, there is evidence that the effects can be contradictory. In the territory of the European part of Russia for 1998–2017, there was no growth of climate-driven yields due to aridification and increasing tension of the thermal regime. At the same time, the increase in thermal resources and the duration of the growing season stimulated the growth of bioclimatic potential [15].

For such a large country as Russia, regional differences in the response of crop production to climate change can be very significant. The regions of Volga and Southern federal districts were national champions in terms of the growth rate of grain yield (2.2–2.6% over 10 years). However, the threat of drought is also most likely in the southeastern part of European Russia [11].

On the contrary, Central Siberia would likely benefit from the climate warming trend. For regions such as Krasnoyarsk Krai, Republic of Khakassia and Tyva Republic, agricultural production could double as the climate warms in the 21st century. In this case, the cultivation of traditional crops such as grain, potatoes and corn for silage may gradually shift 500 km to the north, and in the south of the territories, may be introduced new crops such as apricots, grapes and gourds [16].

Another feature of spatial heterogeneity is revealed in the yield dynamics of different crops. Between 1975 and 2010, the yield of grain and leguminous crops in Russia increased by an average of 4.5%. A decrease in yields was recorded only in the south of the Volga federal district, as well as in the southern and central regions of the Central federal district. Spring barley yields during the same period increased only in the Urals and Siberia, while negative trends were observed in the European part of Russia [17].

Notably, the risks for winter wheat are lower than for spring wheat [15]. Consequently, winter wheat was in favorable growing conditions in 1975–2010; the maximum growth was observed in the North Caucasus and the south of the Volga region. Between 1996 and 2010, there was also a climate-driven increase in the yield of sunflower and sugar beet for the European part of Russia. Corn yields increased in the Volga and Central federal districts but declined in the Southern federal district due to increased drought in the summer [17].

During the past two decades, Russia became one of the major exporters of agricultural production. By 2017–2018, Russia was supplying 10–13% of global exports of total grain and 20–23% of wheat [18]. An important point is that this successful performance on the global market was not driven by extensive development. Russia’s grain harvested area fell from 58 mln ha in 1987–1991 to 41 mln ha in 2017–2019, but at the same time, yields rose significantly from 1.63 tons per ha in 1987–1991 to 2.81 tons per ha in 2017–2019 [18]. The reason was that the economic growth of the 2000s allowed for an increase in the use of fertilizers and higher-quality seeds and machinery, combined with an increase in farm management efficiency [19,20]. All these factors led to the large increase in yields for all main agricultural products over the past twenty years (Figure 1).

Since 2000, the yield of potatoes and vegetables has increased by 1.7 times, grain by 1.8 times and fruit and berry by 2.6 times. Gross harvests are also rising, except for potatoes, which still have not exceeded the level of 1990. The best outcome has been in grain, whose gross harvests have doubled since the 2000s.

The most export-oriented regions of Russia are in the southwest of the country: Krasnodar Krai, Rostov Oblast, and Stavropol Krai. The global warming trend may decrease their grain export volumes by 4–5 mln tons [21]. To overcome these negative effects, it is necessary to carry out some measures of adaptation to future climate change. Development options may include increasing the area of reclaimed land, and changing the structure of crops and tillage methods [21].

1.2. Assessing the Impact of Climatic Conditions on Agricultural Productivity in Russia

The influence of climatic factors on crop yields is considered by representatives of multiple scientific fields: agronomists, meteorologists and economists, who focus on various drivers of crop productivity [22]. Despite the different points of view, most of the studies employ the methods of econometric modelling. The common specification is the regression on yields as a dependent variable with several climatic characteristics as predictors. The most common climatic variables considered are temperature and precipitation. Further differences in the choice of variables are usually due to the growing conditions of a particular crop. For instance, the winter wheat might be more sensitive to autumn and winter temperatures, while the spring grains might be more strongly affected by extreme heat days [23].

The choice of methods for testing the presence of a causal relationship between climate change and agricultural productivity is also highly dependent on the object of study. Focusing on the country level makes it easier to use time series analysis due to the availability of larger datasets. In contrast, regional analysis is usually bound with more data restrictions. This problem is especially crucial for studies on the Russian economy. Socio-economic data on the same territories of Russia and the USSR are not fully compatible. In addition, the economic crisis and hyperinflation of the 1990s made all monetary indicators incomparable to other periods. Most time series of socio-economic indicators of Russia’s development are limited to 20 years of the 21st century or even less. Multiple changes in statistical classifiers during the mentioned period make some of the data inconsistent. This pattern is inherent in many Russian industries [24] and agriculture is not an exception.

