1 Introduction

Soil nutrients and soil organic matter (SOM) play a critical role in soil quality, plant growth, and nutrient cycling in the ecosystem, and they are closely correlated with soil productivity (Hu et al. 2021a; Farooq et al. 2021). Crop development mainly depends on the availability of soil nutrients and SOM. Soil nutrients including nitrogen (N), phosphorus (P), and potassium (K) as well as SOM are recognized as the most critical mineral components for plants and microorganisms (Hu et al. 2021a). In addition, SOM, N, P, and K are closely related to a variety of current environmental issues such as carbon sequestration, carbon neutralization, and eutrophication (Hu et al. 2021b; Huang et al. 2021). Therefore, accurate information related to the spatial–temporal variability and factors determining soil nutrients and SOM in farmland is critical for soil management and agricultural production. However, the farmland ecosystem is usually complex due to varying conditions (e.g., crop types, soil management practices, topography, geomorphology and lithology heterogeneity, climate factors) (Fu et al. 2021; Wu et al. 2021). All of these lead to significant spatial heterogeneity of the soil nutrients and SOM (Yan et al. 2021a, b).

Many researchers have conducted studies to analyze the spatial or spatio-temporal variability of soil nutrients and SOM in various terrestrial ecosystems of different regions around the world (Granger et al. 2021; Zhou et al. 2021). Geostatistical methods have proved to be an effective tool to reveal the spatial variation of soil properties; hence, it has been broadly used to predict spatial distribution of soil properties at various spatial scales (Hu et al. 2019; Sheng et al. 2022). However, the high spatial heterogeneity of soil nutrients and SOM, coupled with the limited observations, leads to tremendous uncertainty in predicting spatial–temporal variation of soil nutrients and SOM, primarily when the study was conducted at a large spatial scale. At the same time, there is growing interest in exploring environmental factors (e.g., climate, relief properties) and anthropogenic activities which determine the variation of soil nutrients and SOM (Souza and Billings 2021). Previous studies have mainly focused on analyzing the role of natural factors such as climate conditions, terrain properties, and lithology for the variation of soil nutrients and SOM. However, few studies have quantified the influence of anthropogenic factors like straw return and chemical fertilizer input on the variation of soil nutrients and SOM in farmland (Farooq et al. 2021; Souza and Billings 2021). Moreover, the correlations between the soil properties and environmental as well as anthropogenic factors are very complex (Peng et al. 2019; Liu et al. 2022); thus, more robust and flexible methods are urgently needed to capture complicated and non-linear relationships between soil properties and their possible controls. Therefore, understanding spatio-temporal variation of soil nutrients as well as SOM, and identifying its potential controls is of great importance and crucial for implementing efficient and reasonable soil management measures to enhance soil fertility and promote agricultural production.

Jiangxi Province has a farmland area of 27,216 km2 and produced 21.9 million tons of grain in 2021. Yet, to meet the increasing demand for food, the large amount of chemical fertilizer is applied and the farmland is intensively cultivated to improve food production. Therefore, it is of great importance to explore the spatio-temporal variation and potential controls of soil nutrients and SOM in the farmland of Jiangxi Province. However, current information on the spatio-temporal variation and possible controls of soil nutrients and SOM is still far from demand.

Therefore, in this study, we aim to explore the spatio-temporal variation of SOM, and available N, P, and K between 2005 and 2012 in farmland throughout Jiangxi Province. In addition, we identified the most important variables (e.g., straw return, chemical fertilizer input, precipitation, temperature, soil properties) which affect the variation of SOM, and available N, P, and K. Finally, specific recommendations were put forward. The results obtained from this study were expected to provide critical information for making scientific and efficient farmland management policies as well as enhancing food production. We hypothesized that (1) the soil nutrients and SOM were largely affected by soil management measures, relief, soil properties, and climatic conditions; (2) the spatial distribution of soil nutrients and SOM is uneven in farmland of Jiangxi Province; (3) substantial spatial–temporal variation of soil nutrient concentration is expected due to the long-term effect of different soil management measures, relief factors, soil properties, and climatic condition; and (4) the machine learning methods could well capture the relationships between the concentrations of soil nutrients and SOM and multiple covariates; hence, it is an excellent tool to identify main factors which determine spatio-temporal of soil nutrients and SOM. This study provides crucial information for understanding the spatio-temporal variations of soil nutrients and SOM in the farmland. In addition, it will allow for the optimization of farmland management strategies and policy decisions.

2 Materials and methods

2.1 Study area

We conducted this study in Jiangxi Province, which is situated in Southern China. The Jiangxi Province is geographically located between 24°29′14′′N–30°04′41′′N and 113°34′36′′E–118°28′58′′E (Fig. 1). The total area of farmland in Jiangxi Province is 27,216 km2, among which the areas of paddy field, irrigated farmland, and dry farmland are 22,705, 41, and 4470 km2, respectively (https://bnr.jiangxi.gov.cn/art/2021/12/29/art_35804_3,810,534.html) (Fig. S1). Jiangxi Province is characterized by a humid subtropical climate, with hot, rainy summers and chilly, dry winters. The mean annual temperature of Jiangxi Province ranges from 16.3 to 19.5 °C, and the annual precipitation varies between 1341 and 1943 mm.

