Fashion & Beauty

Suitability of the TRMM satellite rainfalls in driving a distributed hydrological model for water balance computations in Xinjiang catchment, Poyang lake basin

Suitability of the TRMM satellite rainfalls in driving a distributed hydrological model for water balance computations in Xinjiang catchment, Poyang lake basin
of 11
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
  Suitability of the TRMM satellite rainfalls in driving a distributed hydrologicalmodel for water balance computations in Xinjiang catchment, Poyang lake basin Xiang-Hu Li a , Qi Zhang a, ⇑ , Chong-Yu Xu b,c a State Key Laboratory of Lake Science and Environment, Nanjing Institute of Geography and Limnology, Chinese Academy of Sciences, Nanjing 210008, China b Department of Geosciences, University of Oslo, Oslo, Norway c School of Geographic and Oceanographic Sciences, Nanjing University, China a r t i c l e i n f o  Article history: Received 8 September 2011Receivedinrevisedform21November 2011Accepted 8 January 2012Available online 24 January 2012This manuscript was handled byKonstantineP. Georgakakos, Editor-in-Chief,with the assistance of Hervé Andrieu,Associate Editor Keywords: RainfallTRMMHydrological processWater balanceDistributed hydrological modelXinjiang catchment s u m m a r y Spatialrainfallisakeyinputtodistributedhydrologicalmodels,anditsprecisionsheavilyaffecttheaccu-racy of stream flow predictions from a hydrological model. Traditional interpolation techniques whichobtainthespatialrainfall distributionfromraingaugedatahavesomelimitationscausedbydatascarcityand bad quality, especially in developing countries or remote locations. Satellite-based precipitationproducts are expected to offer an alternative to ground-based rainfall estimates in the present and theforeseeable future. For this purpose, the quality and usefulness of satellite-based precipitation productsneed to be evaluated. The present study compares the difference of Tropical Rainfall Measuring Mission(TRMM) rainfall with rain gauges data at different time scales and evaluates the usefulness of the TRMMrainfall for hydrological processes simulation and water balance analysis at the Xinjiang catchment,located in the lower reaches of the Yangtze River in China. The results show at daily time step TRMMrainfall data are better at determining rain occurrence and mean values than at determining the rainfallextremes,andlargerdifferenceexistsforthemaximaldailyandmaximal5-dayrainfalls.Atmonthlytimescale,goodlinearrelationshipsbetweenTRMMrainfallandraingaugesrainfalldataarereceivedwiththedetermination coefficients ( R 2 ) varying between 0.81 and 0.89 for the individual stations and 0.88 forareal average rainfall data, respectively. But the slope of regression line ranges between 0.74 for Yingtanand 0.94 for Yushan, indicating that the TRMMsatellite is inclined to underestimate the monthly rainfallin this area. The simulation of daily hydrological processes shows that the Water Flow Model for LakeCatchment (WATLAC) model using conventional rain gauge data produces an overall good fit, but thesimulation results using TRMM rainfall data are discontented. The evaluation results imply that theTRMM rainfall data are unsuited for daily stream flow simulation in this study area with desired preci-sions. However, good performance can be received using TRMM rainfall data for monthly stream flowsimulations.Thecomparisonofthesimulatedannual waterbalancecomponentsshowsthatthedifferentrainfall data sources can change the volume value and proportion of water balance components to someextent, but it generally meets the need of practical use.   2012 Elsevier B.V. All rights reserved. 