Response of runoff and nitrogen loadings to climate and land use changes in the middle Fenhe River basin in Northern China

The simulation of runoff and water quality and their response to the changing environment, mainly including climate and land use changes, are particularly important because water with poor quality cannot be directly used for potable water (via desalination) and industrial applications (Panagopoulos 2020, 2021).

Climate change has long-term effects on water resources on a global as well as local scale, and the consequences have recently come to the forefront in government and scientific circles. Climate change includes relative humidity, precipitation, temperature, solar radiation, and so on (Feistel & Hellmuth 2021), and the typical characteristics of climate change focus on shifts in temperature and precipitation. In the coming decades, climate change in all regions will intensify with temperature extremes and abnormal precipitation events becoming more frequent. Climate change has a marked effect on rainfall distribution as rainfall may increase in high latitudes, but decrease in subtropical regions (IPCC 2021). At the same time, economic and social development has resulted in tremendous changes in land use. The developmental status and direction of growth in different regions are different, so the changing trends in land use vary from region to region. These changes significantly influence the hydrological cycle in terms of streamflow and nutrient loading (Milly et al. 2008; Wang & Hejazi 2011), which needs to be incorporated into the water cycle simulation.

Climate change and land use have a direct impact on the water cycle in a watershed. The water cycle process includes rainfall, evapotranspiration (ET), infiltration, and runoff, and global warming increases the intensity of water cycle events (Kerr 2012). The effects of climate change on the water cycle are different in different regions, making humid areas more humid, and arid areas more arid (Trenberth 2011). Climate change is considered to be a major factor in causing extreme weather events, and its complex effects on hydrological processes have been analyzed in a number of previous studies (Gosling et al. 2011; Meaurio et al. 2017). Different land use conditions have different impacts on runoff. For example, in tropical regions, river basins are not particularly sensitive to land use (Lucas-Borja et al. 2020). Simulations have shown that changes in land use can lead to changes in surface runoff and groundwater recharge (Baker & Miller 2013).

The impact of climate and the underlying surface on runoff will further affect the water quality of the river basin, which is discussed relatively scarce in previous researches. Changes in climate factors, such as temperature and rainfall, cannot directly affect the nitrogen loading process, but can indirectly affect water quality by affecting the source and migration of pollutants into the water. Whitehead et al. (2009) believed that changes in rainfall intensity and frequency affected the amount of non-point-source pollutants introduced by runoff. The rainfall intensity determined the amount of pollutants entering the water body, and the rainfall frequency affected the degree of dilution of the pollutants. Water temperature affects lake eutrophication, the secondary release of pollutants from sediment (Lamon et al. 2009), and their migration and transformation, which influences the self-purification capacity of water. These effects will change the physical and biochemical characteristics of a water body (Arnell et al. 2015), but land use has a more direct impact on water quality conditions than climate change. The increase in urban pollutants and the use of agricultural fertilizers will have a major effect on water quality in the basin (Choi et al. 2021).

Climate change scenarios, land use change scenarios, and hydrological models are the key elements to study the response of runoff and nitrogen loadings to the changing environment. The impact of climate and land use changes on runoff and nitrogen loadings are obtained by controlling variables (Khoi et al. 2021, 2022): (1) There are two main ways to obtain future climate scenarios: one is by making reasonable assumptions about precipitation and temperature changes within a certain range, and the other is to use the global climate model (GCM) projections of the future climate (Reshmidevi et al. 2018). (2) Similarly, there are two ways to obtain future land use scenarios: one is to use the land prediction model (system dynamic model) to predict land use change in a specified area (Yu et al. 2011), while the other assumes that the land use change in the region represents the scenario of extreme land use. The actual lifestyle and policy implementation in various regions are different, so it is still controversial to use the land prediction model to generate the future land use situation. (3) Hydrological models are an important tool for simulating future runoff and nutrient loading changes. The Soil and Water Assessment Tool (SWAT) is a semi-distributed, physically based hydrological model that can be used to describe surface, root zone, groundwater, and stream processes (Arnold et al. 1998). The SWAT model is now widely used in many watersheds throughout the world to study the impact of climate change and land use on runoff (Yang et al. 2019; Hajihosseini et al. 2020) and water quality (Delia et al. 2021; Zango et al. 2022). However, its analytical ability in assessing the sensitivity of water quality to climate change and land use is discussed relatively scarcely in the existing studies.

