Patterns of nitrogen and phosphorus pools in terrestrial ecosystems in China 1

Recent increases in atmospheric carbon dioxide (CO 2 ) and temperature relieve the limitation 18 of these twotheir limitations on terrestrial ecosystem productivity, while nutrient availability 19 constrains the increasing plant photosynthesis more intensively. Nitrogen (N) and phosphorus 20 (P) are critical for plant physiological activities and consequently regulates ecosystem 21 productivity. Here, for the first time, we mapped N and P densities and concentrations of 22 leaves, woody stems, roots, litter and soil in forest, shrubland and grassland ecosystems 23 across China, based on an intensive investigation in 41754,865 sites, covering species 24 composition, biomass, and nutrient concentrations of different tissues of living plants, litter 25 and soil. Forest, shrubland and grassland ecosystems in China stored 7665.62 × 10 6 Mg6803.6 26 Tg N, with 7434.53 × 10 6 Mg (96.996635.2 Tg N (97.5%) fixed in soil (to a depth of one 27 metre), and 32.39 × 10 6 Mg27.7 Tg N (0.42%), 59.4%), 57 × 10 6 Mg.8 Tg N (0.78%), 124.21 28 × 10 6 Mg8%), 71.2 Tg N (1.62%) and 14.92 × 10 6 Mg11.7 Tg N (0.192%) in leaves, stems, 29 roots and litter, respectively. The forest, shrubland and grassland ecosystems in China stored 30 3852.66 × 10 6 Mg2806.0 Tg P, with 3821.64 × 10 6 Mg2786.1 Tg P (99.193%) fixed in soil (to 31 a depth of one metre), and 3.36 × 10 6 Mg (0.09%), 14.06 × 10 6 Mg2.7 Tg P (0.36%), 11.47 × 32 10 6 Mg1%), 9.4 Tg P (0.303%), 6.7 Tg P (0.2%) and 2.14 × 10 6 Mg (0.061.0 Tg P (< 0.1%) in 33 leaves, stems, roots and litter, respectively. Our estimation showed that N pools were low in on ecosystem


Introduction
of plant species in one site; is the biomass density of a specific organ of the i th plant species 140 in onethat site, where the plant organ biomass was estimated by allometric equations or 141 harvesting; represents the N or P concentration (g kg -1 ) of the same organ of the i th plant 142 species in that site. Allometric equation methods were adapted to trees and some shrubs (tree-143 like shrubs and xeric shrubs) for biomass estimation, while the biomass of grass-like shrubs and 144 herbs were obtained by direct harvesting. Litter N or P density was litter biomass density (by 145 harvesting) multiplied by litter N or P concentration of each sampling site. The soil N or P 146 density was calculated to a depth of one metre. Soil N or P concentration and bulk density were 147 measured at different depths (0-10, 10-20, 20-30, 30-50, and 50-100 cm) to determine the 148 community-level soil N or P density using Equation (2) where ( )SND (SPD) is the total N or P density of the soil within top 1 m (Mg 152 ha -1 ); n is the total number of soil layers (ranging from one to five) in the i th layer (0-10, 10-20, 153 20-30, 30-50 and 50-100 cm),one site; is the volume percentage of gravel with a diameter > 154 2mm, is the bulk density (g cm -3 ), is the soil N or P concentration (g kg -1 ), and is 155 the depth (cm) of the i th layer. For detailed calculations of species biomass and community-156 level concentrations at each site, please refer to previous studies (Tang et al., 2018a. into the following 13 Vegetationvegetation types: five forest types, i.e., evergreen broadleaf 176 forests, deciduous broadleaf forests, evergreen needle-leaf forests, deciduous needle-leaf 177 forests, broadleaf and needle-leaf mixed forests; four shrubland types, i.e., evergreen 178 broadleaf shrublands, deciduous broadleaf shrublands, evergreen needle-leaf shrublands, and 179 sparse shrublands; and four grassland types, i.e., meadows, steppes, tussocks, and sparse 180 grasslands. 181 182

Prediction the nationwide nutrient pools and distribution patterns 183
We used back-propagation artificial neural network for nutrient density spatial 184 interpolating. The input layer containedrandom forest to predict the nutrient densities and 185 concentrations across China. The predictors included MAT, MAP, longitude, latitude, elevation, 186 EVI and vegetation types (as dummy variables). We established one artificial neural network 187 random forest model for N or P in each component (three plant organs, litter and P in five 188 components,soil layers), respectively. The observation dataIn each model, six variables were 189 randomly grouped into two subsets, 90% data for trainingsampled at each split, and the other 190 10% for500 trees were grown. Larger values of these parameters did not increase validation. 191 When building the artificial network, we used one and two layers, one to 20 hidden neurons per 192 layer, respectively, to find out a model configuration with the best predicting ability. The 193 training and testing process R 2 obviously. Model prediction were repeated 100 times for each 194 configuration. The best predicting model was selected according to the minimal mean root mean 195 square error (RMSE). Then the chosen model was used to predict the nationwide nutrient 196 distribution in corresponding component for 100 times to obtain the average conditions. 197 results. When modelling the nutrient densities in woody stems, we excluded the four 198 grassland types. The vegetation N or P density was the sum of all plant organs, and the 199 ecosystem N or P density was the sum of all components. 200 All densities were log-transformed based on e, and explanatory variables were transformed 201 using the following equation to ensure they were in the same range before modelling. 202 where xi means the i th value of the environmental variables x, and max(x) and min(x) 204 represent the maximum and minimum values of x, respectively. We estimated the relative 205 importance of predictors using the increase in node purity for the splitting variable, which was 206 measured by the reduction in residual sum of squares. The same procedures were repeated for 207 the prediction of N and P concentrations in different components across China. The spatial 208 pattern of N:P ratio was calculated from the predicted N and P density datasets of the 209 corresponding component. 210 The vegetation N or P density was the sum of all plant organs, the soil N or P density was 211 the sum of all soil layers, and the ecosystem N or P density was the sum of all components. The 212 soil depth data across China were obtained from Shangguan et al (2017).The N and P pools in 213 13 Vegetationvegetation types were estimated, respectively. The N and P pools were calculated 214 from the predicted nationwide densities. The predicted N and P densities were in 1 km spatial 215 resolution, so the nutrient stock is the density multiply the grid area (1 km 2 ) for each grid. The 216 nutrient pools of a given vegetation type equals the sum of stocks of the grids belonging to that 217 type. 218 219