1. Introduction Distributed hydrological models have become the main tool tounderstand the hydrological processes and solve practical hydro-logical andwater resources problems. Physically-baseddistributedhydrological models can fulfill the necessity of describing spatialheterogeneity, assessing the impact of natural and human inducedchanges and providing detailed descriptions of the hydrologicalprocesses in watersheds to satisfy various needs in spatial model-ling (Abbott and Refsgaard, 1996). However, these models requirethe spatially distributed data as input to reflect the heterogeneityof base information in the watersheds. The spatial rainfall is oneofthekeyinputsforthesemodels,andtheaccuracyof streamflowpredictions froma hydrological model is heavily dependent on theaccuracy of rainfall inputs (Gourley and Vieux, 2006), therefore,accurate estimate of the rainfall patterns over a catchment and aregion is a great concern (Kurtzman et al., 2009).Conventionalestimatesofdailyarealrainfallcanbeobtainedbyspatial interpolation of rain gauges’ data (Kurtzman et al., 2009).Various interpolation techniques have been proposed for arealrainfall estimations. The isohyetal and Thiessen polygon tech-niques are commonly used techniques of this kind (Guillermoet al., 1985). However, direct application of these techniques mayproduce inaccurate results because of the effects of topographical 0022-1694/$ - see front matter    2012 Elsevier B.V. All rights reserved.doi:10.1016/j.jhydrol.2012.01.013 ⇑ Corresponding author. Address: Nanjing Institute of Geography and Limnology,Chinese Academy of Sciences, 73 East Beijing Road, Nanjing 210008, PR China. Tel.:+86 25 86882102; fax: +86 25 57714759. E-mail addresses: (X.-H. Li), Zhang), (C.-Y. Xu). Journal of Hydrology 426–427 (2012) 28–38 Contents lists available at SciVerse ScienceDirect  Journal of Hydrology journal homepage:  variationandthelimitednumberofavailablerainfallstations(Tae-sombat and Sriwongsitanon, 2009). The geostatistical approaches,in particular the kriging method and inverse distance weighting(IDW) technique, have been widely applied for the estimation of spatially distributed rainfall. But, their results are influenced bythe heterogeneity of the random fields, and the assumption of anisotropic covariance structure in the kriging method is demon-strated to be inappropriate in several articles (Brown et al., 1994;Le et al., 1997; Kibria et al., 2002). Furthermore, most methodsused for interpolating rainfall have a tendency to produce toosmooth rainfall fields, i.e. to underestimate the spatial variability(Creutin and Obled, 1982; Haberland, 2007), which will affect theestimation of extreme values and undermine the strength of thedistributed hydrological models (Skaugen and Andersen, 2010).At the same time, it is both economically and practically impossi-ble to greatly increase the number of rain gages for estimating thespatial rainfall (Taesombat and Sriwongsitanon, 2009). Alterna-tively, the incorporation of satellite-based and weather radar-based (He et al., 2011) rainfall estimates in hydrological modellinghasthepotentialtoimproveourcapabilitytoreduceuncertaintyinrainfall inputs (Sawunyama and Hughes, 2008).Recent development in global and regional satellite-based pre-cipitation products has greatly improved their applicability as in-put to large-scale distributed hydrological models (Stisen andSandholt, 2010) and are expected to offer an alternative toground-based rainfall estimates in the present and the foreseeablefuture (Sawunyama and Hughes, 2008). Such data are especiallyvaluable in developing countries or remote locations, where con-ventional rain gauge data are sparse or of bad quality (Hughes,2006). Furthermore, the near-real-time availability of the satel-lite-based data products makes them suitable for modelling appli-cations where water resources management is crucial and datagatheringandqualityassurancearecumbersome(StisenandSand-holt, 2010). The use of satellite-based information to improve spa-tial rainfall estimates has been widely reported (Hsu et al., 1999;Sorooshian et al., 2000; Grimes and Diop, 2003). Nevertheless, sat-ellites