Therefore, the main purpose of this study is to evaluate the response sensitivities of runoff and nitrogen loadings to precipitation, temperature, and land use changes quantitatively by the SWAT model in the middle of the Fenhe River watershed. We have divided the article into five parts: (1) the previous researches on the impact of climate and land use changes on runoff and nitrogen loading; (2) the introduction and scenario setting of the SWAT model; (3) the description and analysis of the simulation results; (4) the discussion of the simulation results and comparison with previous research results; and (5) the conclusion of the article.

The study area was located from 112 °15′E–113 °36′E to 37 °24′N–38 °18′N in the middle stream of the Fenhe River, which is the second largest tributary of the Yellow River, in Shanxi Province, Northern China. The length of the main stream is 54 km. From the entrance of the Lancun gauging station to the exit of the Erba gauging station, the watershed encompasses an area of 6325 km². The location and stations of the basin are shown in Figure 1.

The middle Fenhe River basin has a temperate continental monsoon climate with seasonal changes. The precipitation is affected by topography, with large changes in annual and inter-annual runoff. From the precipitation data of the 12 rain stations in the watershed, we determined the mean annual precipitation to be 493 mm, with large inter-annual changes. Rain mainly occurred from July to September and the average annual temperature was 11 °C. Because of the large proportion of urban land in the basin from Taiyuan City, which is the capital of Shanxi Province, the urban drainage caused difficulties for the simulation of runoff and nitrogen loading in the study area. Based on data from monitoring the nitrogen loading in four sections of the study area from 2016 to 2017, the main pollution sources and pollutants were determined by the equal pollution load method. The results suggested that nitrogen was the main pollutant, so this study mainly focused on simulating the migration of nitrogen as total nitrogen (TN) and ammonia nitrogen (NH₄-N).

The historically observed daily precipitation and monthly runoff data were available from 2008 to 2017. The nitrogen loading monitoring data of the Lancun and Erba sections and the point-source pollution data from the main urban sewage outlet were provided by the Hydrology Bureau of Shanxi Province. The daily meteorological data were downloaded from the China National Meteorological Information Center (https://data.cma.cn/). The DEM (digital elevation model) with 30 m precision, the land use map of the year 2018 with 30 m precision, and the soil type map with 1 km precision were obtained from the Data Center for Resources and Environmental Science, Chinese Academy of Sciences.

According to the three-level classification system for land use, the distribution of land types in the study area mainly included crop land, forest land, grassland, canals, water bodies, urban land, rural settlements, and unused land. After projection transformation, the distribution of land use in the study area is shown in Figure 2. In the study area, the primary land use types were cultivated land (30.6%), woodland (25.8%), and grassland (34.8%). The water area, urban land, rural land, and unused land accounted for 0.5, 4.7, 3.5, and 0.1%, respectively.

There are six soil types in the watershed: cinnamon soil (24.98%), loessal soil (45.64%), chisley soil (4.09%), acid skeletal soil (12.15%), moisture soil (12.86%), and paddy soil (0.28%). The soil database was established according to the book of Shanxi soil, the Chinese soil database, and the results of previous research on the watershed. The distribution of soil in the study area is shown in Figure 3.

The SWAT model is a process-based, semi-distributive catchment model, which can simulate the impacts of changes in runoff, sedimentation, pollution, climate, and land use (Molina-Navarro et al. 2014; Khatun et al. 2018). In the SWAT, runoff was simulated by the SCS curve-number method, ET was calculated by the Penman–Monteith method, the soil water process was computed by the dynamic water storage method and the groundwater flow included shallow and deep aquifer storage. Nutrients were simulated by the SWAT in the soil profile and shallow aquifer, nutrient routing into the main channel by surface flow and lateral subsurface flow that was then transported downstream with channel flow (Molina-Navarro et al. 2014).

