Introduction

Biogeochemical cycles are impacted by changes in land use cover1,2,3. One of the primary issues in climate change assessment is determining how the nitrogen (N) cycle responds to changes in land cover4,5. The world’s forests cover tropical, temperate, and boreal zones, covering 4.06 billion hectares (ha), around 31% of the world’s area6,7. To date, studies on the biogeochemical effects of forests have mainly focused on changes in carbon (C) sinks and alteration in atmospheric CO2 concentrations through changes in forest cover8,9,10,11,12. However, compared to the forest C cycle, the forest N cycle has received little attention, particularly on a global scale.

Nitrous oxide (N2O), a greenhouse gas produced during the N cycle, can significantly contribute to global warming and stratospheric ozone layer depletion13. Agriculture is the largest anthropogenic source of N2O due to N fertilizer application, about 60% of all anthropogenic emissions, generating 4.1 Tg N/year14. As a result, much N2O emissions research has concentrated on agricultural areas, leaving a gap in data from other types of land use, particularly forest ecosystems15,16. Given that anthropogenic N emissions have enhanced N deposition in forest areas, forest N2O emissions are expected to rise continuously in the future17. Thus, even though both soil N availability and N2O emissions from forests are significantly lower than agriculture, the former is still a large source of N2O on a global scale due to their large area18,19,20. Global scale in-depth research of forest N2O emissions is crucial for future climate change estimates and the creation of effective greenhouse gas-reduction policies.

Previous studies showed that the biogeochemical N cycle is strongly influenced by hydroclimate processes21,22,23,24,25. The two primary processes that control soil N2O emissions in the N cycle, nitrification, and denitrification, are influenced by many environmental conditions26,27,28. Studies have shown that nitrifiers and denitrifiers respond sensitively to hydroclimate factors, such as changes in soil moisture (SM) and temperature, which regulate their metabolic activities and gene abundance29,30,31. Forests maintain the hydrological cycle via biophysical mechanisms that have been widely investigated and demonstrate a change in latitude from tropical to boreal areas6,32. Global hydrologic patterns may affect the N cycle and thus global N2O emissions16,33. However, no studies have systematically coupled the N cycle and hydroclimate, identified the key driving variables, and related them to worldwide forest N2O emissions34. Such information would be relevant for management and decision-making but is still lacking.

To address this gap, here we report a systematic study, using in situ field observations and multi-source remotely sensed and gridded datasets to quantifies assess the impact of hydroclimate on forests (tropical to boreal forests) N2O emissions (Fig. 1). As the biophysical processes affecting the hydroclimate cycle have a distinct latitudinal pattern from the tropics to the boreal area6,32. We aim to (a) investigate whether hydroclimate factors and N2O emissions have a clear latitudinal gradient from the tropical to the boreal forest, and (b) determine the contribution of hydroclimate factors to the N2O emission latitudinal gradient. Addressing these two questions will allow us to more accurately assess the impact of climate change on the land N cycle.

Fig. 1: Distribution of the 69 study sites.
figure 1

Study sites were classified into tropical forests (25°S-25°N, number of sites=22, number of observation samples=61), temperate forests (25°S-50°S and 25°N-50°N, number of sites = 28, number of observation samples = 178), and boreal (north of 50°N, number of sites = 19, number of observation samples = 33) forests, based on their latitude. The circular pie plot represents the percentage of different types of forest sites, with the color harmonized with the color of the points.

Full size image

Results

Aridity index (AI) and N2O fluxes from tropical to boreal forest areas

AI denotes the water balance between the land surface and the atmosphere, and its trend from tropical to boreal forest areas are given in Fig. 2a. AI decreases with latitudes (R2 = 0.39, p < 0.0001, Fig. 2b), and the mean value (Fig. 2c) of tropical forest areas (1.6) is significantly higher than temperate forest areas (1.3) and boreal forest areas (0.9).

Fig. 2: The climate of study sites.
figure 2

a Global patterns of aridity index (AI) across three forest types. The color bar represents the value of AI. b Relationship between latitudes and aridity index. R2 and p values are shown in the figure when significant (p < 0.05). Regression lines (black lines), 95% confidence intervals (gray shading along the line). c Average value (mean ± SD) of AI from tropical to boreal forest areas. Statistically significant differences among forests are represented by different lowercase letters (p < 0.05).

