The Suitability of the Satellite Metrological Inputs Source on the Hydrological Model in a Small Urban Catchment

Distributed/semi-distributed models are considered to be sensitive to the spatial resolution of the data input. In this paper, we take a small catchment in high urbanized Yangtze River Delta, Qinhuai catchment as study area, to analyze the impact of spatial resolution of precipitation and the potential evapotranspiration (PET) on the long-term runoff and flood runoff process. The data source includes the TRMM precipitation data, FEWS download PET data, and the interpolated metrological station data. GIS/RS technique was used to collect and pre-process the geographical, precipitation and PET series, which were then served as the input of CREST (Coupled Routing and Excess Storage) model to simulate the runoff process. The results clearly showed that, the CREST model is applicable to the Qinhuai catchment; the spatial resolution of precipitation had strong influence on the modelled runoff results and the metrological precipitation data cannot be substituted by the TRMM data in small catchment; the CREST model was not sensitive to the spatial resolution of the PET data, while the estimation fourmula of the PET data was correlated with the model quality. This paper focused on the small urbanized catchment, suggesting the influential explanatory variables for the model performance, and providing reliable reference for the study in similar area.


Introduction
With the social and economic development in recent decades, floods and droughts prediction, water resources management, and water supply/control infrastructure construction have been the prime focus of hydrology (Barua et al., 2013;Brown et al., 2015;Kalbus et al., 2011).Watershed models describing the metrological parameters and the surface hydrological process precipitation were widely applied to the achieve the new goals of the modern hydrological research and investigations (Henriksen et al., 2003;Jakeman and Hornberger, 1993;Kvočka et al., 2015;Sivapalan et al., 1996).Accurate metrological data reflecting the spatial and temporal variability are crucial for reliable hydrological modeling (Strauch et al., 2012).Previous researches have pointed out that the resolution, of precipitation data can cause serious errors in model outputs (Andréassian et al., 2001;Bárdossy and Das, 2008;Lopes, 1996).The diversity of commonly used potential evapotranspiration (PET) estimations indicates a variation of almost an order of magnitude and predicts a wide range of runoff changes applied in hydrologic model (Boughton and Chiew, 2007;Ekström et al., 2007;Milly, 2015;Oudin et al., 2005a).
Predicting precipitation based on satellite images has been widely discussed, and several prediction methods were proposed corresponding to various electromagnetic spectrum (Dingman, 2002).
Among them, the Geoestationary Operational Environmental System (GOES) series (Vincente et al., 1998) and the Tropical Rainfall Measuring Mission (TRMM) were adopted mostly (Kummerow et al., 2000).TRMM was launched in November 1997 and works as precipitation monitor in the tropical area for the joint project between the National Aeronautics and Space Administration (NASA) and the Japan Aerospace Exploratory Agency (JAXA) (Kummerow et al., 2000;Rozante et al., 2010).It can provide precipitation products with temporal resolution of 3h and spatial resolution of 0.25° × 0.25° for large-scale distributed hydrological models.numerous attempts to validate TRMM retrievals with ground-based estimates are performed and excellent agreement with gauge measurements on monthly to seasonal timescales at continent/sub-continent or regional scale were recognized (Nair et al., 2009;Nicholson et al., 2003;Stisen and Sandholt, 2010).
Successful application of TRMM precipitation data in distributed hydrological models of large-scale or data scarcity basin has been witnessed (Meng et al., 2014;Rozante et al., 2010;Xue et al., 2013), while its suitability in the small basin is not reported yet.Another conventional estimation of daily areal rainfall can be obtained by spatial interpolation of rain gauges' data.Direct interpolation techniques has been pointed out to neglecting the topographical variation and limited by the distribution of the available of precipitation stations (Taesombat and Sriwongsitanon, 2009).
To represent the potential evaporative demand introduced into a model often poses a dilemma to the scientist and engineers in the past decades, due to its agronomic concept and sampling scarcity (Oudin et al., 2005a).A wealth of studies adopted the empirical (Chiew and Mcmahon, 1992;Chiew and McMahon, 1991;Tait and Woods, 2007) or estimated PET input, which contains only mean values but deviated from the real climatic condition.The Famine Early Warning System (FEWS) projected by the U.S. Agency for International Development (USAID) originally for food security analysis for extensive areas of sub-Saharan Africa, provides PET data globally with spatial resolution 0.25° × 0.25°, and has been adopted in various hydrological models (Liu et al., 2005;Verdin et al., 2000).Validation studies disputed about the appropriate data source of PET.Although many hydrologists have seen no output difference between the models driven by mean PET and PET reflected spatial and temporal variations (Burnash, 1995;Fowler, 2002), PET estimations based on temperature and solar radiation tend to provide the best streamflow simulations in other researches (Oudin et al., 2005a(Oudin et al., , 2005b)).Finding the most adequate PET input to distributed rainfall-runoff models to improve the streamflow simulations would be an interesting issue.
In addition to the data inputs, the model structure is considered to be one source of uncertainty in hydrological simulation (Butts et al., 2004;Montanari and Di Baldassarre, 2013).Plenty of distributed and half distributed have been constructed and applied in multi-scale basins or regions globally, and each of them has advantages.The distributed and semi distributed hydrological models, HBV (Lindström et al., 1997), TOPMODEL (Valeo and Moin, 2000), Mike SHE (Im et al., 2009), SWAT (Franczyk and Chang, 2009), HEC-HMS (Lin et al., 2009), DHSVM (Chu et al., 2010), HGS (Brunner and Simmons, 2012) and GR (Ficchì et al., 2016), for example, have been extensively used to assess the hydrologic processes.How to select the model, to sufficiently reflect the real situation of the catchment and focus on the selected scientific question, is essential in the application of the new powerful research tool.
The objective of this paper is to describe and discuss the reliability of TRMM precipitation and Special attention was paid to the resolution and estimation method of the input metrological data.
Finally, based on the metrological input with highest efficiency, the applicability and accuracy of the CREST model in small urbanized catchment was discussed.Oklahoma (http://hydro.ou.edu) and NASA SERVIR Project Team (www.servir.net), is a distributed hydrological model aims at surface and subsurface runoff and storages simulation in a cell-to-cell algorithm (S.I. Khan et al., 2011;S.I. Khan et al., 2011).The core principle of CREST includes the surface runoff generation processes, the cell-to-cell runoff routing, and the coupling of runoff generation and routing scheme.Variable infiltration capacity curve (VIC) based on kinematic wave assumption is adopted in the surface runoff generation calculation.Multi-linear reservoirs are used to simulate cell-to-cell routing of surface and subsurface runoff separately.Finally the coupling mechanism reconstruct the surface and subsurface water flow process cell-to-cell (Meng et al., 2014;Xue et al., 2013).The grid based structure of CREST enables multi-scale modeling research, as well as detailed and realistic treatment of hydrological variables, e.g.soil moisture (Meng et al., 2014).

