Introduction
Oases are special ecological systems that naturally oppose the encroachment of deserts by staying moist (Berndtsson et al. 1996; Gao et al. 2004; Zhu et al. 2004; Li et al. 2007). Oases provide important habitat refugia for plants, animals, and humans alike. Oases in northwest (NW) China account for only about 5% of the total land area of the region, but give shelter and feed about 95% of the growing population of the area (Gao et al. 2004; Chu et al. 2005). It has been proposed by many scientists and observers of desertification in NW China that rapid economic development in these areas is largely unsustainable and is happening at the very detriment of the oases (Kang et al. 2004; Ma et al. 2005; Ji et al. 2006). Signs of oases’ deterioration in NW China are widely documented in the scientific literature (Yang et al. 2007; Zhang et al. 2008a).
Direct rainfall to the oases in the Shiyang River watershed, NW China, is usually greater than to the surrounding deserts (e.g., 120–170 mm versus 40–60 mm yr−1; Table 1), but the amount is insufficient to be ecologically significant for oasis maintenance. A more significant source of water to the oases is the generation of meltwater in the Qilian Mountains. The meltwater usually flows during the spring-to-summer warming of the mountain glaciers and previous-winter snow cover (Ji et al. 2006). Glacial meltwater currently accounts for about 22% of the total direct supply of inland river water in NW China (Lu et al. 2005). With projected climate change, peak flow from mountain sources is anticipated to shift from spring–summer to mid- to late winter as temperatures in the mountains increase, placing greater pressure on downslope water supplies when water demand for crop production and domestic needs is high (Barnett et al. 2005). Another important source of water to the oases in NW China is the water generated from orographic precipitation during the spring–fall period of the year (Zhu et al. 2004). Orographic precipitation is formed when air masses interact with the Qilian Mountains (Roe 2005).
Gao et al. (Gao et al. 2004) and Chu et al. (Chu et al. 2005) show by model simulation that local circulation driven by daytime thermal heterogeneity between oasis and surrounding deserts [oasis breeze circulation (OBC)] is responsible for conserving water vapor and lowering evapotranspiration (ET) in the oases. They demonstrate that the downdraft associated with the OBC reinforces atmospheric stability over the oasis and causes a shallow cool–wet convective boundary layer to form, stabilizing water losses by the oasis. On an hour-by-hour basis, this fact may well be involved in reducing daytime ET. However, this reduction is inadequate to be important in the long-term sustainability of oases. Daytime losses of water vapor associated with the “oasis effect,” where ET is greater than net radiation, can be quite large (Oke 1987; Kai et al. 1997; Warner 2004), and without water replenishment from external sources the oases would eventually give way to desertification.
The connection between upwind vegetation cover, ET, and atmospheric relative humidity and downwind lifting condensation level of moist air (LCL or cloud base, if lifted) and precipitation is well documented in the scientific literature (e.g., Kimmins 1997; Eltahir 1998; Betts et al. 1999; Kleidon et al. 2000; Freedman et al. 2001; Lawton et al. 2001; Warner 2004; Brunsell 2006; Ray et al. 2006; Pielke et al. 2007; Jones and Brunsell 2009). For example, Betts et al. (Betts et al. 1999) and Freedman et al. (Freedman et al. 2001) provide evidence for an inverse relationship between land surface wetness (and, therefore, ET) and LCL over vegetated areas. Likewise, Ray et al. (Ray et al. 2006) demonstrate by model simulation that deforestation of the lowland and premontane regions of northern Costa Rica (in regions more than 1000 m AMSL) is responsible for the lifting of the cloud bank and, in response, the upward displacement of the base of the cloud forest in the mountains adjacent to the lowlands. They attribute this change to increased air temperature and decreased in-air transfer of humidity from the deforested areas (by virtue of lower ET) prior to the orographic lifting of the air.
On this basis, we believe that precipitation in the Qilian Mountains and the flow of water from the Qilian Mountains to the oases can be influenced by the spring–fall vegetation-to-air exchange of water vapor at the base of the mountains, providing means for long-term cycling of water and possibly oasis self-stabilization and support. The objective of this paper is to test the hypothesis of water recycling and long-term oasis stabilization by using monthly climate surfaces of mean air temperature, relative humidity, and precipitation developed from thermal remote sensing data and trained genetic algorithms, and the Landscape Distribution of Soil moisture, Energy, and Temperature model (LanDSET model; Bourque and Gullison 1998; Bourque et al. 2000) to calculate long-term soil water content (SWC) for two oasis-vegetation scenarios: 1) with vegetation, representing current conditions, and 2) without vegetation. Discrepancies in modeled results should illustrate the extent oasis vegetation is able to control the hydrology and self-stabilization of oases along the Qilian Mountains. In this study, SWC is expressed as a relative value between 0 and 1, where 0 represents the permanent wilting point of the soil and 1 represents the field capacity.