data have biases and random errors that are caused by var-ious factors like sampling frequency, nonuniform field-of-view of the sensors, and uncertainties in the rainfall retrieval algorithms(Nair et al., 2009). It is therefore essential to validate the satellitederived products with conventional rain estimates to quantifythe direct usability of the products (Nair et al., 2009).The Tropical Rainfall Measuring Mission (TRMM) is a joint pro- ject between the National Aeronautics and Space Administration(NASA) and the Japan Aerospace Exploratory Agency (JAXA)launched in November 1997 with the specific objectives of study-ing and monitoring the tropical rainfall (Kummerow et al., 2000;Rozante et al., 2010). It can provide precipitation products withhigh temporal (3h) and reasonably high spatial resolution(0.25   0.25  ) for large-scale distributed hydrological models.There have been numerous attempts to validate TRMM retrievalswith ground-based estimates in the tropics and the mid-latitudes(Nair et al., 2009). Nicholson et al. (2003) used gauge data from a network of 920 stations over West Africa to evaluate TRMM (PR,TMI, 3B43) rainfall products for the year 1998. While TRMM PR and TMI products showed a net tendency to overestimate gaugemeasurements, 3B43 merged product showed an excellent agree-ment with gauge measurements on monthly to seasonal time-scales. Narayanan et al. (2005) validated TRMM 3B42-V5 datawith India Meteorological Department (IMD) rain gauge data andshowed that the satellite algorithm does not pick up very highand very low daily average rainfalls. Rahman and Sengupta(2007) compared the Global Precipitation Climate Project (GPCP),3B42-V5and3B42-V6rainfallproductswiththeIMDgriddeddailyrainfall at grid resolution of 1   1   for the monsoon season. Theirresults showed that GPCP and 3B42-V5 reproduce only the broad-est features of the monsoon rainfall, but spatial patterns of 3B42-V6 data show closest agreement with observed patterns of IMDgauge data except over certain places. Stisen and Sandholt (2010)evaluated five satellite-based rainfall estimates with temporal res-olution of daily and spatial resolution between 8 and 27kmthrough their predictive capability in a distributed hydrologicalmodel. However, most validation studies are performed at conti-nent/sub-continent or regional scale. Therefore fewer studies dealwith the comparison between TRMM rainfall and rain gauge dataat catchment scale, and no evaluation of hydrological processessimulation and water balance analysis using TRMM rainfall datain mesoscale catchments which will provide useful informationfor hydrology studies.Therefore, the objectives of the study are designed to (1) evalu-ateandcomparethetemporalcharacteristicofdailyTRMMrainfalland the spatial distribution of annual rainfall with that of the raingauge data in a mesoscale catchment located in the lower reachesoftheYangtzeRiverinChina.Bydoingso,differentstatisticalmea-sures arecalculatedandthe correlationsof theTRMMrainfall withrain gauge data at monthly time scale are investigated; and (2)cross compare the performance of the TRMM rainfall and raingauge data in driving the Water Flow Model for Lake Catchment(WATLAC) model (Zhang, 2007; Zhang and Werner, 2009; Zhangand Li, 2009) in simulation of daily and monthly hydrological pro-cessesatthecatchment.Emphasiswaspaidtoinvestigatethesuit-ability of the TRMM rainfalls for water balance analysis through adistributed hydrological model at different time scales, althoughsome researchers consider that most satellite-based rainfall esti-mation techniques are better suited at determining rain/no rainsituations compared to actually determining the rainfall amount(Stisen and Sandholt, 2010), and imprecise rainfall amounts andespecially biases are critical in water balance studies (Stisen andSandholt, 2010). This study contributes to the enhancement of knowledge regarding the usefulness of TRMM 3B42-V6 rainfalldata in hydrological modelling studies at catchment scale overvarying time scales.Therestofthispaperisorganizedasfollows.Inthenextsectionwe will provide details of the study area and the data used. In Sec-tion 3, the concept of WATLAC model is briefly described with thehelp of cited references. Major results of this study are presentedand discussed in Sections 4 and 5 summarizes the conclusions. 