Next Article in Journal
Expressway Vehicle Trajectory Prediction Considering Historical Path Dependencies
Previous Article in Journal
Assessing the Belt and Road Initiative’s Impact: A Multi-Regression Model Based on Economic Interaction
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Sustainability
Volume 16
Issue 11
10.3390/su16114695
sustainability-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles thumb_up
...
Endorse textsms
...
Comment
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Modelling Soil Ammonium Nitrogen, Nitrate Nitrogen and Available Phosphorus Using Normalized Difference Vegetation Index and Climate Data in Xizang’s Grasslands

by
Wei Sun
1,
Huxiao Qi
1,
Tianyu Li
1,
Yong Qin
1,
Gang Fu
1,*,
Fusong Han
2,*,
Shaohua Wang
3 and
Xu Pan
4
1
Lhasa Plateau Ecosystem Research Station, Key Laboratory of Ecosystem Network Observation and Modeling, Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing 100101, China
2
College of Urban and Environmental Sciences, Hunan University of Technology, Zhuzhou 412007, China
3
Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing 100094, China
4
Wetland Research Center, Institute of Ecological Conservation and Restoration, Chinese Academy of Forestry, No. 2 Dong Xiaofu, Haidian District, Beijing 100091, China
*
Authors to whom correspondence should be addressed.
Sustainability 2024, 16(11), 4695; https://doi.org/10.3390/su16114695
Submission received: 29 March 2024 / Revised: 9 May 2024 / Accepted: 27 May 2024 / Published: 31 May 2024
Browse Figures
Versions Notes

Abstract

:
There is still a lack of high-precision and large-scale soil ammonium nitrogen (NH4+-N), nitrate nitrogen (NO3−-N) and available phosphorus (AP) in alpine grasslands at least on the Qinghai–Xizang Plateau, which may limit our understanding of the sustainability of alpine grassland ecosystems (e.g., changes in soil NH4+-N, NO3−-N and AP can affect the sustainability of grassland productivity, which in turn may alter the sustainability of livestock development), given that nitrogen and phosphorus are important limiting factors in alpine regions. The construction of big data mining models is the key to solving the problem mentioned above. Therefore, observed soil NH4+-N, NO3−-N and AP at 0–10 cm and 10–20 cm, climate data (air temperature, precipitation and radiation) and/or normalized vegetation index (NDVI) data were used to model NH4+-N, NO3−-N and AP in alpine grasslands of Xizang under fencing and grazing conditions. Nine algorithms, including random forest algorithm (RFA), generalized boosted regression algorithm (GBRA), multiple linear regression algorithm (MLRA), support vector machine algorithm (SVMA), recursive regression tree algorithm (RRTA), artificial neural network algorithm (ANNA), generalized linear regression algorithm (GLMA), conditional inference tree algorithm (CITA), and eXtreme gradient boosting algorithm (eXGBA), were used. The RFA had the best performance among the nine algorithms. Climate data based on the RFA can explain 78–92% variation of NH4+-N, NO3−-N and AP under fencing conditions. Climate data and NDVI together can explain 83–93% variation of NH4+-N, NO3−-N and AP under grazing conditions based on the RFA. The absolute values of relative bias, linear slopes, R2 and RMSE values between simulated soil NH4+-N, NO3−-N and AP based on RFA were ≤8.65%, ≥0.90, ≥0.91 and ≤3.37 mg kg−1, respectively. Therefore, random forest algorithm can be used to model soil available nitrogen and phosphorus based on observed climate data and/or normalized difference vegetation index in Xizang’s grasslands. The random forest models constructed in this study can be used to obtain a long-term (e.g., 2000–2020) raster dataset of soil available nitrogen and phosphorus in alpine grasslands on the whole Qinghai–Tibet Plateau. The raster dataset can explain changes in grassland productivity from the perspective of nitrogen and phosphorus constraints across the Tibetan grasslands, which can provide an important basis for the sustainable development of grassland ecosystem itself and animal husbandry on the Tibetan Plateau.

1. Introduction

Nitrogen and phosphorus are pivotal limiting factors in global alpine ecosystems, serving as crucial nutrient sources for plant growth and development. However, not all forms of nitrogen and phosphorus are readily utilized by plants [1]. Soil ammonium nitrogen (NH4+-N), nitrate nitrogen (NO3−-N) and available phosphorus (AP) are specific nitrogen and phosphorus nutrients directly usable by plants. Their concentrations play a vital role in maintaining the stability of global alpine ecosystems. While actual measurement is undoubtedly the most accurate method, it is accompanied by drawbacks such as time and energy consumption. Model simulation emerges as a valuable approach for obtaining large spatial scale and long time series for soil NH4+-N, NO3−-N and AP. Based on these mentioned above, some model simulation studies on soil NH4+-N, NO3−-N and/or AP have been carried out by predecessors [2,3,4,5,6,7,8]. The integration of machine learning algorithms into soil AP and other soil variables has gained traction [9,10]. These scientific endeavors establish a critical foundation for exploring farmland soil nitrogen and phosphorus management and non-farmland soil nitrogen and phosphorus cycling mechanisms, particularly against the backdrop of a growing global population and various environmental challenges. Despite these advancements, certain uncertainties persist in the current study. Geostatistical methods generally obtain soil NH4+-N, NO3−-N and/or AP on a large spatial scale within the same period but often face challenges in simulating data across time scales limiting the acquisition of longer time series data [2,11]. Although models like HYDRUS-2D are relatively mature models, they generally require not only localization of model parameters before use but also have relatively more model parameters [8]. Additionally, the ongoing debate on which machine learning algorithm performs better in quantifying soil NH4+-N, NO3−-N and/or AP adds another layer of uncertainty [12,13]. Hence, further research is imperative to address these existing gaps.
Alpine grassland ecosystems constitute the primary vegetation cover on the Qinghai–Xizang Plateau and serve as representative examples of global alpine ecosystems [14,15,16]. Numerous studies have been conducted on the relationships between soil NH4+-N, NO3−-N and/or AP and environmental factors in alpine grassland regions of the Qinghai–Xizang Plateau [17,18]. These studies offer crucial scientific and technological insight into the mechanisms underlying changes in available nitrogen and phosphorus nutrition in alpine grasslands of global transformations, as well as their impacts on the structure and function of alpine grassland ecosystems. However, the current research presents three uncertainties that warrant further investigation. First, many studies have quantified the impacts of soil temperature and precipitation, influenced by technical approaches such as warming and precipitation change experiments, on soil available nitrogen and phosphorus. These studies have explored the feedback effects on grassland ecosystem productivity and forage nutrition quality [19,20,21]. Radiation, as a significant climate variable, has been demonstrated to markedly influence the nutritional quality of herbage, herbage stock, plant diversity, soil moisture, soil pH, soil carbon, nitrogen and phosphorus in the Qinghai–Xizang grassland ecosystems [22,23,24,25,26]. The combined influence of temperature, precipitation and radiation appears to explain the productivity and diversity of alpine grassland more than the individual factors of temperature and precipitation [27,28]. Although it remains unconfirmed whether radiation is directly linked to soil available nitrogen and phosphorus, soil available nitrogen and phosphorus are correlated to some extent with forage nutrient quality and nutrient pool, plant diversity, soil carbon, nitrogen and phosphorus [18,29], which seems to indicate that the combined air temperature, precipitation and radiation may also explain the variation of soil available nitrogen and phosphorus in alpine grasslands on the Qinghai–Xizang Plateau. Further research is required due to the lack of relevant studies. Second, several studies indicate that surface spectral information can be leveraged to simulate soil-available nitrogen and phosphorus [6,13], and site-scale investigations on the Qinghai–Xizang Plateau show a significant correlation between normalized vegetation index (NDVI) and soil available nitrogen and phosphorus [30]. However, the extent to which the climate trivariate (temperature, precipitation and radiation) and NDVI can collectively explain variations in soil available nitrogen and phosphorus in alpine grasslands of the Qinghai–Xizang Plateau, compared to the climate trivariate alone, remains unknown due to a lack of relevant studies. Third, while previous research confirms that random forest algorithm is more accurate than other algorithms in quantifying pasture nutrient quality and nutrient pool, plant diversity, soil moisture, soil pH, soil carbon, nitrogen and phosphorus in alpine grasslands of the Qinghai–Xizang Plateau [22,23,24,25,26], it is uncertain whether the random forest algorithm also exhibits superior performance compared to other models in quantifying variations in soil-available nitrogen and phosphorus in alpine grasslands of the Qinghai–Xizang Plateau.
In this study, our aim was to construct machine learning models of soil NH4+-N, NO3−-N and AP in grasslands of the Qinghai–Xizang Plateau, with the objective of facilitating the quantification of the spatio-temporal distribution patterns of soil available nitrogen and phosphorus. The primary goal was to identify the most effective algorithms of soil NH4+-N, NO3−-N and AP at 0–10 and 10–20 cm under fencing and free-grazing conditions in Xizang’s grasslands. The nine algorithms employed in this study were the RFA, generalized boosted regression algorithm (GBRA), support vector machine algorithm (SVA), multiple linear regression algorithm (MLRA), recursive regression tree algorithm (RRTA), artificial neural network algorithm (ANNA), generalized linear regression algorithm (GLRA), conditional inference tree algorithm (CITA) and eXtreme gradient boosting algorithm (eXGBA). According to some previous studies [22,23,24,25,26], we hypothesized that the performances of RFA were stronger than the other eight algorithms in estimating soil NH4+-N, NO3−-N and AP.