Materials and methods
Study area
The Hexi Corridor and the greater Shiyang River regions are some of the driest regions in the world (Shi and Zhang 1995; Ma et al. 2005). These regions are noted for their low precipitation, elevated temperatures, and high evaporative demand (Chen and Qu 1992; Ma et al. 2005). Extreme dryness in these areas is associated with the predominance of the Qinghai–Tibet Plateau to the south (Warner 2004; Sato 2009) and northwesterly or northerly dry continental winds associated with either the Azores high pressure system in summer or the Siberia high in winter (Warner 2004).
Annual precipitation in the greater Shiyang River region varies from over 500 mm in the Qilian Mountains to less than 100 mm in the deserts north of the Qilian Mountains, where mean annual potential evaporation rates can often exceed 2000 mm (Ding and Zhang 2004; Zhang et al. 2008b). In the mountains and plains of the greater Shiyang River region, 52%–71% of the rainfall occurs during the June–August period of the year.
The Shiyang River, one of three main rivers in the Hexi Corridor, has its origin in the Qilian Mountains and flows northeastward to terminate at the Minqin lake district in the east of the Hexi Corridor (Li et al. 2007). The study area is the Upper Shiyang River watershed (USRW) that extends from the Qilian Mountains to the Hongyashan Lake reservoir to the northeast (Figure 1). The USRW occupies a total land area of 14 360 km2. Within-watershed elevations range from 1370 to 5192 m AMSL (Figure 1), giving a mean watershed elevation of 2487 m. Slopes in the watershed vary from 0° to 69°. Cross-USRW mean annual precipitation is less than 200 mm and mean annual temperature is about 8.2°C. Panevaporation rates across the watershed range from 700 to 2600 mm yr−1 (Tong et al. 2007). Total supply of water from the Qilian Mountains is estimated to be 16.6 × 108 m3, 94% of which (i.e., 15.6 × 108 m3) is delivered directly as surface runoff (Tong et al. 2007). Soils in the Liangzhou Oasis (lower USRW; Figure 1) have a light sandy loam texture, with a mean dry bulk density of 1.43 g cm−3 and mean porosity of 52% (Zhang et al. 2008b). Although soils vary across the watershed, their variation has a small effect on watershed surface and subsurface flow characteristics compared to variation in physiographic gradients.
LanDSET model
LanDSET is a grid-based terrain analysis and process model (Bourque and Gullison 1998; Bourque et al. 2000), which integrates several components to compute long-term, spatially variable biophysical attributes of landscapes. The program consists of a topographic analysis, hydrological network extraction, radiation, and soil water balance module. The first step is to activate the terrain analysis module, which computes the primary terrain attributes of slope, aspect, horizon angle, view factor, terrain configuration factor, and catchment drainage area for individual grid cells in a digital elevation model (DEM; Gallant and Wilson 1996; Bourque and Gullison 1998; Bourque et al. 2000). These calculations are used by the hydrological network to create a depressionless DEM (Planchon and Darboux 2001), compute flow directions and flow accumulation, and apply these toward drainage network identification and delineation. The hydrological network extraction module uses the D8 (O’Callaghan and Mark 1984), Rho8 (Fairfield and Leymarie 1991), either algorithms, or alternatively, the grid-based stream tube approach of Costa-Cabral and Burges (Costa-Cabral and Burges 1994).
The radiation module computes the radiative balance from incoming and reflected shortwave radiation and incoming and outgoing longwave radiation (Bourque et al. 2000). This is achieved by using output from the terrain analysis module and parameter values from radiation and vegetation files, for example, midday albedo as a function of land cover; 0.15 for vegetative cover and 0.40 for dry desert sand (Geiger 1965; Oke 1987; Bourque et al. 2000). The algorithms used by the radiation module are based in part on work by Dozier et al. (Dozier et al. 1981), Moore et al. (Moore et al. 1993), Bourque and Gullison (Bourque and Gullison 1998), and Bourque et al. (Bourque et al. 2000). Output from the radiation module is used by the soil water balance module.
Long-term water balance computed by the soil water balance module is based on spatially variable topography, net radiation, and April–October mean relative humidity, air temperature, and accumulated precipitation surfaces. Calculations follow generalized descriptions given in Moore et al. (Moore et al. 1993). It uses topographic indices of flow direction, drainage area, and slope from the terrain attributes module, and produces steady-state wetness index values and effective upslope drainage area. Soil water content at individual grid cells is calculated from the wetness index (Moore et al. 1993). Net radiation from the radiation module (Bourque and Gullison 1998; Bourque et al. 2000) is used to assess potential ET, from which actual ET is calculated based on a Newton–Raphson solution of SWC (Moore et al. 1993). Deep seepage is computed for saturated areas. Gridded output from the individual modules of LanDSET can be viewed and further processed in GIS. The reader is encouraged to consult the papers cited for further information concerning LanDSET model formulation.