2. Study area and data preparation The Xinjiang catchment (27  33 0 –28  59 0 N and 116  23 0 –118  22 0 E)isselectedasthestudyarea,whichisoneofthefiverivercatchments of Poyang Lake (the largest freshwater lake in China)basin located in the lower reaches of the Yangtze River (Fig. 1).The catchment above Meigang Hydrological station covers about15500km 2 and has a subtropical wet climate characterized by amean annual precipitation of 1878mm for the period of 1960–2005 and annual mean temperature of 18  C. The topographyvaries from high mountainous and hilly areas (with a maximumelevation of 2138m.a.s.l) to alluvial plains in the lower reachesof the primary watercourses. The Xinjiang River flows primarilyfrom the east to the west and enters Poyang Lake. The averagestream flow at Meigang station for the 1960–2002 period was578m 3 /s.Based on the digital elevation model (DEM) data of the catch-ment which are derived from the National Geomatics Centre of China, the river networkand physical boundariesof the catchmentare delineated. Landuse map is available fromprevious studies (Yeet al., 2011a,b) as Fig. 2 shows. In the Xinjiang catchment, forest is the mainlandusecovering 84%of the catchment area, followedbycrop land of 10% and Shrubland of 5%. Other land uses such as  X.-H. Li et al./Journal of Hydrology 426–427 (2012) 28–38  29  grassland, water bodies and urban are minor with a total area of 1%. Land use condition is simulated in the model through theparameters of maximum canopy interception which is assumedto be linearly proportional to the Leaf Area Index (LAI) (ZhangandLi,2009).AndLAIforeachvegetationclasscanbederivedfromNational Oceanic and Atmospheric Administration/Advanced VeryHigh Resolution Radiometer (NOAA/AVHRR) Normalized Differ-ence Vegetation Index (NDVI) data through the Simple BiosphereModel Version 2 (SiB2) method (Myneni and Williams, 1994; Sell-ers et al., 1994, 1996; Andersen et al., 2002; Zhou et al., 2006): SR  ¼ 1 þ NDVI1  NDVI  ð 1 Þ FPAR  ¼ FPAR  min þð FPAR  max  FPAR  min Þ  SR   SR  min SR  max  SR  min ð 2 Þ LAI ¼ð 1  F  cl Þ LAI max ln ð 1  FPAR  Þ ln ð 1  FPAR  max Þþ F  cl LAI max FPAR FPAR  max ð 3 Þ where SR is the simple ratio of hemispheric reflectance for the NIR (near-infrared) light to that for the visible light, FPARis the fractionofphoto-syntheticallyactiveradiation, F  cl  isthefractionof clumpedvegetation, SR  min  and SR  max  are SR with 5% and 98% of NDVI popu-lation. The values of NDVI at 5% population are adopted from SiB2for all vegetation types (NDVI 5%  =0.039 globally). FPAR  min  =0.001and FPAR  max  =0.950 consider the satellite-sensed NDVI saturation.LAI max  is the maximum LAI when the vegetation develops fully. Some useful parameters for each vegetation class are shown inTable 1 from Zhou et al. (2006). ThesoilsinthecatchmentareclassifiedaccordingtotheGenet-ic Soil Classification of China, and soil distributions are obtainedfrom a soil survey completed by the Land Management Bureau of  Jiangxi Province, China. Soil types of the catchment are dominatedby paddy soil (47%) and red soil (45%); other types include yellowsoil (6%), latosol (1%) and a spot of yellow–brown 1 soil (0.7%) andpurplish soil (0.3%) as Fig. 3 shows. The properties of every soil typeare determined from the soil survey (Shi et al., 2004) and are shownin Table 2, with porosity ranging from 0.48 to 0.50, field capacityfrom 0.32 to 0.36, and saturated hydraulic conductivity varying from0.60 to 0.90 m/d. Satellite-based rainfall data used in this study are TRMM3B42-V6 daily data from 1 January, 1998 to 31 December, 2003. And forthe comparison of rainfall data between TRMM and rain gauges,we also use the rain gauge data from five national meteorologicalstations namely Yushan, Shangrao, Qianshan, Guixi and Yingtan Fig. 1.  