Fig. 1
figure 1

Map of sampling locations

Full size image

2.2 Data collection and chemical analysis

An extensive survey was conducted in farmland across Jiangxi Province. A total of 16,504 surface (0–20 cm) soil samples were collected between 2005 and 2012 under the framework of the soil testing and formulated fertilization project which was organized by the Department of the Ministry of Agriculture and Rural Affairs of China. The representative soil samples were randomly collected by an “S” shape pattern on each 0.3–0.5 km2 plot. At each plot, five subsamples (with a depth of 20 cm) within a 100-m radius were collected, and thoroughly mixed to obtain a composite sample of around 1 kg after removing the surface litter. The sampling locations were recorded using a portable global positioning system. Then, the tone and large roots were removed. All soil samples were air-dried, crushed using a mortar and a pestle, and then sieved through a 20-mesh nylon screen (1 mm aperture size) to determine the concentration of SOM, and available N, P, and K. The SOM concentration was determined by the dichromate oxidation method of Walkley–Black (Nelson and Sommers 1983). The available N was extracted with KCl and measured by alkaline hydrolysis diffusion method (Spargo and Alley 2008). The available P was extracted using sodium bicarbonate and the concentration was determined by the molybdenum blue method (Bao 2000). The available K was determined by spectrophotometry and flame photometry (Bao 2000; Hu et al. 2021b). Soil pH was analyzed by potentiometer at a soil/water ratio of 1:2.5 referring to the national standard (NY/T 1377–2007) regulated by the Chinese government (Bao 2000). The bulk density was determined using the cutting ring (5 cm diameter and 5 cm height) method (Liu et al. 2021a, b). The cation exchange capacity was measured using the ammonium acetate method (Schollenberger and Simon 1945). The exchangeable calcium and exchangeable magnesium were measured using ammonium acetate extraction-atomic absorption method (Bao 2000). The soil available silicon was measured using the molybdenum blue colorimetric methods (Wang et al. 2021a, b, c).

Other information about soil properties, soil management, and topography was also recorded when conduct soil sampling (Table S1). The raster data of mean annual precipitation, mean annual temperature, and population density with a spatial resolution of 1 km was downloaded from the Resources and Environmental Sciences Data Center, Chinese Academy of Sciences (RESDC 2017). Relief factors, such as slope, aspect, topographic wetness index, topographic position index, and multi-resolution valley bottom flatness, is calculated based on the DEM data issued by the European Space Agency using SAGA GIS (http://saga-gis.org/) software. The Topographic wetness index is a physical indicator to measure the effect of regional topography on runoff flow direction and accumulation. The Topographic position index is a topographic position classification identifying upper, middle, and lower parts of the landscape. The multi-resolution valley bottom flatness is a topographic index designed to identify areas of deposited material at a range of scales, based on the observations that valley bottoms are low and flat relative to their surroundings and that large valley bottoms are flatter than smaller ones. The landform is recorded according to the expert knowledge when sampling. The landowners provide information on the input of N fertilizer, P fertilizer, K fertilizer, and straw return amount.

2.3 Data analysis

The minimum, maximum, mean value, standard deviation, and coefficient of variation (%) of SOM and available N, P, and K in different years were analyzed using the RStudio (R Development Core Team 2013). Hillel (1980) divided the variability into three classes: weak (< 10%), moderate (10–100%), and strong (> 100%) variability, based on the coefficient of variation value. The statistical analysis and correlation test were employed in RStudio (R Development Core Team 2013). The optimized experimental semi-variograms in different years were fitted using GS + 9.0 software (Gamma Design Software, Plainwell, MI, USA). The random forest models were trained using the caret package in RStudio (Liaw 2002). All the maps of soil nutrients and SOM in different years were produced using ArcGIS (ESRI Inc., Version 10.7).

2.4 Geostatistical analysis

We used geostatistical method to analyze the spatial variation and map the spatial distribution of soil nutrients and SOM. The experimental semi-variogram was fitted to indicate the spatial dependence of soil nutrients and SOM. The experimental semi-variogram could be expressed as (Webster and Oliver 2007):

$$y^*(h)=\frac{1}{2N(h)}\sum\limits^{N(h)}_{i=1}[Z(x_i)-Z(x_i+h)]^2$$
(1)

where y (h) represents the semi-variance with spatial lag of h, Z(xi) is the value of variable Z at observed site i, and N(h) means the number of pairs of soil samples with distance lag of h. As one of the most widely used geostatistical methods, the ordinary kriging was used to map the spatial distribution of soil nutrients and SOM using ArcGIS 10.7 software (Hu et al. 2020a).