Belyaeva and Bokusheva [23] developed a panel fixed-effects regression on crop yields for 69 Russian regions over the period of 1955–2012. Using several variables describing seasonal temperature and precipitation, they found that large yield losses were caused by significant heat when daily temperatures exceed 25 °C. The hot days influence the spring wheat and spring barley the most. There was also the positive linkage of grain yields with summer precipitation, though it could not reduce the damage of extreme heat. They also conclude that global warming will lead to a decrease in agricultural productivity in Southern regions of Russia along with a positive effect on spring barley, winter and spring wheat in the Northern and Siberian territories [23].

Siptits et al. [25] built projections of the yield and gross harvest of grain and leguminous crops in the 79 regions of Russia until 2100. They used monthly average temperatures and precipitation level as predictors of yields in 1995–2016. The authors showed that the volume of grain and legume production in Russia will increase up to 170 mln tons by 2100 due to improved agricultural conditions in the Volga and Central Siberia regions.

Ksenofontov and Polzikov [21] made scenario forecasts of grain production and consumption by federal districts of Russia until 2030. The results confirm the negative consequences of global warming on the Southern regions of Russia, but expect the growth of gross harvests of grains and other crops in the Central and Northwestern parts of European Russia.

Pavlova and Sirotenko [17] used multiple linear regressions with differences in yield, temperature and precipitations first developed by Lobell and Field [2] to analyze the influence of climatic change for the 54 Russian regions over 1975–2010. Climatic changes caused an increase in winter wheat yield from 1% in the south of the Central federal district to 17% in the south of the Volga region. In much of European Russia, a tendency to a slight decrease in the yield of spring barley and cereal crops of about 1% in 10 years was revealed. This is compensated by the increase in yields in the Ural and Siberian federal districts.

Tchebakova et al. [16] focused on three regions of central Siberia that might benefit from the global warming trend: Krasnoyarsk Krai and the Republics of Khakassia and Tyva. The authors employed multiple linear regression models on crop yields using the number of days when temperature grew above a base of 5 °C and the ratio of this indicator to the annual precipitation as predictors for the period of 1966–2009. In the next step of the analysis, the forecast for 2090 showed that southern areas of Central Siberia such as forest-steppe, steppe, and semidesert would expand by up to 40% and may become climatically suitable for farming. In addition, Babushkina et al. [26] also proved that the temperature and precipitation conditions in May–July contribute the most to the variability of crop yields in the Khakassia Republic.

Most of the aforementioned econometric studies use only the temperature and precipitation levels as the major predictors and did not consider economic and agronomic variables. In some cases, they were only controlled by the fixed effects terms. Although the results are consistent and generally show a significant increase in yields due to climate change, the absence of other factors may create a significant bias in the estimates. For instance, Ahumada and Cornejo [22] suggested that there are three major groups of factors influencing the soybean yields. The climate group of factors covers variables describing global warming trends and their consequences such as CO₂ concentration and extreme events. Technological factors are captured with the use of fertilizers, modified seeds, machinery, labor and agricultural practices, for example, irrigation. Economic factors include prices on crops, fertilizers, and land.

In this study, we employ fixed effects regressions to assess the retrospective influence of the climatic and other factors on main productivity characteristics for major Russian agricultural products for 2002–2019. This approach allows us to capture the influence of all factors that cause changes in the dynamics of crop yields.

2. Materials and Methods

We used a panel dataset for 77 Russian regions for 2002–2019 (T = 18). Following Ahumada and Cornejo [22], we consider several groups of factors describing the climatic, technological and price dynamics in Russian agriculture (

Table 1. Factors of agriculture productivity in Russia considered in the study.

Group	Variables	Description	Mean	Data Source
Agricultural productivity (dependent)	Yield grain	Yields of the main crops of Russian agriculture: grain, fruit and berry, potato and vegetables, tons per ha	2.1	Russian Federation Unified Interagency Information and Statistical System (EMISS) [27]
	Yield fruit		5.7
	Yield potato		13.5
	Yield vegetables		20.0
	Harvests grain	Gross harvests of the main crops of Russian agriculture: grain, fruit and berry, potato and vegetables, thousands of tons	1270.8
	Harvests fruit		33.8
	Harvests potato		328.7
	Harvests vegetables		159.6
Russian climatic patterns	Temp January	Monthly average air temperature in January, °C	−11.5	The Russian Statistical Yearbook, Rosstat [28]
	D Temp January	Deviation from normal air temperature in January, °C	1.0
	Precipitation January	Amount of precipitation in January, mm	34.6
	D Precipitation January	Ratio to normal amount of precipitation in January, %	111.2
	Temp July	Monthly average air temperature in July, °C	19.4
	D Temp July	Deviation from normal air temperature in July, °C	1.4
	Precipitation July	Amount of precipitation in July, mm	69.8
	D Precipitation July	Ratio to normal amount of precipitation in July, %	95.8
World climatic patterns	CO2	Mean atmospheric carbon dioxide at Mauna Loa Observatory (Waimea, HI, USA), ppm	391.7	Global Monitoring Laboratory of the National Oceanic and Atmospheric Administration [29]
	El Niño	El Niño events (dummy)	0.4	Golden Gate Weather Services [30]
	La Niña	La Niña events (dummy)	0.4
Technology	Harvesters	Number of combine harvesters per 1000 ha of crops (units)	258.5	Russian Federation Unified Interagency Information and Statistical System (EMISS) [27]
	Tractors	Number of tractors per 1000 ha of crop land (units)	6.4
	Mineral fertilizers	Mineral fertilizers applied by agricultural organizations, kg in terms of 100% of nutrients per 1 hectare of crops	35.7
	Organic fertilizers	Organic fertilizers used by agricultural enterprises, tons per 1 hectare of crops	1.5
	Crop land	Crop lands, thousands ha	1004.6
Prices	Price oats	Producer prices in Russia, USD/tons	110.7	FAOSTAT [31]
	Price wheat		142.4
	Price potato		232.9
	Price vegetables		233.4