Location of Xinjiang catchment in Poyang Lake basin and the distribution of stations. 1 For interpretation of color in Figs. 1–3 and 5–7, the reader is referred to the webversion of this article. 30  X.-H. Li et al./Journal of Hydrology 426–427 (2012) 28–38  as Fig. 1 shows. Moreover, other meteorological data includingdaily maximum and minimum temperature, solar radiation, windspeed, and relative humidity are derived from these national sta-tions and used in the study for calculating evapotranspirationandrelatedprocesses. Thesedatahavebeenwidelyusedfordiffer-ent studies previously and the qualities of the data are quite reli-able. We also examined the relation between elevation andrainfall to reflect the difference in mountainous region and in low-lands, but there is no clear evidence that the rainfall changed withelevation in the study region. So, the daily rainfall data are directlyinterpolated to grid (4km  4km) for the whole basin with themethod of Thiessen polygon to satisfy the requirement of the dis-tributedhydrologicalmodel.Inaddition,theobserveddailystreamflow from the Meigang gauging station is available to calibratemodel parameters and validate the simulation results. 3. Hydrological model The WATLAC model (Zhang and Werner, 2009; Zhang and Li,2009),isagrid-basedspatiallydistributedhydrologicalmodelwitheffective computational techniques to simulate complex spatialvariability of surface and subsurface flows. The model wasdesigned to simulate processes including canopy interception,overlandflow, streamflowrouting, unsaturatedsoil water storage,soil lateral flow, soil water percolation to groundwater and satu-rated groundwater flow driven by rainfall and evaporation. Theland surface (including river networks), unsaturated soil layerand saturated groundwater aquifer were coupled in the modeland can reflect the interaction of groundwater and surface water.The most of model parameters can be determined through fieldsurvey or literature values and only few parameters need to beestimated through calibration. The WATLAC model has been suc-cessfully applied for water balance analysis of Fuxian lake catch-ment (Zhang and Werner, 2009), surface–groundwater flowinteractions modelling of Xitiaoxi catchment (Zhang and Li,2009) and assessment of the effects of future climate change oncatchment discharges and lake water level of Poyang lake (Liuet al., 2009; Ye et al., 2011a). Details of model structure were pro-vided in Zhang and Li (2009) and Zhang and Werner (2009) and therefore only a brief description is given here.The WATLAC model first calculates the throughfall  P  n  takinginto account canopy interception which will be evaporated backinto the atmosphere and the maximum soil water storage  S  max .Once the  S  max  is filled, the exceeding throughfall becomes the sur-face runoff. The maximum soil water storage  S  max  is calculated as S  max  ¼ h s  /  ð 4 Þ where  /  is the porosity of the soil;  h s  is the thickness of the simu-lated soil layer (mm). The water that infiltrates into the soil subsequently percolatesdownwardsundergravitytothegroundwatertable,orflowlaterallyclosetothesurfaceassoillateralflow,orelseitmaybeevaporated. Fig. 2.  The landuse map of study area.  Table 1 Landuse threshold parameters from the literatures. Type LAI max  F  cl  NDVI 98%  Root depth (m) Permeable area (%) RoughnessCroplands 7.0 0 0.674 0.7 70 0.101Forests 5.7 0.5 0.721 2.5 60 0.122Shrublands 3.0 1.0 0.674 1.0 80 0.107Grasslands 1.8 0 0.674 0.5 90 0.085Water bodies – 0 0.674 1.3 a 5 0.073Urban and built-up – 0 0.674 0.1 5 0.047 a Root depths for water bodies represent the average water depth (Zhou et al., 2006).  