Generation of LanDSET input data
Genetic algorithms
Development of precipitation fields at fine spatial resolutions (less than or equal to 80 m) for large areas (more than 5000 km2) is commonly based on numerical interpolation of point estimates. In general, the quality of interpolation reflects the quality of the spatial point data as well as the interpolation method used. Many available interpolation methods (e.g., Theissen polygons, kriging, inverse distance weighing, natural neighbors, triangulation, and cubic splines) are incapable of addressing variation in the target measurements as function of extrinsic factors, such as, for example, time elapsed, changes in elevation, slope, direction of the prevailing wind, distance from mountains and coastlines, and overall synoptic conditions, other than the internal relationships contained in the point data themselves because of autocorrelation. Some interpolation methods like cokriging can establish relationships with some external factors, but this is usually limited to only a few covariates. In general, interpolation by most of these methods produces unrealistic values in areas with poor data coverage (with more than a 20-km separation between observation points; Peng et al. 2006) and with strong underlying physical gradients, such as in mountainous terrain (Bourque and Gullison 1998). Precipitation-surface prediction with process-based atmospheric models, such as the Regional Atmospheric Modeling System (Pielke et al. 2007), are possible alternatives to simple interpolation, but their input requirements for normal operation far exceed the meteorological and site information available for this study.
In this paper, we approach the quantification of monthly precipitation and intermediate surfaces with the assistance of genetic algorithms (GAs) designed to use satellite data as input. GAs are linear programming units (computer code) connected together to solve complex problems by relating patterns in input and output variables by machine learning. GAs are based on evolutionary theory of natural selection (Tung et al. 2003); given sufficient iterations the method usually converges to a global solution. GAs are supervised learning systems, requiring that matched inputs and outputs be supplied for pattern recognition. From these matched values, models are created, permitting the prediction of outputs from similar inputs. The models are created as computer code in, for example, C/C++, Fortran90, Pascal, and Assembly, facilitating their incorporation in existing process models. GAs are superior to conventional statistical regression because relationships between input and output values in highly complex, nonlinear datasets are usually difficult to determine, and as a result generating good predictive models with regression techniques can be fairly challenging. GAs are efficient in searching for complex relationships automatically, regardless of how complex the dataset might be.
Development of point models
Point evaluation of temporal trends of dewpoint temperature (Td) and precipitation (R) proceeded by selecting 30 years of weather data (1976–2005) from 29 weather stations in the vicinity of USRW, including 10 stations in the northern portion of Qinghai Province (Figure 1; Table 1), all in vegetated (nondesert) areas. The 30-yr dataset was subdivided into three equal parts of 10 consecutive years of monthly data (i.e., data from 1976–85, 1986–95, and 1996–2005), each part designated for a different modeling operation, all performed within the Discipulus modeling software [Register Machine Learning (RML) Technologies, Inc.]: that is, 1) GA training and model building (1986–95 subdataset); 2) model validation (1976–85 subdataset); and 3) model application (1996–2005 subdataset). Validation is used by Discipulus to prevent overgeneralization of the training data. Further model testing is performed with the application subdataset. For a robust model, the model should be able to reproduce trends in all three subdatasets equally well.
Nonclimate variables also considered included (i) the weather station’s geographic position (i.e., latitude and longitude; in degrees), (ii) DEM-based point elevation at the location of the weather stations (m; Table 1), and (iii) time of year (i.e., month number, 1 through 12). The 80-m resolution DEM of the greater USRW was generated from 3-arcsec point data acquired from the National Aeronautics and Space Administration (NASA) Shuttle Radar Topography Mission (http://srtm.csi.cgiar.org/, last accessed 10 September 2009).
Grid calculation
Monthly DEM-based calculations of Td (eactual) and R proceeded by writing a Fortran90 program that related spatiotemporal patterns in the variables according to (i) an initial RH and T at the upwind edge of the DEM, (ii) changes in topography along a series of transects (or profiles) spaced at 80-m intervals parallel to the direction of the prevailing wind (north-northeast–north-northwest; Figure 4), (iii) presence or absence of vegetation along the transect, (iv) monthly T surface, and (v) GAs for the calculation of Td and R (Figure 2). To facilitate prediction of rainfall in the Qilian Mountains, an artificial rainfall point was added at the highest elevation (5192 m). The point’s monthly rainfall was extrapolated from spatiotemporal trends contained in existing rainfall measurements collected at the 29 weather stations.
Initial RH (∼19%) and T (∼37°C) at the upwind edge [representing daytime desert conditions, based on data from the Dunhuang weather station—station identification (ID) 52418] was used to calculate the initial water vapor content of desert air with respect to its partial vapor pressure (Bourque et al. 2009) prior to modification either by (i) the addition of water vapor by transpiring vegetation or (ii) the loss of water vapor due to orographic lifting and subsequent cooling of the air mass and release of liquid water. Over deserts, eactual was assumed constant; that is, ET over deserts was assumed uniformly small compared to ET over the oasis (Warner 2004). To model variability in surface wind direction along the prevailing NNE–NNW direction (Figure 4) we allowed the transect lines to be placed at angles starting with −22.5° that were gradually increased to +22.5° (with 0° representing north) using a step rotation angle of 0.1°. Movement along each transect and calculations of elevational gradient, Td, eactual, and RH [i.e., 100 × eactual/esat, with esat(T ′ = T) determined with Equation (2)], potentially for thousands of transects for each increment rotation (450 rotations, in total), were based on an implementation of a modified version of Dozier et al.’s (Dozier et al. 1981) forward-direction profile algorithm.