).

The choice of factors is determined by the availability of data and the perceived influence of the variables. We use two main indicators of agricultural productivity (yields and gross harvests) as the dependent variables for the four main crops: grain, potato, vegetables (onion, cabbage, beet, cucumber, tomato, pepper, eggplant, lettuce) and fruit and berry (pomaceous, drupaceous and berries).

The average temperatures and precipitation of the coldest (January) and the warmest (July) months are the most important factors to assess agricultural productivity [14]. As climate change increases the frequency of extreme weather events, we additionally use their deviations from the normal values for the given month. Temperature and precipitation data were obtained from the Russian Statistical Yearbook published by the Federal State Statistics Service of Russia (Rosstat) [28].

We also used several variables describing the global climatic patterns. There is evidence that CO₂ concentration might have a positive influence on plant growth [32,33]. As we do not have clear data for Russia, we presume that the genuine values for Russia are time-correlated with global ones. We used the values of CO₂ concentration from the Mauna Loa Observatory (Waimea, HI, USA) as the baseline [29]. Other greenhouse gases, such as methane, also might influence the growth of the crops [34], but due to the high correlation with CO₂ concentrations and the consequent bias of estimates, they were excluded from the analysis.

Other world climatic variables are the occurrences of El Niño and La Niña events in the Pacific Ocean. These events are usually associated with floods and droughts in southeastern South America [22]. El Niño and La Niña do not directly influence most of the Eurasian continent and generally have an impact on North and South America, Southern Asia, South-Eastern Africa and Australia. Although, the Roshydromet pointed out that El Niño had a significant effect on the record-breaking high temperatures in autumn 2020 in Russia [35]. We consider the presence of these events as dummy variables in each year that might affect both the plant growth conditions and the situation on the global crop markets. The data were obtained from the Golden Gate Weather Services website, managed by J. Null [30].

The group of technological factors is represented by the regional endowment with tractors, combine harvesters and fertilizers, which are the major factors of the positive changes in Russian agricultural sector over the past twenty years [18,19]. Though crop lands have declined significantly from 116.5 mln ha in 1990 to 79.1 mln ha in 2020 with simultaneous increases in yields, we still consider this variable as one of the important predictors. Data for these variables were obtained from the Russian Federation Unified Interagency Information and Statistical System (EMISS), which is a project of Rosstat, the official public statistics body of the Russian Federation [27].

The last group of variables combines producer prices for crops in Russia, which can influence farmers’ decisions about crop management practices [22]. These data were obtained from FAO [31], so they are not fully compatible with the crops that we include in the list of dependent variables.

Various types of models fit the structure of panel data quite effectively. To choose the right one, we followed the approach to model choice described by Dougherty [36]. Since the data sample for the Russian regions is not random, the model with fixed effects should be preferred to the model with random effects. In addition, we also ran an F-test for the null hypothesis that all regions have a common intercept (see Appendix B,

Table A3. Main test summaries for all considered models.

Model No.	Welch Robust Test for Differing Group Intercepts (F)	Heteroskedasticity LR test (χ2)	Wooldridge Test for Autocorrelation (F)
(1)	24.74 ***	930.53 ***	17.66 ***
(2)	15.88 ***	549.60 ***	0.13
(3)	23.09 ***	476.12 ***	3.80 *
(4)	76.61 ***	764.94 ***	24.63 ***
(5)	33.57 ***	1156.22 ***	18.97 ***
(6)	18.70 ***	530.61 ***	0.60
(7)	21.76 ***	422.06 ***	7.38 **
(8)	76.61 ***	744.05 ***	23.91 ***
(9)	78.99 ***	3805.10 ***	13.01 ***
(10)	58.60 ***	2721.01 ***	0.03
(11)	107.86 ***	1681.61 ***	1.11
(12)	85.41 ***	3122.44 ***	21.51 ***
(13)	85.14 ***	3876.63 ***	13.35 ***
(14)	74.32 ***	2873.07 ***	0.01
(15)	106.93 ***	1666.57 ***	0.99
(16)	108.49 ***	3357.24 ***	21.13 ***

*** p < 0.001, ** p < 0.01, * p < 0.1.

). In all cases, the null hypothesis was rejected, so the fixed-effects model is also preferable to the pooled OLS regression. It is notable that the approach of fixed effects panel modelling using a within-groups estimator is quite common in studies of yield dynamics and its dependence on climate change [23,37,38,39].

