The study area boundary was drawn from the first China Brazil Earth Resources Satellite (CBERS-1) images acquired in 2001 with spatial resolution of 20 m by contouring the edge of the oasis. Geographic data on land use, soil type and the channel network were made available by the Office for World Bank Financed Project (irrigation-referenced data from the Akesu River Administrative Agency). There are two canal systems in the study area. Each system supplies water to irrigation area of one part of the study area. The study area is then divided into two sub-areas according to the canal system. In each sub-area, the RE model is performed to represent runoff abstraction. The number of land cover types N was taken to be 4, representing irrigation area, wild land, wasteland and wetland, respectively. Three land use types (vegetation, sandy wasteland and wetland) over the plain are determined from the remote sensing images. The vegetation area was divided into irrigation area and wild land. The irrigation area was collected from local statistic bureau. The wild land area was considered as the difference of remote sensing vegetation area and irrigation area. And the number of crop cover types M was taken to be 7, representing wheat, corn, cotton, paddy, melon, gardens and manmade pasture, respectively. Meteorological data has been collected at Awati, Xinjiang, which lies within the Akesu oasis. The comparison between Awati station and two stations near the study area shows that monthly evapotranspiration ability values are quite similar (Tang et al., 2004). And the runoff-evaporation processes are largely driven by the monitored water diversions rather than meteorological data. So the data from Awati station is used to represent the whole study area. Stream discharge data for Akesu River (Xidaqiao station and Yimapaxia station) have been collected since 1980. The proportionality constant for river loss was then derived from measured stream discharge observed during the 1980s. The diversion from stream is monitored since 1980, but the drainage data are limited to recent years. Twenty three groundwater observation wells were drilled to watch water table in irrigation area and the data were available from 2000.
The RE model introduced above was used to reproduce the monthly runoff-evaporation processes. The streamflow simulation is evaluated in terms of the error variances for the stream flow values. Let xi and fi (i = 1,..., n) denote time series of observations and simulated discharge. The averages are 33#33 = 34#34xi/n and 35#35 = 34#34fi/n. The mean square error between model-simulated fi and the observed xi is:
Set 40#40 = 41#41 and 42#42 = 43#43. The mean square error can be written in a fraction equation (Murphy, 1988): The three terms in the equation (15) are related to overall bias error, amplitude error (through the ratio of the variances) and phase error (through the correlation). The root mean square error, rmse = 58#58, the relative root mean square error, rrmse = rmse/33#33, and the mean square skill score, MSSS = 1 - e2/59#59, are also calculated. The root mean square error in the simulation period is 20.5 m3/s and the relative root mean square error is 0.21. The mean square skill score MSSS is 0.986, showing the model performed well in the study area. The errors related to overall bias error, amplitude error and phase error is 1.7%, 22.1% and 76.2%, respectively. Fig.6a shows reproduced and observed streamflow to downstream from 1999 to 2002. Generally, the stream flow is fairly well reproduced, except that the magnitude of the peak in 2002 is subject to larger errors. The simulated irrigation area water table depth of two sub-areas is consistent with observation in annual average but doesn't agree well with monthly observation (Fig.6b1 and 6b2), probably because elevation is not well considered in the model and groundwater observations are too scattered to represent the average situation in the study area. In this study, streamflow and water table reproduction are not the goal per se, instead, our purpose in evaluation of the reproduced hydrographs is to provide evidence that the model is reproducing a reasonable water budget.The river water from upstream was dispersed or consumed to downstream (Co), riverway loss (Ce + Cg), net diversion (Cd) and riverway storage change (dCs). Fig.7a shows the reproduced monthly river water dispersion and consumption from 1999 to 2002, where the water volume is represented as water depth over whole study area. This graph is a stacked area graph, which displays the trend of the contribution of each item over time. The net diversion, which is dispersed to irrigation area, occupies more than half of the stream inflow especially in low water season (from November to May) when almost all the stream flow is diverted to irrigation area. That shows the manmade runoff absorbing process is obvious in this area. The river loss is also considerable in flood season (July and August), and the river loss item can be negative in low water season because of the regeneration water. Averagely, 8% of the stream inflow becomes river loss in flood season, i.e. natural runoff abstraction. Expectably, the riverway storage change is relatively small throughout the period.
The diverted canal water was dispersed or consumed to irrigation area (Ic), canal surface and surrounding saturated zone (Ec), and groundwater (Gc). Fig.7b shows the monthly canal water dispersion and consumption from 1999 to 2002, where the water volume is represented as water depth over irrigation area. Less than half of diversion arrives at field. About ten percent of the diversion becomes evaporation from canal surface and surrounding saturated zone and near half of the diversion penetrates through to groundwater. Although canal water dispersion is a regionalized process which is mainly determined by canal type, canal system distribution and management level etc., the results still show that canal loss can be a considerable factor in manmade runoff abstraction process.
Fig.8 shows the water consumptions in wild land (Ewild), waste land (Ewaste), wetland (Ewetland) and groundwater exchange (X), where the water volume is represented as water depth over whole non-irrigation area. The groundwater from irrigation area occupies nearly 70% of the total water consumption in non-irrigation area. It is non unexpected that the water consumption in non-irrigation area originates from irrigation area because the rainfall is not enough to support the ecosystem.
The resulting water budget of the study area is summarized in Table 1. More than seventy percent of stream inflow is consumed as evapotranspiration back into atmosphere and the remaining less than thirty percent flows downstream. The runoff-evaporation processes are the dominating hydrological processes in the Akesu oasis. About half of total water consumption is occupied by evapotranspiration in irrigation area, and another 13% is evaporated back to atmosphere through the canal networks. It shows that the hydrological cycle in the oasis is highly impacted by human activity. The water consumed in ecological valuable areas such as wild land and wetland, which is mainly supplied by the groundwater from irrigation area, is 21% of total consumption. This indicates that runoff absorbing processes are very important for watershed ecosystem in typical RAA.