Spatial and seasonal variations of hydrological responses to climate and land-use changes in a highly urbanized basin of Southeastern China

Climate and land-use changes are two major factors that significantly affect the watershed hydrology cycle. It is essential for regional water resource management to quantitatively assess the respective hydrological impact of these two factors. In this study, the Soil and Water Assessment Tool (SWAT) was constructed to quantify the contributions of climate and land-use changes to runoff at the annual and seasonal time scales in the Qinhuai River basin (QRB), where significant urbanization occurred from 1986 to 2015. Moreover, based on the partial least squares regression, the specific impact of individual land-use change on major hydrological components was evaluated at the subbasin scale. The results showed that: (1) the predominant patterns of land-use change in the QRB included the transformations from paddy fields to urban areas and dry lands, forest to dry lands and dry lands to urban areas; (2) the flood seasonal precipitation series and all air temperature series had significant increasing trends over 1986–2015, and annual and seasonal runoff series had significant increasing trends and had an abrupt change point in 2001; (3) the average annual, flood seasonal, and non-flood seasonal runoff increased 238.5, 130.2 and 108.3 mm, of which land-use change was responsible for 77.6, 55.1, and 104.8% of the increases, respectively, while climate change was responsible for 22.4, 44.9, and 4.8%, respectively and (4) the hydrological response to land-use change showed an obvious decrease in actual evapotranspiration (ET) and significant increases in surface runoff and baseflow. The decrease of ET and increase of baseflow could be attributed to the conversion patterns from paddy fields and forest to dry lands, while the conversions from paddy fields and dry lands to urban areas caused a remarkable increase in surface runoff in the QRB. The study demonstrated that these practicable approaches were beneficial for the more unbiased views of the hydrological responses to climate change and land-use changes in the highly urbanized basin, which were also critical for the sustainable development of regional water resource and future landuse planning.

Some recent studies showed that land-use change had large impacts on the hydrological process in the highly urbanized basin (Hao et al. ; Kalantari et al. ; Wang et al. b). These studies revealed that land-use change led to decreasing ET and increasing runoff, especially as impervious areas consistently expanded. However, few studies evaluated the specific effects of changes in individual land-use classes on the hydrological process for a highly urbanized basin. There are still valid research questions as to how mutual conversion between several land-use types influences each hydrological component. Without such accurate assessment, the impacts of each land-use change pattern on hydrological components may be underestimated, overestimated, or even misunderstood (Nie et al. ). In view of this deficiency, this study applied the hydrological model to quantitatively assess the runoff response to climate and land-use changes, and further accurately investigated the contribution of each transformation pattern of land use to changes in major hydrological components combining the partial least squares regression (PLSR) analysis.
This accurate assessment will improve the predictability of hydrological consequences of land-use changes and is thus crucial for future land-use and water resource planning and management for the highly urbanized basin.
The objectives of this study are to quantify hydrological responses to climate and land-use changes in the QRB, a highly urbanized basin, and further evaluate the specific effects of variations in individual land-use type on major hydrological components. To achieve these objectives, temporal change trends of the hydro-meteorological series from 1986 to 2015 and a land-use change matrix were assessed.
Then, the hydrological process was simulated to quantitatively isolate the individual contribution of climate and land-use changes on annual and seasonal runoff based on the calibrated SWAT model. Finally, the spatial hydrological responses to each transformation pattern of land use were assessed by the PLSR.

DATA SOURCES AND METHODOLOGY
The Qinhuai River basin The Qinhuai River is a typical tributary of the lower Yangtze River. It goes through Nanjing city and runs into the Yangtze River at the Wudingmeng sluice and the Inner Qinhuai sluice (Figure 1). The QRB is located between 118 39 0 E to 119 19 0 E, and 31 34 0 N to 32 10 0 N with a length of 36.6 km and a drainage area of 2,631 km 2 . The QRB is located in a semi-humid area with a subtropical monsoon climate. The average annual precipitation is 1,116 mm, which is concentrated in the flood season from June to September. The mean annual air temperature is 15.4 C.
The observed long-term  mean annual runoff depth is about 448.4 mm. Over the years, the Qinhuai River, called 'the mother river of Nanjing', has played an important role in ecosystem services, including drought and flood prevention and crop irrigation.