In its generalized form, the specification of the obtained model is as follows:

$$A{P}_{cit}={\beta }_{1}R{C}_{it}+{\beta }_{2}W{C}_{it}+{\beta }_3{T}_{it}+{\beta }_2{P}_{it}+{\alpha }_i+u_{it},$$

(1)

where AP is the agricultural productivity indicator (yield or gross harvest), RC stands for the list of Russian climate indicators (see Table 1), WC combines the world climate variables, T is for the group of technological variables and P denotes the prices vector. As for the indices, c = 1, …, 4 stands for a specific crop, i = 1, …, 77—Russian regions and t = 1, …, 18—time period. ${\beta }_i$ is the vector of model parameters, ${\alpha }_i$ reflects the fixed effects for Russian regions and $u_{it}$ is the error term.

Using a lot of variables to explore the yields and harvest dynamics is good in terms of interpretation and decreasing the model error, but also might cause the multicollinearity problem. Though the panel data analysis itself reduces the multicollinearity problem compared with cross-sectional OLS [40], the relationships between variables can still cause some bias. In our sample, obvious linkages exist between temperature and precipitation variables with their deviations. To avoid the instability and bias of the estimates due to the high correlation between them, we assessed two specifications for each crop and productivity indicator. The other variables were added in each model one by one, and the coefficients remained stable.

There are several other well-known problems that occur while modelling the panel regression. In general, the structure of data causes autocorrelation and heteroskedasticity. To analyze autocorrelation, we used the test developed by Wooldridge [41] and implemented in Stata by Drukker [42]. To test heteroskedasticity in panel models, we used the likelihood ratio test suggested by Wiggins and Poi [43]. The testing results (Appendix B, Table A3) show strong evidence of the presence of heteroskedasticity in all models. The correlation over time mostly appears in models with grain and vegetable yields and harvests as the dependent variables. The same tests for potatoes are less significant and the models on fruit yields and harvests show no autocorrelation at all. We employed a common way to avoid possible negative effects of these problems by using the heteroskedasticity and autocorrelation-consistent (HAC) standard errors [44,45].

All calculations and visualizations were made using Stata [46] and tables with regression results were obtained via the asdoc package for Stata [47].

3. Results

Table 2. Yield model results with initial climate variables, 2002–2019.

	(1)	(2)	(3)	(4)
	Yield Grain	Yield Fruit	Yield Potato	Yield Vegetables
Temp January	0.0162 ***	0.0606 ***	0.0977 ***	0.0756 ***
	(0.029)	(0.184)	(0.216)	(0.213)
Temp July	−0.0499 ***	0.0513	−0.3658 ***	−0.0948 *
	(0.071)	(0.31)	(0.541)	(0.558)
Precipitation January	−0.0024 ***	0.0019	−0.0106 **	−0.0186 ***
	(0.007)	(0.046)	(0.048)	(0.054)
Precipitation July	−0.0003	−0.0021	−0.0018	0.0018
	(0.004)	(0.018)	(0.02)	(0.022)
Crop land	0.0007 ***	0.0026 **	−0.0004	−0.001
	(0.002)	(0.011)	(0.012)	(0.017)
Mineral fertilizers	0.0108 ***	0.0174 **	0.0288 ***	0.0328 **
	(0.026)	(0.076)	(0.081)	(0.164)
Organic fertilizers	0.1385 ***	−0.074	0.1439	−0.4272 **
	(0.321)	(1.403)	(1.202)	(2.099)
Harvesters	−0.000001 ***	−0.000005 ***	0.000005 ***	0.00001 ***
	(0)	(0)	(0)	(0)
Tractors	0.0026	0.0389	0.0971 *	−0.0747
	(0.095)	(0.608)	(0.556)	(1.081)
CO2	0.0116 ***	0.0914 ***	0.1151 ***	0.1451 ***
	(0.028)	(0.181)	(0.149)	(0.248)
El Niño	0.103 ***	−0.0258	0.0173	−0.4281 **
	(0.223)	(1.256)	(1.287)	(1.78)
La Niña	0.1267 ***	−0.3133 **	0.0215	−0.3262
	(0.325)	(1.318)	(1.402)	(2.362)
Price wheat	0.0001
	(0.005)
Price oats	0.0004
	(0.006)
Price potato		0.0025 ***	0.008 ***	0.0079 ***
		(0.008)	(0.009)	(0.013)
Price vegetables		−0.0007	−0.0078 ***	−0.0047 ***
		(0.009)	(0.01)	(0.012)
Const	−2.7251 **	−34.1663 ***	−24.2807 ***	−32.5473 ***
	(12.261)	(72.367)	(60.093)	(105.415)
Observations	1223	1200	1214	1214
R-squared	0.537	0.283	0.484	0.429

Robust standard errors are in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1.

presents the results for the initial values of the temperature and precipitation as predictors of yields for different crops.