Study area
The framework of CREST model is shown in Fig. 2.

Kriging interpolation
Kriging interpolation technique, named after the South-African mining engineer who developed this method in the 1950s, was adopted in our research (Matheron, 1963).It was originally proposed to evaluate the natural resources storage, and recently are widely used in constructing grid data from point data series, or predicting data at location with no data based on the spatial autocorrelation of observed data after improvement (Aalto et al., 2012;Zhu and Li, 2009).The basic formula are as follows,   In order to test the accuracy of Kriging interpolation.The data from the eighth precipitation station, Xiajiabian were adopted.We abstracted the interpolated precipitation data from the distribution map according to the coordination of Xiajiabian, and then compared it with the gauged precipitation.The mean absolute error and relative error were around 1 mm/day and -1.84% respectively.This proved the high reliability of the Kringing interpolation.In addition, comparison between the averaged station precipitation and the TRMM precipitation were made and the results were shown in Fig. 5.The TRMM data tends to overestimated the storm volume in general.The relative error between the two series was around 68.79%.evaluating the other two series.The PET data can be estimated by the empirical formal, based on the energy and temperature (Hargreaves and Allen, 2003;Hargreaves and Samani, 1985).
Estimation based on Blaney-Criddle method, Abtew method and Hargreaves were carried out, and the results indicated that the Hargreaves estimation shows the highest agreement with the gauged data.Therefore, we mainly focus on the explanation of Hargreaves method.
Hargreaves methodology was first proposed in 1975, further developed in 1985, and has been proven to be very efficient and accurate in PET estimation (Archibald and Walter, 2014;Klein et al., 2015).The formula is as follows, = ∆ .( + 17.8) where PET is the daily evapotranspation mm/day; SP is the potential solar radiation (kJ/m 2 /day); KET is the calibration coefficient; ΔT is the daily temperature range (Tmax-Tmin) (°C); T is the daily averaged temperature (°C).
A modification was made to the synthetic PET based on the empirical coefficient correction method, to get the final daily PET of the current study area.Fig. 6 gives the trend lines of the three PET series.The estimated and downloaded data fit the gauged data well in the troughs time period.
Pervading over estimation occurred at the peak value position, especially from 2002 to 2007.The Hargreaves estimated data has relatively higher quality, compared with the downloaded one.Moreover, the simulated date shows understand estimation trend in general, despite the results of storm events were in relatively better accordance.Compared the three simulation, T1, T2 and S, the influence of the precipitation data source was much higher than that of the PET data.The low resolution and high over estimation of the TRMM precipitation data lead to the over estimation of the runoff.While the under estimation of the model T2 and S were resulted from the over estimation PET data.Although the FEWS PET data has lower spatial resolution and higher error combined to the gauged PET data, the model results were comparable to the model S.This indicated the insensitivity of the CREST model to the PET data source.
Both monthly and daily simulation were carried out to the three models.

