Soil indigenous nutrients increase the resilience of maize yield to climatic warming in China

County-level data of maize yield and planting area were obtained from the Chinese Academy of Agricultural Sciences (http://www.caasbase.cn/). The dataset includes data for the period 1980–2010 from 1767 counties with records for longer than 20 years and from 477 counties with records for shorter than 20 years. Long-term recorded data were used to establish the yield-climate relationship.

We obtained the maize calendar dataset from the China Meteorological Data Service Center (https://data.cma.cn/). This dataset includes the sowing and maturity dates from 272 agrometeorological stations. We then interpolated the station level data to a raster surface by using ArcGIS to obtain the sowing and maturity dates for each county (figure S1 (available online at stacks.iop.org/ERL/15/094047/mmedia)). Maize was divided into spring maize and summer maize according to the sowing dates (figure S2). We used crop calendar data to compute cumulative or average values in the maize growing season for each climate variable.

Climate data were extracted from the AgMERRA dataset (Ruane et al 2015). AgMERRA comprises gridded daily climate data at 0.25° spatial resolution over the period 1980–2010. Gridded daily climate variables were averaged for each county by its boundary. Mean temperature (T), total precipitation (P), and average solar radiation (R) of the maize growing season over the period 1980–2010 were then calculated for each county.

We used data on soil indigenous nutrients from the Basic Nutrients Dataset of Soil Testing and Fertilization Recommendation (NATESC 2015) to measure the relationship between yield change caused by elevated temperatures and soil. The data cover the field observations from 2563 counties. Soil was sampled at 120–8500 sites in each county, depending on the area of cropland in a given county. The larger the area of cropland, the more sampling sites. The values of soil indigenous nutrients in each county are the average of the measurements at different sites. The soil indigenous nutrients include total nitrogen (STN), available phosphorus (SAP), and available potassium (SAK). For counties without soil indigenous nutrient data, we used the comprehensive gridded soil characteristics dataset of China (Wei et al 2013).

A summary of the variables and the abbreviation used in this paper are presented in table 1.

Linear regression models were used to assess the climate and yield trends from 1980 to 2010 for each county. Only the significant trends were considered (p < 0.05). We averaged county-level trends into provincial level.

The impact of climate variables on maize yield was estimated using a multiple regression model (equation (1)) for 1767 counties. This approach was used in accordance with previous studies (Schlenker and Lobell 2010, Zhu et al 2019).

$$\begin{align}{\text{ln(}}{Y_{i,t}} {\text{)}} = & {{\mathrm{\beta }}_{\text{0}}}{\text{ + }}{{\mathrm{\beta }}_{\text{1}}}t{\text{ + }}{{\mathrm{\beta }}_{\text{2}}}{t\,^{\text{2}}}{\text{ + }}{{\mathrm{\beta }}_{\text{3}}}{T_{i,t}}{\text{ + }}{{\mathrm{\beta }}_{\text{4}}}T_{i,t}\,^{\text{2}}{\text{ + }}{{\mathrm{\beta }}_{\text{5}}}{P_{i,t}} \nonumber \\ & {\text{ + }}{{\mathrm{\beta }}_{\text{6}}}P_{i,t}\,^{\text{2}}{\text{ + }}{{\mathrm{\beta }}_{\text{7}}}{R_{i,t}}{\text{ + }}\varepsilon\end{align} \tag{ 1 }$$

where ln(Y_i,t) is the natural logarithm of yield at county i in year t. The quadratic time trend (β₁t and β₂t²) was used to remove the effects of overall technological progress (e.g. artificial fertilization, and cultivar shift) on maize yield (Schlenker and Lobell 2010, Zhao et al 2017). The quadratic terms of T and P simulate the nonlinear response of yield to temperature and precipitation change, respectively. R represents the average radiation during the maize growing season. is the model error.