The climatic variables significantly influence all yields of crops within our study. The average temperature of January is highly significant and has the expected positive regression coefficient, which means that the softening temperatures in January through 2002–2019 contributed to higher yields. A warming of 1 °C in January on average caused an increase in grain yield by 0.016 tons per ha, fruit and berries yield by 0.06 tons per ha, potato yield by 0.098 tons per ha and vegetable yield by 0.076 tons per ha.

In contrast, higher July temperatures lead to droughts, which is reflected in negative coefficients. Although precipitation in Russia has been increasing over the past 40 years [11], in the model obtained, only January precipitation is significant for most crops except fruit.

CO₂ concentration has a strong positive effect on the growth of all crops, confirming the original hypothesis. Nevertheless, the specifics of the data used in the model force us to be cautious about the values of the coefficients because of possible bias.

The influence of the El Niño and La Niña events on crop yields is quite controversial. We found a strong positive influence of El Niño on grain, but a significant negative influence on vegetables. The La Niña dummy also showed a positive linkage for grain, but negative for fruit. This is quite consistent with other studies. For example, for Southeastern USA, El Niño has a negative impact on corn, winter vegetables, and strawberries but a positive influence on winter wheat. During La Niña years in this region, wheat, tomato and green pepper yields are higher, while the pasture crops and subtropical fruits yields might decrease [8]. In Canada, which has similar climatic conditions to Russia, the El Niño Southern Oscillation has also caused an increase in wheat and barley yields in recent years [48].

Cropland is significant only for grain and fruit. This is due to the largest share being grain land—more than half of all arable land. Over the past 20 years, grain acreage has increased slightly from 45.6 mln ha in 2000 to 46.6 mln ha in 2019. At the same time, the area under vegetables decreased from 7.4 to 5.2 mln ha, under potatoes from 2.8 to 1.3 mln ha and under fruit and berries from 0.8 to 0.5 mln ha.

The fertilizer variables showed different results. During the study period, the use of organic fertilizers increased by 1.8 times, and minerals—almost 3 times. This may be the reason for the more obvious and intense positive effect of mineral fertilizers on yields. The results for organic fertilizers are contradictory: they showed a strong but opposite effect on grains and vegetables. Though the effects of the organic fertilizer on crop nutrition are still not well-studied [49], the negative coefficient on vegetable yield is hard to explain as the expected influence should be positive [50].

The other technological factors also do not show the expected results. Tractors have no significant coefficients except for a weak positive influence on potato yields. Estimates for the combine harvesters are even more contradictory and show significant negative effects for the grain and fruit, but positive effects for the vegetables and potato. However, the coefficients are very small, which means that the possible impact is not large.

The prices influence on yields is still questionable in the literature. Lobel et al. [51] concluded that the estimates of the dependence of yields from prices vary a lot due to the strong interrelationships between prices, technology and weather factors. Miao et al. [52] found a significant influence of prices on corn yields, but not for soybean yields. In model (1), we did not find any impact of prices on grain yields. However, potato prices have a strong positive influence not only on potato but also on other crops. Negative significant coefficients for vegetable prices may be the result of incomplete comparability of data from different sources or due to the influence of unaccounted factors.

The worst results are obtained for regression (2) on fruit and berries. The number of significant variables is the smallest and the R-squared is only 0.28. For the other models, the obtained R-squared value is more reasonable and consistent with similar studies [14,23,25].

Finally, we calculated the individual effects for the specific regions. The most significant positive effects over 2002–2019 were obtained for Southwestern Russia, especially for the Caucasian regions. For example, for grain yields (model 1), the highest values were found for the Republic of North Ossetia–Alania (3.06 tons per ha), the Kabardino-Balkarian Republic (2.62), Republic of Adygea (2.36), the Karachay-Cherkess Republic (1.12) and the Republic of Dagestan (1.09). These results are consistent with other studies [14,23] and are explained by the fact that these regions are the most productive in agricultural terms due to a favorable climate.

On the contrary, negative values for grain yields are inherent to the majority of Siberian regions such as Krasnoyarsk Krai (−2.94), Tomsk Oblast (−1.15), Omsk Oblast (−1.02), and Irkutsk Oblast (−0.34). However, moderate positive values were obtained for the Altay Krai (0.34) and Khakassia Republic (0.05), which are the most developed agricultural regions of South Siberia [16,53].

The central part of Russia mostly shows moderate positive values of region-specific effects: Ivanovo Oblast (0.51), Kaluga Oblast (0.50) Bryansk Oblast (0.40), Belgorod Oblast (0.16), and Kostroma Oblast (0.08). According to other studies, these regions will benefit from global warming, but this effect will be limited [17,21]. All these findings indicate the uncaptured effects of difference in soil productivity that greatly affect yield potential in Russian regions [54].

The next list of models contains the same dependent variables (yields) but uses the deviations of climate variables as predictors (

Table 3. Yield model results with deviations of climate variables, 2002–2019.

	(5)	(6)	(7)	(8)
	Yield Grain	Yield Fruit	Yield Potato	Yield Vegetables
D Temp January	0.0072 ***	0.0843 ***	0.1143 ***	0.0629 ***
	(0.02)	(0.134)	(0.164)	(0.153)
D Temp July	−0.0422 ***	0.0065	−0.1963 ***	−0.0674 **
	(0.055)	(0.185)	(0.268)	(0.287)
D Precipitation January	−0.0001	−0.0005	−0.002 *	−0.0045 ***
	(0.002)	(0.012)	(0.011)	(0.017)
D Precipitation July	−0.0002	−0.001	0.0035 **	0.0025
	(0.003)	(0.014)	(0.016)	(0.019)
Crop land	0.0007 ***	0.0027 **	−0.0005	−0.0009
	(0.002)	(0.011)	(0.012)	(0.017)
Mineral fertilizers	0.0107 ***	0.0163 **	0.0271 ***	0.032 *
	(0.027)	(0.076)	(0.071)	(0.163)
Organic fertilizers	0.1346 ***	−0.0657	0.1186	−0.4309 **
	(0.318)	(1.371)	(1.203)	(2.089)
Harvesters	−0.000001 ***	−0.000005 ***	0.000005 ***	0.00001 ***
	(0)	(0)	(0)	(0)
Tractors	0.0004	0.0419	0.0861 *	−0.0817
	(0.095)	(0.581)	(0.515)	(1.056)
CO2	0.0105 ***	0.1014 ***	0.1287 ***	0.151 ***
	(0.03)	(0.187)	(0.149)	(0.248)
El Niño	0.11 ***	−0.081	0.1047	−0.424 **
	(0.223)	(1.284)	(1.328)	(1.827)
La Niña	0.1848 ***	−0.1508	0.2926 **	−0.1801
	(0.34)	(1.276)	(1.379)	(2.26)
Price wheat	0.002 ***
	(0.006)
Price oats	−0.0019 ***
	(0.007)
Price potato		0.0017 **	0.006 ***	0.0066 ***
		(0.008)	(0.008)	(0.013)
Price vegetables		−0.0009	−0.0067 ***	−0.0043 ***
		(0.009)	(0.011)	(0.014)
Const	−3.5163 ***	−37.7165 ***	−37.8716 ***	−37.6349 ***
	(12.595)	(75.309)	(61.279)	(106.598)
Observations	1222	1198	1212	1212
R-squared	0.53	0.3	0.472	0.429

Robust standard errors are in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1.

).

The results for models 5–8 are similar to those for models 1–4, and the coefficients for the same predictors are rather stable. However, some discrepancies should be discussed. The ratio to the normal amount of precipitation in January is not significant for grain apart from the previous results with the initial indicator. Additionally, the same ratio for July became significant for potato. In this model, potato is also affected by La Niña, but the coefficient for fruit has become insignificant. The major difference with previous results is in the prices. In model (5), wheat price and oat price have significant but opposite effects on grain yields. This proves the multicollinearity of prices with weather and other factors as discussed above. A positive link with wheat was expected, as it is one of the main Russian crops. The shares of winter and spring wheat in the total area of grains and legumes are 34% and 26% in 2019, respectively. The share of oats is only 5%, so the possible significant negative effect of its prices on grain yields might be explained by the other unaccounted factors.

We also obtained the fixed effects models for harvests (see

Table A1. Harvests model results with initial climate variables, 2002–2019.