2.5 Random forest

The random forest is a classical and widely used machine learning method developed by Breiman (2001). It builds a large number of decision trees (500 trees were used in this study) and then combines all the tree to generate a random forest. The final prediction results are the averaged value produced by different individual decision tree. The random forest could model the linear or non-linear relationships between the dependent variable (SOM, and available N, P, and K in this study) and independent variables (like soil management factors, climate condition, terrain properties, soil properties in this study) (Hu et al. 2020b; Munnaf and Mouazen 2022). Moreover, it could quantify the importance of independent variables in the constructed model, according to how much worse the prediction would be if the independent variables were randomly permuted (Chen et al. 2019; Jia et al. 2020; Yan et al. 2020). Breiman (2001) provides the detailed principle of the random forest.

In this study, 24 explanatory variables thought to affect SOM, and available N, P, and K were chosen using an extensive literature review and expert knowledge (Table S1). These variables covered the lithology, climate, relief, soil management, and soil properties. The random forest model was constructed to calculate the relative importance of these variables. The covariates used in this study and corresponding sources are listed in Table S1.

In addition, the relative importance of different variables was calculated by the mean increase in prediction error for out-of-the-bag data, which occur as a result of randomly permuting each variable while leaving all others unchanged (% IncMSE) when running the machine learning models (Xie et al. 2021). The formula is as follows:

$${W}_{i}=\frac{{g}_{i}}{\sum_{n=1}^{m}{g}_{i}}$$
(2)

where \({W}_{i}\) was the relative importance of the \(ith\) factor, \({g}_{i}\) was the value of %IncMSE for the \(ith\) factor, and m was the number of covariates.

3 Results

3.1 Descriptive statistics of soil nutrients and SOM

The average concentration of SOM slightly dropped by 3.6% and decreased from 30.9 to 29.8 g/kg in 2012 when compared with 2005 (Table S2). The mean concentration of available N grown by 7.8% and increased from 155.8 mg/kg in 2005 to 167.9 mg/kg in 2012. In terms of available P, the averaged concentration dropped by 16.4% and decreased from 23.8 mg/kg in 2005 to 19.9 mg/kg in 2012. The mean concentration of available K slightly improved by 5.1% and increased from 86.0 mg/kg in 2005 to 90.6 mg/kg in 2012. The coefficient of variations of SOM and available N, P, and K varied from 30.7 to 74.7% between 2005 and 2012, which indicates moderate variability of soil nutrients and SOM concentration in the farmland of Jiangxi Province (Hillel 1980).

The mean pH value in the farmland of Jiangxi Province varied between 5.1 and 5.3, which belongs to the acidic grade as regulated by the Office of the National Soil Survey in China (1998). The mean concentration of cation exchange capacity changed between 6.0 and 6.4 cmol/kg. As indicated by the standard issued by the Office of the National Soil Survey in China (1998), it illustrates that farmland soil in Jiangxi Province has a weak ability to maintain fertility. In addition, a right-skewed trend was detected for the histogram of SOM and available N, P, and K, pH, and cation exchange capacity (Table S2, Fig. S2). This means the value of the SOM and available N, P, and K, pH, and cation exchange capacity deviate from the normal distribution. It also indicates that the mean values of the SOM and available N, P, and K, pH, and cation exchange capacity tend to higher than corresponding median values due to existing of a few extremely highly values (Fig. S2).

3.2 Temporal trend of soil nutrients and SOM

The boxplot of SOM and available N, P, and K and fitted temporal trends of mean between 2005 and 2012 are presented in Fig. 2. The averaged concentration of SOM and available P showed insignificant temporal trend between 2005 and 2012 (p = 0.93), whereas the mean concentration of available N (p < 0.05) and available K (p < 0.01) presented a significant increasing trend.

Fig. 2
figure 2figure 2

Temporal trend of soil available nitrogen (a), available phosphorus (b), available potassium (c), soil organic matter (d), pH (e), cation exchange capacity (f) in the farmland of Jiangxi Province. The upper figures are the boxplot of soil properties in each year. The blue point indicates the mean value of soil properties in each year. The solid orange line indicates the fitted linear temporal trend of each soil property. The grey area means the 95% confidence interval of the fitted temporal trend

Full size image

3.3 Geostatistical analysis of soil nutrients and SOM

The optimized fitted semi-variogram models for the SOM and available N, P, and K in different years are listed in Table S3. The high R2 value of most of experimental semi-variogram models indicate that the fitted semi-variogram models could well reveal the spatial variation of soil nutrients and SOM (Table S3). The exponential model gives the best fit for the experimental semi-variogram models of soil nutrients and SOM in most of the years.