2. Materials and Methods

2.1. Study Area

The study area is located at Xizang’s grassland regions and the locations of all sampling points are shown in Figure S1. The elevation ranged from 2924 m to 5330 m and from 4279 m to 5261 m under free-grazing conditions and fencing conditions, respectively (Figure S1). The longitude and latitude were 79.69–95.68° E and 28.37–33.17° N, and 83.25–92.01° E and 29.28–33.17° N under free-grazing conditions and fencing conditions, respectively (Figure S1).

2.2. Soil Sampling and Analyses

Under grazing conditions, soils at 0–10 cm were collected in 2013, 2015–2016 and 2018–2020, and soils at 10–20 cm were collected in 2013, 2015 and 2019–2020. Under fencing conditions, soils at 0–10 cm were collected in 2013–2020, and soils at 10–20 cm were collected in 2013, 2015 and 2020.
The ammonium bicarbonate extraction molybdenum antimony resistance colorimetric method was used to measure AP [31,32]. For each soil sample, 20 g of soil fresh soils were extracted with 100 mL K2SO4 (2 mol L−1) and filtered with 0.45 μm filter membrane. Then, the filtered liquid was used to measure NH4+-N and NO3−-N on a LACHAT Quikchem Automated Ion Analyzer [31,32]. There were 280, 270 and 290 and 240, 240 and 257 soil NH4+-N (mg kg−1), NO3−-N (mg kg−1) and AP (mg kg−1) data at 0–10 cm and 10–20 cm under free-grazing conditions, respectively (Tables S1–S6). There were 250, 270 and 276 and 170, 190 and 190 soil NH4+-N (mg kg−1), NO3−-N (mg kg−1) and AP (mg kg−1) data at 0–10 cm and 10–20 cm under fencing conditions, respectively (Tables S1–S6). Soil NH4+-N was 0.06–97.15 mg kg−1 and 0.09–39.37 mg kg−1 at 0–10 cm and 10–20 cm under free-grazing conditions. Soil NO3−-N was 0.12–82.28 mg kg−1 and 0.02–77.58 mg kg−1 at 0–10 cm and 10–20 cm under free-grazing conditions. Soil AP was 0.98–38.02 mg kg−1 and 2.18–38.34 mg kg−1 at 0–10 cm and 10–20 cm under free-grazing conditions. Soil NH4+-N was 0.09–53.12 mg kg−1 and 0.12–33.14 mg kg−1 at 0–10 cm and 10–20 cm under fencing conditions. Soil NO3−-N was 0.24–53.92 mg kg−1 and 0.11–47.06 mg kg−1 at 0–10 cm and 10–20 cm under fencing conditions. Soil AP was 1.71–39.23 mg kg−1 and 1.80–40.32 mg kg−1 at 0–10 cm and 10–20 cm under fencing conditions.

2.3. Statistical Analyses

The sources and accuracy of climate data are detailed in previous studies [25]. Monthly air temperature, precipitation and radiation data were used to estimate soil NH4+-N, NO3−-N and AP under fencing conditions. In contrast, monthly air temperature, precipitation, radiation and normalized difference vegetation index data were used to estimate soil NH4+-N, NO3−-N and AP under free-grazing conditions. According to previous studies [24,25], the sample function of R 4.2.2 software was used to separate the original data (soil NH4+-N, NO3−-N or AP, and their corresponding independent variables) into two groups. One group (n = 30) was used for model accuracy test, and the rest was used for model construction. The randomForest function of the randomForest package can provide random forest algorithm, which was used for the construction of random forest models of soil NH4+-N, NO3−-N and AP. The randomForest function can output R2 and mse values. Three steps were carried out to obtain the optimal random forest model. In the first step, 100 cycles were performed to obtain the ntree dataset, and the ntree parameter was set to 1000 in each cycle. The mtry parameter was set to 1–3 (potential NH4+-N, NO3−-N or AP) or 1–4 (actual NH4+-N, NO3−-N or AP) for each cycle in both the first and second steps. In the second step, we also carried out 100 cycles of randomForest function. In each cycle, the value between the mean and median (the mean and median values also included) of the ntree dataset obtained from the first step was entered as the input value of the ntree parameter. The ntree and mtry parameters of the random forest model with the maximum R2 value and minimum mse were treated as the optimal combination of ntree and mtry. In the third step, we performed a while loop, and the ntree and mtry parameters were set to be the optimal ntree and mtry obtained in the second step. The qualification was when the model R2 was greater than the R2 of the random forest model with the optimal combination of ntree and mtry obtained in the second step, or mse was less than the mse of the random forest model with the optimal combination of ntree and mtry obtained in the second step. Generalized boosted regression models of soil NH4+-N, NO3−-N and AP were constructed in the two steps. In the first step, we carried out 100 loops trying to find the optimal n.trees parameter. Each loop was composed of two processes, including gbm and gbm.perf processes, which were carried out using the gbm and gbm.perf functions of the gbm package, respectively. The parameters of distribution, cv.folds and n.trees of the gbm function were set to ‘gaussian’, 2 and 1000 in the gbm process, respectively. The parameter of the method of the gbm.perf was set to be ‘cv’ in the gbm.perf process. Each loop can yield a n.trees value, indicating that the first step ends up with a total of 100 n.trees values. In the second step, we carried out one gbm operation, in which parameters of distribution, cv.folds and n.trees were set to ‘gaussian’, 2 and the median value of the 100 n.trees obtained in the first step, respectively. Recursive regression tree models of soil NH4+-N, NO3−-N and AP were constructed based on the rpart function of the rpart package, and the parameter of method and cv was set to be ‘anova’ and 0.01, respectively. The fit function of the rminer package was used to develop artificial neural network, generalized linear regression, conditional inference tree and eXtreme gradient boosting models of soil NH4+-N, NO3−-N and AP, and the parameter of the model was set to ‘mlp’, ‘cv.glmnet’, ‘ctree’ and ‘xgboost’, respectively. The parameter of task, search and scale of the fit function of the rminer package was set to be ‘reg’, ‘heuristic5’ and ‘none’, respectively. The svm function of the e1071 package was used to obtain support vector machine models of soil NH4+-N, NO3−-N and AP. The lm function of the stats package was used to construct multiple linear regression models of soil NH4+-N, NO3−-N and AP. For SVA and MLRA, all the parameters were set to be default values. Linear slope and R2 value, relative bias and root-mean-square error (RMSE) were treated as the model accuracy evaluation indices. Packages of ggpubr and ggplot2 were used to plot [23]. All statistical analyses were performed on the software of R.4.2.2.

