1. Introduction
Nitrogen is a common element in nature, and controls the processes, functions, and diversity of ecosystems [
1]. As a result of the industrial revolution and human activities, new sources of nitrogen, such as fertilizer application, crop cultivation, nitrogen deposition, and food inputs, have had a significant impact on nitrogen cycling at the global and the regional level [
2,
3,
4,
5]; in some cases, natural nitrogen fixation is exceeded [
6]. However, when nitrogen inputs exceed the nitrogen uptake and storage capacity of regional ecosystems, the remaining nitrogen enters watersheds, with long-term negative impacts [
7,
8], such as groundwater contamination, soil acidification, biodiversity decline, water eutrophication, and hypoxia [
9]. The increasing nitrogen concentrations and fluxes in watersheds have become major factors contributing to the eutrophication of lakeshores, estuaries, and coastal areas [
4,
5,
6,
7,
8,
9].
Driven by resource scarcity, population expansion, and economic growth, human activities greatly interfere with the natural nitrogen cycle [
10], and it has been shown that the nitrogen input from human activities has exceeded the natural nitrogen fixation in most river watersheds in China, such as the watershed of the Yangtze River and the watershed of the Pearl River [
11]. For example, in 2010, 61.6% of the 26 state-controlled key lakes (reservoirs) in China had a water quality of V (the lowest grade), with total nitrogen being one of the main pollutants; in addition, inorganic nitrogen has become one of the common pollution indicators in the major sea areas of China [
12]. Overall, excessive nitrogen input from human activities has become an important factor restricting environmental sustainability in China.
The high load of anthropogenic nitrogen input has attracted widespread attention from researchers, organizations, and the international scientific community, to study the anthropogenic biogeochemical cycling of nitrogen for sustainable development [
13,
14]. Quantitative source analysis of nitrogen in watershed ecosystems has become an international research hotspot, guiding studies on the optimization of nitrogen management in ecosystems at the watershed scale [
15,
16].
The most widely used model for accounting for anthropogenic nitrogen inputs into watersheds is the NANI model proposed by [
17] and consisting of four components: atmospheric nitrogen deposition, nitrogen fertilizer consumption, crop nitrogen fixation, and food/feed imports, which represent the main types of nitrogen inputs due to human activities. The calculation results can be used to characterize the impacts of nitrogen produced by human activities on watersheds. The NANI model and its improved methods are widely used at various watershed and regional scales in the U.S., Europe, and Asia [
18,
19,
20,
21]. Han et al. estimated NANI for 99% of the world from 1961 to 2009 and suggested that the global NANI levels dramatically increased due to higher net nitrogen inputs in Asia; furthermore, the NANI values in Asia, Europe, and North America were above average [
2]. To calculate NANI across United States watersheds, a consistent method was proposed and showed that yield-based estimation of NANI differed significantly from area-based estimation [
18]. Hong et al. investigated regional variation parameters for the NANI Calculator Toolbox and observed significant regional variation in NANI across the Baltic Sea Basin, where the associated relationship with riverine nitrogen fluxes suggested that NANI contributed to a portion of the riverine nutrient fluxes [
19]. Swaney et al. calculated NANI in Indian watersheds, showing that agricultural fertilizer is the main source of NANI, followed by crop nitrogen fixation, but the regression relationship between NANI and riverine nitrogen fluxes is problematic, due to the limited availability of river data [
20]. A study about the spatiotemporal differences of NANI in mainland China from 1981 to 2009 revealed that the NANI had more than doubled [
21].
To guide nitrogen management at the macroscopic scale, assessing the impacts of human activities on nitrogen inputs into watersheds is of great significance. The indicators involved in the NANI accounting model include the urban population, the rural population, the number of livestock, crop yield, nitrogen fertilizer consumption, compound fertilizer use, planting area, and energy consumption, among others, with a total of 45 indicators [
22]. Data for these indicators are usually obtained from local or national statistical yearbooks. At longer time scales, the data usually have heavy missingness because of changes in the accounting of statistical yearbooks or omissions in data aggregation. The large number of indicators involved increases the difficulty of NANI accounting, and the missing data make NANI accounting less accurate. In this paper, we attempt to build a NANI prediction HSVC model based on as few predictor variables as possible for the Yangtze River Basin, China.The HSVC model is a generalization of the variable coefficient model, aiming to determine the spatial variations of data by allowing the coefficients to be functions of location. Since the model has obvious indigenous effects on analyzing spatial heterogeneity, and the coefficient function has a strong flexibility, it has been widely used in environmental science, ecology, and epidemiology [
23,
24,
25,
26].
In this study, a hierarchical spatially varying coefficient process regression (HSVC) model was built to elaborate the mechanisms of the predictor variables per gross domestic product (PGDP) and population density (PD) for NANI, and to forecast the annual values of NANI within the watershed of the Yangtze River. The main objectives were as follows: (1) to analyze the spatiotemporal variations of NANI, PGDP, and PD; (2) to explore the relationship between watershed NANI, PGDP, and PD, and to compare the fitting and prediction accuracy of HSVC, GP, and DLM models; (3) to predict the values and 95% interval predictions of NANI for the years 2025 and 2030. This study not only provides a simple and easy-to-use method for NANI prediction but also provides a watershed nitrogen management strategy, which is useful for nitrogen pollution control.
2. Study Area and Data Sources
The Yangtze River Basin, displayed in
Figure 1, is located between 90°–122° E and 24°–35° N, covering an area of approximately 1.8 million km
2 [
27]. It originates southwest of the Tang-gu-la Mountains on the Qinghai-Tibet Plateau and flows through 11 provinces and autonomous regions, flowing into the East China Sea in Shanghai. According to the characteristics of the Yangtze River basin, the watershed of the Yangtze River is divided into 11 sub-basins; namely, Jinshajiang (JSJ), Mingtuojiang (MTJ), Wujiang (WJ), the Upper mainstream region (UM), Jialingjiang (JLJ), Dongtinghu (DTH), the Middle mainstream region (MM), Hanjiang (HJ), Poyanghu (PYH), the lower mainstream region (LM), and Taihu (TH) [
22]. These sub-basins range in area from approximately 36,900 km
2 (TH) to 483,000 km
2 (JSJ). The map of the Yangtze River Basin and its subcatchments was created using ArcGIS 10.2.
In previous research, Cui et al. [
28], studied the linear relationship between NANI, per capita GDP (PGDP), and population density (PD). In addition, the correlation coefficients between PGDP and PD were below 0.7, and we did not consider collinearity, as per reference [
29]. Therefore, we directly adopted PGDP and PD as explanatory variables. The data for modeling were derived from a published master’s thesis [
22], including the response variable NANI and its four components; the predictor variables GDP; and the population in each subcatchment for the years 1980, 1985, 1990, 1995, 2000, 2005, 2010, and 2012, with a total of 88 samples. To ensure normality, NANI was modeled on the square root scale, PGDP and PD were centralized. Model fitting was performed using data between 1980 and 2010; for validation, data from 2012 were used.