Data Model validation and uncertainty and validation 220
To evaluate the model performance, we calculated the linear relationship between the observed 221 validation data (10% of the dataset by random sampling) and predicted data that was estimated 222 based on training data (90% of the dataset by random sampling) for 100 times with the selected 223 models for every component. TheWe then calculated means of validation R 2 , slopes and 224 intercepts of thesethe 100 relationships were estimated using standard major axis regression..

Mapping of N and P densities in China's terrestrial ecosystems 281
All models of the N and P densities of different components performed well (, with the 282 validation R 2 ranging from 0.55 to 0.78 for plant organs and litter (Fig. 4), especially those for 283 the woody stems (R 2 = 0.81 and 0.69 for N and P densities, respectively)from 0.47 to 0,62 for 284 soil layers (Fig. 5). As to the concentration models, the validation R 2 varied from 0.45 to 0.63 285 for plant organs and litter (R 2 =0.66Fig. S2), and 0.62 for N and P densities, respectively).from Mongolia (< 0.01 Mg N ha -1 ).. The woody stem and litter N densities showed the similar 296 patterns to thosethat of the leaves, (Fig. 6c & g), whereas that in rootsroot N density was high 297 in the Mount Tianshan, Mount Alta, Qinghai-Tibetan Plateau, northeastern mountainous area 298 and the eastern Inner Mongolia steppe (Fig. 56e). The vegetation N density was relatively highhigher in eastern China, eastern Qinghai-Tibetan Plateau, Mount Tianshan and Mount Alta, 300 ranging from 0.5 to 2.5 Mg N ha -1 . (Fig. 7a). The soil and ecosystem N densities were low in 301 northern China except the Changbai Mountains, Mount Tianshan and Mount Alta, but high in 302 the eastern Qinghai-Tibetan Plateau and the Yunnan Province (Fig. 67c & e). 303 The P densities in leaves, woody stems, roots, litter and litterthe whole vegetation showed 304 similar patterns to the N densities in the corresponding components, respectively. (Fig. 6b, d, f 305 & h; Fig 7b). However, soil and ecosystem P densities were high in western and northern China 306 but low in eastern and southern China, but low at high altitudes in the Qinghai-Tibetan Plateau 307

Performance and uncertainty of density models 347
The accuracy of the density models varied among different components. Soil interpolation 348 models Models for soil showed poorestrelatively poorer accuracy (R 2 =0.38than models for 349 Nplant organs and 0.27 for P) among these models,litter (Fig. 4 & 5), partly because that soil N with soil depths (Fig. 5 and S3). The models preformed best for the stem N and P, because The predicted SDs were relatively higher in high-latitudes and high-altitudes, such as the 363 northeastern mountainous area and the Qinghai-Tibet Plateau, probably because of the lower 364 sampling density. Meanwhile, the temperature in these regions was about the lower limit of the 365 temperature range in our dataset, which could consequently lead to the weaker validity of the 366 prediction results in such cold regions. 367 368 5.2It is also noteworthy that the validation R 2 of the density models were higher than those 369 of the concentration models for plant organs and litter (Fig. 4 & S2), which was opposite for 370 soil layers (Fig. 5 and S3). They might reflect that biomass were more constrained by the 371 selected factors in this study than nutrient concentrations in vegetation, while bulk density was 372 less affected than nutrient concentrations in soil.

Potential driving factors of the N and P densities in various components
The distribution and allocation of N and P pools in ecosystems were largely determined by 392 vegetation types and climate. The difference in the spatial patterns of nutrient pools could reflect 393 the spatial variation in local vegetation. For example, it is obvious that the regions covered by 394 forests tend to have higher the aboveground nutrient densities than those covered by other types, 395 while the regions covered by sparse shrublands tend to have the lowest nutrient densities (Fig.  396 3). Despite its decisive influences on vegetation types, climate also impacts greatly on the 397 nutrient utilization strategies of vegetation (Kirilenko and Sedjo, 2007;Poudel et al., 2011). For 398 example, in southeastern China with higher precipitation and temperature, forests tend to allot 399 more nutrient to organs related to growth, for example, leaves that perform photosynthesis and 400 stems that related to resource transport and light competition (Zhang et al., 2018). These 401 influences were reflected in our models (Fig. S8-S11). In the models of densities for plant 402 organs and litter, vegetation types and climate variables showed higher relative importance.  Hou, E., Luo, Y., Kuang, Y., Chen, C., Lu, X., Jiang, L., Luo, X., and Wen, D.: Global meta-analysis 545 shows pervasive phosphorus limitation of aboveground plant production in natural terrestrial 546 ecosystems, Nat Commun, 11, 637, https://doi.org/10.1038/s41467-020-14492-w, 2020.   787600.4 32.39427.7 59.57157.8 124.20971.2 3.3572.7 14.0579.4 11.4666.7 See table 1 for abbreviations.  The topographic map of China (u) is also shown..