Storm simulation
The simulation during flood season or storm event is of great significance, especially for the small urban catchment confronting increasing flood risk.Six classical storm events, three of which were from calibration term and the other three from validation term, were picked up to test the simulation quality of model T2 and S in this part.The results were shown in Tab. 5 and Fig. 10.In general, the trend lines of T2 and S simulated runoff were nearly overlapped with each other, and reflected the trend of the observed data.The averaged NSCE and r of the six events were 0.84 and 0.89 for T2 model, and 0.86 and 0.90 for S model respectively.The averaged MAE were 67.30 and 62.09 mm/d, which confirmed the high quality of the FEWS PET data driven model.The estimated flood peak volume were lower than observed data, while the flood rising and recession process were mostly over estimated, which resulted in relatively short and wide flood hydrograph.
Tab. 5.The efficiency of storm runoff simulation   As shown in Fig. 11, the Bias of the estimation have a clearly increase trend with the growing of peak volume, with the determination coefficient (R 2 ) of 0.78 and 0.69 for T2 and S. When the peak flow was under 1000 mm, the underestimation occurred, while over estimation happened only when the peak reach to the 1200 mm.The model seems to be more applicable to the medium peak event, with volume around 1000-1200 mm.Plenty of literature about the suitability of TRMM precipitation data in hydrologic models are available.The TRMM data were normally adopted in large scale catchment and TRMM-based calculated hydrographs are comparable with those obtained using station data, with better monthly performance and discounted daily performance (Collischonn et al., 2008;Huffman et al., 2007;Meng et al., 2014;Nicholson et al., 2003).In our study, the TRMM data driven CREST model reflected the runoff process in rather poor quality in monthly scale, which become worse in daily scale.According to the comparison TRMM data tend to over estimated the precipitation, especially during the storm event in this area.The substitution of gauged data with TRMM data is proven to be impractical in this small catchment, despite the relatively homogeneity of topography and climate condition inside the catchment.

PET data and its uncertainty
Various estimation of PET based on metrological parameters were proposed by previous researchers.The referenced data from evaporating dish was a point data, and is not able to describe the spatial variation of the catchment.The wide applied one based on temperature and solar radiation, such as Hargreaves estimation, was proven to be more reliable.On the contrary, the estimation based only on temperature, for example Blaney-Criddle PET data showed lower agreement with real data (Fig. 12).When we drove the CREST model with the Blaney-Criddle PET data, the efficiency declined compared with the Hargreaves and FEWS based model, especially during the storm event simulation.In conclusion, despite the spatial distribution of PET data was revealed in the interpolated-estimated data, the simulated results from these data were not improved much.The model was more responsive to the PET from different estimation method, instead of the spatial resolution of PET.The sensitivity of CREST model to precipitation data was strong.TRMM precipitation was not sufficient to substitute the station gauged data in Qinhuai catchment even in simulation with monthly step.
CREST model was not responsive to the spatial resolution of PET data.Both FEWS and gauge-estimated PET data input provided satisfied runoff output.Instead, the estimation method of the PET data was influential to the model performance.
Compared with other distributed/semi-distributed models, CREST produced acceptable long-term runoff series, and high accurate flood runoff simulation.
The above discussion and conclusions indicate the necessity of further develop of satellite precipitation monitor, both in spatial resolution and data precision, especially during storm event.
The new generation of satellite product, GPM, is expected to improve the availability and accuracy of precipitation estimation (Wang, 2015).Accordingly, the hydrological simulation driven by GPM precipitation will be improved.The largely improvement of satellite precipitation would be combined with, but never completely replace the conventional precipitation data.
The data quality of PET series, mainly depend on the estimation formula and sample approach.
Given most hydrological model is not responsive to the spatial resolution, but the data quality of PET, to improved the satellite monitor PET in the future would be sufficiently improve the performance of multi-scale hydrological simulation.
tropical cyclone and subtropical high press belt, has severely increased flood disaster, especially in the downstream.The main soil types includes yellow-brown soil, purple soil, limestone soil, paddy soil, and gray fluvo-aquic soil.The paddy filed and dry land occupy most area of the basin, while the woodland, impervious surface, water cover the rest.Data collected from the seven rain gauging stations and two stream flow gauging stations at the outlets of the basin were adopted for the current study.The location of the study area and gauge stations, elevation, and river network pattern are given in Fig. 1.

