Terrestrial nutrient exports and environmental changes explain eutrophication trends in fty large lakes of Yangtze Plain, China

11 Over the past two decades, lakes in Yangtze Plain have suffered from serious eutrophication, in some regions with 12 increased frequency of cyanobacteria blooms over years. In this study, we investigated the underlying causes of 13 eutrophication using a combination of process-based ecosystem modelling and statistical data analysis. We found 14 that terrestrial nutrient exports with runoff have significantly increased from 1979 to 2018 in Yangtze Plain, directly 15 linked to the enhanced usage of chemical fertilizer for crops. Based on statistical analyses of environmental variables, 16 terrestrial nutrient exports and satellite-observed probability of eutrophication occurrence (PEO), we separated the 17 studied fifty lakes into five classes with similarities in environmental and nutrient variations, and attributed key 18 factors in controlling the temporal changes of PEO. The results showed that the satellite-observed PEO trends in five 19 classes could be largely linked to the terrestrial nutrient exports and environmental changes. Specifically, we found 20 agricultural activities can explain the observed eutrophication trends in western lakes where lake catchments are 21 dominated with arable and natural land, and the reduced discharge of industrial wastewater was found to be linked 22 to the declining trends in eutrophication for eastern lakes where the green growth of industrialization were promoted 23 from 2003 to 2011. These findings highlight the importance of sustainable management of agriculture and 24 industrialization to overcome eutrophication issues in this region. of the roles of activities in degradation and water the approach of combining process-based

has been mainly linked to nutrient over-enrichment, mainly nitrogen (N) and phosphorus (P) elements because they that diffuse nitrogen sources from agricultural and urban systems contributed 90% to river nitrogen exports, while 52 point sources contributed 52% to river phosphorus exports in Taihu Lake based on in-situ measurements. The 53 temporal dynamics of algal blooms were primarily associated with nutrient (total dissolved nitrogen and phosphorus) 54 concentration in Taihu Lake from 1993 to 2015 (Zhang et al. 2018). Moreover, the nitrogen and phosphorus 55 concentration in lakes were monitored as discrete points to demonstrate the roles of nutrient loads in eutrophication 56 these previous findings only focused on individual and several hyper-eutrophic lakes for short periods, which made 58 it difficult to provide comprehensive understanding of factors in regulating regional patterns of lake eutrophication. 59 Furthermore, plants actively participate in terrestrial water and nutrient cycling, and these abovementioned studies cannot consider the dynamic roles of terrestrial ecosystems in influencing runoff and leached nutrients from land to 61 and is influenced by daily NPP and crop development stages (Olin et al. 2015). Dead leaf, wood and root enter soil 108 litter C and N pools. In the model, there are in total 11 soil carbon pools with different decomposition rates (Parton 109 et al. 1993;Parton et al. 2010). The C and N transfers among the C pools are influenced by soil temperature, water 110 content and C-N ratios of receiving pools (Smith et al. 2014). Vegetation takes up mineral N via roots for growth and 111 development. Apart from plant uptake, soil N is influenced by atmospheric deposition, plant biological fixation, 112 nutrient gaseous losses and decomposition of SOM, and can be leached with runoff in the forms of dissolved organic 113 (DON) and inorganic (DIN) nitrogen. DON is computed daily as a fraction of the decomposition rate of the soil 114 microbial SOM pool and is affected by percolation, while DIN is based on daily percolation and available mineral N. 115 116 Crops are initialized with C mass from seeds, they also have an initial N content based on the CFT's C:Nmin value. 117 Fertilizer and manure were applied three times during the growing period and the application rates at each time were allocation at different development stages can be found in (Olin et al. 2015). 121

122
In this study, we used 11 CFTs to represent the majority of crops in Yangtze Plain. Hybrid and super-hybrid rice are 123 widely distributed in this region, characterized by high grain yield in a short growing season. The high grain yield is 124 associated with enhanced photosynthesis rates due to higher chlorophyll and Rubisco contents in leaves than normal 125 rice (Yuan et al. 1998). Therefore, we calibrated the relationship between leaf N content and the maximum catalytic 126 capacity of Rubisco (Vmax) in LPJ-GUESS via optimization of grain yield (Haxeltine and Prentice 1996). 127 (1) 128 where N represents the foliage N content, 25 m V is the maximum catalytic capacity of Rubisco at 25℃. p and N0 are 129 empirical parameters. The parameter, p, is largely reduced to increase the rubisco capacity for super-hybrid rice (Table  130 A1). Furthermore, super-hybrid rice has higher a specific leaf area (SLA, in Table A1), compared with normal rice 131  (Table S1), where three crop types were found as crop rotations, with combinations of two crops 159 cultivated in one year: 1) early-season rice and late-season rice; 2) winter wheat and summer maize; 3) winter wheat 160 and soybean. The details of the interpolation of these station-based data are described in section S3. 161 162

Evaluation of LPJ-GUESS 163
Three types of datasets were used to evaluate the performance of LPJ-GUESS simulation: 1) GOSIF GPP product; 164 2) GIMMS LAI3g product, and 3) field-observed grain yield. The annual global GOSIF GPP products from 1992 to 165 products cover the period of 1982 to 2011, with a spatial resolution of 0.25° (Zhu et al. 2013). The observed annual 167 grain yield for ten main crops was obtained from 89 stations across the Yangtze Plain from 2000 to 2018 (Liu et al. 168 2010). Due to limited yield data for each crop, the average grain yield for each crop type was calculated to represent 169 regional mean grain yield for comparison with simulated grain yield in Yangtze Plain.