Climate change can alter the hydrological cycle and runoff processes, which will change nitrogen loading and transport, and the degree of eutrophication. To predict and evaluate the hydrological cycle changes under climate change, the same land use conditions were employed, but with different climate variables set as simulation scenarios. Our research mainly focused on the impact of precipitation and temperature changes because these can directly influence water balance and are the most important factors in the hydrological cycle. Climate change scenarios included single changes and combinations of factor, as shown in Table 1. S0 represents the baseline scenario without climate and land use changes. Sixteen single factor change scenarios, including eight temperature change scenarios and eight precipitation change scenarios, were set up. The temperature change ranged from a reduction of 2 °C with an interval of 0.5 °C to an increase of 2 °C. The precipitation change ranged from a decrease of 10% with an interval of 2.5% to an increase of 10%. Furthermore, simultaneous changes in both temperature (±2 °C) and precipitation (±10%) constituted four combination scenarios. The calibrated and validated SWAT model was further forced by the climate change scenarios to determine the sensitivity of runoff and nitrogen loading to climate changes.

Land use influences the hydrological cycle by changing the underlying surface environment, which influences runoff, nitrogen transport, loading, and eutrophication. To investigate the impacts of different land use conditions on runoff and nutrient loads, a series of extreme land use situations were set as scenarios (Table 2). An extreme land use scenario means that the coverage for each type of land use is 100%. By changing the land use during 2008 to the extreme situation, the calibrated and validated SWAT model was driven by the land use change scenarios to analyze the sensitivity of runoff and nitrogen loading to land use changes.

To test the influence of runoff, 12 parameters were chosen to perform trial calculations for sensitivity analysis. The value of t-Stat, P-value, and sensitivity order of runoff parameters are shown in Table 3. The five most important parameters in order of sensitivity were SOL_K, CN2, ESCO, ALPHA_BF, and CH_N2.

RS4, BIOMIX, SOL_NO3, SOL_ORGN, NPERCO, and RCHRG_DP were chosen as the nitrogen loading parameters based on the calibrated runoff parameters. The ranking of sensitivity parameters for nitrogen loading are shown in Table 4. It can be seen that SOL_NO3 was the most sensitive parameter, with a remarkable influence on NH₄-N and TN loading. RS4 was more sensitive for NH₄-N loading, and SOL_ORGN was more sensitive for TN loading. These sensitivity parameters were further calibrated and validated.

The performance of the runoff simulations during the calibration (January 2008–December 2012) and validation (January 2013–December 2017) periods is shown in Figure 4. R², NSE, and Re were 0.84, 0.79, and −7.06%, respectively, during the calibration period, and 0.86, 0.79, and −2.8%, respectively, during the validation period. The runoff simulation performance demonstrates the validity of the established model for the simulation of nitrogen loading in the watershed.

Nitrogen loading parameters were calibrated and validated in 2016 and 2017 respectively, due to data limitations. For the NH₄-N loading simulation, R², NSE, and Re were 0.73, 0.53, and −0.83%, respectively, during the calibration period, and 0.79, 0.56, and −1.16%, respectively, during the validation period. For the TN simulation, R², NSE, and Re were 0.71, 0.54, and −2.34%, respectively, during the calibration period, and 0.78, 0.56, and 6.86%, respectively, during the validation period. Thus, the model showed a statistically acceptable performance for NH₄-N and TN loading (Figure 5).

The calibrated SWAT model was used to simulate runoff under the different scenarios (Table 1). The annual mean runoff under different single factor scenarios was compared with the baseline scenario (S0) in Figure 6. Results suggested that the increased temperature contributed to the downward trend of runoff, and that runoff decreased slowly when the temperature change range increased. Increase in precipitation contributed to the upward trend of runoff, and the runoff increased more as the precipitation change increased. Furthermore, it was found that the response of runoff to temperature and precipitation was reversed under a similar change range. By comparing the runoff change ratio between scenario T8 and T1, T7 and T2, T6 and T3, and T5 and T4, we showed that runoff was more sensitive to temperature decrease than to temperature increase. In comparison, the runoff change ratio between scenario P8 and P1, P7 and P2, P6 and P3, and P5 and P4 indicated that runoff was more sensitive to increased precipitation than to decreased precipitation.

The change ratios for runoff under the combined factor scenarios are shown in Figure 7. The runoff increased by 11.53% under scenario S21 (P+10%/T−2 °C), which showed the largest increase among all scenarios. Temperature increase and precipitation decrease resulted in decreased runoff, and runoff showed the largest decrease (7.22%) under scenario S12 (P−10%/T+2 °C). Comparing T1, P8, and S0, we found that the decreased range of runoff with a 2 °C temperature increase was smaller than the increased range of runoff with a 10% precipitation increase. Similarly, by comparing T8, P1, and S0, we showed that the increased range of runoff with a 2 °C temperature decrease was smaller than the decreased range of runoff when precipitation was decreased by 10%. Apparently, on consideration of the comprehensive effect, the runoff increased under S21 while it decreased under the S12 scenario.

The differences in runoff response to single factor and combination factor changes are compared in Table 5. Results suggested that the sensitivity of runoff to precipitation and temperature may be subject to the effect of combining factors. In the combined factor scenario, S21, for example, the difference in runoff change ratio between S21 and P8 (P+10%) plus T1 (T−2 °C) was the largest among all the scenarios. This indicated that runoff was more sensitive to precipitation increase and temperature decrease, which is similar to the results from the single factor scenarios. However, the difference in runoff change ratios from combined factor scenarios, S11 and S22, compared to the single factor effects was relatively small, which suggests that the combined factors effect was equal to the linear superposition of the single factor effects. In contrast, the runoff change ratios of the combined factor scenarios, S12 and S21, were larger than the sums of the single factor effect changes. The reason appears to be that the runoff is more sensitive to temperature decrease and precipitation increase; hence, the effects of the precipitation and temperature factors make the combined influence larger than for single factors.

The calibrated model was used to simulate runoff under different extreme land use scenarios and to compare the monthly average runoff from 2008 to 2017 (Figure 8(a)). It can be seen that the impact of different types of land use on runoff was significantly different. In the seven extreme (100% coverage) land use scenarios, runoff was reduced under cropland and water body scenarios. The reduction in runoff was greatest in the extreme water body scenario with a reduction of 41.11%. Runoff increased under the other extreme land use scenarios and was largest in the extreme urban land use scenario, with an amplification of 129.44%. Furthermore, the runoff changes of different land use types in the flood season (May–October) and non-flood season (January–April and November–December) were compared. The change in runoff in the flood season was significantly greater than that in the non-flood season, but the directions of change in the flood season and non-flood season were consistent; the flood season only played a role in amplifying the degree of change.

The runoff results under eight different land use scenarios, including baseline S0, are shown in Figure 8(b). Different land uses will increase or decrease runoff, but the changes with time under the eight land use scenarios were basically similar; the influence of land use on runoff was not instantaneously formed. It affects the water cycle process of the basin by changing multiple factors such as evaporation, infiltration, surface-water depth and soil moisture profile.

Nitrogen loading under single factor change scenarios of temperature and precipitation was compared with baseline, S0 (Figure 9). Nitrogen loading showed a negative correlation with temperature, decreasing as the temperature increased. In contrast, the precipitation and nitrogen loading were positively correlated in that the nitrogen loading increased when the precipitation increased. In addition, nitrogen loading was directly related to runoff, increasing as runoff increased; hence, increases in temperature and precipitation promote runoff and increased nitrogen loading. The responses of nitrogen loading to temperature changes were quantified and compared under the same temperature change range, as obtained from the nitrogen loading change ratios of T8–T1, T7–T2, T6–T3, and T5–T4 scenarios. Nitrogen loading was more sensitive to temperature decrease. Similarly, the responses of nitrogen loading to increases and decreases in precipitation were quantified and compared under the same precipitation change range by comparing the results of nitrogen loading change ratios under P8–P1, P7–P2, P6–P3, and P5–P4 scenarios. When the precipitation range was changed by ±5%, the response of nitrogen loading increased and decreased in relative accordance. When the precipitation range was changed by ±10%, the nitrogen loading was more sensitive to precipitation increase under the same precipitation change range.

The sensitivity of nitrogen loading under combined precipitation and temperature scenarios was estimated using the calibrated model and associated parameters (Figure 10). Under the combined precipitation and temperature scenarios, the response of nitrogen loading showed the same trend as for the single factor scenarios. The nitrogen loading decreased as the temperature increased and precipitation decreased. The largest increase in nitrogen loading occurred under S21 (P+10%/T−2), while the largest loading decrease occurred under S12 (P−10%/T+2). Comparing T8 and P8 with S0, the increase in nitrogen loading under a 10% precipitation increase was smaller than the nitrogen loading decrease under a 2 °C temperature increase. Moreover, the nitrogen loading was decreased under the combination scenario of S11 (P+10%/T+2). In addition, the comparison of the S22 and baseline scenarios showed that the decrease in nitrogen loading as precipitation was decreased by 10%, smaller than the increase in nitrogen loading due to a temperature decrease of 2 °C. For the combination of precipitation and temperature, S22 (P−10%/T−2), the nitrogen loading increased. The nitrogen loading of P1 and P8 was compared with T8 and T1, respectively. It was found that nitrogen loading was significantly influenced more by temperature changes of ±2 °C than precipitation changes of ±10%. A comparison of Figures 7–9 with Figures 7–10 shows that NH₄-N loading is more sensitive to precipitation and temperature than TN, and the degree of change of NH₄-N loading is larger than TN.

The variation in nitrogen loading was different under single factor and combined factor scenarios, and the sensitivity of the nitrogen loading response to precipitation and temperature affected the response to the combined factors. The effect of precipitation and temperature are additive or subtractive (Table 6). Nitrogen loading is more sensitive to temperature change than to variation in precipitation, which differs from the results in the case of runoff. Table 6 shows that the effects of precipitation and temperature on nitrogen loading are in the same direction under the S21 and S12 scenarios. The influence of combined factors was larger than for single factors. For example, the nitrogen loading change ratios were 14.47% for S21 and −13.72% for S12 under the combined factor scenarios and 13.69 and −12.55% under single factor scenarios; hence, the interaction of precipitation and temperature can magnify the combined influence. When the effects of precipitation and temperature were in different directions, the nitrogen loading change ratios were −7.81% (S11) and 8.61% (S22) due to the combined influence of the factors.

The degree of nitrogen loading under different extreme land use scenarios was estimated by using a calibrated model and appropriate parameters (Figure 11), and NH₄-N and TN showed similar sensitivity under the different land use scenarios. NH₄-N and TN decreased only in extreme waters, where NH₄-N decreased by 68.21% and TN decreased by 65.88%. In the cropland, forest, grassland, and unused land scenarios, NH₄-N and TN increased, but the amplification was within 10% and not significant. In the extreme urbanization case, NH₄-N and TN had the largest increase in the land use simulation scenario, in which NH₄-N increased by 101.00% and TN increased by 93.71%. With extreme rural land development, NH₄-N increased by 44.07% and TN increased by 40.72%. The change of nitrogen loading between flood season and non-flood season was also compared in the simulation. The degree of nitrogen loading in flood season was significantly greater than that in non-flood season. The simulation results for the seven kinds of land use showed that the change range of NH₄-N was slightly higher than that of TN, indicating that NH₄-N was more sensitive to land use than TN.

The migration of NH₄-N and TN with time under eight different land use scenarios (including baseline, S0) is shown in Figure 12, and the changes in NH₄-N and TN were basically the same under the different land use scenarios. The eight curves representing the migration process of NH₄-N under different land uses are basically similar, showing differences only from June to August in 2016. The three curves representing cultivated land, grassland, and unused land showed some fluctuation, which may be related to the species and growth of plants, but the results for TN are similar to those for NH₄-N.

In this study, the simulation accuracy of nitrogen loading is always lower than that of runoff, because the emission data of nitrogen loading can only be measured in several fixed locations, which may not include all sources of nitrogen pollution in the basin. Furthermore, the total amount of nitrogen loading is obtained rather than the concentration in the SWAT model; hence, the error of runoff simulation will further amplify the simulation error of nitrogen loading (Shrestha et al. 2012). For these reasons, the results of nitrogen loading simulation need to be further optimized.

The simulated responses of runoff to climate change in this paper are consistent with the conclusions of many scholars (Qi et al. 2009; Shang et al. 2019). The influence of temperature on runoff is due to the relationship among temperature, precipitation, and evaporation (Findell et al. 2011; Oueslati et al. 2017), which is not analyzed in this article; the influence of precipitation on runoff is reflected by factors such as evaporation and infiltration in the basin (Liu et al. 2008). On the other hand, the impact of urbanization on runoff was consistent with other published reports in that increasing urban and rural land use increased runoff (Marhaento et al. 2017; Zhang et al. 2018). This may be due to an increase in impervious areas, resulting in a decrease in infiltration in the hydrological cycle. Our results on the influence of cropland, forestland, and grassland on runoff are similar to those of some scholars (Shang et al. 2019), but others reported the opposite effect (Molina-Navarro et al. 2014; Wang et al. 2017). This seeming contradiction may be related to the different geographical locations of the study areas, which have a significant impact on runoff generation and confluence processes.

At present, many scholars have reached conclusions similar to ours, that elevated temperature and reduced rainfall will reduce nitrogen loading (Fan & Shibata 2015; Shrestha et al. 2018). The sensitivity of nitrogen loading to temperature is more significant than to precipitation, because the intensity of nitrification and denitrification decreases as the temperature decreases and lead to the nitrogen load increases. Temperature and precipitation do not impact nitrogen loadings independently, so the combined effects will be amplified or reduced with their internal interaction. In addition, there have been more research and general conclusions about the effects of urbanization on nitrogen loading (Chang 2008; Tu 2009). The effects of land use on nitrogen loading are not only reflected in the change of nitrogen loading when runoff changes, but also in the influence of different land use types. Changes in human activities will also affect nitrogen loading. For example, in the simulation of the extreme cropland scenario (L1), factors, such as crop types and farming methods, need to be considered together (Khan et al. 2021), and in the context of extreme urbanization, urban development types and greening situations must be comprehensively examined. The results varied because of the complexity of the situation. Nitrogen loading is also associated with sediment transport rates, since both outcomes are closely related to soil erosion (Marshall & Randhir 2008), so one future research goal will be to investigate the impact of sediment transport on nitrogen loading.

There are some areas to be further studied in this article. There are uncertainties in the simulation results caused by the uncertainty of the model parameters, model structure, input data (He et al. 2019; Karakouzian et al. 2020) climate change, and land use change (Kay et al. 2008; Chang et al. 2017). How these uncertainties impact runoff and nitrogen loadings need further study. Furthermore, the impact of climate change and land use on runoff and nitrogen loading is complex and compound, which includes many processes such as evaporation, infiltration, and water vapor transport. Future studies may get better results if more elements are considered for comprehensive analysis.

This study evaluated the response sensitivity of runoff and nitrogen loading to climate change and land use change by establishing 20 climate change scenarios and seven extreme land use scenarios, using a calibrated SWAT model by the SWAT-CUP. The R² and NSE were greater than 0.84 and 0.79 for runoff simulation and greater than 0.71 and 0.53 for nitrogen loading simulation; the absolute value of Re was less than 7.06% for runoff and less than 6.86% for nitrogen loading. Thus, the simulation performance of the SWAT fulfilled the requirements of runoff and nitrogen loading sensitivity analysis. The method proposed in this article is also applicable to other regions, especially in the basin where nitrogen loading is the main pollution factor. For the choice of water quality indicators, it should be judged according to the development of the region, to make a meaningful judgment on regional development.

From the standpoint of temperature change, runoff and nitrogen loading were more sensitive to temperature decrease than temperature increase, but the opposite was true for the effects of precipitation change. Compared with temperature changes of ±2 °C, the response of runoff to precipitation changes of ±10% was greater, but the effects on nitrogen loading were the opposite. When the influence of precipitation and temperature was in the same direction (both increase or both decrease), the interaction between the two factors further enhanced the combined effect, while the combined impact was weaker than the single factor effect when the direction of influence of these two factors was opposite. In simulations comparing the effects of the seven extreme land use scenarios with baseline, the runoff and nitrogen loading decreased the most in the extreme water body scenario (runoff decreased by 41.11% and nitrogen loading decreased by more than 65%); in the extreme urban scenario, runoff and nitrogen loading increased the most (runoff increased by 129.44% and nitrogen loading increased by more than 90%). Runoff increased under five scenarios and decreased under two scenarios, and nitrogen loading increased under six scenarios and decreased under one scenario. Runoff and nitrogen loading were more sensitive to land use in flood season than in non-flood season.

The effectiveness of the SWAT model for runoff and nitrogen loading simulation has been further verified. The response sensitivity of runoff and nitrogen loading under different climate and land use conditions will provide useful guidance for future water resource management.

This study was sponsored by the Natural Science Foundation of Shanxi Province, China (201901D111060), the National Key Research and Development Program of China (2019YFC0408601), and the Project of Shanxi Provincial Department of Water Resources, China (202125033). The authors would like to thank the editors and reviewers for their valuable comments and suggestions.

The authors declare there is no conflict.

Data cannot be made publicly available; readers should contact the corresponding author for details.