A PARAMETER ESTIMATE FOR THE LAND SURFACE MODEL VIC WITH HORTON AND DUNNE RUNOFF MECHANISM

FOR RIVER BASINS IN CHINA

ZHENGHUI XIE

Institute of Atmospheric Physics, ChineseAcademy of Sciences
Beijing,100029, China

FEI YUAN, QIAN LIU

Institute of Atmospheric Physics, ChineseAcademy of Sciences
Beijing,100029, China

This paper presents a parameter estimate of the land surface model VIC to simulatestreamflow for river basins in China by a methodology for model parametertransfer that limits the number of basins requiring direct calibration, where thenew surface runoff parameterization that represents both Horton and Dunne runoffgeneration mechanisms with the framework of considering subgrid spatial scale soilheterogeneity in VIC is applied. The mainland area of China is represented by 4355cells with a resolution of 50 ×50 km2 for each cell and grouped into climate zones. Initially, some of model parameterswere calibrated for nine catchments,and those for thenine catchments were transferred within the climate zones. The transferred parameters werethen used to simulate the water balance in six other catchments and river basinsin China. The simulated daily runoff of VIC-3L with transferred and un-calibratedparameters is routed to the outlets of the catchments, and compared to the monthly-observed streamflow at the related gauge stations. As a whole,the parameter transferapproach reduced the bias and relative root mean squared error (RRMSE) and increased the Nash-Sutcliffe model efficiency coefficients for each individual catchment, and the parameter transfer scheme improved the streamflow simulation.Subsequent recalibrationof all basins further enhanced the modeling performance. Results show that the model forthe transferred parameters can simulate the observations accurately.

INTRODUCTION

As a macro-scale hydrological model, the Variable Infiltration Capacity (VIC) model [1,2,3,4] has been used to simulaterunoff over large river basins. One practical problemin application of the model in river basins in China is the determination ofmodel parameters. Studies have shown that land surface models could performwell if their model parameters were appropriately estimated on the basis of calibrationswith observations but perform poorly if their model parameters arenot calibrated properly. To better applythe VIC model to simulate streamflow for river basins in China, we presenta parameter estimation of the land surface model VIC by a methodology ofmodel parameter transfer that limits the number of basins requiring directcalibration, where the new surface runoff parameterization that representsboth Horton and Dunne runoff generation mechanisms with the frameworkof considering sub-grid spatial scale soil heterogeneity is applied.

MODEL IMPLEMENTATION AND DATA SOURCES

VIC-3L model

Liang et al. [1] developed a two-layer Variable Infiltration Capacity (VIC-2L) model which includes two different time scales (i.e., fast and slow) forrunoff to capture the dynamics of runoff generation. To better representquick bare soil evaporation following small summer rainfall events, a thin soillayer is included in VIC-2L [2], and VIC-2L becomes VIC-3L.Liang and Xie [3] developed a new parameterization to represent the infiltration excess runoff mechanism in VIC-3L and combined it effectively withthe original representation of saturation excess runoff mechanism [4]. To apply the VIC-3L including the new surface runoff mechanismto river basins in China, the model parameters have to be determined.

Data and model parameters

The VIC-3L model requires three types of data sets,which are vegetation,soil, and forcing data. In this study, vegetation, soil, and forcingdata needed to apply VIC-3L are prepared at 50 × 50 km2resolution forthe river basins in China.

Vegetation data setsare derived based on AVHRR [5] andLDAS information. Soil parameter sets are deduced from the NOAA global 5-min soil data and the work of Cosby et al. [6] and Rawls et al. [7]. For detailed information about theVIC-3L vegetation and soil parameters, please refer to Su and Xie [8].

The forcing data are obtainedthrough interpolation methods (minimum distance method and inverse distance square method)

based on 740 meteorological stations,which contain 11 years of daily precipitation and air temperature data from1980 to 1990. Some of forcings for two gauge stations are from 1990 to 2000.

CALIBRATION AND PARAMETER TRANFER

Before conducting numerical simulations, some model parameters of VIC-3Lneed to be calibrated since they cannot be determined well based on the availablesoil information. These are the depths of three soil layers (D1, D2 and D3),the exponent (B) of the VIC-3L soil moisture capacity curve which describesthe spatial variability of the soil moisture capacity, and the three parametersin the ARNO subsurface flow parameterization (i.e., Dm, Dsand Ws)[1].

Classification of climate zones

Climatic characteristics were selected as the basis for the transfer of calibrated parameters under the premise that hydrological processes and the parametersused to describe them are more similar within than between different climatezones. Based on the forcing data, the mainland area of China was grouped into six climate zones according to Köppenclassification rules [9].Figure 1 shows the main river basins in China and the climate zones.