	(9)	(10)	(11)	(12)
	Harvests Grain	Harvests Fruit	Harvests Potato	Harvests Vegetables
Temp January	15.4025 ***	0.5579 ***	3.7313 ***	0.8116 **
	(44.364)	(1.595)	(9.963)	(3.467)
Temp July	−51.7595 ***	0.9495 ***	−10.8224 ***	2.0981
	(108.957)	(3.496)	(27.912)	(15.061)
Precipitation January	−2.6993 **	−0.0035	−0.4077 ***	−0.192 **
	(13.413)	(0.337)	(1.519)	(0.748)
Precipitation July	−0.1394	0.0142	0.0721	−0.0113
	(4.179)	(0.111)	(0.912)	(0.372)
Crop land	2.4679 ***	−0.0006	0.0454	0.088 *
	(5.135)	(0.079)	(0.669)	(0.503)
Mineral fertilizers	5.8926 *	0.0751	0.9139 ***	0.9481 *
	(32.81)	(0.815)	(3.443)	(5.491)
Organic fertilizers	72.5909	−1.637	0.8567	−5.0638
	(508.942)	(16.311)	(47.994)	(55.087)
Harvesters	0.0009 ***	−0.000004	0.0002 ***	−0.00004
	(0.003)	(0)	(0)	(0)
Tractors	38.377 **	−0.5036	−3.3623	−6.8964
	(157.477)	(5.36)	(20.293)	(44.039)
CO2	16.0717 ***	0.096	−4.5795 ***	−0.7724
	(55.543)	(1.683)	(7.561)	(7.293)
El Niño	191.214 ***	0.0446	−6.3009	−5.565 *
	(458.572)	(15.08)	(60.121)	(30.697)
La Niña	279.8807 ***	−2.6475 **	−8.6199	−7.4179 *
	(590.989)	(12.447)	(58.001)	(42.647)
Price wheat	−0.1503
	(5.908)
Price oats	1.3552
	(8.536)
Price potato		−0.0037	0.2577 ***	0.0634 **
		(0.056)	(0.427)	(0.245)
Price vegetables		−0.0203 *	−0.3027 ***	−0.0805 ***
		(0.109)	(0.437)	(0.237)
Const	−7251.25 ***	−4.6897	2358.5331 ***	380.3244
	(26,086.591)	(656.768)	(3237.168)	(3300.697)
Observations	1231	1201	1214	1214
R-squared	0.31	0.05	0.21	0.156

Robust standard errors are in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1.

and

Table A2. Harvests model results with deviations of climate variables, 2002–2019.

	(13)	(14)	(15)	(16)
	Harvests Grain	Harvests Fruit	Harvests Potato	Harvests Vegetables
D Temp January	9.756 ***	0.509 ***	3.4507 ***	1.0222 ***
	(35.503)	(1.486)	(6.961)	(2.732)
D Temp July	−48.3434 ***	0.2357	−6.2703 ***	−0.9026 *
	(113.41)	(1.725)	(12.081)	(4.795)
D Precipitation January	−0.0251	−0.0004	−0.0669	−0.0591 **
	(3.63)	(0.076)	(0.433)	(0.27)
D Precipitation July	0.075	0.0034	0.1953 **	−0.0632 *
	(4.061)	(0.121)	(0.76)	(0.375)
Crop land	2.4734 ***	0.0013	0.0491	0.0953 *
	(5.19)	(0.078)	(0.647)	(0.507)
Mineral fertilizers	5.7554 *	0.0698	0.8595 **	0.935 *
	(32.551)	(0.814)	(3.341)	(5.508)
Organic fertilizers	66.8484	−1.5756	−0.1808	−4.7257
	(503.773)	(16.135)	(45.331)	(54.259)
Harvesters	0.0007 *	−0.000006	0.0002 ***	−0.00001
	(0.004)	(0)	(0.001)	(0)
Tractors	36.017 **	−0.4862	−3.7842 *	−6.8222
	(156.762)	(5.265)	(20.519)	(43.031)
CO2	15.2656 ***	0.1446	−4.1975 ***	−0.6169
	(56.563)	(1.842)	(7.478)	(7.392)
El Niño	198.654 ***	−0.6754	−4.055	−7.2557 **
	(464.878)	(15.953)	(60.196)	(28.448)
La Niña	350.4988 ***	−1.5857	0.2316	−3.9752
	(759.373)	(9.981)	(57.966)	(41.454)
Price wheat	1.5665 **
	(7.77)
Price oats	−0.6992
	(10.609)
Price potato		−0.0087	0.1891 ***	0.0475 *
		(0.058)	(0.351)	(0.24)
Price vegetables		−0.0233 **	−0.2653 ***	−0.0638 ***
		(0.097)	(0.457)	(0.224)
Const	−8192.8086 ***	−12.1816	1947.9513 ***	347.4659
	(26,134.194)	(715.743)	(3263.291)	(3331.251)
Observations	1230	1200	1212	1212
R-squared	0.311	0.053	0.201	0.157

Robust standard errors are in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1.

in Appendix A). The main results are quite similar to the yields models, but the R-squared values are lower. The major differences in estimates were revealed in the technological group of factors. The organic fertilizers are not significant at all in models 9–16. The positive effect of mineral fertilizers is also less significant and appears mainly in grain. The coefficients for combine harvesters and tractors are more reasonable in these models than for yields. These variables positively influence gross harvests of grain and potato.