The temperature sensitivity of maize yield (equation (2)) was measured by the partial derivative of equation (1) (Chen et al 2017, Zhu et al 2019).

$$\begin{equation}{S_{Y,T}}{\text{ = }}\frac{{\partial Y}}{{\partial T}}{\text{100% = (}}{{\mathrm{\beta }}_{\text{3}}}{\text{ + 2}}{{\mathrm{\beta }}_{\text{4}}}\overline {{T_i}} {\text{)100% }}\end{equation} \tag{ 2 }$$

where $\overline {{T_i}}$ represents the mean of T during the period 1980–2010 in county i. β₃ and β₄ are the parameters of county i in equation (1). S_Y,T is the temperature sensitivity of maize yield, representing the yield change (%) for 1 °C warming. We evaluated the S_Y,T of 1767 counties using equations (1) and (2).

No significant trends in precipitation were detected in most of the meteorological stations and regions of China (Zhang et al 2011, Sun et al 2014b). Therefore, we did not consider the yield sensitivity to precipitation.

We employed a machine learning method, random forest (Breiman 2001), to evaluate the relationship between temperature sensitivity and soil indigenous nutrients. The package 'randomForest' (Liaw and Wiener 2002) that was built for R language was used on the input data with the control parameters ntree = 500 (appropriate number of trees to ensure the model has stable and low error rates) and mtry = 1 (default number of variables considered for splitting) in our study. In each tree, 36% of samples are left out and these are called out-of-bag (OOB) data. The OOB data are used to estimate the generalization error. The importance of a variable is interpreted as an increase in the mean square error (IncMSE; %) due to random permutation of that variable. The dependent variables were S_Y,T, and the independent variables were STN, SAP, and SAK. We conducted the Random Forest algorithm for both spring maize and summer maize. We used 1767 samples to build the random forest model. For the other 477 counties without S_Y,T, we simulated it by inputting the data on STN, SAP, and SAK into the random forest model.

Warming was more pronounced than the change in precipitation and solar radiation. Significant trends in the temperature of the maize growing season were observed in 1386 counties, with an increase at the rate of 0.36 °C per decade. Large areas of northern, central, and eastern China showed positive trends in temperature (figure 1 (a)). For spring maize, 15 of the 22 Provinces showed significant warming trends (table S1). However, no significant warming trend was found in southwest China at the provincial scale. For summer maize, the largest warming rate was observed in Beijing (0.49 °C per decade). Only Shandong (SD) and Yunnan Provinces had no significant warming trend.

Zoom In Zoom Out Reset image size

Only 218 counties showed significant trends in precipitation at an average increase rate of 28.4 mm per decade. Positive trends were found in parts of northeastern China, while negative trends were found in a narrow part of eastern and southwestern China (figure 1(b)). At the provincial level, Heilongjiang (HLJ) Province significantly became drier (−24.0 mm per decade) during the spring maize growing season, and Gansu (GS, −40.1 mm per decade) and Sichuan (−42.2 mm per decade) Provinces became drier during the summer maize growing season (table S1). However, precipitation increased by 50.8 mm per decade in SD Province.

Downward trends in solar radiation occurred in a part of northwestern, southwestern, and eastern China, while upward trends occurred in a confined area of central China (figure 1(c)). The average trend of 280 counties was −0.32 MJ m⁻² d⁻¹ per decade. For spring maize, solar radiation significantly decreased in Ningxia (NX) and GS Provinces (table S1). For summer maize, Xinjiang (XJ) was the only Province with a reduced rate in solar radiation.