Figure1. The river basins in China, and climate zones of China

Table 1. Selected river basins

River basin / Catchment / Predominant climate zones / Area(km2)
upstream of gauge
Category 1
Yellow River / Qinan / Continental climate with cool summer / 9,805
Nanhechuan / Continental climate with cool summer / 1,3580
HaiheRiver / Xiahui / Continental climate with hot summer / 5,340
Xiabao / Continental climate with hot summer / 4,040
Yangtze River / Wuhouzhen / Rainy, mid latitude climate / 3,092
Madao / Rainy, mid latitude climate / 3,415
HuaiheRiver / Xixian / Rainy, mid latitude climate / 10,190
Bantai / Rainy, mid latitude climate / 11,280
HeiheRiver / Zhamashike / Continental climate with short cool summer / 4,589
Category 2
Yellow River / Yangjiaping / Continental climate with cool summer / 14,124
HaiheRiver / Xiangshuibao / Continental climate with hot summer / 14,140
Yangtze River / Hanzhong / Rainy, mid latitude climate / 9,329
Chadianzi / Rainy, mid latitude climate / 1,683
HuaiheRiver / Luohe / Rainy, mid latitude climate / 12,150
HeiheRiver / Yingluoxia / Continental climate with short cool summer / 10,009

Selected catchments

Table 1 lists the fifteen selectedcatchments in China for calibration and parameter transfer. Nine catchments in category 1 in Table 1, calledthe primary catchments, were calibrated to produce parameter cluster for each climate zone. We tried to select at least one catchment for each climate zone, but due to the unavailability of streamflow data for dry and cold climate zone and for tropical climate zone, no catchments were selected for these two climate zones, and parameters for these zones were set to be default values without calibration.

As shown in Table 1, six catchments in category 2 (secondary catchments) were initially modeledusing parameters transferred from the calibrated primary catchments. Observedand simulated streamflow for these catchments were compared to determinethe effectiveness of the parameter transfer method.The secondary catchments were then recalibrated in a second stage of the study.This second stage served two purposes. First, calibration of the secondarycatchments served to further evaluate the effectiveness of the parameter transferprocess. Ideally, this calibration should result in minimal improvement ofmodel results, which would indicate that the parameter transfer process washighly successful. Second, the second stage calibration ensured that the estimatesof water balance components were the best possible, given the modeland meteorological forcings. In a final step, the calibrated parameters from allcatchments were transferred to the remaining land surface grid cells in riverbasins in China to allow estimation of the continental and water balance.

Model calibration

In this study, model calibration focused on fitting streamflow data, since other model-predicted water storage and flux components(e.g. soil water storage, snow cover, evapotranspiration) are rarely observedat spatial and temporal scales suitable for direct comparison with theoutput from macroscale hydrological models.Calibration was performed manually and focused on matching the totalflow volume and the shape of the monthly hydrograph. Relative error (Er) between simulated and observed mean annualrunoff, and the Nash-Sutcliffe coefficient (Ce) were selected as the criteria for model.

Parametertransfer scheme

The infiltration parameter (B) and the depths of the three-soil layer (D1, D2 and D3), and the ARNO model parameters Dm, Dsand Wswere calibrated and thentransferred to the river basins in China. Parameters were transferred based on climate zone. The detail transfer scheme is described as follows: (1) Two catchments in theYellow River Basin are selected to calibrate the above model parameters, the six parameters for the two catchments are averaged respectively as the correspondingparameters for the zone of continental climate with cool summer. (2) Two catchments in the HaiheRiver Basin are selected to calibrate, andthe parameters for the two catchments are averaged respectively as the correspondingparameters for the zone of continental climate with hot summer.(3) Because of the unavailability of enough streamflow data, only one catchmentin the HeiheRiver Basin is selected to calibrate, and the parametersfor the catchment are set to those corresponding parameters for the zone ofcontinental climate with hot summer. (4) Most of area in the HuaiheRiverBasin and the YangtzeRiver Basin belongs to the zone of rainy, mid latitudeclimate. Two catchments in the HuaiheRiver Basin are selected to calibrate,and these parameters for the two catchments are averaged respectively as thecorresponding parameters for the zone of rainy and mid latitude climate locatedin the HuaiheRiver Basin. Two catchments in the YangtzeRiver Basinare also selected to calibrate, and the parameters for the two catchments areaveraged respectively as the corresponding parameters for the zone of rainyand mid latitude climate located in the YangtzeRiver Basin. Parameters forthe rainy and mid latitude climate zone north of the HuaiheRiver Basin andthe YangtzeRiver Basin are set to that for the HuaiheRiver Basin; parametervalues for the climate zone south of these two river basins are equivalent tothat for the YangtzeRiver Basin. (5) The zone of tropical climate has similarclimatic characteristics as those in rainy and mid latitude climate zone.Therefore, the parameters for the zone of tropical climate are set to be thecorresponding parameters for the YangtzeRiver Basin.(6)Sincestreamflowdata for the zone of dry and cold climate is not available, default values ofB, D1, D2, Dm, Dsand Wsfor the area are set to be 0.3,0.1, 0.5, 2.0, 0.02, 8.0, and 0.8 respectively.

To evaluate the effectiveness of the parameter transfer process and to providethe best possible water balance estimates, the secondary catchments were thenfurther calibrated after transferring the parameters from all of the calibratedcatchments to the remaining land surface grid cells.

SIMULATED RESULTS
Primary Catchments

The VIC model also provides a default parameter set, namely base case parameter set, which can be a parameter substitute when no calibration is performed. Comparisons were made between the results of base case and calibration. Figure 2 shows the observed and the simulated mean monthly hydrographs for the nine primary catchments based on base case (no calibration) and calibrated parameter sets. The model performance was considerably better for the calibration parameters than the parameters without calibration. The model in base case commonly largely overestimated the streamflows for Qinan, Nanhechuan, Xiahui and Xiabao catchments and underestimated the streamflow for Zhamashike catchment, but the modeled streamflows using calibrated parameters fitted the observed well. The model in base case and calibration case both provided good simulation results for Wuhouzhen, Madao, Xixian and Bantai catchments, while the simulated streamflows in calibration case were closer to observed ones as compared with the simulated results in base case. Table 2 lists the results for the nine primary catchments based on base case and calibration. In general,calibration improved the results in all instances, although in some cases thefinal calibration was still unsatisfactory, especially for arid basins such as theHaiheRiver, which flows through a region with strong human activities. Calibration reduced the mean bias from to 111.1% to 9.1%and the relative root mean squared error (RRMSE) from 43.9% to 8.8%. Theefficient coefficients (CE) for calibration are all higher than 70% except Xia Bao station (21.8%) in Haihe river basin. These results indicated that after calibration the VIC model could perform good streamflow simulation for the nine primary basins and the calibrated parameters could be used reasonably to transfer over the secondary catchments.

Figure 2. Mean monthly hydrographs of observed and simulated flow (base case and calibrated) for the primary river basins

Table 2. Calibration and parameter transfer statistics

River
basin / Gauge
station / Base case / Parameter transfer / Calibration
CEa / RRMSEb / Biasc / CEa / RRMSEb / Biasc / CEa / RRMSEb / Biasc
Category 1
Yellow / Qinan / -1961.7 / 94.6 / 270.5 / — / — / — / 70.9 / 11.3 / 11.5
Nanhechuan / -694.1 / 48.6 / 133.1 / — / — / — / 70.9 / 9.3 / 13.9
Haihe / Xiahui / -1150.8 / 94.7 / 248.4 / — / — / — / 71.3 / 14.3 / 3.1
Xiabao / -4765.4 / 79.5 / 197.4 / — / — / — / 21.8 / 7.7 / -1.4
Yangtze / Wuhouzhen / 76.2 / 15.3 / -20.4 / — / — / — / 97.0 / 5.5 / 4.1
Madao / 66.0 / 15.5 / -35.2 / — / — / — / 97.9 / 3.8 / 0.9
Huaihe / Xixian / 76.4 / 11.0 / -6.4 / — / — / — / 83.9 / 9.0 / 3.7
Bantai / 78.2 / 11.8 / 29.9 / — / — / — / 84.1 / 10.1 / 21.8
Heihe / Zhamashike / 58.1 / 23.8 / -58.2 / — / — / — / 89.2 / 8.2 / -21.5
Category 2
Yellow / Yangjiaping / -646.9 / 47.8 / 132.4 / 84.3 / 6.9 / -11.8 / 86.6 / 6.4 / -3.2
Haihe / Xiangshuibao / -52613.7 / 162.6 / 437.6 / -3666.0 / 43.5 / 87.8 / -299.4 / 14.2 / 1.0
Yangtze / Hanzhong / 72.2 / 15.1 / -28.1 / 95.4 / 6.1 / -2.3 / 95.7 / 5.9 / 0.6
Chadianzi / 77.1 / 13.8 / -26.5 / 97.6 / 4.4 / 9.8 / 96.6 / 5.3 / 8.7
Huaihe / Luohe / 83.2 / 8.8 / 14.8 / 79.0 / 9.9 / 18.0 / 86.7 / 7.9 / 8.2
Heihe / Yingluoxia / 72.7 / 22.0 / 10.9 / 74.6 / 24.5 / 73.6 / 89.3 / 13.7 / 24.7

a CE - Nash-Sutcliffe model efficiency coefficient: , with and the simulated and observed flow in month i.

bRRMSE – relative root mean squared error, defined as: .

c bias, defined as:

Secondary Catchments

Figure 3 shows the observed and the simulated mean monthly hydrographs for the six secondary catchments in base case, parameter transfer and recalibration. For the remaining six basins, theparameter transfer process improved the simulated flow volume in four cases(Yangjiaping, Xiangshuibao, Hanzhong, and Chadianzi) with the absolute value of bias being reduced from 132.4% to 11.8%, 437.6% to 87.8%, 28.1% to 2.3% and 26.5% to 9.8% respectively, and resulted in a little worse change in two cases (Luohe, and Yinghuoxia) with the absolute value of bias being increased from 14.8% to 18% and 10.9% to 73.6%. The transferred parametersreduced the relative (monthly) RRMSE for all cases except Luohe and Yingluoxia catchments, and increasedall the efficiency coefficients(CE) other than that of Luohe catchment. As a whole, the parameter transfer scheme improved the runoff simulation for the secondary catchments. Subsequent recalibrationof all basins further enhanced the modeling performance. Although Xiangshuibao catchment in the HaiheRiver Basin was involved in intense human activities where runoff simulation was a tough task, simulation for Xiangshuibao catchment in recalibration case was still improved. The recalibrated model reduced the average RRMSE from 22.0% to 8.9%and the average absolute value of bias from 33.9% to 7.3%, and increased the mean efficiency coefficient(CE) from86.2% to 91.0%, where the CEs of Xiangshuibao catchment were not in statistics.

Figure 3. Mean monthly hydrographs of observed and simulated flow (base case, parameter transfer and recalibrated) for the secondary river basins

CONCLUSIONS

In this paper, a parameter transfer scheme for VIC-3L is given to simulate streamflow for river basins in China, which is represented by 4355cells with a resolution of 50 × 50 km2and was grouped by climate zone, and model parameters were transferred within zones.The transferred parameters were then used to simulate the water balance inriver basins in China. The simulated daily runoff of VIC-3L with transferredparameters and un-calibrated parameters was routed to the outlets of theriver basins, and compared to the monthly-observed streamflow at the relatedcatchments. Results show that the model for the transferred parameters cansimulate the observations well and the proposed parameter transfer frameworkis promising in estimating the VIC model parameters for data-sparse areas inChina.

ACKNOWLEDGEMENTS

This work was supported by the National Natural Science Foundation of China(Grant Nos. 40275023), the National Key Planning Development Project forBasic Research (Grant Nos. 2001CB309404), the Hundred Talents Program ofthe ChineseAcademy of Sciences, and the Knowledge Innovation Key Projectof Chinese Academy of Sciences(Grant No. KZCX2-SW- 317).

REFERENCES

[1]Liang, X., Lettenmaier, D. P., Wood, E. F. and Burges, S. J., “A simple hydrological based model of land surface water and energy fluxes for general circulation models”, Journal of Geophysical Research. Vol. 99, No. D7, (1994), pp 14,415-14,428.

[2]Liang, X., Wood, E. F. and Lettenmaier, D. P., “Surface soil moisture parameterization of the VIC-2L model: Evaluation and modifications”, Global and Planetary Change, Vol. 13, (1996), pp 195-206.

[3]Liang, X. and Xie Z., “A new surface runoff parameterization with subgrid-scale soil heterogeneity for land surface models”, Advances in Water Resources, Vol.24, (2001), pp 1,173-1,193.

[4]Xie Z., Su. F., Liang, X., et al., “Applications of a surface runoff model with Horton and Dunne runoff for VIC”, Advances in Atmospheric Sciences. Vol. 20, No.2, (2003), pp 165-172.

[5]Hansen,M.,DeFries, R., Townshend, J. R. G. and Sohlberg, R., “Global land cover classification at 1km resolution using a decision tree classifier”,International. Journal of Remote Sensing, Vol. 21, (2000), pp 1,331-1,365.

[6]Cosby, B. J., Hornberger, G. M., Clapp, R. B. and Ginn, T.R.,“A statistical exploration of the relationships of soil moisture characteristics to the physical properties of soils”, Water Resources Research, Vol. 20, (1984), pp 682-690.

[7]Rawls, W. J., Ahuja, L. R., Brakensiek, D. L. and Shirmohammadi, A., “Handbook of Hydrology”, McGraw-Hill Inc., (1993).

[8]Su, F. and Xie Z., “A model for assessing effects of climate change on runoff of China”, Progress in Natural Science, Vol.13, No. 9, (2003), pp 701-707.

[9]Hubert, B., Francois, L., Warnant, P. and Strivay, D., “Stochasticgeneration of meteorological variables and effects on global models of waterand carbon cycles in vegetation and soils”, Journal of Hydrology, Vol. 212-213, (1998), pp 318-334.

1