Results

Model calibration and validation results

The raster-based Xin’anjiang model captured the seasonal trend of discharge and extreme flows in heavy rainfall events (Fig. 2). Discharge in Hengtangcun was well simulated with a correlation coefficient (R2 ) and Nash-Sutcliffe efficiency coefficient (NS ) value of 0.90 and 0.81 in calibration period and withR2 and NS value of 0.93 and 0.81 in validation period respectively. Several discharge peaks (e.g., on Aug. 10, 2009 and Jun. 14, 2011) were well described.
Fig. 2 Daily observed and simulated discharge at Hengtangcun station during model calibration (2009-2011) and validation periods (2012).
Simulated water level from NDP model has been verified using data from a typical polder close to the study area, which performances were acceptable with the NS value of 0.73 and 0.50 for the calibration and validation periods (Huang et al., 2018a). More details can be found in the Supporting Information.

The different hydrological responses to climate change and land use between mountain and lowland artificial watersheds

We compared the simulated hydrological processes between mountain and lowland artificial watersheds under current and each future climatic and land use conditions. To quantify the influence of future climate change, we changed temperature and precipitation parameters corresponding to climate change scenarios from CMIP6 for the 2030s, 2050s, 2070s and 2090s, and kept the other parameters of the two models unchanged. To quantify the influence of future land use, we evaluated discharge variations for the 2050s under the developed land use scenarios with the current change rate, using the identical climate data from 2009 to 2012.

Hydrological responses to climate variability

Monthly discharge and evaporation calculated from the 2009-2012 basic data were used to compare the different hydrological responses to climate variability between mountain and lowland artificial watersheds. Fig. 3(a) showed the slope (4.75) of evaporation-temperature linear correlation in lowland watersheds was larger than in mountain watersheds (2.61), accounting for the same growth in temperature would cause larger evaporation increment in lowland. Fig. 3(b) showed anR2 of 0.65 in mountain watersheds was larger than that in lowland watersheds (0.12). This may because precipitation was the unique water resource in mountain watersheds while irrigation also controlled discharge in lowland artificial watersheds.
Fig. 3 The linear correlation of evaporation-temperature (a) and discharge-precipitation (b) in mountain watersheds (blue points) and lowland artificial watersheds (orange points).
In case that temperature and precipitation changed simultaneously in future climate change scenarios, the mean annual runoff change was more significant in mountain (10~200%) than in lowland (10~60%) (Fig. 4). In terms of season, autumn and winter were significantly related to the increment in mountain watersheds, and summer and autumn (rice seasons) had a largest contribution in lowland watersheds. The R2between precipitation variability and mean annual runoff change was 0.78 in mountain, comparing to 0.96 in lowland. TheR2 between annual average temperature and runoff variability was negligible (0.05) in mountain comparing with that (0.33) in lowland. Therefore, the dominated factor for runoff variation under climate change scenarios was precipitation.
Fig. 4 The annual and seasonal runoff variation under future climate scenarios. RCP2.6_20, RCP4.5_20, RCP8.5_20 were the simulations for the 2030s; RCP2.6_40, RCP4.5_40, RCP8.5_40 were the simulations for the 2050s; RCP2.6_60, RCP4.5_60, RCP8.5_60 were the simulations for the 2070s; RCP2.6_80, RCP4.5_80, RCP8.5_80 were the simulations for the 2090s.

Hydrological responses to land use

In order to learn the hydrological processes response to the different land use conditions, we subdivide the mountain watersheds into several sub-watersheds based on DEM, and the lowland artificial watersheds into several polders based on satellite images. Different metrics and thresholds have been used to characterize discharge regimes (Berihun et al., 2019). The annual mean runoff coefficient (α) is the average value of the ratio of annual runoff depth to annual precipitation depth. In lowland artificial watersheds, discharge is the total amount of seepage, culvert drainage and flood drainage, excluding the inflowing irrigation. A higher value of α correlates with the difficulty in absorption of precipitation by soil. The coefficient of variation (CV) is the ratio of standard deviation to average value. Here it estimates the variation extent of daily discharge in each sub-watershed (polder) as results of different land use conditions.
It can be found that α in mountain watersheds (0.41) was generally larger than that in lowland watersheds (0.36). It had a relatively weak negative correlation with the ratio of cultivate land in mountain (R2 =-0.28), compare to the relatively strong negative correlation in lowland (R2 =-0.85). In addition, the CV of discharge was generally larger in lowland watersheds (2.39) than in mountain watersheds (1.02), which was positively related to the ratio of cultivate land in the former (R2 =0.55) while negatively related to the ratio in the latter (R2 =-0.37). This is probably because in lowland artificial watersheds, the surface runoff flowing in or out of the cultivate land were controlled by pumping stations, generating sharper discharge hydrograph with higher peak-values and lower minimum-values (Yan et al., 2018). In terms of the ratio of water area, the CV of discharge had a negative relation (R2 =-0.83) with it in lowland watersheds comparing to non-significance in mountain watersheds, partly accounting for the capacity of regulating flow of ponds in polders. With respect to the ratio of residential area, we found that α having an extremely strong positive correlation (R2 =0.99) with it in lowland. The other result like α and CV value of discharge in mountain watersheds were positively correlated to the slope (R2 =0.73 and 0.82 respectively), implying that sloping regions more likely lead to the productions and variations of surface runoff (Fig. 5).
Fig. 5 The description of discharge characteristics and the land use conditions in mountain watersheds and lowland artificial watersheds respectively (Left); The different responds of surface runoff to the ratio of land use types and slope classes between mountain watersheds and lowland artificial watersheds (Right).
Note: α: the annual mean runoff coefficient; CV: the coefficient of variation of surface runoff; Farm: cultivated land; W: water area; R: residential area; For: forest land; G: grass land.
Monthly and annual runoff variations between two watersheds were compared to evaluate the different effects of future land use conversions. For the 2050s, keeping the current rate of land use change, the most dramatic variations of annual runoff in mountain watersheds were found under the scenarios of converting cultivate land, forestland or grassland into residential area (increasing 7.8%, 3.5% and 1.7% respectively). Moreover, under the above three scenarios, seasonal runoff showed increasing trend especially in May. The most significant changing of annual runoff in lowland artificial watersheds was found in the scenarios of converting cultivate land into residential area or into water area (increasing 22.0% and 2.1% respectively). The seasonal runoff showed a dramatic increase except in winter (from Dec. to Feb.) when converting cultivate land into residential area. It showed a slight increasing trend except in summer when converting cultivate land into water area because of the reduced irrigation. Overall, the effects of land use change on runoff were generally more significant in lowland artificial watersheds than that in mountain watersheds (Fig. 6).
Fig. 6 The different effects of land use conversions on monthly (a, b) and annual (c, d) runoff between mountain watersheds (a, c) and lowland artificial watersheds (b, d). The value of (c) and (d) represents annual runoff in X axis to that in Y axis.
Note: Farm: cultivated land; For: forest land; G: grass land; R: residential area; W: surface water area.

Water balance at mountain and lowland artificial watersheds under future scenarios

Land use change scenarios were generated along the same storylines as climate change scenarios based on RCP 2.6 for the 2050s to assess the additive impacts of the two stressors. In mountain watersheds, climate changes itself led to an increase by 0.7 mm and 312.9‬ mm in annual evaporation and annual runoff respectively, probably due to the increment in rainfall. When combined with land use change, the water balance components varied slightly on the whole. Therefore, water balance in mountain watersheds were more sensitive to climate change. The land use change had weak enhancing effects on climate impacts (Mountain: S1, S2, S3 in Fig. 7).
In lowland artificial watersheds, the water input was larger than water output in summer, opposite to autumn, which related to the special water management and drainage rhythm of polders. As to evaporation component, it sharply declined by 266.6 mm when combining the effect of land use change, mainly due to the absence of rice evapotranspiration (Evaporation: S1, S2 in Fig. 7). Irrigation showed‬ a decrease of 77.5 mm when rising 24.4% of annual precipitation (Irrigation: S0, S1 in Fig. 7), and decline became more apparent with the cultivate land shrinking (Irrigation: S1, S2 in Fig. 7). Pump drainage notably increased by 213.8 mm in response to the increment in precipitation (Pump: S0, S1 in Fig. 7), and when combining land use change, the capacity of retaining water of increased ponds can mitigate 57.2 mm of flood discharge (Pump: S1, S2 in Fig. 7). Culvert drainage increased by 25.8 mm with precipitation (Culvert: S0, S1 in Fig. 7). Moreover, cultivate land shrinking was direct to the reduction of water holding, consequently the culvert drainage increased another 111.7 mm (Culvert: S1, S2 in Fig. 7). Seepage increased abruptly by 138.6 mm under future land use change scenarios (Seepage: S1, S2 in Fig. 7). In summary, the land use change had strong combined effects on climate factors in lowland artificial watersheds.
Fig. 7 Water balance simulations of mountain and lowland artificial watersheds under climate change and land use scenarios for the 2050s. S0: Historical climate scenario + a baseline land use scenario. S1: RCP2.6 climate scenario + a baseline land use scenario. S2: RCP2.6 climate scenario + developed land use scenario for the 2050s.