Data sources
The data of this study mainly consisted of meteorological and hydrological data, and topography, soil, and land-use data. The meteorological data were acquired from two standard meteorological stations between 1986 and 2015 at the daily scale, and daily rainfall data were collected from the other five rain-gauge stations across the QRB (Figure 1).
The series data of potential evapotranspiration (PET) were calculated using the FAO Penman-Monteith method

Methodology
In this study, three approaches were combined to quantitatively assess the hydrological response to climate and landuse changes in the study basin. The approaches included the trend and abrupt change test, the SWAT hydrological model, and the PLSR method.

The trend and abrupt change test
The Mann-Kendall trend (M-K) test is a non-parametric method and has been recommended by the World Meteorological Organization to detect monotonic trends of hydro-meteorological variables such as runoff, precipitation, air temperature, and PET (Chebana et al. ). The statistic S and standardized test value Z of the M-K test are calculated following Zuo et al. (). If the value of Z is greater than or equal to 2.56, the trend is significant at the significance of 0.01; if the value of Z is greater than or equal to 1.96, the trend is significant at the significance of 0.05. In addition, the trend slope β is generally jointed with the M-K test to explain the change in long-term time series. A positive value of β indicates an upward trend, whereas a negative value represents a downward trend.
The abrupt change is one of the important external manifestations of climate change and hydrological process variability, and mainly reflects the occurrence of extreme events or severe anthropogenic activities (Nayak & Villarini ). Hence, detecting the abrupt change point is considered as the key step to be tackled in hydrological timeseries analysis (Xie et al. ). The non-parametric Pettitt test was applied to determine the change point in this study, which can quantify the statistically significant level of change points for varying meteorological and hydrological series. This method uses the Mann-Whitney statistic function U t, N and considers the samples x 1 , …, x t and x tþ1 , …, x N to be independently and identically distributed.
N represents the sample size. U t , N is calculated as follows: If time t is satisfied by K(t) ¼ max(U t,N ), then t represents an abrupt change point and the formula for the significant level of P of the change point is as follows: If P 0.05, then detected abrupt change point is considered to be statistically significant.

The SWAT model
The SWAT is a physically based, semi-distributed and con- In addition, the procedures of parameter calibration, verification, and sensitivity analysis in the SWAT model can be conducted by SWAT Calibration and Uncertainty Programs (SWAT-CUP). The sensitivity analysis of the parameters is determined by the t-statistics and the p-value. A larger absolute value of the t-statistic represents greater sensitivity of the parameters, and the p-value, closer to zero, is more sensitive.
Furthermore, the performances of the SWAT model are evaluated by the coefficient of determination (R 2 ), the Nash-Sutcliffe coefficient, and percent bias (PBIAS).