4. Discussions and Conclusions

In this paper, we made the first attempt to estimate the influence of different factors on agricultural productivity in Russian regions over the past two decades. Most previous studies have focused only on two factors: temperature and precipitation. In contrast, we considered a large set of variables, including Russian and global climatic conditions, technological change, and producer prices for major crops. The results obtained are quite reasonable and consistent with other studies and common knowledge.

All regressions prove the significant influence of global and local climate conditions on yields and gross harvests for main crops. The January temperature and its deviation from the normal values have a strong positive contribution to the increase in yields and gross harvests for all crops. According to Roshydromet estimates, the climate in Russia is warming about 2.5 times more intensely than the global average [11]. The softening of winters in Russia could be an important factor in the future development of agriculture, especially for the Ural, Volga, and Siberian regions. This result for Russia is consistent with studies on other countries with similar climatic conditions. Projections to the middle of the 21st century show a significant expansion of winter crops in Finland [6] and Canada [9].

The other side of the global warming trend is droughts. The forecasted increase in climate aridity poses additional risks to crop yields [15]. In our models, the July temperatures have a significant and strong negative influence on most of the specifications. The main risks to crop production in Russia are increased aridity in the southwestern regions, which are currently the main producers of agricultural products, and the increased negative impact of pests and crop pathogens, which may spread their habitat to other regions.

We found a strong negative effect of winter precipitation for all crops, but almost no effect of July precipitation, except for potato yield in model (7). Since most of the literature proves the existence of a positive relationship between yield and summer precipitation [23], this result can be explained by the unaccounted influence of precipitation in other months. In addition, there are spatial differences in crop response to precipitation due to a combination of latitudinal temperature gradient, altitudinal precipitation gradient, and the technological water infrastructure such as hydrological network and irrigation system [26].

As was mentioned above, the geographical location of Russia makes it very sensitive to global climate changes. Our modelling results support this assumption since global climate variables such as El Niño and La Niña events have a significant impact on agriculture in Russia, especially on grain yields and gross harvests. The CO₂ concentrations also show a strong positive significant effect, which is consistent with other studies in the literature [22,55]. However, there is also evidence of a possible effect of decreased protein concentration in cereal crops [56,57].

Although technological factors have been the main drivers of Russian agricultural development over the past 20 years, their influence is not obvious in our models. The influence of crop land and machinery is limited and, in some models, controversial. This may be due to poor data quality and aggregation errors. Assessing the impact of these factors requires further research, which would consider disaggregated data on sown areas and combine harvesters by type of crop.

Fertilizer application made the most obvious contribution to the increase in yields and gross harvests. Mineral fertilizers are used more intensively in Russia, and according to modeling results, they had a sustained positive impact on yields, while the impact of organic fertilizers was limited. Nevertheless, the potential of the fertilizer market in Russia is not explored. The average volume of mineral fertilizers used in Russia is 48 kg/ha, while in European countries, the annual consumption of mineral fertilizers per 1 ha of arable land averages 149 kg [58]. In addition, the use of combinations of mineral and organic fertilizers can further increase yields [59] and accumulation of soil organic carbon [60].

We found that producer price factors are related with weather variables, which is consistent with other studies [51]. The use of temperature and precipitation deviations from normal values made wheat price a significant predictor of grain yield in model (4) in contrast to the initial weather conditions in model (1). However, the positive impact of potato prices on yield and gross harvest and the negative impact of vegetable prices were stable and significant in all obtained models. In our opinion, the negative coefficients are mainly explained by a gradual decrease in vegetable prices since 2010, along with the positive dynamics of crop yields.

Although we confirmed the main research hypothesis of positive impact of climate change on agricultural productivity in Russia, the fraction of variance unexplained is still significant in all obtained models. Directions for further analysis and refinement of modelling results can be formulated as follows:

• Data set supplementation with other factors of agricultural productivity. There are unaccounted factors that might bias the estimations, such as labor, use of modified seeds or irrigation practices [22]. This is especially important for interregional comparison. For instance, the Southwestern regions of Russia are export-oriented and, consequently, may employ more agricultural practices to increase yields.
• Use of more disaggregated data for both dependent and explanatory variables. The contradictory results for croplands and combine harvesters not divided by crop type have already been discussed above. This is also true for grain yield. For instance, winter and spring wheat also differ in their response to climatic conditions [15,17].
• Using additional weather data. Although weather conditions during the coldest and warmest months of the year explain the dynamics of yields and gross harvests relatively well, models that take into account temperatures and precipitation in all seasons of the year are more accurate [23].
• Considering the spatial heterogeneity. Due to its vast territory, Russia is a very heterogeneous country in many dimensions [61], and climate change trends have a multidirectional impact on Russian agriculture in different regions [21].
• Testing the different specifications. There is some evidence of the nonlinear nature of the influence of climatic conditions on yield, so it is possible to consider logarithmic or polynomial specifications [23,38].