Statistical analysis 189
We applied statistical analyses for determining possible drivers of the satellite-observed PEO changes. Seven 190 explanatory variables (listed in Table S4)

Model evaluation 204
To evaluate the performance of LPJ-GUESS model, the simulated annual crop yield, leaf area index (LAI) and 205 gross primary production (GPP) were compared with observation-based estimates. The crop yields for ten main 206 crop types, averaged for the Yangtze Plain area, agreed well with observed values, with both mean relative errors 207 (MRE) and root mean squared errors (RMSE) between simulated and observed grain yields lower than 10% 208 (Fig. 2). Similar spatial patterns appeared between the simulated and satellite-derived LAI and GPP estimates 209 (Fig. S2). The magnitudes of simulated LAI were in good agreement with GIMMS LAI3g estimates (Fig. S1a), 210 while LPJ-GUESS overall underestimated GPP compared with GOSIF GPP (underestimation on average ~20%, 211 considered acceptable for such a regional simulation with relatively low MRE and RMSE (~23%). 215

Growing anthropogenic nutrient application in Yangtze Plain 222
The total N and P applications from chemical fertilizer and manure have clearly increased over the past forty years 223 (Fig. 3), with average rates of 2.75 kg N ha -1 yr -2 and 0.51 kg P ha -1 yr -2 , respectively, resulting in a three-fold increase 224 over the study period. However, due to differences in the paces of the two elements, the N:P ratio shows a decreasing

Temporal and spatial variability of leached nitrogen 233
We used LPJ-GUESS to simulate and quantify the impact of this enhanced application of fertilizer and manure on 234 the nutrient surplus leached as potential nutrient sources for lake water eutrophication. In the Yangtze Plain, the 235 simulated leached N from agricultural and natural land has increased significantly during the study period, with a 236 considerably lower rate for natural (~0.2 kg N ha -1 yr -2 , Fig. 4b) than for agricultural ecosystems (~4.5 kg N ha -1 yr -237 2 , Fig. 4a), due to external inputs from fertilizer and manure to the latter. Considerable spatial differences in N 238 leaching from cropland (CLN) and natural land (NLN) existed. Higher N leaching was simulated for the cropland 239 area in the catchment of Dongting Lake (longitude range 110°E to 114°E), which was linked to higher fertilizer rates. 240 A decrease of N leaching was simulated in the eastern parts of Yangtze Plain (Fig. 4c). NLN distributed differently 241 between the south and north sides of Yangtze Plain (Fig. 4d), mainly because of differences in N sources from Individually, all studied lake catchments were characterized by significantly increasing trends for CLN, but with 245 different start points for the CLN increases (Fig. 4e): The CLN values for lake catchments in the western region (L01-246 L06) started to increase already around 1990, much earlier than the catchments for eastern lakes (L45-L50, starting 247 approximately 2005), suggesting that the nutrient status in western lakes to be potentially affected earlier than in 248 eastern lakes. Similarly, NLN for all studied lake catchments, except for Donghu Lake (L18), has significantly 249 increased with generally lower rates than CLN values over the past forty years (Fig. 4f). Seven lake catchments 250 dominated by natural vegetation exhibited higher NLN (i.e., L28, L30, L32, L35, L37, L38, L43) than the surrounding 251 lakes, collectively distributed in the middle regions of Yangtze Plain (near Poyang Lake). 252 Kendall trend test, p < 0.05) annotated with '*'. The ID numbers of lakes are listed in Table S2. were dominated by increasing trends of PEO, with the strongest increases occurring in class III (Fig. 5c)

. Lakes of 278
Class II are mainly located at the eastern parts of Yangtze Plain, where lake catchments undergo rapid industrialization. 279 Lakes and their associated catchments of Class III cover a large area of Yangtze Plain (50.6%) and include the two 280 largest lakes of this region (i.e., Poyang Lake, L35 and Dongting Lake, L07 in Fig. 1 Fig 6), and for each class, different drivers were responsible for the PEO changes. Across 302 all lakes (Fig. 6f), we found PEO variations to be positively linked to agricultural nutrients NP and CLN, but 303 negatively linked to NLN, highlighting that agricultural activities could be responsible to the overall changes of lake 304 PEO in Yangtze Plain. At the class level, Class II showed clear positive linkage of PEO with IW and T, but negative 305 linkage to the agricultural related indices (i.e., CLN, NP &AP) and total runoff, and the PEO changes for all other 306 classes appear to be positively linked to the agricultural activities. In Class II, the lakes are dominated by negative 307 trends of PEO (Fig. 5c), and the PLSR model provides many possible drivers for this (Fig. 6b). The decreasing trends 308 in IW (Fig. S5) are expected to be the mechanistic driver of PEO dynamics, while for the other four variables with 309 significantly negative coefficients and positive trends in each variable (Fig. S5), the PLSR model provides 310 correlations between these explanatory variables and PEO in Class II lakes that are mechanistically unlikely (Fig.  311   5c). Furthermore, the large magnitude of IW changes observed in Class II lakes might suggest that the decreasing 312 trends of PEO were mainly explained by the decreased exports of IW for the study period (Fig. S5), and that increased 313 export of nutrients from cropland, although correlated, counterbalances this effect. Green industrial growth and 314 wastewater treatment plants seem to have efficiently limited nutrient exports to local water bodies (Lyu et al. 2016). 315 Notably, the attribution of driving factors to PEO changes based on these five classes could also be influenced by the 316 spatial variations among lakes within each class, as the lakes have been grouped based on the dominant environmental 317 and nutrient variables (PCA results). Therefore, we further examined the temporal trends of the significant factors within the same class. We found that lakes within the same class often showed similar trends in explanatory variables 320 and PEO as indicated by the regression coefficients in the PLSR models. For instance, the positive coefficients 321 between PEO and IW in Class II lakes (Fig. 6b) were also seen as negative trends for both PEO and IW when we 322 look at individual lakes. These alignments of trends and correlations support that the environmental and nutrient 323 variables are mainly linked to the inter-annual changes in PEO in these five classes (Fig. S5). For Class III, there are 324 different trends of T and PEO, although they showed positive coefficients, which in this case, indicates potential 325 impacts from spatial variability of lakes in this class.

Contributions of anthropogenic activities to eutrophication 335
In this study, we have attempted to determine possible causes of PEO for studied lakes using statistical methods. The 336 results from these methods alone are unavoidably influenced by the correlations between individual variables, and/or 337 by similar long-term trends in several explanatory variables that do not differ greatly between studied lake catchments. 338 In such cases, the used statistical methods may result in correlations that do not allow differentiating between possible 339 underlying mechanisms.

341
Through our analysis, we found out that lake eutrophication in Yangtze Plain is mainly associated with anthropogenic 342 stresses, consisting of agriculture and industrialization. Industrialization-affected eutrophic lakes collectively 343 distribute in the eastern parts of Yangtze Plain, while agricultural activities have large impacts on middle and western 344 lake catchments with large cropland distribution in the catchments, most of which are located upstream of Dongting 345 Lake (L7 in Fig. 4). Some lakes in the middle parts are characterized by PEO changes induced via large variability 346 of land cover fraction, including two river-connected lakes (i.e., Dongting and Poyang Lake) for which PEO 347 dynamics were not only affected by drivers in the local catchment, but also by upstream land use changes (Dai et al.

Limitations and Uncertainties 367
Terrestrial N and P sources were considered as nutrient elements for phytoplankton growth and development (Yi et  . We expect that leached P depends on not only P fertilizer use, but also plant uptakes and soil supplies, similar to N processes (Fig. S6), which suggests 371 that a further consideration of ecosystem processes on leached P and PEO will be needed. The P manure data, 372 generated from N manure data via N-P ratios in animal excrements, are not fully independent from N manure inputs 373 but are still important to include here as the nutrient sources in order to assess the contributions of P exports (i.e., of fish baits were directly cast into aquaculture area for feeding fish. The surplus nutrients enable to contribute to 383 more severe eutrophication (Guo and Li 2003). These limitations will need to be further explored to obtain a complete 384 picture of nutrient sources available for phytoplankton communities and to reduce anthropogenic impacts on lake 385 water quality and eutrophication. 386 387

Conclusion 388
In this study, we combined a dynamic vegetation model LPJ-GUESS with statistical analysis to reveal factors 389 controlling eutrophication status in fifty large Yangtze lakes. During the past forty years, the agriculture-related 390 nutrient exports (N&P) have significantly increased for all studied lakes, mainly linked to the enhanced chemical 391 fertilizer usage. The leached nitrogen from natural land are much smaller than cropland. The averaged nitrogen 392 leaching from cropland in the catchment of Dongting Lake is apparently greater than the middle-and eastern-parts 393 of Yangtze Plain, while the leached nitrogen from natural land revealed considerable spatial variations from the north 394 to south sides of Yangtze Plain linked to different climatic conditions. The overall PEO trends in the studied lakes