In mountainous terrain, eactual was permitted to be reduced as the air was displaced upward and temperatures dropped (Roe 2005). Internal temperature of the rising air was assumed to equilibrate with monthly T at its point of displacement. Upward displacement of the air resulted in the condensation and precipitation of excess water vapor (in locations where eactual > esat) causing the RH of the air to remain at 100%, at which time eactual was forced to equal esat. On descent, increases in T resulted in a drop in RH. However, increases in RH soon followed with the air’s interaction with vegetation on its way down, providing T were high enough to support ET. Final spatial calculation of Td, eactual, and RH entailed calculating the average of potentially 450 values (for each variable) calculated at each DEM grid point with each step rotation along the series of parallel transects.
Grids of monthly T were derived by (i) averaging three to four 8-day composites of Moderate Resolution Imaging Spectroradiometer (MODIS) images of daytime land surface temperature (LST; MOD11A2 product at 1-km resolution) for each month from April through October, and (ii) adjusting the MODIS LST image by comparing the station-measured monthly daytime air temperatures and geographically corresponding point estimates of MODIS-based LST (after Hassan et al. 2007). Following calculation of T, corresponding grids of monthly mean Tmin and Tmax were then calculated with regression-based equations derived from station temperature data. MODIS images used in this experiment were all from 2005 and were assumed to represent normal temperature distributions over oases, mountains, and desert surfaces. Current presence–absence of vegetation in the greater USRW was determined from a multispectral analysis of a Landsat-7 Enhanced Thematic Mapper Plus (ETM+) summer image of the greater USRW (similar to Figure 1) with maximum likelihood classification available with the MultiSpec Application software version 5.2 (Purdue University, Indiana).
Grids of R, T, and RH generated for each month (April–October) were then either summed (in the case of R) or averaged (in the case of T and RH) to yield grids representative of the growing season. An example of a seasonal T grid for daytime conditions for the April–October period is provided in Figure 3b.
Application of LanDSET
Grids of R, T, RH, and mean April–October period net radiation calculated with the radiation module of LanDSET (Bourque et al. 2000) were subsequently used as input in the LanDSET calculation of growing-season SWC (Table 2). In the calculation of SWC we assumed soil conditions to be uniform throughout the USRW because of the lack of digitized soil cover data for the area. LanDSET parameter values in the calculation of SWC appear in Table 2.
A second series of grids were generated for R, T, RH, and SWC under assumed conditions of complete vegetation removal from the Liangzhou Oasis to examine the extent oasis vegetation controls the water balance and soil water storage in the USRW. Removal of vegetation from the Liangzhou Oasis was addressed in the combined GA-based model (in Figure 2) by turning off the GA calculation of Td for vegetation and permitting only decreases in desert eactual to occur as a result of orographic lifting. In this simulation, we did not alter the T surface. The lower T associated with the oasis (arrow, Figure 3b) can cause RH in the “without vegetation” simulation to increase slightly in the location of the oasis because of the inverse relationship between RH and T, which can, in turn, lead to a slight increase in R over the oasis. All of these effects can culminate to slightly higher predicted SWC in the location of the oasis, but this increase is minor when compared to the effect associated with disrupting the mass exchange of water vapor from the oasis to the atmosphere by removing vegetation and by allowing desert conditions (dry air) to dominate.
Results and discussion
Point evaluation of trained GA based on weather station data
Figure 5 shows the observed versus GA-predicted data plots for the training, validation, and application subdatasets for the calculation of Td. In all three cases, the results were generally the same, suggesting that the model constructed was sufficiently robust to reproduce multiyear time series of monthly Td of independent datasets as good as or better than those produced during model training. The degree to which the variability in the three 10-yr subdatasets was explained by the model ranged from 97% for both the training and validation subdatasets to 96% for the application subdataset. Standard error of estimates (SEEs) for the three cases were within ±0.2°C of each other, at roughly 1.9°C. No significant bias was evident in the predictions for all three cases; regression slopes were all within ±0.03 of the 1:1 correspondence line (slope = 1.0 and y intercept = 0.0; p statistics > 0.05; Figure 5) and the y intercepts were not significantly different from zero (p statistics > 0.05).
Figure 6 provides a comparison of time series of monthly precipitation for the 1996–2005 period (application subdataset) for five weather stations in vicinity of the USRW, that is, Yongchang (station ID 52674), Wuwei (52679), Gulang (52784), Wushaoling (52787), and Menyuan (52765; Table 1). Of the five stations, Wuwei received the least amount of precipitation during the 1996–2005 (10 yr) period, that is, 1808.3 mm, while Menyuan received the most, at 5054.8 mm. Generally good agreement was obtained between measured and modeled values of monthly precipitation (Figure 6). The GA for R was able to explain on average 80% (ranging from 75% to 86%) of the variability in measured monthly precipitation. Precipitation predictions for Menyuan station were the best among the five stations considered.
Spatial application of GAs and orographic lifting principles
Figure 7 gives RH and R as a function of the two land-cover scenarios: 1) with vegetation (Figure 7a for RH and Figure 7c for R) and 2) without vegetation in the lower USRW (Figure 7b for RH and Figure 7d for R). In the scenario where vegetation is included, RH tends to be high throughout most of the USRW (50%–70%). The high RH in the USRW (Figure 7a), translates in a lower LCL (Freedman et al. 2001) and heavier precipitation in the lower slopes of the USRW (Figure 7c). In contrast, without vegetation, the RH remains low (20%–30%; Figure 7b), requiring that the air be lifted to higher elevations for precipitation to form (Figures 7d and 8). This situation has the potential to create a significant precipitation (water) deficit in the USRW (Figure 8).
LanDSET calculation of SWC
Figure 9 provides the LanDSET calculations of SWC with and without vegetation in the lower USRW (Figures 9a and 9b). With vegetation and associated heavy precipitation in the lower slopes of the USRW, SWC in the Liangzhou Oasis tends to remain quite favorable for plant production (Figure 9a). In the Liangzhou Oasis, SWC ranges from 0.2 to 0.8. In contrast, without vegetation, modeled SWC in the USRW and Liangzhou Oasis drop by 22%–78% (Figure 9c); the greatest drop occurs at the base of the Qilian Mountains where water in this area pools prior redistribution to the lower and flatter portion of the USRW. These drops suggest that, with vegetation removal, oasis conditions can degenerate quickly permitting area desertification. Simulated SWC in the desert ranges from 0.02 to 0.09.
The relationships between actual distribution of vegetation and favorable soil water conditions in the Liangzhou Oasis match exceptionally well (Figure 9d). This suggests that physiographic conditions at the base of the Qilian Mountains have an important role in the long-term distribution of surface and subsurface water to the center of the USRW. Slight mismatch between SWC and vegetation cover (above the Liangzhou Oasis) is related to the fact that the input to the calculation of SWC is based on 1-km resolution maps, while the vegetation cover is derived from a 30-m resolution Landsat-7 image. The portion of the USRW without vegetation (left and right of the Liangzhou Oasis, lower USRW; Figure 9d) receives little to no water from the upper reaches of the USRW because of the drainage characteristics of the watershed. Rainwater to these nonvegetated parts of the USRW is insufficient (∼0.2 mm day−1 or ∼40 mm over the growing season) to make up for the high evaporative demand of the atmosphere and support oasis vegetation. As a result, natural expansion of the oasis into these areas is not feasibly possible with current climatic and physiographic conditions. Obviously, artificial methods (e.g., pumps, building of diversion channels, irrigation) can be employed to extend the boundaries of the oasis but at a tremendous cost, both financially and ecologically, by promoting, for example, soil salinization (Youhao et al. 2007) and other environmental problems.
With vegetation present in the oasis, there is a natural tendency for the oasis to self-regulate and resist the encroachment of the surrounding desert by staying moist. Changes that promote the replacement of high water vapor–producing vegetation cover with vegetation that is not so accommodating will result in a deterioration of the oasis’ ability to resist desertification (Bruelheide et al. 2003). Feedback mechanisms of oasis self-support introduced in this paper are most likely the same mechanisms that promote self-support and long-term stabilization of the chain of oases along the base of the Qilian Mountains (Figure 1). Increases in temperature in the deserts and mountains with climate warming may cause a similar change in the LCL and displacement of the high-intensity precipitation band in the Qilian Mountains to what is observed with the simulated removal of vegetation in the lower USRW. Climate change, land-cover transformations, water resource mining, and rapid population growth in NW China will no doubt intensify desertification of oases in this region.
Concluding remarks
The paper provides a realistic spatial calculation of soil water content for the USRW in northwest China. USRW calculations of SWC are based on the application of the LanDSET process model and gridded inputs of air temperature derived from MODIS land surface temperatures, LanDSET calculation of net radiation, relative humidity, and precipitation. Spatiotemporal estimates of dewpoint temperature, relative humidity, and precipitation are generated with genetic algorithms integrated in a Fortran90 program (Figure 2). In the LanDSET calculation of SWC, we assume soil texture in the USRW to be uniform; that is, elevational gradients in the USRW are considered more important in the redistribution of water in the watershed. Soil water content distribution mirrors the distribution of vegetation in the lower watershed extremely well. Soil water conditions are shown to decrease with the simulated removal of vegetation. This reduction is not considered recoverable because without vegetation there is a significant loss of atmospheric water vapor returning to the watershed as precipitation. This back and forth exchange of water vapor and liquid water from the vegetation to the Qilian Mountains and from the Qilian Mountains to the Liangzhou Oasis as surface or subsurface flow is considered an important mechanism by which long-term oasis stabilization can be achieved. The model shows that vegetation removal could potentially affect within-watershed precipitation patterns by reducing the exchange of water vapor from the land surface to the air, by increasing the air’s lifting condensation level, and by causing high-intensity precipitation to fall at higher elevations and at reduced levels. Results in this paper are consistent with modeled observations using process-based models (Lawton et al. 2001; Ray et al. 2006; Pielke et al. 2007).