Full size image

N2O fluxes (R2 = 0.14, p < 0.0001) and the cumulative (R2 = 0.28, p < 0.0001) emissions significantly decreasing with latitudes (Figs. 3a, b and S1a). The value of N2O fluxes in tropical forests, temperate forests, and boreal forests is 45.4 ± 97.4, 37.7 ± 74.5, and 11.6 ± 24.1 μg·m2·h−1, respectively (Fig. 2b). The mean value of cumulative N2O in tropical forests, temperate forests, and boreal forests is 2.5 ± 2.1, 1.0 ± 1.1, and 0.5 ± 0.7 kg·ha−1, respectively (Figure S1b).

Fig. 3: Forest N2O fluxes emission globally.
figure 3

a Global patterns of N2O fluxes. b Relationship between latitudes and N2O fluxes. R2 and p values are shown in the figure when significant (p < 0.05). Regression lines (black lines), 95% confidence intervals (gray shading along the line). Data of N2O fluxes are logarithmically (LN) transformed before regression analysis. c Average value (mean ± SD) of N2O fluxes from tropical to boreal forests. Statistically significant differences among forests are represented by different lowercase letters (p < 0.05).

Full size image

N2O flux emissions were highly positively correlated with cumulative emissions (R2 = 0.95, p < 0.0001, Fig. S2), implying that the processes that drive N2O fluxes and cumulative emissions are similar. As a result, in this work, we focus primarily on the results of N2O flux emissions.

Variations in hydroclimate factors along latitude and forest types

We found a clear latitudinal gradient in all hydroclimate factors in this study (Fig. 4). Mean annual incoming shortwave radiation (Rin, R2 = 0.81, p < 0.0001), mean annual temperature (MAT, R2 = 0.67, p < 0.0001), mean annual vapor pressure deficit (VPD, R2 = 0.54, p < 0.0001), mean annual precipitation (MAP, R2 = 0.61, p < 0.0001), actual annual evapotranspiration (ET, R2 = 0.87, p < 0.0001), and Runoff (R2 = 0.30, p < 0.0001) exhibit significantly negative relationships with latitudes (Fig. 4a–f). These factors varied significantly among forest types (p < 0.05), consistently showing tropical > temperate > boreal. SM (R2 = 0.14, p < 0.01) and water-filled pore space (WFPS, R2 = 0.21, p < 0.0001) showed significantly positive relationships with latitudes (Fig. 4g, h). SM and WFPS are not statistically different in temperate and boreal forests (p > 0.05), while they are significantly greater in tropical forests (p < 0.05).

Fig. 4: Variations in hydroclimate factors along latitude and forest types.
figure 4

a Rin (mean annual incoming solar radiation). b MAT (mean annual temperature). c VPD (mean annual vapor pressure deficits). d MAP (mean annual precipitation). e ET (actual annual evapotranspiration). f Runoff. g SM (soil moisture). h WFPS (water-filled pore space). R2 and p values are shown in the figure when significant (p < 0.05). Regression lines (black lines), 95% confidence intervals (gray shading along the line). The average value (mean ± SD) of different variables of tropical (blue), temperate (orange), and boreal (red) forests are shown in each figure. Statistically significant differences among forests are represented by different lowercase letters (p < 0.05).

Full size image

Interaction between hydroclimate factors is shown in Figs. S3–7. Solar radiation is crucial in determining land surface temperature and atmospheric moisture requirements, as MAT (R2 = 0.64, p < 0.0001), VPD (R2 = 0.53, p < 0.0001), and ET (R2 = 0.69, p < 0.0001) are significantly increased with Rin (Fig. S3). Rin can affect ET directly and can also increase ET indirectly by affecting MAT and VPD (Fig, S4). ET significantly increased with MAT (R2 = 0.54, p < 0.0001) and VPD (R2 = 0.41, p < 0.0001) (Fig. S5). As ET increases, SM (R2 = 0.15, p < 0.01) and WFPS (R2 = 0.16, p < 0.0001) decrease significantly (Fig, S6). In addition, soil moisture was significantly influenced by runoff (Fig. S7). These results explain why soil moisture increases with latitudes since solar radiation, temperature, and VPD decrease with latitudes, lowering ET and thereby preserving more water in the soil.