3. Results

3.1. Evaluation of Soil Nutrient Modeling Algorithms

The RFA, MLRA and RRTA but not the other six algorithms provided the R2 values during model building (Tables S1–S3). Among the three algorithms, the R2 values of the RFA (>0.77) were always the largest, but those of the MLRA (<0.55) were mostly the smallest (Tables S1–S3). In addition, the explanatory power of the MLRA exceeded 50% for less than half of the cases (Table S2). In other words, the RFA with air temperature, precipitation and radiation as independent variables can explain the most variations of soil NH4+-N, NO3−-N and AP under fencing conditions. The RFA with air temperature, precipitation radiation and NDVI as independent variables can explain the most variations of soil NH4+-N, NO3−-N and AP under free-grazing conditions. The tree numbers of the RFA and GBRA were obviously greater than the support vector numbers of the SVMA, but the first two were almost equal (Tables S1, S4–S5). The ANN, GLR, CIT and eXGB were all dependent on the rminer package and provided the error parameters (Table S6). These four algorithms had different model-building capabilities because their model-building errors were different (Table S6). In addition, only from these error values, it was not entirely possible to judge which of these four algorithms was the best (Table S6).

3.2. Soil Nutrient Modeling Accuracy under Different Conditions

The absolute values of the relative bias between modeled soil NH4+-N, NO3−-N and AP based on the eXGBA and observed soil NH4+-N, NO3−-N and AP were the largest among the nine algorithms (Table 1). In addition, the absolute values of the relative bias between modeled soil NH4+-N, NO3−-N or AP based on the GBRA, MLRA, SVMA, RRTA, ANNA, GLRA or CITA and observed soil NH4+-N, NO3−-N or AP were >10% for some cases (Table 1). In contrast, the absolute values of the relative bias between modeled soil NH4+-N, NO3−-N or AP based on the RFA and observed soil NH4+-N, NO3−-N or AP were <9% (Table 1).
For most cases, the RMSE values (<3.40 mg kg−1) between modeled soil NH4+-N, NO3−-N or AP based on the RFA and observed soil NH4+-N, NO3−-N or AP were the lowest among the nine algorithms (Table 2). For at least 5 out of the 12 cases, the RMSE values between modeled soil NH4+-N, NO3−-N or AP from the MLRA, SVMA, RRTA, ANNA, GLRA, CITA or eXGBA and observed soil NH4+-N, NO3−-N or AP were greater than 5.00 mg kg−1 (Table 2). In addition, the RMSE values between modeled soil NH4+-N, NO3−-N or AP based on the GBRA and observed NH4+-N, NO3−-N or AP were >3.50 mg kg−1 for some cases (Table 2).
The linear slopes and R2 values between the modeled soil NH4+-N at 0–10 cm based on the RFA and observed soil NH4+-N at 0–10 cm were the greatest among the nine algorithms under fencing or grazing conditions, respectively (Figure 1 and Figure 2, Table 3). The linear slopes and R2 values between the modeled soil NH4+-N at 10–20 cm based on the MLRA, SVMA, RRTA, ANNA, GLRA or CITA, and observed soil NH4+-N at 10–20 cm were <0.80 and <0.85, and <0.88 and <0.78 under the fencing and grazing conditions, respectively (Figure 1 and Figure 2, Table 3). In addition, the linear slope and R2 values between the modeled soil NH4+-N at 10–20 cm based on the RFA, GBRA or eXGBA and observed soil NH4+-N at 10–20 cm were 0.96 and 0.93, 0.99 and 0.91 and 0.50 and 0.92 under the fencing condition, respectively (Figure 1, Table 3). The linear slope and R2 values between the modeled soil NH4+-N at 10–20 cm based on the RFA, GBRA or eXGBA and observed soil NH4+-N at 10–20 cm were 1.00 and 0.96, 1.00 and 0.94 and 0.50 and 0.94 under the fencing condition, respectively (Figure 2, Table 3). The linear slopes and R2 values between the modeled soil NO3−-N at 0–10 cm based on the MLRA, SVMA, RRTA, ANNA, GLRA or CITA and observed soil NO3−-N at 0–10 cm were <0.63 and <0.90 under the fencing conditions, respectively (Figure 3, Table 3). In addition, the linear slope and R2 values between the modeled soil NO3−-N at 0–10 cm based on the RFA, GBRA or eXGBA and observed soil NO3−-N at 0–10 cm were 0.97 and 0.97, 0.87 and 0.96 and 0.48 and 0.97 under the fencing condition, respectively (Figure 3, Table 3). The linear slope and R2 values between the modeled soil NO3−-N at 10–20 cm based on the RFA, GBRA or eXGBA and observed soil NO3−-N at 10–20 cm were 0.92 and 0.95, 0.90 and 0.96 and 0.43 and 0.96 under the fencing condition, respectively (Figure 3, Table 3). The linear slopes and R2 values between the modeled soil NO3−-N at 0–10 cm and 10–20 cm based on the RFA and observed soil NO3−-N at 0–10 cm and 10–20 cm were the greatest among the nine algorithms under grazing conditions, respectively (Figure 4, Table 3). The linear slopes and R2 values between the modeled soil AP at 0–10 cm and 10–20 cm based on the RFA and observed soil AP at 0–10 cm and 10–20 cm were also the greatest among the nine algorithms under fencing conditions, respectively (Figure 5 and Figure 6, Table 3).

4. Discussion

4.1. Random Forest Models with the Highest Accuracy

Aligned with the hypothesis, random forest algorithm exhibited superior performance in estimating soil NH4+-N, NO3−-N and AP using climate data and/or normalized difference vegetation index among all the nine algorithms. This observation is consistent with prior studies that highlighted the enhanced efficacy of the random forest algorithm in quantifying soil moisture, soil pH, plant species α-diversity, forage nutrition quality and quantity using climate data and normalized difference vegetation index than the other algorithms in Xizang’s grasslands [22,23,24,25]. Moreover, this finding further strengthened the advantages and prospects of random forest in quantifying key indicators of alpine grassland ecosystems on the Qinghai–Xizang Plateau.

4.2. The Accuracy of the Models in This Study versus That of Previous Studies

The accuracies of the random forest models of soil NH4+-N, NO3−-N and AP established in this study were at least comparable to the results of previous studies [4,7,9,12,13,33]. For example, the simulated values of soil NH4+-N and NO3−-N obtained by the HYDRUS-2D model can explain 34–99% and 14–98% variation of observed soil NO3−-N and NH4+-N, respectively [8]. The RMSE values between simulated AP obtained by geostatistics technique and observed AP were ≥10.20 mg kg−1 [34] and ≥6.98 mg kg−1 [35]. Simulated AP based on the transfer learning prediction algorithm can only explain ≤79% variation of observed AP [36]. Simulated AP based on the visible–near-infrared spectroscopy can only explain 69–82% variation of observed AP, and their RMSE values were ≥3.79 mg kg−1 [6]. Simulated AP based on three machine learning algorithms (random forest, boosting tree, gradient boosting tree) can explain 82–86% variation of observed soil AP [10]. Simulated AP based on the HJ-1A HSI can explain 67% variation of observed soil AP, and their RMSE was 15.31 mg kg−1 [37].