X.-H. Li et al./Journal of Hydrology 426–427 (2012) 28–38  31  Thegroundwaterrechargerate R G , iscomputedasafunctionof thedrainablesoilwater,saturatedsoilhydraulicconductivityandshal-lowaquiferconductivity, similartothat inNeitschet al. (2002). Anempirical parameter  b 1  ( b 1 P 0) is introduced in the computation,through which the magnitude of the groundwater recharge can beadjustedand a larger value will result ina greater groundwater re-chargerate. Generally, it shouldbe set inthe range of 0.0–10.0andcanbebestestimatedinmodelcalibration.The soil lateral flow  R L  is calculated using a function of soil dra-inable water, soil hydraulic conductivity, soil slope length andslope gradient as that used in SWAT (Neitsch et al., 2002). Also,an empirical parameter  b 2  ( b 2 P 0) is introduced to reflect themagnitude of soil lateral value. This parameter is usually in therangeof0.0–10.0formostcasesandcanonlybeestimatedinmod-el calibration.Actual evapotranspiration calculation adopts the same ap-proach as that in USACE (2000), i.e., the total evapotranspirationis a sum of various components from canopy storage, soil storageand shallow groundwater. The potential evapotranspiration usedas the up limit of the actual evapotranspiration is calculated usingthe Penman–Monteith approach (Xu et al., 2006).OverlandflowroutesaregeneratedfromDEMbytheD-8meth-od considering time lag effects when the overland flow is trans-ferred from overland to known waterways. Stream flow routingis simulated using the Muskingum method. The saturated ground-water flow is simulated through MODFLOW-2005 (Harbaugh,2005) which was integrated in WATLAC and can achieve the inter-actionwithsurfacewaterflow,i.e.ontheonehand,thegroundwa-ter recharge calculated from the surface water model is passed totheMODFLOWforgroundwaterflowmodelling;ontheotherhand,groundwater table simulated from MODFLOW is used in surfacewater model to update the thickness of the soil column (Zhangand Li, 2009).The model parameters are automatically optimized by the PEST(Parameter ESTimation) optimization tool (Doherty, 2004). PEST isa robust and efficient model-independent parameter estimationsoftware, which uses the Gauss–Marquardt–Levenberg algorithmtoidentifytheparametersetthatgivestheleastsumofsquarediffer-ence between simulated and observed data, and has been widelyused for groundwater-surface water optimization problems (Keat-ingetal.,2003).Themodelperformanceisevaluatedusingstatisticalanalyses of model outputs. Evaluation criteria, e.g., Nash–Sutcliffeefficiency ( E  ns ) and determination coefficient ( R 2 ) are used to mea-surethecapabilityandreliabilityofthemodelindescribingtheob-servedprocesses. Inaddition, for evaluationof systematic errors inmodel simulation, the relative runoff depth error ( DE  ) is also ana-lysed.Thevaluesof  E  ns  and DE   arecalculated, respectively,as E  ns  ¼ 1  X ni ¼ 1 ð Qobs i  Qsim i Þ 2 = X ni ¼ 1 ð Qobs i  Qobs Þ 2 ð 5 Þ DE  ¼ X ni ¼ 1 ð Qsim i  Qobs i Þ = X ni ¼ 1 Qobs i  100 %  ð 6 Þ where  Qobs i  is the observed stream flow at step  i ;  Qsim i  is the sim-ulated streamflowat step  i ; and  Qobs  is the mean observed streamflow over all time steps; and  n  is the total time step. 4. Results and discussions 4.1. Validation of TRMM rainfall with rain gauges data For the comparison of rainfall data between rain gauges andTRMM, we first analyse several statistical indices of two types of  Fig. 3.  The soil type map of study area.  Table 2 The property of each soil type from the soil survey. Soil type Porosity Field capacity Saturated  K   (m/day)Red soil 0.48 0.34 0.67Latosol 0.47 0.34 0.60Yellow soil 0.50 0.35 0.79Yellow–brown soil 0.50 0.36 0.90Paddy soil 0.46 0.33 0.63Purplish soil 0.48 0.32 0.8632  X.-H. Li et al./Journal of Hydrology 426–427 (2012) 28–38
Similar documents
View more...
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks