UNCERTAINTY OF THE WATER RESOURCES ASSESSMENT

IN THE YELLOWRIVER BASIN

DAWEN YANG1, 3, CHONG LI2, GUANGHENG NI3 AND HEPING HU3

1Department of Civil Engineering, University of Tokyo, Tokyo 113-8656, Japan

2China Institute of Water Resources and Hydropower research, Beijing, China

3Department of Hydraulic Engineering, TsinghuaUniversity, Beijing 100084, China

For assessing the water resources variability under long-term changes of climate and land cover, a distributed, physically based hydrological model is necessary since it can represent the spatial distribution of related river basin properties and can examine the impact of local land cover change on the basin hydrological cycle. Before the quantitative assessment of water resources, it is important to understand the possible uncertainties in the hydrological simulation. The present study applies a distributed model to the YellowRiver basin and focuses on exploring the uncertainty from the meteorological input, land use and groundwater initial conditions.

INTRODUCTION

In a consequence of human activity and climate fluctuation the water resources shortage and eco-environment degradation in the YellowRiver basin of China became essential problems since the end of 20th century. In particular, the serious drying-up of the main river along the lower reaches during the 1990s has drawn a lot of attention from all over the world. The Yellow River Conservancy Commission (YRCC, is the Yellow River authority under the Ministry of Water Resources) is promoting an integrated enhanced management of water resources in this basin. Key issues related to the water resources management include the availability and variability of natural water resources, impact of water soil conservation on the basin hydrological cycle, sustainability of eco-environment and potential risk of floods.

The atmosphere, land cover and human activity are the driving force and important variable respectively for the changes in hydrological cycle. The records of river runoff usually are the integrated results of all changes in atmosphere and catchment surface and the artificial direct uses of river water. For quantitative evaluation of these changes individually, a distributed physically-based hydrological model is needed since it can represent the spatial distribution of related basin properties and can examine the impacts of local changes on the basin hydrological cycle. Precipitation usually shows a considerable spatial variation over a large river basin. In practice, use of the traditional gauge measurement of precipitation as the input to hydrological models, it needs a spatial interpolation. There are different methods for interpolating the point data into a gridded data. It may introduce additional uncertainty during the process of interpolation. Bias of precipitation can introduce significant errors in model predictions [1].

The use of hydrological model for evaluating the effect of changes in land cover is not an easy work, because of the lack of continuous data and difficulty in determining 'correct' parameter values for both pre-change and post-change catchment conditions [1]. Uses of the hydrological modeling approach to simulating the impact of land-use changes on stream flow, it usually alters the proportions of existing vegetation types in the catchment, rather than introducing new species which may have given problems in obtaining the ‘correct’ parameters for the hydrologic models [1].

Initial conditions, in particular the groundwater water initial condition can change the result of hydrological simulation. In the natural condition, groundwater water level keeps dynamically stable, but it decreases under the over-pumping condition. Therefore it is also necessary to examine the effects of groundwater changes on the river runoff.

The present study applies a distributed physically-based model to the YellowRiver basin and focuses on exploring the uncertainty from the meteorological input, land use and groundwater initial condition.

STUDY AREA AND DATA AVAILABILITY

Target Catchment

The Yellow River, the second longest river basin in China, is the target basin in the present study. It originates from the Tibetan plateau, wanders through the northern semiarid region, crosses the loess plateau, passes through the eastern plain, and finally discharges into the BohaiGulf (see Fig. 1). The Yellow River flows about 5,500 km in distance in the main course and accumulates 753,000 km2 of drainage area. About 100 million people live within the catchment, and it consists of 1200 million ha of farmland of which nearly half is irrigated by the Yellow River.

Figure 1. The Yellow River basin

From the origins to the river mouth, the Yellow River experiences three typical landforms, the Qingzhang (Tibet) high plateau with elevations from 2,000 to 5,000 meters, the loess plateau and midstream tributaries with elevations from 500 to 2,000 meters, and the alluvial plain in the eastern part. The climate conditions vary from cold to temperate zones, and change from arid and semi-arid to semi-humid regions. The main irrigation areas are located in the northern part, in the tributaries of the midstream and on both sides along the lower reaches (see Fig. 1).

Data Availability

The geographical information concerning the YellowRiver basin used in this research is obtained from a number of global data sets. The digital elevation data of 1-km resolution is obtained from the USGS HYDRO1k data set which is available. Land cover is obtained from the USGS Global Land Cover Characteristics Data Base Version 2.0. This dataset has a spatial resolution of 1-km. Besides the global dataset, another land use map of 1:250000 scale is obtained from ChineseAcademy of Sciences. For estimating the leaf-area-index (LAI), a monthly NDVI with 8-km resolution is obtained from the DAAC of GSFC/NASA. This dataset is available from 1982 onwards. The soil type and the texture data are obtained from the Digital Soil Map of the World and Derived Soil Properties [2]. It is developed at 5-minute resolution using the FAO-UNESCO soil classification. The soil properties used for the hydrological simulation including the porosity, the saturated hydraulic conductivity, and the other soil water parameters corresponding to each soil type in this map are obtained from the Global Soil Data Task [3]. The water-retention relationship and unsaturated hydraulic conductivity are represented by Van Genuchten’s formula, and the parameters are available in this data set.

Table 1.Available data sets for the Yellow River

Data set / Scale / Source / Content/quality
DEM / 1000-m / USGS / Global
Land cover / 1000-m / USGS / Global, 24 types, 1990
1:250000 / CAS / China, 6 classes, 25 types, 1990
Soil / 5-km / FAO / Global, texture
10-km / Global Soil Task / Global,
Meteorological data / Point / China Meteorological Administration / Daily, precipitation, maximum, minimum and mean air temperature, wind speed, relative humidity, sunshine hours and pan evaporation
River discharge / Point / Ministry of Water Resources, China / Daily or monthly

The meteorological data are obtained from the China Meteorological Administration, which is available at a daily temporal resolution at 108 gauges in the YellowRiver basin. The discharge data collected is from the “Hydrological Year Book” published by the Hydrological Bureau of the Ministry of Water Resources of China. The data availability in the Yellow River is summarized in Table 1.

METHODOLOGY

The present study employs a distributed model for estimating the natural runoff in the YellowRiver basin, in which only natural hydrological processes are included (there is no consideration of irrigation or reservoir control). The model uses a grid system with a 10-km spatial resolution, and runs in hourly time steps. The methodology used for constructing this model includes a basin subdivision scheme, a sub-grid parameterization scheme, a physically-based hydrological simulation on hillslope and a kinematic wave flow routing method.

The catchment is the minimum unit for implementing water resources management. For a large river basin, it needs to be subdivided into sub-basins and to simulate the variability of water resources in each sub-basin. For subdividing the YellowRiver basin, the Pfafstetter scheme [4] is applied in the present study. In the present application, a total of 137 sub-basins have been identified in the upstream of the Huayuankou gauge. Since the model uses a 10-km grid, the heterogeneity inside a grid affects the hydrological processes, and therefore, a sub-grid parameterization is necessary. The sub-grid parameterization used in this research includes representations of the sub-grid variabilities in topography and land-cover. The topographical parameterization uses the catchment geomorphologic properties, which represents a grid by a number of hillslopes. The hillslopes located in a 10-km grid are grouped according to the land cover types. The hydrological simulation is carried out for each land cover group. The hillslope is a fundamental computational unit for hydrological simulation. A physically-based model is used for simulating the hillslope hydrology. The hydrological processes included in this model are the snowmelt, the canopy interception, evapotranspiration, infiltration, surface flow, subsurface flow and the exchange between the groundwater and the river [5]. The runoff generated from grid is the lateral inflow into the river at the same flow interval. Flow routing in the river network is solved using the kinematic wave approach.

The topography and soil are treated as being constant over time. The model uses the land cover of the 1990 as the base map and considers annual and seasonal changes in vegetation using remotely-sensed NDVI data. The 10-km gridded atmospherical forcing data used in the hydrological simulation is interpolated from the point dataset. For the precipitation data, both angular-distance weighting method [6] and nearest distance method [12] are used. For specifying an appropriate initial groundwater level for the natural condition, the present study carries out a 20-years test run from 1981 to 2000 for achieving a stable groundwater level. Then, the groundwater level and soil moisture contents at the end of the test run are used as the initial conditions for simulating the natural hydrological cycle. Besides the natural groundwater conditions, the present study tests also the decline in groundwater level during the 1980s and 1990s. The groundwater level at the beginning of each year in the natural condition is saved as the base. Then the groundwater level is declined from the base by 4 meters in 1980s and by 8 meters in 1990s for the urban areas and the agricultural areas with no surface water irrigation, and is used as the initial condition for testing change of the river runoff. In a summary, the hydrological simulations carried out in the present study include a base simulation, a test for land cover, a test for precipitation and a test for groundwater (see Table 2).

Table 2.Hydrological simulations designed in the present study

Case / Simulation conditions
Case1: Base simulation /
  1. The 10-km gridded precipitation data is generated using the angular-distance weighting method;
  2. The land cover data is from USGS and is grouped into 9 categories including water bodies (0.7%), urban areas (0.1%), bare land (1.17%), forest (4.55%), irrigated cropland (6.48%), non-irrigated cropland (16.73%), grassland (46.96%), shrub (23.3%), and wetland (0.01%);
  3. Uses the natural groundwater conditions. The groundwater level after a 20-yrs test run is used as the initial condition and no artificial decline is made during the simulation.

Case 2: Test for land cover /
  1. The land cover data is from CAS and is grouped into 9 categories including water bodies (0.55%), urban areas (0.63%), bare land (8.86%), forest (6.59%), irrigated cropland (6.13%), non-irrigated cropland (20.11%), grassland (51.29%), shrub 5.5%), and wetland (0.34%);
  2. The other conditions are the same as the base simulation.

Case3: Test for precipitation /
  1. The 10-km gridded precipitation data is generated using the nearest distance method;
  2. The other conditions are the same as the base simulation.

Case4: Test for groundwater /
  1. Uses the declined groundwater level (by 4 m in 1980s and by 8 m in 1990s for the urban areas and the agricultural areas with no surface water irrigation) as the initial condition at the beginning of each year;
  2. The other conditions are the same as the base simulation.

RESULTS AND DISCUSSIONS

Model Calibration and Validation

A 5-year test run from 1981 to 1985 is carried out for calibrating the model parameters. One of the calibrated parameters is the snowmelt factor in the temperature-based snowmelt equation. Another calibrated parameter, the hydraulic conductivity of the groundwater, is calibrated by checking the base flow in different sub-basins. Model validation is carried out from 1986 to 1990 in the upstream of the Tangnaihai gauge (see Fig. 1), where the direct human activity is negligible, and the snowmelt runoff and groundwater flow are the main sources of river discharge.

Figure 2. Comparison of simulated and observed river discharge at the Tannaihai gauge

Figure 2 shows a comparison between the simulated and observed daily discharges at the Tangnaihai gauge for both the calibration and validation periods. Based on the daily discharges, the ratio of the absolute error to the mean and the Nash coefficient are calculated to be 19% and 0.88 respectively for the calibration period and to be 17% and 0.89 respectively for the validation period. A good agreement between the simulated daily hydrograph with the observed and a consistency of the simulations in both the calibration and validation periods are achieved. The water balance error is -2% in the calibration period and 0.2% in the validation period.

Model Sensitivity Analysis

Using the calibrated model parameters, the hydrological simulations are carried out from 1981 to 2000 for these four cases listed in Table 2. The annual water balance of twenty years average is checked and summarizedin Table 3. Figure 3 shows the monthly hydrographs of twenty-year means for the four cases. Regarding the annual runoff, in general, the precipitation had the largest effect, the groundwater had the smallest effect. The changes in land cover and groundwater level made decreases in annual runoff, but the precipitation made an increase in annual runoff. The effect in the upstream of Lanzhou gauge is smaller than that in the downstream of Lanzhou gauge.

Table 3.Annual water balance for twenty years average

Case / Case 1
(Base) / Case 2
(Land cover) / Case 3
(Precipitation) / Case4
(Groundwater)
Upstream of the Lanzhou / P (mm/yr) / 434.3 / 434.3 / 440.5 (1.4%) * / 434.3
E (mm/yr) / 310.6 / 318.4 / 299.2 / 310.6
Q (mm/yr) / 119.6 / 111.1 (-7.1%) / 140.3 (17.3%) / 119.6 (0.0%)
S (mm/yr) / 4.1 / 4.8 / 1.0 / 4.1
Downstream of the Lanzhou / P (mm/yr) / 423.9 / 423.9 / 422.9 (-0.2%) / 423.9
E (mm/yr) / 378.8 / 394.1 / 369.3 / 367.7
Q (mm/yr) / 41.8 / 30.3 (-27.5%) / 52.6 (25.8%) / 34.9 (-16.5%)
S (mm/yr) / 3.3 / -0.5 / 1 / 21.3
Whole basin / P (mm/yr) / 427.2 / 427.2 / 428.6 (0.3%) / 427.2
E (mm/yr) / 355.7 / 368.7 / 345.8 / 349.3
Q (mm/yr) / 67.7 / 57.3 (-15.4%) / 81.9 (21.0%) / 61.7 (-8.9%)
S (mm/yr) / 3.8 / 1.2 / 0.9 / 16.2

* The number in bracket is the relative difference from the base simulation (case 1); P: the annual precipitation, E: the annual actual evapotranspiration, Q: the annual runoff, and S: the annual change of water storage in the catchment.

Figure 3.The 20-year mean monthly hydrographs for the four cases

Figure 4. Comparison of the spatial distribution of precipitation interpolated using (a) the angular-distance weighting method and (b) using the nearest distance method

The change in land cover from the case 1 to case 2 made an increase in evapotranspiration. This is mainly because of the increases in forest, cropland (both irrigated and non-irrigated) and grassland. The area fraction of shrub from the land cover map of the USGS is much higher than that from the CAS data. But it was treated as very sparse shrub in the hydrological simulation. The nearest distance interpolation method made a slight increase in annual precipitation in the upstream of the Lanzhou gauge. However, it made a large increase in annual runoff in both upstream and downstream of the Lanzhou gauge. This is due to the spatially high concentration of rainfall (see Fig. 4). Decrease in groundwater level made an increase in the recharge into groundwater, and therefore, the annual runoff was decreased.

CONCLUSION

The annual runoff in the YellowRiver basin is less than 20% of the annual precipitation. Therefore the accuracy of estimating the actual evapotranspiration is critical for the confidence of water resources assessment. It was found that the land cover and spatial concentration of precipitation had significant effects on the actual evapotranspiration. Impact by the groundwater level on the river runoff was found to be mainly from the change of recharge. With deeper groundwater level, higher recharge was required. This caused the decrease in river runoff.

ACKNOWLEDGEMENT

This research was partially supported by the Core Research for Evolutional Science and Technology (CREST) program of Japan Science and Technology Agency (JST). The authors would like to appreciate their grant in aid on this research.

REFERENCE

[1]Nandakumar N. and R.G. Mein (1997). Uncertainty in rainfall-runoff model simulations and the implications for predicting the hydrologic effects of land-use change, Journal of Hydrology, 192: 211-232

[2]FAO (2003). Digital soil map of the world and derived soil properties, Land and Water Digital Media Series Rev. 1.

[3]IGBP-DIS (2000). Global Soil Data Products CD-ROM. International Geosphere-Biosphere Programme, Data and Information System, Potsdam, Germany. Available from Oak RidgeNationalLaboratoryDistributedActiveArchiveCenter, Oak Ridge, Tennessee, U.S.A. [

[4]Yang, D. and K. Musiake (2003). A continental scale hydrological model using distributed approach and its application to Asia. Hydrological Processes, 17: 2855-2869.

[5]Yang, D., T. Oki, S. Herath, and K. Musiake (2002). A Geomorphology-Based Hydrological Model and Its Applications. In V.P. Singh & D.K. Frevert (ed.) Mathematical Models of Small Watershed Hydrology and Applications, Water Resources Publications, Littleton, Colorado. Chapter 9, 259-300.

[6]New, M., M. Hulme and P. Jones (2000). Representing twentieth-century space-time climate variability. Part II: Development of a 1961-96 monthly grids of terrestrial surface climate, Journal of Climate, 13: 2217-2238.

1