Fig. 1 .
Fig. 1.Location of the Qinhuai basin and the distribution of hydrological stations 2.2 CREST Model

Fig. 2 .
Fig. 2. Core components of CREST: (a) vertical profile of a cell including rainfall-runoff generation, evapotranspiration, sub-grid cell routing and feedbacks from routing; (b) variable infiltration curve of a cell; (c) plane view of cells and flow directions; and (d) vertical profile along several cells including sub-grid cell routing, downstream routing and subsurface runoff redistribution from a cell to its downstream cells.
) here, F(x,y) refers to the estimated value on the point with cooridination (x,y), n represents the number of discrete points, is the value of each point, means the weight of the points, is the weight coefficient, menns lagrange multiplier, is the variation coefficient, data of the catchment were downloaded from the hydrologic data center of the United States Geological Survey (USGS, http://hydrosheds.cr.usgs.gov/dataavail.php).The original digital elevation model (DEM), flow area chart (FAC) and the river flow direction (FDR) data were corrected and rectified in Arcgis 10.1, and finally tailored according to the border of study area (Fig.3).

Fig. 3 .
Fig. 3.The basic geographical information of the study area, (a) DEM, (b) FAC, (3) FDR.The river network data were abstract from the 1:50000 digital topographic map (DGM), with the help of thematic map (TM image) with the resolution of 30 m from 2008.Finally 724 rivers were abstracted, among which, Jurong river and Lishui river are the main branch rivers of the catchment, with drain area of 1280 km 2 and 902 km 2 respectively.

Fig. 6 .
Fig. 6.The PET data from gauage station (solid line), Hargreaves estiamtion (dashed line) and FEWS downloaded (dotted line).The interpolation processing based on the IDW The source, resolution, interpolation method and the number of the data source about the precipitation and PET data used to drive the model were summarized in table.

Fig. 9 .
Fig. 9. Comparision of the daily observed and simulated runoff series of the Model-S.

Fig. 10 .
Fig. 10.The observed (solid lines) and simulated runoff series of model T2 (dotted lines) and S (dashed lines) during storm event.The simulated and observed peak flow data were shown in Tab. 6.The FEWS PET driven model produced peak flow results with higher accuracy, with a MAE of 118.48mm, 23mm less that the error of model S. The simulated peak volume is -10.52% and -12.13% lower than observed data averagely, from model T2 and S respectively.The only over estimation occurred during the highest flood peak, with volume around 1214 mm in July 2007.

Fig. 11 .
Fig. 11.The estimated errors and trend lines of simulated peak flow of model T2 (round dot, solid line) and S (square dots, dashed line).

Fig. 12 .
Fig. 12.The evaporating dish PET against Blaney-Criddle and Hargreaves estimated PET, the solid line is the ideal fitting line, and the dotted line is the real fitting line.

Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 13 August 2016 doi:10.20944/preprints201608.0134.v1 estimated
PET data in the hydrological modeling in a small urbanized catchment in comparison with the gauged data, based on CREST model (Coupled Routing and Excess Storage model).
The daily simulated results of T1 model was hundreds times higher or lower than the observed value, indicating very less practical significance.Therefore it was not shown in this paper.The monthly simulation efficiency of T1 has been discussed in former part.For the reason of non-repetition, only the monthly simulation efficiency of model T2 and S were shown here (Tab.4).Slightly superiority of models S was found according to Tab. 4.

Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 13 August 2016 doi:10.20944/preprints201608.0134.v1
Only four same flood event simulation based on HEC-HMS and CREST were available, the volume and simulated error of which have been shown in Tab. 8.When we compared the Bias and MAE of each flood and the averaged condition, the CREST simulated volume is much closer to the observed data (Tab.8).While during peak simulation, the HEC-HMS model presented higher accuracy (Tab.8).Compared with the HEC-HMS, CREST model require less catchment underlying surface information and has superior efficiency in model construction and calibration.Despite the relative simpler of structure, the CREST model provide very high quality results in the flood simulation.