Variations in soil physical and chemical properties along latitude and forest types

We also tested for soil factor’s latitude pattern that might explain some of the variations in the magnitude of N2O fluxes. (Figure S8 and S9). soil organic carbon content (SOC, R2 = 0.12, p < 0.0001), total nitrogen content (TN, R2 = 0.077, p < 0.01), ratio of SOC to TN (C/N, R2 = 0.032, p < 0.05), nitrate nitrogen content (NO3−, R2 = 0.12, p < 0.0001), and silt content (R2 = 0.23, p < 0.0001), exhibit significantly positive relationships with latitudes (Fig, S8a–e), while clay content (R2 = 0.25, p < 0.0001), bulk density (BD, R2 = 0.11, p < 0.0001), and altitude (R2 = 0.12, p < 0.05) significantly negative correlation with latitudes (Fig. S8f–h). No significant (p > 0.05) correlation of soil ammonia nitrogen content (NH4+), pH, and sand content with latitudes was observed (Fig. S9). Among these soil factors, only clay showed significant differences among the three forest types, with size classes showing tropical > temperate > boreal forests (p < 0.05, Fig. S8f).

Relationships among environmental factors and N2O fluxes

Further, we investigated the correlation of N2O fluxes with different factors (those with significant latitudinal gradients in the above results). Linear regression analysis of hydroclimate factors and soil physicochemical properties on N2O fluxes are showed in Fig. 5. Except for SM, all hydroclimate parameters have a significant correlation with N2O fluxes, and only WFPS has a negative influence (p < 0.05). Of the 11 soil factors, only 4 were significantly correlated with N2O fluxes. Soil NO3− and clay are positively related to N2O fluxes (p < 0.05), while C/N and silt revealed opposite relationships with N2O fluxes (p < 0.05). In this study, a total of 11 factors were significantly correlated with N2O fluxes, of which 7 hydroclimate factors (accounted for 63.6%).

Fig. 5: Correlation of environmental factors and N2O flux emissions.
figure 5

a Linear regression analysis of environmental factors and N2O flux emissions. Only the fitted lines with significant (p < 0.05) correlation are shown. Regression lines (dashed lines), 95% confidence intervals (shading along the line). Data are logarithmically transformed before regression analysis. b R2 of linear regression analysis of each factor with N2O flux emissions, less than 0 means negative correlation, and more than 0 means positive correlation. *p < 0.05; **p < 0.01; ***p < 0.001.

Full size image

Relative contribution of environmental factors to N2O fluxes

Finally, we used random forest to examine the relative contributions of the 11 parameters listed above to N2O fluxes (Fig. 6). The random forest analysis found that the most crucial variable affecting N2O fluxes was the hydroclimate factors (i.e., MAP, ET, Rin), followed by C/N. Among hydroclimate factors, the relative contribution rank of each factor was: MAP > ET > Rin > C/N > MAT > WFPS > VPD. For soil factors, the relative contribution rank of each factor was C/N > NO3− > clay > silt. Overall, random forest analysis indicated that the key factor that determined the N2O flux was the hydroclimate, which cumulative contribution amounted to 78.2%.

Fig. 6: The relative influence of hydroclimate and soil factors on N2O flux emissions.
figure 6

The importance values (contribution, %) are derived from a random forest analysis. Data are logarithmically transformed before analysis. Barplots and donut charts indicate the relative influence of different factors on N2O flux emissions.

Full size image

Of four (C/N, NO3−, clay, and silt) key soil factors, we found that soil texture (clay and silt) had a significant effect on water condition (runoff, ET, and WFPS, Fig. 7). Runoff and ET decreased with silt (R2 = 0.052, p < 0.05; R2 = 0.22, p < 0.0001, respectively, Fig. 7a, b), while increased with clay (R2 = 0.32, p < 0.0001; R2 = 0.32, p < 0.0001, respectively, Fig. 7d, e). WFPS is positively correlated with silt (R2 = 0.32, p < 0.0001, Fig. 7c), and has a negative correlation with clay (R2 = 0.19, p < 0.05, Fig. 7f).

Fig. 7: Relationship between soil texture and water condition.
figure 7

Linear regression analysis of the effect of soil silt content on runoff (a), ET (b), and WFPS (c). Linear regression analysis of the effect of soil clay content on runoff (d), ET (e), and WFPS (f). R2 and p values are shown in the figure when significant (p < 0.05). Regression lines (dashed lines), 95% confidence intervals (shading along the line).

Full size image

Discussion

Many environmental factors influence N2O emissions, resulting in exceptionally high temporal and spatial variability in fluxes28,30,35,36. It is one of the main reasons for the persistent difficulties in suppressing N2O fluxes37,38. Previous studies have shown that patterns of hydroclimate may affect soil N dynamics, such as availability and flux emissions16,21,22,23,24. Forests cover a large area, and their N2O emission cannot be ignored. In addition, forests’ hydroclimate processes showed a latitudinal transition from tropical to boreal areas6,39. However, the study of the impact of hydroclimate on N2O emissions latitude gradient at the global scale is still largely unknown34. Our study fills this gap, on a global scale we found that hydroclimate factors explained 78.2% of forest N2O emission variations, which decreased from tropical to boreal forests. The four potential main mechanisms (Fig. 8) are as follows: First, the frequency of drought-rewetting decreases with latitude, resulting in a decrease in N2O fluxes from tropical to boreal forests. A recent study found that high atmospheric demand in the tropics induces excessive evapotranspiration, which accelerates the drawdown of soil moisture, resulting in a higher likelihood of flash drought40. Rewetting following a drought usually induced large, quick pulses in N2O fluxes “hot moment”30,41,42. A study found that hot moments lasted only a fourth of the study period, but could provide more than half of the N2O fluxes in the soil43. The major reason for a hot moment is drought-induced nitrogen substrate accumulation and their subsequent accessibility following rewetting44. The response of hydroclimate processes to climatic differs across forests, leading to a latitudinal gradient of N2O emissions. Our results confirm the above hypothesis, since MAP, Rin, and ET are the three largest relative contributors to N2O emissions among many environmental factors (Fig. 6). Second, the texture of the soil can affect hydroclimate conditions and consequently N2O emissions45,46,47. In general, soil texture influences N2O emissions by determining the likelihood of anaerobic or aerobic soil conditions prevailing48,49. Soil texture, for example, can influence water drainage during rainfall and drought, and hence the type of oxic and anoxic microsites of soil30,50. N2O emission was significantly enhanced in clayey soils, owing to its high water retention and low aeration, which favored to produce of anaerobic conditions, enhancing denitrification and rising N2O fluxes51,52. Soil clay decreases from tropical to boreal areas (Figure S8), which is a key factor contributing to the latitudinal gradient of N2O emission. Third, rising temperatures in tropical areas may speed the decomposition of organic materials in soils, providing energy and substrates to bacteria for denitrification and nitrification and therefore increasing N2O emissions47,53. Previous research has demonstrated that warming increases the rate of N cycling, resulting in an increase in soil N availability54,55. Soil temperature is usually coupled with moisture in regulating N2O emission, as higher temperatures can enhance microbial activity, while adequate soil moisture creates anaerobic conditions that promote denitrification56. Therefore, the highest N2O emissions occur in tropical forest zones under moist and warm conditions. Fourth, evapotranspiration (including evaporation and transpiration) decreases from tropical forests to boreal forests and can affect the latitudinal gradient of N2O emissions. Studies have found that transpiration contributes to increasing N2O by plant roots absorbing N2O from soil water and releasing it into the atmosphere, thereby increasing N2O in the surrounding soil15,57. The transport of N2O within the transpiration stream is a mechanism for emissions from tree stems and canopies. The similar latitude pattern between N2O fluxes and evapotranspiration in this study supports this hypothesis.

Fig. 8: Conceptual diagram of how hydroclimate process influences N2O fluxes.
figure 8

a Hydroclimate processes dominate soil drought and rewetting. b A comparative conceptual diagram of the four main potential mechanisms dominating N2O flux emissions in different forests. The dashed arrows of different colors represent the pathways of influence of different factors on N2O, the thicker the arrow, the greater the influence. Orange, purple, dark blue, and light blue represent the pathways of temperature, clay, wet and dry alternation, and ET effect on N2O flux, respectively. The red solid arrows represent N2O fluxes. Rin incoming shortwave radiation, MAT mean annual temperature, VPD vapor pressure deficit, MAP annual precipitation, ET evapotranspiration, WFPS water-filled pore space. The background image was Created with BioRender.com (https://biorender.com/), and some elements were sourced from the Vecteezy database (https://www.vecteezy.com/).