The nugget variance (C0) represents the experimental error and field variation within the minimum sampling interval (Xia et al. 2019; Hu et al. 2022). The sill is commonly considered to be the variogram value where the variogram points or function flatten off at increasing distance (Webster and Oliver 2007; Xia et al. 2021). The nugget to sill ratio (C0/(C + C0), %) represents the proportion of spatial variation caused by random factors. When the C0/(C + C0) ratio is less than 25%, it means strong spatial dependence. When the C0/(C + C0) ratio is between 25 and 75%, it means moderate spatial dependence. When the C0/(C + C0) ratio is larger than 75%, it means weak spatial dependence (Webster and Oliver 2007; Hu et al. 2017). As shown in Table S3, the C0/(C + C0) ratio of different soil nutrients and SOM in different years varied between 50.0 and 89.9%. Among which, the C0/(C + C0) ratio of available N in 2012 and available P in 2005 and 2012 is the smallest, while the C0/(C + C0) ratio of available K in 2008 is the largest. This indicates that the soil nutrients and SOM showed weak to moderate spatial dependency in the farmland of Jiangxi Province.

The spatial range of soil nutrients and SOM in different years changed between 12.8 and 166.2 km. The spatial range of available K in 2011 was the smallest, while the spatial range of SOM in 2012 was the largest. The spatial range of available K showed increasing trend, while the spatial range of SOM, available N, and available P violently fluctuated between 2005 and 2012.

3.4 Spatial–temporal variation of soil nutrients and SOM

3.4.1 Spatial–temporal variation of SOM

Changes were observed for the SOM concentration in different regions between 2005 and 2012 (Fig. 3). The high value of SOM concentration mainly appeared in the central part of Jiangxi Province, while the low value of SOM concentration mainly observed in Northern and Southern regions. As shown in Fig. 3a, the value of SOM clearly decreased from 2005 to 2012 in farmland of the central part of Jiangxi Province, especially in the plain around Poyang Lake, while the SOM concentration in the farmland of the eastern, western, and southeastern parts of Jiangxi Province increased.

Fig. 3
figure 3

Spatio-temporal variation of soil organic matter (SOM) in the farmland of Jiangxi Province

Full size image

3.4.2 Spatial–temporal variation of available nitrogen

The general spatial pattern of available N in farmland of Jiangxi Province keeps stable between 2005 and 2012 (Fig. 4). The high value of available N was mainly detected in the central part of Jiangxi Province, while the low value majorly discretely distributed in western, northern, and southern regions of Jiangxi Province. Great difference was observed for the temporal change of the concentration of available N in different regions. The available N in farmland in the central part of Jiangxi Province decreased. In contrast, the available N in the western and southern region increased between 2005 and 2012 (Fig. 4i).

Fig. 4
figure 4

Spatio-temporal variation of alkali-hydrolyzable nitrogen (AN) in the farmland of Jiangxi Province

Full size image

3.4.3 Spatial–temporal variation of available phosphorus

Between 2005 and 2012, the high concentration of available P in farmland is mainly observed in the central and eastern parts of Jiangxi Province, while the low value was primarily detected in the Northern part and plain around the Poyang Lake (Fig. 5). In addition, the spatial pattern of available P concentration considerably changed from 2005 to 2012. The available P concentration in farmland in the southern, western, and eastern parts of Jiangxi Province slightly increased. Especially, the available P concentration in farmland in the southeastern part greatly increased, while available P concentration in the farmland of the central region decreased from 2005 to 2012.

Fig. 5
figure 5

Spatio-temporal variation of available phosphorus (AP) in the farmland of Jiangxi Province

Full size image

3.4.4 Spatial–temporal variation of available potassium

Clearly, change was taken place for the spatial pattern of available K concentration in farmland of Jiangxi Province from 2005 to 2012 (Fig. 6). Concentration of available K in northern part kept at high level while in southern and western and eastern part kept at low level between 2005 and 2012. In addition, the high concentration and low value of available K are crosswise distributed in the central part of Jiangxi Province. In terms of temporal trend, the available K concentration in the central part of Jiangxi Province increased, while available K concentration in most regions of the southern, western, and eastern parts decreased between 2005 and 2012 (Fig. 6i).

Fig. 6
figure 6

Spatio-temporal variation of available potassium (AK) in the farmland of Jiangxi Province

Full size image

3.4.5 General spatial pattern of soil nutrients and SOM during whole the study period

By compiling all the soil samples collected between 2005 and 2012, we produced the maps of different soil nutrients and SOM during whole the study period (Fig. 7). Our results indicate that the high values of SOM are mainly observed in farmland of the central part of Jiangxi Province, while the low values of SOM were mostly detected in northern part throughout the study period (Fig. 7a). In terms of available N, the high values are mainly located in the central part and plains around the Poyang Lake, while the low value is mainly situated in the southern and northern parts of Jiangxi Province (Fig. 7b). Regarding available P, the high values are primarily distributed on the central and southwestern part of Jiangxi Province. In contrast, the low value is primarily observed in the western and eastern part (Fig. 7c). For the available K, the high values are concentrated in the central parts and plains around Poyang Lake and the low values are concentrated in the eastern and southeastern parts of Jiangxi Province.

Fig. 7
figure 7

Spatial pattern of soil organic matter (SOM) (a), alkali-hydrolyzable nitrogen (AN) (b), available phosphorus (AP) (c), available potassium (AK) (d) in the farmland of Jiangxi Province which compiled all the soil samples collected between 2005 and 2012

Full size image

3.5 Changes of soil fertilizer grades in the farmland of Jiangxi Province

As regulated by the national standard of nutrients classification issued by the Chinese government, the soil samples were classified into six grades: class 1 (extremely high), class 2 (very high), class 3 (high), class 4 (moderate), class 5 (low), and class 6 (very low) (Table S4). The proportion of soil samples with SOM concentration at class 1 and class 2 decreased from 2005 to 2012, while the proportion of class 3, class 4, and class 5 increased. The SOM concentration of most of soil samples belongs to class 1 and class 4 from 2005 to 2012.

The ratio of soil samples with available N concentration which belongs to classes 1, 4, 5, and 6 slightly decreased, while the proportion of soil samples with available N concentration which belongs to classes 2 and 3 increased from 2005 to 2012 (Table S5). In terms of the available P, the proportion of soil samples with available P concentration which belongs to class 1 dramatically dropped by 11.7%. In contrast, the proportion of soil samples which belong to class 5 slightly decreased and the proportion of soil samples which belong to classes 2, 3, and 4 increased. For available K, the proportion of soil samples with available K concentration which belongs to classes 1, 2, 5, and 6 increased, while the proportion of soil samples which belong to classes 3 and 4 decreased. It is worth noting that the apparent lack of K fertility is detected in farmland of Jiangxi Province since the available K concentration in most soil samples belongs to classes 3, 4, 5, and 6. Lacking K would lead to the delay or advance of the natural growth and development process of crop hence cause reduction of quality and yield of crop. Therefore, applying K fertilizer is recommended to enhance the K nutrient supply for the crop.

3.6 The effects of different covariates on soil nutrients and SOM

We also quantified the relative importance of variables for affecting soil nutrients and SOM variation in farmland of Jiangxi Province (Fig. 8), and listed the top ten most important variables for different soil nutrients and SOM (Table 1). The results indicate that straw return amount, mean annual precipitation, and mean annual temperature were the most important variables for affecting variation of SOM. At lower magnitude, elevation, population density, the input of N fertilizer, the input of phosphate fertilizer, the input of potash fertilizer, available silicon, and multi-resolution valley bottom flatness also affected variation of SOM. These factors were ranked as the top ten variables for affecting variation of SOM. Concerning available N, the mean annual precipitation, straw return amount, mean annual temperature, population density, elevation, N fertilizer, potash fertilizer, available silicon, multi-resolution valley bottom flatness, and phosphate fertilizer were the top ten most important variables. In terms of available P, the mean annual precipitation, mean annual temperature, available silicon, elevation, population density, N fertilizer, phosphate fertilizer, potash fertilizer, multi-resolution valley bottom flatness, and exchangeable calcium were identified as most important variables which control variation of available P. Concerning available K, mean annual temperature, mean annual precipitation, elevation, population density, crop rotation system, soil available silicon, potash fertilizer, phosphate fertilizer, N fertilizer, and pH have the largest influence on variation of available K. Overall, the climate (e.g., mean annual temperature, mean annual precipitation) and soil management (e.g., straw return amount, population density, crop rotation system, N fertilizer, phosphate fertilizer, potash fertilizer) had larger effects on variation of soil nutrients and SOM.

Table 1 Top ten most important variables (variables with 10 largest value of % IncMSE) for affecting variation of different soil nutrients. The yellow color represents the anthropogenic factors, the green color represents the climate factors, the brown color represents the relief factors, the purple color represents the soil properties
Full size table
Fig. 8
figure 8

The relative importance of different variables which affect variation of soil organic matter, available nitrogen, available phosphorus, available potassium based on % IncMSE (MAP means mean annual precipitation, MAT means mean annual temperature, Ele means elevation, PD means population density, NF means input of nitrogen fertilizer, PF means input of phosphate fertilizer, KF means input of potash fertilizer, ASi means available silicon, MRVBF means multi-resolution valley bottom flatness, EMg means exchangeable magnesium, ECa means exchangeable calcium, CRS means crop rotation system, CEC means cation exchange capacity, PM means parental material, TLD means tillage layer depth, TWI means topographic wetness index, ST means soil type, SC means soil class, LF means landform

Full size image

4 Discussion

4.1 Comparison with others regions in China

Overall, most soil samples had concentrations of SOM, and available N and P at or above the level of class 3 (high grade). This indicates the farmland in Jiangxi Province could well supply SOM and available N and P for crop growth. In contrast, the concentration of available K in most soil samples belongs to or below the level of class 4 (moderate grade). This reveals relatively low concentrations of available K in farmland soils, and more K fertilizer input is urgently needed, although the average concentration of available K increased by 5.1% during the study period.

In this study, we compared soil pH, SOM, and available N, P, and K in other regions of China by lumping all the data collected from different years (Table S6). As shown in Table S6, the mean soil pH (5.21) in the farmland of Jiangxi Province was lower than these of Hunan Province and Guizhou Province and close to the soil pH in farmland of Zhejiang Province. The mean SOM in Jiangxi Province was30.9 g/kg and belongs to a very high degree (Table S4). It was higher than the mean SOM concentration in Sichuan Province, Jiangsu Province, Fujian Province, and Beijing City, while lower than those in Guizhou Province, Hunan Province, Zhejiang Province, and Northeast China. The mean available N in the farmland of Jiangxi Province was 164.3 mg/kg and belongs to extremely high grade, which was the highest among all the regions listed in Table S6. The mean available P concentration in farmland soils of Jiangxi Province was 20.6 mg/kg and belongs to a very high grade. It was higher than those in Sichuan Province, Guizhou Province, Jiangsu Province, and Zhejiang Province, while lower than those in Liaoning Province, Hunan Province, Jilin Province, Fujian Province, Beijing City, and Northeast China. The mean available K in the farmland soil of Jiangxi Province is 87.43 mg/kg and belongs to the moderate grade (Table S4). It was higher than those in Sichuan Province, Guizhou Province, Jiangsu Province, and Zhejiang Province but lower than those in Liaoning Province, Hunan Province, Jilin Province, Fujian Province, Beijing City, and Northeast China. Overall, the soil nutrients and SOM in farmland soils of Jiangxi Province are at a high level.

4.2 Effects of climate factors on soil nutrients and SOM

Our results indicate that the climate factors (mean annual precipitation, mean annual temperature) have the most significant effect on the variation of soil nutrients and SOM in the farmland of Jiangxi Province (Figs. 8 and 9). Many researchers have proved that temperature and precipitation could greatly affect the decomposition process of SOM through the alteration of soil temperature, hydrological cycle, soil moisture, and microbial activity (Pregitzer and King 2005; Yan et al. 2021a, b). Among which, the changes in soil moisture were proved as a consequence of climate change and which could then alter soil nutrient availability as well as soil–plant microbial interactions (Emmett et al. 2004). Yuan et al. (2017) revealed that soil carbon, N, and P concentrations generally decreased with water addition in manipulative experiments but increased with annual precipitation along environmental gradients. Osland et al. (2018) reported that at the regional scale, the climate has a considerable influence on SOM. Yu et al. (2018) demonstrated that the SOM concentration decreased with the increase of mean annual precipitation in subtropical China. Li et al. (2020) revealed a moderately negative correlation between SOM and mean annual temperature. 

Fig. 9
figure 9

Temporal trend of annual application rates (kg/ha) of nitrogen fertilizer (a), phosphate fertilizer (b), and potash fertilizer (c) per hectare in the farmland of Jiangxi Province (kg/ha)

Full size image

Moreover, the temperature and precipitation could also affect the plant cover and crop system in different regions, resulting in changes of the plant uptake of soil nutrients, which then indirectly change the cycles of soil nutrients. Pregitzer and King (2005) reported that plant nutrient uptake is influenced by changes in soil temperature. Matias et al. (2011) indicated that higher precipitation could boost microbial and plant-nutrient uptake, and hence achieve nutrient balance. All of these studies confirmed the crucial role of climate factors on the variation of soil nutrients and SOM.

4.3 Effects of soil management measures on soil nutrients and SOM

Our results found that soil management factors such as straw return and chemical fertilizer input essentially affected the variation of soil nutrients and SOM in the farmland of Jiangxi Province (Figs. 8 and 9). Many studies have confirmed that straw return could improve soil fertility, alleviate nutrient leaching, and increase crop yield and nutrient use efficiency (Wang et al. 2021a, b, c; Cui et al. 2022; Wu et al. 2022). Crop straw is a post-harvest waste material that can function as organic fertilizer. It is an important source of SOM and soil nutrients like N, K, magnesium, and sulfur in farmland (Zhu et al. 2010). Wang et al. (2015) found that straw return significantly increased the SOM and total N concentration by 10.1% and 11.0%, respectively. Zhang et al. (2021a, b) also found that straw return was the most effective way to conserve N in soil and could significantly reduce N runoff by 11.6%. Cui et al. (2022) revealed that straw return significantly increased the concentrations of SOM and available P over 4-year period.

However, Zhu et al. (2010) found a negative effect of straw return under initial high concentration of SOM, and a positive impact under initial moderate level on crop yield. In addition, the way of straw return to the field and the balance of soil nutrients can also affect the effectiveness of straw return (Huang et al. 2021; Hu et al. 2022). Moreover, excessive straw return can lead to imbalanced soil carbon and N and P stoichiometry (Jin et al. 2020). Thus, further surveys and deeper analyses are necessary to highlight the effect of straw return on soil nutrients and SOM.