Acknowledgments
This study was jointly funded by the Chinese Meteorological Administration, Lanzhou Regional Climate Centre of the Gansu Provincial Meteorological Bureau (GMB), Lanzhou, China (Natural Science Foundation of China under Grant 40830957), and the Faculty of Forestry and Environmental Management, University of New Brunswick (UNB), New Brunswick, Canada. We are grateful to Sun Landong, Han Tao, Fang Feng, and Wang Yirong of GMB for their assistance with retrieving and preparing the data for this paper. We would also like to acknowledge NASA for providing MODIS and SRTM DEM data free of charge and Meng Fan-Rui of the Faculty of Forestry and Environmental Management, UNB, and D. Edwin Swift, Natural Resources Canada, Canadian Wood Fibre Centre for reviewing the manuscript prior to its submission to Earth Interactions. Finally, we would like to acknowledge the helpful editorial comments received from three anonymous reviewers.
REFERENCES
Barnett, T. P., J. C. Adam, and D. P. Lettenmaier. 2005. Potential impacts of a warming climate on water availability in snow-dominated regions. Nature 438:303–309.
Berndtsson, R., K. Nodomi, H. Yasuda, T. Persson, H. S. Chen, and K. Jinno. 1996. Soil water and temperature patterns in an arid desert dune sand. J. Hydrol. 185:221–240.
Betts, A. K., M. Goulden, and S. Wofsy. 1999. Controls on evaporation in a boreal spruce forest. J. Climate 12:1601–1618.
Bourque, C. P-A. and J. J. Gullison. 1998. A technique to predict hourly potential solar radiation and temperature for a mostly unmonitored area in the Cape Breton Highlands. Can. J. Soil Sci. 78:409–420.
Bourque, C. P-A., F-R. Meng, J. J. Gullison, and J. Bridgland. 2000. Biophysical and potential vegetation growth surfaces for a small watershed in northern Cape Breton Island, Nova Scotia, Canada. Can. J. For. Res. 30:1179–1195.
Bourque, C. P-A., W. Buchanan, and Q. K. Hassan. 2009. An analysis of habitat suitability and charcoal-yielding characteristics of five exotic tree species intended for bioenergy production in Jamaica. J. Biobased Mater. Bioenergy 3:46–56.
Bruelheide, H. Coauthors 2003. Vegetation changes in a river oasis on the southern rim of the Taklamakan Desert in China between 1956 and 2000. Phytocoenologia 3:801–818.
Brunsell, N. A. 2006. Characterization of land-surface precipitation feedback regimes with remote sensing. Remote Sens. Environ. 100:200–211.
Chen, L. H. and Y. G. Qu. 1992. Rational Development and Utilization of Water and Soil Resources in Hexi Region. Science Press, 420 pp.
Chu, P. C., S. Lu, and Y. Chen. 2005. A numerical modeling study on desert oasis self-supporting mechanisms. J. Hydrol. 312:256–276.
Cios, K. J., W. Pedrycz, R. W. Swiniarski, and L. A. Kurgan. 2007. Data Mining: A Knowledge Discovery Approach. Springer, 606 pp.
Costa-Cabral, C. and J. Burges. 1994. Digital elevation model networks (DEMON): A model of flow over hillslopes for computation of contributing and dispersal areas. Water Resour. Res. 30:1681–1692.
Ding, H. and J. Zhang. 2004. Relationships between sustainable development and water resources in arid oases area—An example of Hexi Corridor. J. Arid Land Resour. Environ. 18:50–55.
Dozier, J., J. Bruno, and P. Downey. 1981. A faster solution to the horizon problem. Comput. Geosci. 7:145–151.
Eltahir, E. A. B. 1998. A soil moisture–rainfall feedback mechanism 1. Theory and observations. Water Resour. Res. 34:765–776.
Fairfield, J. and P. Leymarie. 1991. Drainage networks from grid digital elevation models. Water Resour. Res. 27:709–717.
Freedman, J. M., D. R. Fitzjarrald, K. E. Moore, and R. K. Sakai. 2001. Boundary layer clouds and vegetation–atmosphere feedbacks. J. Climate 14:180–197.
Gallant, J. C. and J. P. Wilson. 1996. TAPES-G: A grid-based terrain analysis program for the environmental sciences. Comput. Geosci. 22:713–722.
Gao, Y., Y. Chen, and S. Lu. 2004. Numerical simulation of the critical scale of oasis maintenance and development in the arid regions of northwest China. Adv. Atmos. Sci. 21:113–124.
Geiger, R. 1965. The Climate near the Ground. Harvard University Press, 611 pp.
Hassan, Q. K., C. P-A. Bourque, F-R. Meng, and W. Richards. 2007. Spatial mapping of growing degree days: An application of MODIS-based surface temperatures and enhanced vegetation index. J. Appl. Remote Sens. 1:013511. doi:10.1117/1.2740040.
Hewson, E. W. and R. W. Longley. 1951. Meteorology, Theoretical and Applied. John Wiley and Sons, 468 pp.
Hubbard, K. G., R. Mahmood, and C. Carlson. 2003. Estimating daily dew point temperature for the northern Great Plains using maximum and minimum temperature. Agron. J. 95:323–328.