Maize yield in China has been constantly increasing over the past decades. Significant trends were observed in 1570 counties (89% of maize growing counties) at an average increase rate of 120.4 kg ha⁻¹ yr⁻¹. Positive trends (1559 counties) occurred extensively in maize planting areas, especially in northeastern and northwestern China (figure 1(d)). Remarkable enhancements were found in XJ (237.4 kg ha⁻¹ yr⁻¹), HLJ (167.8 kg ha⁻¹ yr⁻¹), NX (155.9 kg ha⁻¹ yr⁻¹), Jilin (JL, 153.4 kg ha⁻¹ yr⁻¹), and Inner Mongolia (IM, 153.0 kg ha⁻¹ yr⁻¹) Provinces. Another 17 Provinces recorded an increased yield of 100–140 kg ha⁻¹ yr⁻¹. Spring maize (124.3 kg ha⁻¹ yr⁻¹) had stronger elevated trends in yield than summer maize (113.1 kg ha⁻¹ yr⁻¹).

Figure 2 shows the distributions of S_Y,T of spring maize (figure 2(a)) and summer maize (figure 2(b)) simulated by equations (1) and (2). The change in yield was concentrated in ±20% for 1 °C warming for both spring maize and summer maize. The area weighted average yield of spring maize in 1077 counties and summer maize in 690 counties decreased by 1.1% and 2.9% for 1 °C warming, respectively. Summer maize is more vulnerable than spring maize. S_Y,T showed a clear spatial pattern. The positive impacts of increased temperature were distributed in the northern areas of Northeast, Central, and South China, and the negative impacts mainly occurred in the southern areas of Northeast China, Loess Plateau, and North China Plain (figure S3).

Zoom In Zoom Out Reset image size

The random forest model was built with data from 1767 counties. It revealed that STN was the most important predictor of S_Y,T for spring maize (IncMSE = 40.2%), followed by SAK (IncMSE = 17.7%) and SAP (IncMSE = 16.8%). The most important predictor for summer maize was SAK, followed by STN and SAP, and their IncMSEs were 12.8, 5.9, and 2.2%, respectively. The R² of the model was 0.78 for spring maize and 0.73 for summer maize. The high R² indicated that the soil indigenous nutrients regulated the response of maize yield to temperature increase and explained the spatial pattern of S_Y,T well.

The partial dependence plots visualized the relationship between S_Y,T and soil indigenous nutrients (figure 3). Most of STN (95%) of spring maize and summer maize were concentrated from 0.52 to 2.52 and from 0.61 to 2.14 g kg⁻¹, respectively. With the increase of STN, the response of spring maize yield to warming changed from negative (yield loss) to positive (yield increase) (figure 3(a)), while summer maize did not show a consistent dependency (figure 3(d)). In agreement with the order of IncMSE, SAP also did not show a consistent dependency (figures 3(b) and (e)). SAK of spring maize and summer maize were mainly distributed from 88.9 to 93.7 and from 74.1 to 197.7 mg kg⁻¹, respectively. As SAK increase, the risk of yield loss increased in a certain range of SAK (figures 3(c) and (f)).

Zoom In Zoom Out Reset image size

We calculated the S_Y,T of an additional 477 counties that had less than 20 years of records through the random forest model. Compared to S_Y,T simulated by climate variables (figure 2), S_Y,T simulated by soil indigenous nutrients was distributed in a narrower interval (figure 4). Counties with S_Y,T between ±10% account for 89% and 90% of the total for spring maize and summer maize, respectively (figure 4). The average yield losses of spring maize and summer maize were 1.8% and 3.3% for 1 °C warming, respectively. For all maize planting areas, 751 counties showed a positive response to warming (3.8% °C⁻¹), but 1493 counties showed negative responses (−5.8% °C⁻¹). The area-weighted average maize yield in 2244 counties decreased by 2.6% for 1 °C warming. Based on the average of the provincial scale, maize yield in 25 of the 31 Provinces decreased with increasing temperature (table S2).