The application of chemical fertilizer could also greatly affect soil nutrient variation. Generally, the input of chemical fertilizer could improve soil nutrient supply. However, as presented in Fig. 9, the application amount of N and K fertilizer per hectare showed a significantly decreasing trend, while the P fertilizer gave a non-significantly decreasing trend during the study period. This is inconsistent with the temporal trend of available N and K in the study area (Fig. 2). It indicates an increasing utilization efficiency of N and K fertilizer in Jiangxi Province. It is also revealed that the soil nutrient and SOM variation are also affected by some other factors, such as soil pH and the microbial community, as well as application strategy of chemical fertilizer (Zhalnina et al. 2015). Unreasonable fertilization and over-fertilization can lead to adverse effects such as an imbalance of soil nutrients or even a decline of soil fertility (Fulford et al. 2018). Therefore, more attention should be paid to increasing the utilization efficiency of chemical fertilizers instead of overuse of chemical fertilizers, hence reducing the related environmental risks and economic costs of the landowner (Zhao et al. 2014). Specifically, chemical fertilizers combined with straw return can improve the availability of the nutrient, which reduces the input of chemical fertilizers and thus plays a positive role in maintaining nutrient balance (Wang et al. 2022).

Crop rotation is another major factor, which determines the variation of soil nutrients and SOM (Figs. 8 and 9). It can maintain soil fertility, improve soil physicochemical properties, increase soil microbial diversity, and affect various natural processes such as N-enrichment of the soil by leguminous plants (Malobane et al. 2020; Town et al. 2022). Using different species in rotation allows for increased SOM and adds nutrients to the soil (Malobane et al. 2020; Hu et al. 2022). Song et al. (2016) reported an increased soil organic carbon concentration in the rice–wheat rotation system compared to the conventional plow system. Haruna and Nkongolo (2019) revealed that no-till management and corn-soybean rotation significantly improve total carbon compared with continuous corn and soybean production. Malobane et al. (2020) showed that crop rotation increased soil N by 6.0% in marginal soils of South Africa. Town et al. (2022) found that different crop rotations affect the bacterial and fungal communities in the root, rhizosphere, and bulk soil, and impact soil microbial processes, which then essentially affect the availability of soil nutrients and crop yield. However, further work is needed to explore and compare the effects of various crop rotations on soil nutrients, SOM, and crop yield.

4.4 Effects of soil properties on soil nutrients and SOM

Compared with climate factors and soil management measures, the soil properties like soil available silicon, exchangeable Ca, and pH had slight but significant effects on the variation of soil nutrients and SOM (Figs. 8 and 9). As shown in Fig. 10, the soil available Si was significantly and positively correlated with available K. Meanwhile, it was significantly and negatively correlated with SOM and available N and P, which is consistent with the results of Yanai et al. (2016). Silicon is the second most common element in the Earth’s crust, and it has been recognized as beneficial for crop production by alleviating various biotic and abiotic stresses (Skinner 1979; Wang et al. 2020). Silicon could modify the uptake and acquisition of nutrients differently for different plant species (Islam and Saha 1969; Greger et al. 2018). According to Wallace (1989) and Ma et al. (1990), the uptake of N, P, K, Mg, and Ca by the plants is influenced by silicon in different ways. De Tombeur et al. (2020) also confirmed that the SOM and soil available Si are closely related to phytolith concentration, releasing silicon in soil solution through dissolution. 

Fig. 10
figure 10

The correlation between soil nutrients and soil properties. SOM, soil organic matter; AN, alkali-hydrolyzable nitrogen; AP, available P; AK, available potassium; MAP, mean annual precipitation; MAT, mean annual temperature; ASi, available silicon; ECa, exchangeable calcium; EMg, exchangeable magnesium; CEC means cation exchange capacity

Full size image

The exchangeable Ca is significantly and positively correlated with SOM and available N and K. In contrast, it was significantly but negatively linked to available P. High concentrations of exchangeable calcium will hinder the uptake of soil nutrients by crops since the cation Ca2+ can directly compete with other soil nutrients on adsorption sites (Mei et al. 2016).

Moreover, the pH correlated significantly and positively with available K while negatively with SOM and available N, which is consistent with previous studies, such as Zhao et al. (2011) and Qian et al. (2015). This may attribute to the fact that the microbial decomposition and transformation of straw returning to the field could increase the concentration of SOM. It can also lead to the increase of cellulose, lignin, polysaccharide, and humic acid, which may decrease soil pH. The mean value of soil pH in Jiangxi Province was 5.2 and belongs to the acidic grade (Table S6). Moreover, Guo et al. (2018) reported that excessive fertilizer application and acid rain intensity led to an overall 0.6-unit decrease of soil pH over the farmland of Jiangxi Province between the 1980s and the 2010s. Soil pH significantly influences the concentration of various nutrients and SOM in the soil by affecting the availability of soil nutrients and microbial activity (Zhalnina et al. 2015; Chen et al. 2019). Wright et al. (2009) reported that soil pH strongly affected the availability of soil nutrients. For example, the increase of soil pH contributes to the rise of negative charge of the colloid and the change of the colloid-adsorbed chaperone ions, which leads to the improvement of K availability (Qian et al. 2015).