Ji, X., E. Kang, R. Chen, W. Zhao, Z. Zhang, and B. Jin. 2006. The impact of the development of water resources on environment in arid inland river basins of Hexi region, northwestern China. Environ. Geol. 50:793–801.
Jones, A. R. and N. A. Brunsell. 2009. Energy balance partitioning and net radiation controls on soil moisture–precipitation feedbacks. Earth Interactions 13.[Available online at http://EarthInteractions.org].
Kai, K., M. Matsuda, and R. Sato. 1997. Oasis effect observed at Zhangye Oasis in the Hexi Corridor, China. J. Meteor. Soc. Japan 75:1171–1178.
Kang, S. Coauthors 2004. The impacts of human activities on the water-land environment of the Shiyang River basin, an arid region in northwest China. Hydrol. Sci. J. 49:413–427.
Kimmins, J. P. 1997. Forest Ecology: A Foundation for Sustainable Management. 2nd ed. Prentice Hall, 596 pp.
Kleidon, A., K. Fraedrich, and M. Heimann. 2000. A green planet versus a desert world: Estimating the maximum effect of vegetation on the land surface climate. Climatic Change 44:471–493.
Lawton, R. O., U. S. Nair, R. A. Pielke Sr., and R. M. Welch. 2001. Climatic impact of tropical lowland deforestation on nearby montane cloud forests. Science 294:584–587.
Li, X., D. Xiao, X. He, W. Chen, and D. Song. 2007. Evaluation of landscape changes and ecological degradation by GIS in arid regions: A case study of the terminal oasis of the Shiyang River, northwest China. Environ. Geol. 52:947–956.
Lu, A., Y. Ding, H. Pang, L. Yuan, and Y. He. 2005. Impact of global warming on water resource in arid area of northwest China. J. Mt. Sci. 2:313–318.
Ma, J. Z., X. S. Wang, and W. M. Edmunds. 2005. The characteristics of groundwater resources and their changes under the impacts of human activity in the arid northwest China—A case study of the Shiyang River basin. J. Arid Environ. 61:277–295.
Moore, I. D., T. W. Norton, and J. E. Williams. 1993. Modelling environmental heterogeneity in forested landscapes. J. Hydrol. 150:717–747.
O’Callaghan, J. F. and D. M. Mark. 1984. The extraction of drainage networks from digital elevation data. Comput. Vision Graph. Image Process. 28:323–344.
Oke, T. R. 1987. Boundary Layer Climates. 2nd ed. Methuen, 372 pp.
Peng, G., J. Li, Y. Chen, A. P. Norizan, and L. Tay. 2006. High-resolution surface relative humidity computation using MODIS image in peninsular Malaysia. Chin. Geogr. Sci. 16:260–264.
Pielke Sr., R. A., J. Adegoke, A. Beltrán-Przekurat, C. A. Hiemstra, J. Lin, U. S. Nair, D. Niyogi, and T. E. Nobis. 2007. An overview of regional land-use and land-cover impacts on rainfall. Tellus 59B:587–601.
Planchon, O. and F. Darboux. 2001. A fast, simple and versatile algorithm to fill the depressions of digital elevation models. Catena 46:159–176.
Ray, D. K., U. S. Nair, R. O. Lawton, R. M. Welch, and R. A. Pielke Sr.. 2006. Impact of land use on Costa Rican tropical montane cloud forests: Sensitivity of orographic cloud formation to deforestation in the plains. J. Geophys. Res. 111:D02108. doi:10.1029/2005JD006096.
Roe, G. H. 2005. Orographic precipitation. Annu. Rev. Earth Planet. Sci. 33:645–671.
Sato, T. 2009. Influences of subtropical jet and Tibetan Plateau on precipitation pattern in Asia: Insights from regional climate modeling. Quat. Int. 194:148–158.
Shi, Y. and X. Zhang. 1995. The influence of climate changes on the water resources in arid areas of northwest China. Sci. China 25:968–977.
Tong, L., S. Z. Kang, and L. Zhang. 2007. Temporal and spatial variations of evapotranspiration for spring wheat in the Shiyang River basin in northwest China. Agric. Water Manage. 87:241–250.
Tung, C. P., S. Y. Hsu, C. M. Liu, and J. S. Li. 2003. Application of the genetic algorithm for optimizing operation rules of the Li Yu Tan reservoir in Taiwan. J. Amer. Water Resour. Assoc. 39:649–657.
Warner, T. T. 2004. Desert Meteorology. Cambridge University Press, 595 pp.
Yang, X., Z. Ding, X. Fan, Z. Zhou, and N. Ma. 2007. Processes and mechanisms of desertification in northern China during the last 30 years, with a special reference to the Hunshandake Sandy Land, eastern Inner Mongolia. Catena 71:2–12.
Youhao, E., J. H. Wang, P. Yan, H. Han, and D. Zhang. 2007. Assessment of land salinization change using remote sensing techniques in Minqin Basin, northwest China. Proc. IEEE Int. Geoscience and Remote Sensing Symp., Barcelona, Spain, IEEE, 3154–3157, doi:10.1109/IGARSS.2007.4423514.
Zhang, B. Z., S. Z. Kang, F. Li, and L. Zhanghttp. 2008b. Comparison of three evapotranspiration models to Bowen ratio-energy balance method for a vineyard in an arid desert region of northwest China. Agric. For. Meteor. 148:1629–1640.
Zhang, X., X. Wang, and P. Yan. 2008a. Re-evaluating the impacts of human activity and environmental change on desertification in the Minqin Oasis, China. Environ. Geol. 55:705–715.
Zhu, Y., Y. Wu, and S. Drake. 2004. A survey: Obstacles and strategies for the development of ground-water resources in arid inland river basins of western China. J. Arid Environ. 59:351–367.
Chain of oases along the base of the Qilian Mountains (green patches denoted by Δ; main figure), west-central Gansu, NW China, including the Upper Shiyang River watershed study area (outlined in red). The vegetated area at the bottom of the USRW forms the Liangzhou Oasis (inset). Area to the left of the oasis is predominantly desert. The lake at the outlet of the watershed is the Hongyashan Lake reservoir. White areas in the Qilian Mountains (inset) are either glaciers, patches of snow, clouds, or combinations thereof. Images are based on draping Landsat-7 ETM+ pseudocolor scenes of the USRW and surrounding areas (Geocover 2000) over an 80-m resolution DEM of the area.
Citation: Earth Interactions 13, 13; 10.1175/2009EI286.1
Description of the spatial calculator (Fortran90 program) of dewpoint temperature (Td; °C), actual water vapor pressure (eactual; hPa), relative humidity (RH; %), and precipitation (R; mm month−1). Here Td and R surfaces are based on DEM point-by-point calculations performed with trained genetic algorithms, each with four input variables (all defined as grids, except month). Gridded output from the procedure is used as input in the LanDSET calculation of soil water content.
Citation: Earth Interactions 13, 13; 10.1175/2009EI286.1
Distribution of (a) cloud-free solar radiation (MJ m−2) for the month of July calculated with the LanDSET model and (b) average MODIS-based 8-day composites of air temperature (°C) for the April–October period. Markers in both images coincide with the locations of five weather stations in vicinity of the USRW, i.e., A—Yongchang, B—Wuwei, C—Gulang, D—Wushaoling in Gansu, and E—Menyuan in Qinghai. The white arrow in (b) points to the lower temperatures associated with the Liangzhou Oasis (lower USRW).
Citation: Earth Interactions 13, 13; 10.1175/2009EI286.1
Wind direction frequencies (%) as a function of time of day and month at Wuwei weather station (station ID: 52679; Liangzhou Oasis) for 2004.
Citation: Earth Interactions 13, 13; 10.1175/2009EI286.1
Observed vs GA-predicted data plots for the training, validation, and application subdatasets, applied to the calculation of dewpoint temperature (Td; Figure 2). Model performance statistics are provided for each of the 10-yr subdatasets. The regression equation gives an indication of model bias (i.e., deviation from the 1:1 correspondence line); and r 2 is the degree to which the model explains the variability in monthly Td.
Citation: Earth Interactions 13, 13; 10.1175/2009EI286.1
Observed vs GA-modeled time series of monthly precipitation (mm) for five stations in the vicinity of the USRW (1996–2005, application data only). The five stations are at (a) Yongchang (station ID 52674), (b) Wuwei (52679), (c) Gulang (52784), (d) Wushaoling (52787), and (e) Menyuan (52765; Table 1). Observed values are represented by the closed circles and the simulated values are represented by the continuous lines.
Citation: Earth Interactions 13, 13; 10.1175/2009EI286.1
Projected growing-season distribution of mean relative humidity (%) (a) with and (b) without vegetation and total precipitation (mm) (c) with and (d) without vegetation. Weather station markers are provided as reference points: A—Yongchang, B—Wuwei, C—Gulang, D—Wushaoling in Gansu, and E—Menyuan in Qinghai.
Citation: Earth Interactions 13, 13; 10.1175/2009EI286.1
Projected distribution of net growing-season reductions in precipitation (mm day−1) in the USRW associated with the simulated removal of vegetation. The arrow refers to the net movement of the high precipitation band and LCL with the simulated removal of vegetation in the lower USRW. Weather station markers are provided as reference points: A—Yongchang, B—Wuwei, C—Gulang, D—Wushaoling in Gansu, and E—Menyuan in Qinghai.
Citation: Earth Interactions 13, 13; 10.1175/2009EI286.1
Projected distribution of LanDSET calculation of growing-season SWC (a) with and (b) without vegetation. (c) The net reduction in SWC associated with the removal of vegetation in the lower USRW. Weather station markers are provided as reference points: A—Yongchang, B—Wuwei, C—Gulang, D—Wushaoling in Gansu, and E—Menyuan in Qinghai. (d) An overlay of the current vegetation distribution (outlined) over the distribution of SWC calculated with vegetation present in the lower USRW.
Citation: Earth Interactions 13, 13; 10.1175/2009EI286.1
List of weather stations, their coordinates, elevation, and mean total annual precipitation (R) based on 1976–2005 data.
Grid input and parameter values for the LanDSET calculation of growing-season soil water content in the greater USRW.