Zoom In Zoom Out Reset image size

Great variation in soil indigenous nutrients existed across different regions (figure 5). Northeastern and southwestern China showed higher STN content (figure 5(a)). The STN of HLJ, JL, Tibet, and Guizhou Provinces exceeded 2.00 g kg⁻¹ (table S2). However, the STN of northwestern China and the North China Plain was lower, even less than 1.00 g kg⁻¹ in Shanxi (SX), Shaanxi, NX, SD, Henan, and Hebei Provinces. High SAP content was seen in northeastern, southwestern China, and North China Plain (figure 5(b)). SD and SX Provinces showed the highest (26.8 mg kg⁻¹) and lowest (12.5 mg kg⁻¹) SAP content, respectively. SAK content was higher in northern China than in southern China (figure 5(c)). Most Provinces in the south showed less than 100 mg kg⁻¹ of SAK, while in the north, values over 120 mg kg⁻¹ of SAK were seen (table S2).

Zoom In Zoom Out Reset image size

We built a framework to measure the capability of soil indigenous nutrients to mitigate the impact of warming on maize yield. We classified different effects of soil indigenous nutrients on S_Y,T into seven levels: extreme vulnerability (S_Y,T ≤ − 10% per °C), moderate vulnerability (−10% per °C < S_Y,T ≤ − 5% per °C), low vulnerability (−5% per °C < S_Y,T ≤ − 1% per °C), marginal (−1% per °C < S_Y,T ≤ 1% per °C), low resilience (1% per °C < S_Y,T ≤ 5% per °C), moderate resilience (5% per °C < S_Y,T ≤ 10% per °C), and high resilience (S_Y,T > 10% per °C). The areas of extreme vulnerability were mainly distributed in the Loess Plateau (figure5(d)). The areas of moderate and low vulnerability were concentrated in the XJ and IM Provinces, North China Plain, and in the southern regions of Northeast China. Resilience mainly occurred in the IM, Hunan, Guangxi Provinces, northern regions of Northeast China, and the western part of Southwest China. The spatial distribution of S_Y,T resembled that of STN.

In the context of climate warming, maize yields in regions with high STN generally increased, while the risk of yield reduction appeared in the regions with high SAK (table S2). In addition, the SAP concentration is likely not closely related to the warming-related yield change (table S2).

Summer maize is more vulnerable to climate warming than spring maize. The vulnerable area of summer maize and spring maize were 8.47 Mha and 7.97 Mha, respectively, accounting for 72.1% and 53.6% of the total planting area (figure S4). The areas of summer and spring maize with resilience accounted for 15.9% and 36.1%, respectively. Overall, the vulnerable area and the resilience area accounted for 62% and 27% of the total maize-planted area in China.

We averaged the STN, SAP, and SAK for each level of S_Y,T to obtain more insights into what kind of soil indigenous nutrient conditions increase the resilience of maize yield (table 2). S_Y,T of spring maize increased with increases in STN and SAP. The yields decreased by more than 10% per °C in the regions where STN and SAP were less than 0.91 g kg⁻¹ and 13.4 mg kg⁻¹, respectively. In regions where STN exceeded 1.80 g kg⁻¹ and SAP exceeded 19.6 mg kg⁻¹, spring maize showed resilience to the impacts of warming. STN and SAP in maize planting areas with high resilience to warming were twice as high as in areas that were extremely vulnerable to warming. About 31% of the 1457 counties were resilient to the negative impacts of warming because of favorable soil indigenous nutrients. In barren spring maize planting regions, the negative effects of temperature rise can be reduced and yield can be improved by increasing the application of nitrogen and phosphorus fertilizer.

S_Y,T of summer maize increased with an increase in STN (table 2). Yield decreased by more than 10% for 1°C warming in the regions with STN less than 1.10 g kg⁻¹. Summer maize obtained resilience to warming in regions with an STN of more than 1.33 g kg⁻¹. However, only 18% of the 787 counties were resilient for temperature rise because of appropriate soil indigenous nutrients. Maize yield benefitted from a relatively low SAK. Increasing the use of nitrogen fertilizer and appropriately decreasing potassium fertilizer can increase the resilience of summer maize against the negative impacts of elevated temperature.