Full size image

There are two important implications of this study. One of them is that it can be used to validate the N2O emission model results and provide key parameters for the simulation. Models are considered to be efficient research tools for calculating national and global N2O emissions5,36,58. However, there are large uncertainties in model calibration and validation, as well as sizable differences when simulating the spatial and temporal variations of N2O emissions with various models59,60. Overall, our results based on in situ field experiments validate the findings of some recent models19, showing N2O emissions of tropical > temperate > boreal forests. To date, despite significant research efforts, accuracy in predicting N2O generation and emission remains unsatisfactory, whether empirical or process-based models are used38,61. Because models must account for a large range of complex regulatory characteristics when simulating worldwide N2O emissions62,63. Our research discovered that hydroclimate factors such as Rin, VPD, and ET are important in predicting forest N2O emissions, and that incorporating these components into the model may assist to enhance its accuracy. Another important implication is to enable new research horizons for N2O field experiments. Over the past decades, many studies about the mechanisms regulating N2O emissions have focused on soil factors64,65,66. Currently, research on the influence of climate parameters on N2O emissions focuses mostly on temperature and rainfall, which are insufficient to define the impact of climate, especially in the context of global climate change67,68. And yet, despite our results indicating hydroclimate factors (e.g., Rin, ET, runoff, and VPD) potentially high contribution to global forests’ N2O emission, in situ studies of these factors on N2O production are rare. In addition, the hydrological cycle’s acceleration or intensification in response to global warming will be a persistent and challenging problem in climate study69. Therefore, we suggest that it is time to expand our focus from land to the atmosphere and that it is now necessary to integrate N2O emissions with more atmospheric factors, as the effects of atmospheric changes on soils will become increasingly evident.

Two uncertainties in this study are primarily caused by the data and methodology. The soil bacteria produce the majority of the world’s N2O during the N cycle35,70,71. Denitrification and nitrification are the two key processes of soil N2O production72. A recently global-scale study has been conducted that soil-moisture atmospheric feedback in hydroclimate processes plays an importance role in the balance between nitrifiers and denitrifiers16. However, the paucity of information on functional genes in this study creates uncertainty in the microbiological mechanisms by which hydrological cycles alter N2O emissions. An in-depth understanding of climate change’s effects on the coupling of soil characteristics, microbial populations, and N2O emission pathways is critical for driving accurate biogeochemical models and identifying the dynamics of ecosystem N2O emissions under climate change73,74. To have a better mechanistic knowledge of the reactions of soil N2O emissions to forest hydroclimate, more research should be conducted on the regulation of nitrifiers and denitrifiers. Forest types are classified according to latitude rather than biome, which creates uncertainty in this study. Moreover, plant species can influence soil microclimate and structure, N effectiveness, and root respiration, thus affecting N2O emissions75,76,77,78. For example, root secretions can stimulate denitrification activity and so increase N2O generation in the soil47.The response of different species to hydroclimate and their influence on it processes are different79,80,81, which could affect soil N2O emissions. For instance, varying plant species canopies have varying effects on hydroclimate processes, which can ultimately affect soil N2O emissions43. Therefore, the mechanisms driving the changes in N2O emissions by plant species at different latitudes need to be investigated further.

Our study finds for the first time that hydroclimate highly explains forest N2O emissions at a global scale. We found that N2O fluxes emissions significantly decreasing with latitudes, and the value of N2O fluxes in tropical forests (45.4 ± 97.4 μg·m2·h−1) > temperate forests (37.7 ± 74.5 μg·m2·h−1) > boreal forests (11.6 ± 24.1 μg·m2·h−1). The hydroclimate factor contributed 78.2% of N2O emissions, whereas the soil factor contributed 21.8%. Among the hydroclimate factors, MAP, ET, Rin, and runoff contributed the most to N2O emissions. Our results provide new ideas for future field experiments, and we should improve the study of the effects of hydroclimate factors (such as evapotranspiration and solar radiation) on N2O emissions. The key hydroclimate factors we found can be useful for N2O models, and incorporating these dynamics may improve the prediction of global N2O emissions. Overall, recognizing the primary causes of global forest N2O emissions is critical for addressing the climate change dilemma.