Additionally, we found no significant relationship between the cation exchange capacity and SOM (Table 1), which is consistent with the result reported by Petersen et al. (1996). However, some other researchers reported significant relationships between cation exchange capacity and SOM (Lourenco et al. 2021; Zhao et al. 2022). The relationship between cation exchange capacity and SOM may also be affected by many other factors, such as clay type, parental material, and soil texture (Jackson et al. 1986; Solly et al. 2020). This may lead to the various relations between cation exchange capacity and SOM reported by different studies. Thus, further investigation is still necessary to make this issue more clearly.

4.5 Effects of relief on soil nutrients and SOM

In this study, we found relief factors (e.g., elevation and Multi-resolution valley bottom flatness) had small but significant effects on soil nutrients and SOM. The relief factors could indirectly affect the concentrations of soil nutrients and SOM through redistribution of water and radiation in soil, and migration of soil nutrients and SOM (Fiedler et al. 2004; Moser et al. 2009). For example, in soils with high moisture, irons play a critical role in P adsorption, retention, and release (Aldous et al. 2005), while Al can likewise affect P availability. Bruland and Richardson (2005) confirmed that the decreasing inorganic N and total P concentrations leads to the micro-topographic gradient from higher to lower elevations. Clemens et al. (2010) found that the middle and lower slope positions are the most prone to erosion and long-term intensive tillage led to poor soil fertility in this position. Furthermore, the topographical factors can also affect delineating of agricultural management zones and the selection of crop planting systems (Pilesjo et al. 2005); hence affect the soil nutrient recycling.

4.6 Limitations and recommendations

Although we made some progress in this study, there are still several limitations, which need to be overcome in further work. Firstly, in this study, different number of soil samples is taken in different years. The soil sampling locations also varied in different years, which may pose bias on the spatio-temporal variation prediction. Secondly, some studies have reported apparent intra-annual variations in soil nutrients and SOM (Conyers et al. 1997; Winterdahl et al. 2014). For example, Conyers et al. (1997) revealed a complex and fluctuating intra-annual temporal trend of soil water content, pH, and exchangeable aluminum. The soil samples in this study were collected in different months over several years. Nevertheless, due to limited data availability, we can only explore the inter-annual variability of soil nutrients and SOM, and hard to explore the intra-annual variation of nutrients and SOM. In addition, soil nutrients and SOM are affected by many factors, and the interactions between these factors are not fully understood (Hu et al. 2022). Moreover, in this study, we analyzed the spatial–temporal variation of soil nutrients and SOM by separately producing maps of soil nutrients and SOM in different years using the geostatistical method. This way neglects the temporal dependence on soil nutrients and SOM. Finally, some information, such as the amount of straw return and input of chemical fertilizers, is provided by the landowners, which may lead to bias on the data due to record errors, although many measures have been taken to ensure the reliability and quality of the dataset.

Nevertheless, our results could still provide important and valuable implications for soil management and sustainable agriculture development. First of all, our study confirmed the practical and vital role of straw in enhancing soil fertility and soil carbon sequestration in the farmland. This is crucial for archiving sustainable agriculture, carbon neutralization strategy, and regulating climate changes, and could also contribute to realizing sustainable development goals adopted by the United Nations (Huang et al. 2020; Liu et al. 2021a, b). Secondly, our results indicate that the soil nutrients and SOM in the farmland of Jiangxi Province are at a high grade. As presented in Figs. 2 and 9, reduced use of chemical fertilizers did not lead to a reduction in availability of nutrients and SOM. This enlightens us that environmentally friendly methods like straw return, coupled with appropriate agricultural management measures, can fill the gaps between the supply of soil nutrients and chemical fertilizer input. Finally, the nutrients and SOM usually showed clear intra-annual variation during different growth stages of crops (Conyers et al. 1997; Xiang et al. 2022); thus, more attention should be paid to monitor the intra-annual variation of soil nutrients and SOM. When the study is focused on analyzing the inter-annual variability of soil nutrients and SOM, the sampling time should be as consistent as possible to reduce the bias caused by intra-annual variation of soil nutrients and SOM.

5 Conclusion

This study revealed that SOM and available P showed an insignificant decreasing trend, while the available N and K showed a significant increasing trend in the farmland soils across Jiangxi Province during the study period. The soil nutrients and SOM showed weak to moderate spatial dependence in the farmland. In addition, clear spatio-temporal variability was detected for soil nutrients and SOM. Referring to the national standard of China, the SOM and available N and P in most soil samples are kept at a relatively high level. In contrast, the concentration of available K in most soil samples is kept at a relatively low level. This highlights the urgent need for K fertilizer in the farmland. Finally, climate and soil management factors have dominant impacts on soil nutrient variation. The straw return is widely applied in farmland due to its benefit on improving soil fertility and enhancing carbon sequestration.