4.3. Effect of Soil Depth on the Interpretation of Soil Nutrient Variation by Climate Data and/or Normalized Difference Vegetation Index

The explanatory power of climate data and/or normalized difference vegetation index regarding the variations in soil NH4+-N, NO3−-N and AP were associated with soil depth, and it did not necessarily decrease with the increasing soil depth (Tables S1–S3). Moreover, the accuracies of soil NH4+-N, NO3−-N and AP were also related to soil depth (Table 1, Table 2 and Table 3, Figure 1, Figure 2, Figure 3, Figure 4, Figure 5 and Figure 6). These findings align with similar observations from previous studies conducted both on the Qinghai–Xizang Plateau [24,25] and outside the Qinghai–Xizang Plateau [8]. Several factors may contribute to this phenomenon. Firstly, the sample size can influence the relationships between soil NH4+-N, NO3−-N and AP and temperature and precipitation [19,38]. In accordance with previous studies [24,25], the sample size decreased with the increasing soil depth, and identical variables at different depths were independently divided into model construction and accuracy test datasets. Secondly, varying ranges of soil NH4+-N, NO3−-N and AP may result in different relationships with environmental factors [21,32]. Consistent with earlier studies [39,40,41,42], the value ranges of soil NH4+-N, NO3−-N and AP were dependent on soil depth in this study. Thirdly, soil moisture and temperature may impact soil NH4+-N, NO3−-N and AP, and both factors can vary with soil depth [25,43,44,45,46]. Moreover, the impact of climate change on soil moisture differs with soil depth. Fourthly, soil pH may be linked to the balance between soil NH4+-N and NO3−-N [47,48,49]. Both soil pH and the influence of climate change on soil pH vary with soil depth [50]. Lastly, normalized difference vegetation index captures land surface information [49], and air temperature observation generally pertains to the air temperature at a height of 2 m. Therefore, in theory, soils closer to the surface should be more closely associated with the normalized difference vegetation index and air temperature. Precipitation needs to be transported from the surface of soils to deeper soils through infiltration. Solar radiation initially reaches the land surface and then penetrates deeper soils through thermal conduction. Hence, the relationships between precipitation and radiation and soil NH4+-N, NO3−-N and AP should be also related to soil depth.

4.4. Effect of Combined Climate Data and Normalized Difference Vegetation Index on Soil Nutrient Variation

The combined use of climate data and normalized difference vegetation index may offer a more comprehensive explanation for variations in soil NH4+-N, NO3−-N and AP than climate data alone (Table S1). This discovery aligns with findings from prior studies, illustrating that the inclusion of both climate data and normalized difference vegetation index can better account for variations in soil moisture, forage nutrition quality and quantity compared to relying solely on climate data [22,25]. Conversely, a different study suggested that the joint consideration of climate data and normalized difference vegetation index may be less effective in explaining variations in soil pH compared to using climate data alone [24]. In addition, the effectiveness of climate data combined with normalized difference vegetation index in explaining plant species α-diversity depends on the specific indices used, such as Shannon or Pielou [23]. Therefore, more independent variables do not necessarily improve the close relationships between dependent variables and independent variables. Moreover, similar to some previous studies [22,23,24,25], models of soil NH4+-N, NO3−-N and AP with more independent variables also did not always lead to higher model accuracy. Instead of focusing on expanding the number of model parameters, optimizing model screening processes may offer a more effective approach to enhance the accuracy of dependent variables.

4.5. Uncertainty Analyses

Previous studies have suggested that there are some uncertainties when using climate variables and NDVI to model plant and soil variables in terrestrial ecosystems. It included the accuracy of NDVI itself (such as easy saturation), the accuracy of climate variables (such as the error caused by spatial interpolation), the spatial–temporal matching error between dependent variables and independent variables, the acquisition time of dependent variables and topographic factors (such as slope) [25,26]. In this study, the random forest models of soil NH4+-N, NO3−-N and AP at only 0–10 cm and 10–20 cm were constructed, and soil depth can alter the model accuracies of these soil variables. Therefore, whether climate data and NDVI can simulate soil NH4+-N, NO3−-N and AP at the deeper soil layers (>20 cm) needs to be further studied, and whether the performance of random forest model is still better than the other models for soil NH4+-N, NO3−-N and AP at the deeper soil layers needs to be further studied. The random forest models constructed in this study were based on the observational data of grasslands in Xizang, so the accuracy of these models extruded to other regions of the Qinghai–Xizang Plateau (i.e., outside the Tibet Autonomous Region) and other alpine regions in the world remains to be further studied. In addition to the contents of soil NH4+-N, NO3−-N and AP, the storages of soil NH4+-N, NO3−-N and AP are also very important soil variables, but this study did not model soil NH4+-N, NO3−-N and AP storages, so it is necessary to strengthen the research on these soil storages in the future.

5. Conclusions

This study should be the first study that modeled soil NH4+-N, NO3−-N and AP at 0–10 cm and 10–20 cm under both fencing and free-grazing conditions based on nine algorithms in Xizang’s grassland areas. The performances of random forest algorithm were better than the other algorithms. The absolute values of relative bias and RMSE values between simulated soil NH4+-N, NO3−-N and AP based on random forest algorithms were ≤8.65% and ≤3.37 mg kg−1, respectively. Simulated NH4+-N, NO3−-N and AP based on random forest algorithm explained 91–98% variations of observed NH4+-N, NO3−-N and AP. The linear slopes between simulated NH4+-N, NO3−-N and AP based on random forest algorithm and observed NH4+-N, NO3−-N and AP were 0.90–1.00. Therefore, the random forest models of soil NH4+-N, NO3−-N and AP established in this study can have high prediction accuracies, and can be used to build soil NH4+-N, NO3−-N and AP database in grassland areas over the Xizang Plateau. The random forest models constructed in this study can obtain long-term series dataset of soil available nitrogen and phosphorus in alpine grasslands on the entire Qinghai–Tibet Plateau, so it is convenient to discuss how soil available nitrogen and phosphorus limit the productivity of alpine grasslands on the entire Tibetan Plateau, thus probably affecting the sustainable development of animal husbandry and the sustainability of wildlife protection.

Supplementary Materials

The following supporting information can be downloaded at: https://www.mdpi.com/article/10.3390/su16114695/s1, Figure S1: Sampling sites; Table S1: Random forest algorithm parameters of soil NH4+-N, NO3−-N and AP; Table S2: Multiple linear regression parameters of soil NH4+-N, NO3−-N and AP; Table S3: Recursive regression tree parameters of soil NH4+-N, NO3−-N and AP; Table S4: Generalized boosted regression parameters of soil NH4+-N, NO3−-N and AP; Table S5: Support vector machine parameters of soil NH4+-N, NO3−-N and AP; Table S6: Artificial neural network (ANN), generalized linear regression (GLR), conditional inference tree (CIT) and eXtreme gradient boosting (eXGB) errors of soil NH4+-N, NO3−-N and AP.

Author Contributions

G.F.: conceptualization, investigation, writing—original draft preparation and writing—review and editing; W.S., H.Q., T.L., Y.Q., F.H., S.W. and X.P.: writing—review and editing. All authors contributed to the article. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by Chinese Academy of Sciences Youth Innovation Promotion Association [2020054], Science and Technology Department of Tibet Autonomous Region Base and Talent Program Project “Adaptation and Impact of traditional grazing on grassland ecosystem Uncertainty”, Lhasa Science and Technology Plan Project [LSKJ202422], China National Natural Science Foundation [31600432], Xizang Autonomous Region Science and Technology Project [XZ202301YD0012C; XZ202202YD0009C; XZ202201ZY0003N; XZ202101ZD0007G; XZ202101ZD0003N] and Construction of Zhongba County Fixed Observation and Experiment Station of First Support System for Agriculture Green Development.

Institutional Review Board Statement

Not applicable.

Informed Consent Statement

Not applicable.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon request.

Conflicts of Interest

The authors declare no conflicts of interest.

References

  1. Xu, X.; Ouyang, H.; Kuzyakov, Y.; Richter, A.; Wanek, W. Significance of organic nitrogen acquisition for dominant plant species in an alpine meadow on the Tibet plateau, China. Plant Soil 2006, 285, 221–231. [Google Scholar] [CrossRef]
  2. Xiao, Y.; Xie, G.; An, K. Comparison of interpolation methods for content of soil available phosphor. Chin. J. Eco-Agric. 2003, 11, 56–58. [Google Scholar]
  3. Xiao, Y.; Lu, X. Geostatistics-based spatial variability of soil nutrients in Fengqiu County of Henan Province. Bull. Soil Water Conserv. 2012, 32, 180–184. [Google Scholar]
  4. Ji, W.; Rossel, R.A.V.; Shi, Z. Accounting for the effects of water and the environment on proximally sensed vis-NIR soil spectra and their calibrations. Eur. J. Soil Sci. 2015, 66, 555–565. [Google Scholar] [CrossRef]
  5. Liu, H.; Cao, L.; Zeng, J. Spatial distribution of soil nitrogen in gully hillsides of Sejila Mountain, Southeastern Tibet. Acta Ecol. Sin. 2016, 36, 127–133. [Google Scholar]
  6. Fang, X.; Wang, W.; Jin, X.; Qi, H.; Li, S. Study on visible-near infrared spectroscopy for detection of available phosphorus in soil. Jiangsu J. Agric. Sci. 2019, 35, 1112–1118. [Google Scholar]
  7. Xu, X.; Ban, R.; Jin, Y.; Zhang, Y.; Gao, Y.; Xu, H. Nitrogen migration characteristics in farmland in plain river network area based on HYDRUS-1D model. J. Agro-Environ. Sci. 2022, 41, 400–410. [Google Scholar]
  8. Cui, H.; Zhou, Q.; Han, N.; Zhang, B. Simulating water and nitrogen dynamics in drip-irrigated soil based on the HYDRUS-2D model. J. Irrig. Drain. 2023, 42, 57–66. [Google Scholar]
  9. Qi, H.; Li, S.; Karnieli, A.; Jin, X.; Wang, W. Prediction method of soil available phosphorus using hyperspectral data based on PLS-BPNN. Trans. Chin. Soc. Agric. Mach. 2018, 49, 166–172. [Google Scholar]
  10. Jin, X.; Zhu, X.; Li, S.; Wang, W.; Qi, H. Predicting soil available phosphorus by hyperspectral regression method based on gradient boosting decision tree. Laser Optoelectron. Prog. 2019, 56, 131102. [Google Scholar] [CrossRef]
  11. Mou, X.; Sun, Z.; Liu, X. Spatial distribution patterns of nitrate nitrogen and ammonia nitrogen in typical tidal marsh soils of Yellow River Delta. Bull. Soil Water Conserv. 2012, 32, 256–261. [Google Scholar]
  12. Li, W.; Zhang, S.; Zhang, Q.; Dong, C.; Zhang, S. Rapid prediction of available N, P, and K content in soil using near-infrared reflectance spectroscopy. Trans. Chin. Soc. Agric. Eng. 2007, 23, 55–59. [Google Scholar]
  13. He, D.; Chen, X.; Ren, J.; Wang, M. Detection of soil available P content based on near-infrared spectra. Trans. Chin. Soc. Agric. Mach. 2015, 46, 152–157. [Google Scholar]
  14. Klein, J.A.; Harte, J.; Zhao, X.Q. Experimental warming, not grazing, decreases rangeland quality on the Tibetan Plateau. Ecol. Appl. 2007, 17, 541–557. [Google Scholar] [CrossRef] [PubMed]
  15. Wang, J.; Li, M.; Yu, C.; Fu, G. The change in environmental variables linked to climate change has a stronger effect on aboveground net primary productivity than does phenological change in alpine grasslands. Front. Plant Sci. 2022, 12, 798633. [Google Scholar] [CrossRef] [PubMed]
  16. Han, F.; Yu, C.; Fu, G. Non-growing/growing season non-uniform-warming increases precipitation use efficiency but reduces its temporal stability in an alpine meadow. Front. Plant Sci. 2023, 14, 1090204. [Google Scholar] [CrossRef] [PubMed]
  17. Ren, F.; Yang, X.X.; Zhou, H.K.; Zhu, W.Y.; Zhang, Z.H.; Chen, L.T.; Cao, G.M.; He, J.S. Contrasting effects of nitrogen and phosphorus addition on soil respiration in an alpine grassland on the Qinghai-Tibetan Plateau. Sci. Rep. 2016, 6, 34786. [Google Scholar] [CrossRef] [PubMed]
  18. Zha, X.J.; Tian, Y.; Ouzhu; Fu, G. Response of forage nutrient storages to grazing in alpine grasslands. Front. Plant Sci. 2022, 13, 991287. [Google Scholar] [CrossRef] [PubMed]
  19. Zhang, X.Z.; Shen, Z.X.; Fu, G. A meta-analysis of the effects of experimental warming on soil carbon and nitrogen dynamics on the Tibetan Plateau. Appl. Soil Ecol. 2015, 87, 32–38. [Google Scholar] [CrossRef]
  20. Zhang, K.P.; Shi, Y.; Jing, X.; He, J.S.; Sun, R.B.; Yang, Y.F.; Shade, A.; Chu, H.Y. Effects of short-term warming and altered precipitation on soil microbial communities in alpine grassland of the Tibetan Plateau. Front. Microbiol. 2016, 7, 1032. [Google Scholar] [CrossRef]
  21. Zhang, Y.; Dong, S.K.; Gao, Q.Z.; Liu, S.L.; Ganjurjav, H.; Wang, X.X.; Su, X.K.; Wu, X.Y. Soil bacterial and fungal diversity differently correlated with soil biochemistry in alpine grassland ecosystems in response to environmental changes. Sci. Rep. 2017, 7, 43077. [Google Scholar] [CrossRef] [PubMed]
  22. Han, F.; Fu, G.; Yu, C.; Wang, S. Modeling nutrition quality and storage of forage using climate data and normalized-difference vegetation index in alpine grasslands. Remote Sens. 2022, 14, 3410. [Google Scholar] [CrossRef]
  23. Tian, Y.; Fu, G. Quantifying plant species α-diversity using normalized difference vegetation index and climate data in alpine grasslands. Remote Sens. 2022, 14, 5007. [Google Scholar] [CrossRef]
  24. Dai, E.; Zhang, G.; Fu, G.; Zha, X. Can meteorological data and normalized difference vegetation index be used to quantify soil pH in grasslands? Front. Ecol. Evol. 2023, 11, 1206581. [Google Scholar] [CrossRef]
  25. Wang, S.; Fu, G. Modelling soil moisture using climate data and normalized difference vegetation index based on nine algorithms in alpine grasslands. Front. Environ. Sci. 2023, 11, 1130448. [Google Scholar] [CrossRef]
  26. Wang, S.; Qi, H.; Li, T.; Qin, Y.; Fu, G.; Pan, X.; Zha, X. Can normalized difference vegetation index and climate data be used to estimate soil carbon, nitrogen, phosphorus, and their ratios in Xizang‘s grasslands? Front. Earth Sci. 2024, 11, 1340020. [Google Scholar] [CrossRef]
  27. Huang, S.; Fu, G. Impacts of climate change and human activities on plant species α-diversity across the Tibetan grasslands. Remote Sens. 2023, 15, 2947. [Google Scholar] [CrossRef]
  28. Li, S.; Fu, G. Impacts of anthropogenic activities and climate change on forage nutrition storage in Tibetan grasslands. Plants 2023, 12, 2735. [Google Scholar] [CrossRef] [PubMed]
  29. Liu, J.L.; Zhan, W.; Huang, X.Y.; Tang, D.; Jin, S.F.; Zhu, D.; Chen, H. Nitrogen addition increases topsoil carbon stock in an alpine meadow of the Qinghai-Tibet Plateau. Sci. Total Environ. 2023, 888, 164071. [Google Scholar] [CrossRef]
  30. Fu, G.; Zhang, H.R.; Sun, W. Response of plant production to growing/non-growing season asymmetric warming in an alpine meadow of the Northern Tibetan Plateau. Sci. Total Environ. 2019, 650, 2666–2673. [Google Scholar] [CrossRef]
  31. Rui, Y.C.; Wang, Y.F.; Chen, C.R.; Zhou, X.Q.; Wang, S.P.; Xu, Z.H.; Duan, J.C.; Kang, X.M.; Lu, S.B.; Luo, C.Y. Warming and grazing increase mineralization of organic P in an alpine meadow ecosystem of Qinghai-Tibet Plateau, China. Plant Soil 2012, 357, 73–87. [Google Scholar] [CrossRef]
  32. Zong, N.; Shi, P.L. Enhanced community production rather than structure improvement under nitrogen and phosphorus addition in severely degraded alpine meadows. Sustainability 2019, 11, 2023. [Google Scholar] [CrossRef]
  33. Jia, S.-Y.; Yang, X.-L.; Li, G.; Zhang, J.-M. Quantitatively determination of available phosphorus and available potassium in soil by near infrared spectroscopy combining with recursive partial least squares. Spectrosc. Spectr. Anal. 2015, 35, 2516–2520. [Google Scholar]
  34. Liu, Z.; Chen, X.; Jing, Y.; Zhang, J.; Huang, Q.; Li, Q. A study on spatial variance of soil nutrients on the farmland scale in dryland of red soil based on geostatistics. J. Soil Sci. 2013, 44, 392–397. [Google Scholar]
  35. Wang, X.; Shen, Q.; Gao, F.; Zhang, X.; Zhang, S.; Wang, L. Comparing interpolation methods to predict the spatial distribution of soil available phosphorus in the Mollisol watershed of Northeast China. Res. Soil Water Conser. 2021, 28, 33–40. [Google Scholar]
  36. Zheng, W.; Li, S.; Han, Y.; Shi, S.; Zhu, X.; Jin, X. Study on transfer learning prediction methods for soil available phosphorus NIR. J. Instrum. Anal. 2020, 39, 1274–1281. [Google Scholar]
  37. Wang, L.; Gu, X.; Wang, B.; Wang, Y.; Chen, Y.; Gao, L. Study on county-wide cultivated soil available phosphorous content remote sensing mapping based on HJ-1A HSI. Chin. J. Soil Sci. 2015, 46, 1314–1320. [Google Scholar]
  38. Chen, Y.; Feng, J.; Yuan, X.; Zhu, B. Effects of warming on carbon and nitrogen cycling in alpine grassland ecosystems on the Tibetan Plateau: A meta-analysis. Geoderma 2020, 370, 114363. [Google Scholar] [CrossRef]
  39. Chen, X.-M.; Wu, H.-S.; Sun, J.-H. Time-spatial variability of ammonium and nitrate in farmland soil of Taihu Lake region. Huan Jing Ke Xue= Huanjing Kexue 2006, 27, 1217–1222. [Google Scholar]
  40. Sun, Z.-G.; Liu, J.-S.; Chen, X.-B. Spatial distribution patterns of nitrate nitrogen and ammonia nitrogen in typical Calamagrostis angustifolia wetland soils of Sanjiang Plain. Bull. Soil Water Conserv. 2009, 29, 66–72. [Google Scholar]
  41. Rui, Y.C.; Wang, S.P.; Xu, Z.H.; Wang, Y.F.; Chen, C.R.; Zhou, X.Q.; Kang, X.M.; Lu, S.B.; Hu, Y.G.; Lin, Q.Y.; et al. Warming and grazing affect soil labile carbon and nitrogen pools differently in an alpine meadow of the Qinghai-Tibet Plateau in China. J. Soil Sedim. 2011, 11, 903–914. [Google Scholar] [CrossRef]
  42. Zhang, Y.; Dong, S.K.; Gao, Q.Z.; Liu, S.L.; Zhou, H.K.; Ganjurjav, H.; Wang, X.X. Climate change and human activities altered the diversity and composition of soil microbial community in alpine grasslands of the Qinghai-Tibetan Plateau. Sci. Total Environ. 2016, 562, 353–363. [Google Scholar] [CrossRef] [PubMed]
  43. Sun, Z.; Liu, J. Study on vertical movement of nitrate nitrogen and ammonia nitrogen in typical Calamagrostis angustifolia wetland soils of Sanjiang Plain. J. Soil Water Conserv. 2007, 21, 25–30. [Google Scholar]
  44. Xu, Y.; Zhang, J.; Han, S.; Wang, S.; Wang, C.; Wang, S. Spatial heterogeneity of soil inorganic nitrogen in a broadleaved-Korean pine mixed forest in Changbai Mountains of Northeast China. J. Appl. Ecol. 2010, 21, 1627–1634. [Google Scholar]
  45. Bai, E.; Li, S.L.; Xu, W.H.; Li, W.; Dai, W.W.; Jiang, P. A meta-analysis of experimental warming effects on terrestrial nitrogen pools and dynamics. New Phytologist. 2013, 199, 441–451. [Google Scholar] [CrossRef] [PubMed]
  46. Xu, W.F.; Yuan, W.P.; Dong, W.J.; Xia, J.Z.; Liu, D.; Chen, Y. A meta-analysis of the response of soil moisture to experimental warming. Environ. Res. Lett. 2013, 8, 044027. [Google Scholar] [CrossRef]
  47. Barnard, R.; Le Roux, X.; Hungate, B.A.; Cleland, E.E.; Blankinship, J.C.; Barthes, L.; Leadley, P.W. Several components of global change alter nitrifying and denitrifying activities in an annual grassland. Funct. Ecol. 2006, 20, 557–564. [Google Scholar] [CrossRef]
  48. Zhu, H.; Chen, X.; Du, Z. Spatial variability of soil nitrogen and pH assessed by the method of geostatistics. J. Soil Sci. 2010, 41, 1086–1090. [Google Scholar]
  49. Suleymanov, A.; Abakumov, E.; Alekseev, I.; Nizamutdinov, T. Digital mapping of soil properties in the high latitudes of Russia using sparse data. Geoderma Reg. 2024, 36, e00776. [Google Scholar] [CrossRef]
  50. Sun, W.; Li, S.; Zhang, G.; Fu, G.; Qi, H.; Li, T. Effects of climate change and anthropogenic activities on soil pH in grassland regions on the Tibetan Plateau. Glob. Ecol. Conserv. 2023, 45, e02532. [Google Scholar] [CrossRef]
Sustainability 16 04695 g001
Figure 1. Comparison of modeled and observed soil ammonium nitrogen (NH4+-Np, mg kg−1) for (a,b) RFA, (c,d) GBRA, (e,f) MLRA, (g,h) SVMA, (i,j) RRTA, (k,l) ANNA, (m,n) GLRA, (o,p) CITA and (q,r) eXGBA under fencing conditions. The solid lines are the linear regression lines between the modeled and observed potential ammonium nitrogen. The x and y axes are observed and simulated values, respectively. RFA, random forest algorithm; GBRA, generalized boosted algorithm; MLRA, multiple linear regression algorithm; SVMA, support vector machine algorithm; RRTA, recursive regression tree algorithm; ANNA, artificial neural network algorithm; GLRA, generalized linear regression algorithm; CITA, conditional inference tree algorithm; eXGBA, eXtreme gradient boosting algorithm.
Figure 1. Comparison of modeled and observed soil ammonium nitrogen (NH4+-Np, mg kg−1) for (a,b) RFA, (c,d) GBRA, (e,f) MLRA, (g,h) SVMA, (i,j) RRTA, (k,l) ANNA, (m,n) GLRA, (o,p) CITA and (q,r) eXGBA under fencing conditions. The solid lines are the linear regression lines between the modeled and observed potential ammonium nitrogen. The x and y axes are observed and simulated values, respectively. RFA, random forest algorithm; GBRA, generalized boosted algorithm; MLRA, multiple linear regression algorithm; SVMA, support vector machine algorithm; RRTA, recursive regression tree algorithm; ANNA, artificial neural network algorithm; GLRA, generalized linear regression algorithm; CITA, conditional inference tree algorithm; eXGBA, eXtreme gradient boosting algorithm.
Sustainability 16 04695 g001
Sustainability 16 04695 g002
Figure 2. Comparison of modeled and observed soil ammonium nitrogen (NH4+-Na, mg kg−1) for (a,b) RFA, (c,d) GBRA, (e,f) MLRA, (g,h) SVMA, (i,j) RRTA, (k,l) ANNA, (m,n) GLRA, (o,p) CITA and (q,r) eXGBA under grazing conditions. The solid lines are the linear regression lines between the modeled and observed actual ammonium nitrogen. The x and y axes are observed and simulated values, respectively. The abbreviations are the same as in Figure 1.
Figure 2. Comparison of modeled and observed soil ammonium nitrogen (NH4+-Na, mg kg−1) for (a,b) RFA, (c,d) GBRA, (e,f) MLRA, (g,h) SVMA, (i,j) RRTA, (k,l) ANNA, (m,n) GLRA, (o,p) CITA and (q,r) eXGBA under grazing conditions. The solid lines are the linear regression lines between the modeled and observed actual ammonium nitrogen. The x and y axes are observed and simulated values, respectively. The abbreviations are the same as in Figure 1.
Sustainability 16 04695 g002
Sustainability 16 04695 g003
Figure 3. Comparison of modeled and observed soil nitrate nitrogen (NO3−-Np, mg kg−1) for (a,b) RFA, (c,d) GBRA, (e,f) MLRA, (g,h) SVMA, (i,j) RRTA, (k,l) ANNA, (m,n) GLRA, (o,p) CITA and (q,r) eXGBA under fencing conditions. The solid lines are the linear regression lines between the modeled and observed potential nitrate nitrogen. The x and y axes are observed and simulated values, respectively. The abbreviations are the same as in Figure 1.
Figure 3. Comparison of modeled and observed soil nitrate nitrogen (NO3−-Np, mg kg−1) for (a,b) RFA, (c,d) GBRA, (e,f) MLRA, (g,h) SVMA, (i,j) RRTA, (k,l) ANNA, (m,n) GLRA, (o,p) CITA and (q,r) eXGBA under fencing conditions. The solid lines are the linear regression lines between the modeled and observed potential nitrate nitrogen. The x and y axes are observed and simulated values, respectively. The abbreviations are the same as in Figure 1.
Sustainability 16 04695 g003
Sustainability 16 04695 g004
Figure 4. Comparison of modeled and observed soil nitrate nitrogen (NO3−-Na, mg kg−1) for (a,b) RFA, (c,d) GBRA, (e,f) MLRA, (g,h) SVMA, (i,j) RRTA, (k,l) ANNA, (m,n) GLRA, (o,p) CITA and (q,r) eXGBA under grazing conditions. The solid lines are the linear regression lines between the modeled and observed actual nitrate nitrogen. The x and y axes are observed and simulated values, respectively. The abbreviations are the same as in Figure 1.
Figure 4. Comparison of modeled and observed soil nitrate nitrogen (NO3−-Na, mg kg−1) for (a,b) RFA, (c,d) GBRA, (e,f) MLRA, (g,h) SVMA, (i,j) RRTA, (k,l) ANNA, (m,n) GLRA, (o,p) CITA and (q,r) eXGBA under grazing conditions. The solid lines are the linear regression lines between the modeled and observed actual nitrate nitrogen. The x and y axes are observed and simulated values, respectively. The abbreviations are the same as in Figure 1.
Sustainability 16 04695 g004
Sustainability 16 04695 g005
Figure 5. Comparison of modeled and observed soil available phosphorus (APp, mg kg−1) for (a,b) RFA, (c,d) GBRA, (e,f) MLRA, (g,h) SVMA, (i,j) RRTA, (k,l) ANNA, (m,n) GLRA, (o,p) CITA and (q,r) eXGBA under fencing conditions. The solid lines are the linear regression lines between the modeled and observed potential available phosphorus. The x and y axes are observed and simulated values, respectively. The abbreviations are the same as in Figure 1.
Figure 5. Comparison of modeled and observed soil available phosphorus (APp, mg kg−1) for (a,b) RFA, (c,d) GBRA, (e,f) MLRA, (g,h) SVMA, (i,j) RRTA, (k,l) ANNA, (m,n) GLRA, (o,p) CITA and (q,r) eXGBA under fencing conditions. The solid lines are the linear regression lines between the modeled and observed potential available phosphorus. The x and y axes are observed and simulated values, respectively. The abbreviations are the same as in Figure 1.
Sustainability 16 04695 g005
Sustainability 16 04695 g006