SWAT model application
The impacts of climate and land-use changes on hydrology were quantitatively separated by scenario analysis based on the SWAT output of four scenarios (Table 1). The study period was divided into two time slices: 1986-2001 was con- the difference between Q S1 and Q S2 was the effect of the climate change on runoff. Similarly, the difference between the third simulation (Q S3 ) and fourth simulation (Q S4 ) could also be considered as the effect of the climate change on runoff.
Therefore, the impacts of climate change on runoff were calculated by the following formula: Furthermore, the effects of land-use change on runoff can be determined by applying the difference between the first and third simulations, or between the second and fourth simulations, as in the following formula: The combined effects of climate and land-use changes on runoff were equal to the sum of the individual impacts.
The total changes were recorded: Hence, the percentage contributions of climate and land-use changes in the variations in runoff were calculated as follows: i.e. evapotranspiration, surface flow, and baseflow. Two main indices, goodness of fit (R 2 ) and goodness of prediction (R 2 cross ), are calculated to assess the validity and strength of the PLSR model. If the R 2 value is greater than 0.8 and the R 2 cross value is greater than 0.5, the PLSR model has a good predictive ability.
In the PLSR model, the variable importance of the projection (VIP) and regression coefficients (RCs) were used to explain the relative importance of each independent variable.
It is thus possible to determine which land-use change pattern most strongly interacts with the hydrological components. The     Table 3.
The performance of the SWAT model in the QRB is shown in Table 4 and Figure 5. There was good agreement between the observed and simulated runoff at the monthly level for both the calibration and validation periods. As shown in Table 4, the coefficient of determination (R 2 ) and the Nash-Sutcliffe coefficient (ENS) coefficient were higher than 0.85 in the calibration and        sub-basins of the QRB, as shown in Figure 6. In addition, the corresponding changes of primary hydrological components for each sub-basin under the fixed 1986-2015 climate condition were calculated based on the SWAT model, as shown in Figure 7.
A summary of the PLSR is provided in Tables 7 and 8.
The PLSR models were developed for annual and seasonal different response variables: actual evapotranspiration as ET, surface runoff as SQ, and baseflow as BF. These PLSR models could be considered to have a good predictive power due to the R 2 value being greater than 0.8 and the R 2 cross value being greater than 0.5 (Table 7). Each PLSR model included two significant components based on the cross-validation for response variables (Table 7). In addition, the VIP values of the PLSR models indicated the comprehensive importance of each pattern of land-use change for hydrological components. The predictors with VIP values more than 0.9 were considered to be of great importance for predictions. For the annual ET model, a higher VIP value was observed for the pattern of transformation from forest to dry lands (VIP ¼ 1.45), followed by transformation from paddy fields to dry lands (VIP ¼ 1.02). Specifically, the RC of forest to dry lands was À0.46, and the RC of paddy fields to dry lands was À0.32. The seasonal ET model had the same results as the annual ET model, which indicated that the decreases of ET could be attributed to the transformations of paddy fields and forest to dry lands during the past 30 years in the QRB. For annual, flood seasonal and nonflood seasonal SQ models, the more important predictors were paddy fields to urban areas (VIP ¼ 1.28, 1.31, and 1.29, respectively) and dry lands to urban areas (VIP ¼ 0.92,1.29,and 1.30,respectively). The patterns of paddy fields to urban areas and dry lands to urban areas had positive RCs (Table 7). As expected, transformation from paddy   lands not only decreases the ET, but also allows infiltration and groundwater exchange. In this way, it is credible that replacing paddy fields and forest with dry lands plays a key role in the increase of baseflow during urbanization in the QRB. Future studies should examine the mechanism of baseflow change during the urbanization to confirm the influence of the dry land expansion on baseflow and groundwater change, which will be explored to enhance our understanding of hydrological response to land-use change in a highly urbanized basin.

CONCLUSIONS
This study quantified the effects of climate and land-use changes on annual and seasonal runoff in the QRB, a highly urbanized region in the Yangtze River Delta. The good performance of the SWAT model indicated that the SWAT model could be used to evaluate the contributions of climate and land-use changes to runoff change at annual and seasonal time scales by simulating hydrological processes of the study basin. Moreover, the PLSR method was adopted to evaluate the influence of the individual pattern of land-use change on major hydrological components. From the results of this study, the main conclusions are as follows: (1) The predominant directions of land-use change were converting paddy fields and forest to urban areas and dry lands. The diminishing portion of paddy fields (decrease of 13.9%) and forest (decrease of 7%) contributed to the expansion of urban areas (increase of 9.3%) and dry lands (increase of 9.1%) over the study period.
(2) The results of the M-K test indicated that the flood sea- respectively. In addition, the contributions of land-use