Material and methods

Data collection

The N2O emissions dataset was sampled from published articles. First, we searched Google Scholar (https://scholar.google.com), Web of Science (http://apps.webofknowledge.com), and China National Knowledge Infrastructure Database (http://www.cnki.net) databases for publications containing the keywords “nitrous oxide” OR “N2O” AND “forest”, “forestland”, “plant” OR “vegetation” up to December 30, 2022. Second, to ensure site comparability and mitigate selection bias, we applied the following criteria when extracting peer-reviewed publications: (a) inclusion of only field studies conducted in forest sites without anthropogenic N fertilization treatments; (b) location information of field experiments must be provided in the article; (c) consistent soil sampling from a standardized depth across all sites (between 0 ~ 20 cm); (d) standardization of units for each measured factor (e.g., N2O fluxes emission standardized to μg N m−2 h−1, cumulative N2O emission standardized to kg N ha−1); and (e) ensure that soil factors were measured as consistent method as possible across all sites, such as using oven drying method for SM and KCl extracts method for soil available N. Finally, we compiled 272 observation samples of 69 study sites from 47 articles (references are listed in the supplementary information). These study sites cover a large forest area (28°15′46″S–67°26′26″N and 135°13′48″W–155°14′23E) (Fig. 1), including tropical forests (25°S-25°N, 22 sites, 61 observation samples), temperate forests (25°N-50°N and 25°S-50°S, 28 sites, 178 observation samples), and boreal forests (North of 50°N, 19 sites, 33 observation samples). Among the different forests, 31.9% of tropical forests, 40.6% of temperate forests, and 27.5% of boreal forests (Fig. 1).

Selection and description of characteristics

From all selected field studies, we collected N2O fluxes or cumulative emissions data and the following parameters: coordinates (latitude and longitude), hydroclimate factors (MAT, MAP, SM, and WFPS), soil factors (clay, silt, and sand contents, pH, BD, TN, SOC, C/N, NH4+, and NO3−). Get Data Graph Digitizer (www.getdata-graph-digitizer.com/) was used to extract data from graphical plots.

Remote sensing and gridded data

We extracted multi-year mean AI, VPD, Rin, and ET from remote sensing and gridded data based on the coordinates of forest sites to investigate how hydroclimate affects N2O emissions.

The water balance between the land surface and the atmosphere is represented by AI82. AI data were collected from the Climatic Research Unit time-series dataset (CRU TS3.21), which is derived from a wide network of weather station observations83.

Land ET plays a crucial role in the water cycle and the soil moisture-atmosphere feedback loop in the climate system84. ET was obtained from Zhang, et al. 85, which used a satellite remote sensing-based technique to calculate global terrestrial ET.

VPD is determined by the combination of atmospheric temperature and humidity, indicates atmospheric drought, and is used as a measure of atmospheric water demand86. VPD was calculated from land air temperature and dew point temperature based on ERA-Interim dataset87.

Rin represents energy availability and is important in dominating land surface temperature, VPD, and ET88. Rin was obtained from Climatic Research Unit of the National Centers for Environmental Prediction dataset (CRUNCEP, version 7)89.

Surface runoff is closely linked to precipitation and is a key part of the water cycle90,91. In this work, runoff is calculated using the method of Lehner, et al. 92 as precipitation minus evapotranspiration.

Data analysis

Nine hydroclimate factors and eleven soil factors were utilized to investigate the primary causes of N2O emissions (Table 1). First, analysis of variance (ANOVA) was used to examine the changes in N2O emissions (fluxes and cumulative emissions) and AI among various forests. Unless otherwise specified, p < 0.05 is often taken to signify statistically significant differences. Second, the correlation of each factor with latitude was analyzed by linear regression. The significance of linear regression analysis was calculated by the student’s t-test. All statistical analyses were performed using SPSS 22.0 (Armonk, NY, USA), p values less than 0.05 indicate statistical significance (*p < 0.05, significant; **p < 0.01, highly significant; and ***p < 0.001, extremely significant). Third, linear regression analysis was used to examine the relationship between N2O fluxes and environmental factors. These factors have been determined to have significant latitudinal gradients in the second step of the analysis. Finally, we used the random forest to quantify each factor’s relative importance in predicting the N2O fluxes. Factors used for the random forest analysis were all those that proved to be significantly correlated with N2O fluxes by linear regression analysis in the third step. The random forest analysis was assessed with Python (version 3.7.3) package “sklearn”.

Table 1 Summary of variables information
Full size table