Figure 6. Comparison of modeled and observed soil-available phosphorus (APa, mg kg−1) for (a,b) RFA, (c,d) GBRA, (e,f) MLRA, (g,h) SVMA, (i,j) RRTA, (k,l) ANNA, (m,n) GLRA, (o,p) CITA and (q,r) eXGBA under grazing conditions. The solid lines are the linear regression lines between the modeled and observed actual available phosphorus. The x and y axes are observed and simulated values, respectively. The abbreviations are the same as in Figure 1.
Figure 6. Comparison of modeled and observed soil-available phosphorus (APa, mg kg−1) for (a,b) RFA, (c,d) GBRA, (e,f) MLRA, (g,h) SVMA, (i,j) RRTA, (k,l) ANNA, (m,n) GLRA, (o,p) CITA and (q,r) eXGBA under grazing conditions. The solid lines are the linear regression lines between the modeled and observed actual available phosphorus. The x and y axes are observed and simulated values, respectively. The abbreviations are the same as in Figure 1.
Sustainability 16 04695 g006
Table 1. The relative bias (%) between modeled and observed soil ammonium nitrogen (NH4+-N, mg kg−1), nitrate nitrogen (NO3−-N, mg kg−1) and available phosphorus (AP, mg kg−1).
Table 1. The relative bias (%) between modeled and observed soil ammonium nitrogen (NH4+-N, mg kg−1), nitrate nitrogen (NO3−-N, mg kg−1) and available phosphorus (AP, mg kg−1).
SceneVariablesSoil Depth (cm)RFAGBRAMLRASVMARRTAANNAGLRACITAeXGBA
FencingNH4+-Np0–104.11 1.52 −14.31 −25.55 1.64 −14.31 −12.36 −13.94 −47.41
10–203.62 8.96 −16.71 −27.56 −3.05 −16.71 −14.35 0.49 −38.29
NO3−-Np0–10−0.33 −4.50 −14.75 −27.51 −18.69 −14.75 −4.20 −19.56 −49.93
10–20−5.62 −5.90 4.85 −30.55 −15.93 3.33 6.28 6.28 −50.80
APp0–106.02 10.32 8.26 −19.57 6.62 8.26 8.14 13.79 −46.81
10–206.93 4.10 −1.99 −11.78 0.61 −1.99 2.82 −1.70 −47.89
GrazingNH4+-Na0–10−8.44 −15.87 −19.38 −25.73 −24.07 −27.04 −21.96 −18.21 −55.47
10–206.73 10.65 3.19 −11.73 18.13 3.19 −2.69 −1.10 −40.14
NO3−-Na0–108.65 6.81 −0.56 −9.89 4.09 −0.56 −14.50 −3.34 −50.33
10–20−2.42 −7.12 −9.94 −36.13 −16.69 −18.25 −16.26 −9.40 −51.10
APa0–101.23 −3.59 −16.67 −15.81 −6.22 −16.67 −1.14 −5.20 −51.36
10–20−1.69 −2.07 −6.99 −16.89 −6.79 −6.99 −5.47 −7.41 −51.96
Note: RFA, random forest algorithm; GBRA, generalized boosted algorithm; MLRA, multiple linear regression algorithm; SVMA, support vector machine algorithm; RRTA, recursive regression tree algorithm; ANNA, artificial neural network algorithm; GLRA, generalized linear regression algorithm; CITA, conditional inference tree algorithm; eXGBA, eXtreme gradient boosting algorithm.
Table 2. The RMSE (mg kg−1) between modeled and observed soil ammonium nitrogen (NH4+-N, mg kg−1), nitrate nitrogen (NO3−-N, mg kg−1) and available phosphorus (AP, mg kg−1).
Table 2. The RMSE (mg kg−1) between modeled and observed soil ammonium nitrogen (NH4+-N, mg kg−1), nitrate nitrogen (NO3−-N, mg kg−1) and available phosphorus (AP, mg kg−1).
SceneVariablesSoil Depth (cm)RFAGBRAMLRASVMARRTAANNAGLRACITAeXGBA
PotentialNH4+-Np0–101.28 1.85 3.76 3.07 2.65 3.76 4.46 2.95 3.65
10–200.71 0.86 2.03 1.37 1.77 2.03 2.23 1.49 1.41
NO3−-Np0–102.29 2.92 9.89 9.74 6.27 9.89 9.66 9.18 7.16
10–202.12 2.09 7.55 7.55 4.19 7.68 7.69 7.69 5.50
APp0–102.35 2.81 7.72 8.11 5.45 7.72 8.01 5.55 6.28
10–203.37 3.85 4.35 3.63 3.51 4.35 4.49 4.33 5.36
ActualNH4+-Na0–102.28 2.91 5.70 4.32 3.99 6.77 6.20 4.43 5.78
10–201.22 1.38 3.44 2.69 3.05 3.44 3.84 3.61 2.80
NO3−-Na0–103.14 4.13 12.00 10.97 5.40 12.00 13.25 12.58 10.32
10–202.44 2.83 11.84 11.82 5.83 11.91 12.57 5.22 8.55
APa0–102.06 2.74 7.64 6.23 5.02 7.64 7.86 6.55 7.13
10–201.92 2.49 4.81 3.70 3.26 4.81 4.74 3.39 5.97
Note: RFA, random forest algorithm; GBRA, generalized boosted algorithm; MLRA, multiple linear regression algorithm; SVMA, support vector machine algorithm; RRTA, recursive regression tree algorithm; ANNA, artificial neural network algorithm; GLRA, generalized linear regression algorithm; CITA, conditional inference tree algorithm; eXGBA, eXtreme gradient boosting algorithm.
Table 3. The R2 between modeled and observed soil ammonium nitrogen (NH4+-N, mg kg−1), nitrate nitrogen (NO3−-N, mg kg−1) and available phosphorus (AP, mg kg−1).
Table 3. The R2 between modeled and observed soil ammonium nitrogen (NH4+-N, mg kg−1), nitrate nitrogen (NO3−-N, mg kg−1) and available phosphorus (AP, mg kg−1).
SceneVariablesSoil Depth (cm)RFAGBRAMLRASVMARRTAANNAGLRACITAeXGBA
PotentialNH4+-Np0–100.960.920.680.810.830.680.550.810.93
10–200.930.910.450.840.580.450.330.700.92
NO3−-Np0–100.970.960.470.500.840.470.490.560.97
10–200.950.960.360.410.890.340.340.340.96
APp0–100.950.930.500.460.750.500.470.750.93
10–200.910.870.820.870.890.820.800.820.89
ActualNH4+-Na0–100.940.910.640.840.870.500.630.810.92
10–200.960.940.600.770.720.600.500.560.94
NO3−-Na0–100.970.950.570.660.910.570.510.530.95
10–200.980.970.430.510.930.440.360.920.97
APa0–100.970.950.610.750.830.610.570.700.92
10–200.970.940.790.900.910.790.800.900.94
Note: RFA, random forest algorithm; GBRA, generalized boosted algorithm; MLRA, multiple linear regression algorithm; SVMA, support vector machine algorithm; RRTA, recursive regression tree algorithm; ANNA, artificial neural network algorithm; GLRA, generalized linear regression algorithm; CITA, conditional inference tree algorithm; eXGBA, eXtreme gradient boosting algorithm.
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Sun, W.; Qi, H.; Li, T.; Qin, Y.; Fu, G.; Han, F.; Wang, S.; Pan, X. Modelling Soil Ammonium Nitrogen, Nitrate Nitrogen and Available Phosphorus Using Normalized Difference Vegetation Index and Climate Data in Xizang’s Grasslands. Sustainability 2024, 16, 4695. https://doi.org/10.3390/su16114695

AMA Style

Sun W, Qi H, Li T, Qin Y, Fu G, Han F, Wang S, Pan X. Modelling Soil Ammonium Nitrogen, Nitrate Nitrogen and Available Phosphorus Using Normalized Difference Vegetation Index and Climate Data in Xizang’s Grasslands. Sustainability. 2024; 16(11):4695. https://doi.org/10.3390/su16114695

Chicago/Turabian Style

Sun, Wei, Huxiao Qi, Tianyu Li, Yong Qin, Gang Fu, Fusong Han, Shaohua Wang, and Xu Pan. 2024. "Modelling Soil Ammonium Nitrogen, Nitrate Nitrogen and Available Phosphorus Using Normalized Difference Vegetation Index and Climate Data in Xizang’s Grasslands" Sustainability 16, no. 11: 4695. https://doi.org/10.3390/su16114695

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop