# A General Method to Improve Runoff Prediction in Ungauged Basins Based on Remotely Sensed Actual Evapotranspiration Data

^{1}

^{2}

^{3}

^{4}

^{5}

^{6}

^{7}

^{*}

## Abstract

**:**

## 1. Introduction

## 2. Materials and Methods

#### 2.1. Study Site and Materials

#### 2.1.1. Study Site and Data Sources

_{P}), and runoff data are available from 1948 to 2003 in the MOPEX datasets (ftp://hydrology.nws.noaa.gov, accessed on 22 July 2022). Daily RS-ET data at the gridded resolution of 8 km are derived by AVHRR NDVI from 1983 to 2006 [21], and have been spatially averaged to basin scale. The ET estimates agree well with the observed tower fluxes, with R

^{2}above 0.6 and RMSE below 45 W m

^{−2}for all towers. The ET dataset is demonstrated to be a long-term continuous global ET record with relatively high accuracy, which has been extensively applied to research on assessment of regional ET and climatology anomalies, drought monitoring, and evaluation of global water balance change [31,32,33,34,35,36].

^{−1}; R

^{2}= 0.91).

^{2}) used in this study are located in the continental United States with a diversity of climate characteristics, from semi-arid to temperate and tropical humid regions. Figure 1 shows the average annual precipitation (mm) for the 208 selected MOPEX basins, which varies greatly over the geographical region. Specifically, the precipitation is above 2000 mm on the west coast with temperate maritime climate, but less than 400 mm in arid regions in the central US with temperate continental climate and desert climate. Overall, average annual precipitation shows a clear east–west (subtropical humid climate zone, temperate continental climate zone, plateau mountain climate, and desert climate zone) gradient with decreasing values toward the west coast, and demonstrates another north–south (subtropical humid climate zone, temperate continental climate zone) gradient in the eastern U.S with increasing values toward the south coast.

#### 2.1.2. Data Implementation

_{P}, runoff, and RS-ET data from 1983 to 2003 in the selected 208 MOPEX basins. Specifically, the calibration period is from 1983 to 1996 and the validation period is from 1997 to 2003. Data in the first 60 days are sacrificed for model warming-up.

#### 2.2. Methodology

#### 2.2.1. RR Models

_{P}considering the availability of data materials in the study. Considering the above factors, three lumped conceptual RR models are selected, namely, Xinanjiang, GR4J, and SIMHYD. A schematic overview of the three RR models is presented in Figure 3.

- (1)
- Xinanjiang model

- (2)
- SIMHYD

_{P}, and the latter is calculated as a linear function of the actual storage SMS.

- (3)
- GR4J

_{n}) and the level S in the production storage. E

_{n}is determined by subtracting ET

_{P}from precipitation.

_{1}(mm) is the maximum capacity of the production store.

#### 2.2.2. Model Calibration, Regionalization, and Evaluation

- (1)
- Model calibration

^{3}s

^{−1}) on the ith day, respectively; and n is the number of days.

- (2)
- Parameter regionalization

- (3)
- Model evaluation

_{Q}and NSE

_{E}refer to the NSE value of discharge and ET, respectively:

^{3}s

^{−1}); $E{T}_{RS,i}$ and $E{T}_{sim,i}$ are the RS and the simulated ET values (mm) on the ith day, respectively; $\overline{E{T}_{RS}}$ is the average RS-ET (mm); n is the number of days. NSE ranges from −∞ to 1. The value of NSE should be positive, and higher values indicate better performance [56]. A value of zero for NSE indicates that the observed mean $\overline{{Q}_{obs}}$ is as good as the model-simulated ${Q}_{sim,i}$ result, while negative values indicate that the observed mean is a better predictor than the model. Note that since the evapotranspiration sub-module is replaced directly by RS-ET data in Scheme 2 and Scheme 3, NSE

_{E}is calculated only in Scheme 1 to evaluate the ET simulation efficiency of the original RR models.

#### 2.2.3. Modeling Scheme

- (1)
- Scheme 1

_{Q}and NSE

_{E}, respectively.

- (2)
- Scheme 2

- (3)
- Scheme 3

_{E}of Scheme 1 not only indicates the ET simulation efficiency of the original models, but also reflects the compatibility of model structure with RS-ET data and therefore affects the efficacy of using RS-ET. Accordingly, Scheme 3 utilizes RS-ET as direct input on the premise that NSE

_{E}of Scheme 1 exceeds the respective threshold values (NSE

_{thr}) of three RR models. Otherwise, on the condition that NSE

_{E}of Scheme 1 is below the NSE

_{thr}, RS-ET is not used and ET is modeled in the same way as in Scheme 1.

## 3. Results and Discussion

#### 3.1. Conventional Streamflow Prediction

_{E}above 0.48. GR4J performs well for streamflow but poor for ET, with the 75th percentile NSE

_{E}below 0.2 and 50th percentile below 0.1, which results in a lower NSE

_{E}threshold of GR4J than that of Xinanjiang and SIMHYD models. For SIMHYD, the streamflow performance is the worst of three models, but the ET performance is between that of the above two models. Generally, SIMHYD simulates both streamflow and ET poorly, with the 75th percentile NSE

_{Q}below 0.5 and NSE

_{E}below 0.4. Generally, the relative performance of NSE

_{Q}is GR4J > Xinanjiang > SIMHYD, while in terms of NSE

_{E}, Xinanjiang > SIMHYD > GR4J. For all three models, the regionalization results in ungauged basins are worse than simulation results in gauged basins.

#### 3.2. Selection of RS-ET Utilization Threshold

_{thr}is selected, considering two factors: (i) the relative superiority level (superior basin percent) of Scheme 2 over Scheme 1, and (ii) the number of basins for analysis (basins with NSE

_{E}values exceeding NSE

_{thr}). The former factor is determined by the representation of evapotranspiration processes in the RR model, and the latter is affected by the overall ET simulation performances of the model.

_{Q}

_{2-1}= NSE

_{Q}

_{2}− NSE

_{Q}

_{1}) versus NSE

_{E}of Scheme 1 in the calibration period (the result in the validation period is similar and is therefore not presented here).

_{Q}

_{2-1}> 0) in 85.1% of the total 208 basins. The NSE

_{E}of Scheme 1 exceeds the NSE

_{thr}values of 0.2, 0.3, and 0.4 in 174, 169, and 142 basins, respectively. It can be seen that Scheme 2 outperforms Scheme 1 in 92.0%, 92.3%, and 95.8% of the corresponding 174, 169, and 142 basins, respectively. Similar analysis is conducted for SIMHYD in Figure 4b, which shows that Scheme 2 outperforms Scheme 1 in 51.9% of the total 208 basins and the superiority is actually marginal. The NSE

_{E}of Scheme 1 exceeds the NSE

_{thr}values of 0.2, 0.25, and 0.3 in 120, 183, and 49 basins, of which the corresponding basin proportion of Scheme 2 surpassing Scheme 1 is 63.3%, 67.7%, and 69.4%, respectively. Similarly, for GR4J in Figure 4c, Scheme 2 outperforms Scheme 1 only in 42.3% of the total 208 basins, indicating that using RS-ET as direct input fails to improve model performance in most basins. However, Scheme 2 surpasses Scheme 1 in 61.7%, 67.7%, and 84.6% of the basins where the NSE

_{E}of Scheme 1 exceeds 0.05, 0.1, and 0.2, respectively. Generally, the superiority of using RS-ET over the conventional method is more significant with higher NSE

_{thr}values. In this vein, NSE

_{E}of Scheme 1 could be used as a measurement for acceptability of RS-ET data as direct input.

_{E}of Scheme 1 exceeds 0.3 in 169 basins and Scheme 2 surpasses Scheme 1 for 92.3% of these basins. For SIMHYD, NSE

_{E}of Scheme 1 exceeds 0.25 in 83 basins and Scheme 2 surpasses Scheme 1 for 67.5% of these basins. For GR4J, NSE

_{E}of Scheme 1 exceeds 0.1 in 62 basins and Scheme 2 surpasses Scheme 1 for 67.7% of the basins. However, the relative superiority can be modest with lower NSE

_{thr}, and the number of basins (especially for GR4J) can be insufficient for analysis with higher NSE

_{thr}. Taking into account both the relative superiority level and the basin numbers for analysis, NSE

_{E}values of 0.3, 0.25, and 0.1 are eventually chosen as the NSE

_{thr}for the Xinanjiang, SIMHYD, and GR4J models, respectively.

#### 3.3. Streamflow Prediction Using RS-ET

_{Q}values decreasing by approximately 0.1 compared with the results of validation. Moreover, the inter-quartile range (the spacing between edges of the box) is wider in regionalization, showing larger variability of NSE

_{Q}results. As for modeling schemes, Scheme 3 performs the best for both the simulation and regionalization. Specifically, Scheme 2 surpasses Scheme 1 significantly for the Xinanjiang model, but performs only slightly better or even worse than Scheme 1 for SIMHYD and GR4J models. The results indicate that using RS-ET data as direct input may not definitely improve streamflow simulation performances, which is consistent with the results of most past studies [12,58]. However, Scheme 3 outperformed both Scheme 1 and Scheme 2 almost in all cases, though more significantly for the Xinanjiang model than for the other two models. Generally, the performance differences between Scheme 2 and Scheme 3 are marginal, as inferred from Figure 5.

_{Q}) of Scheme 2 and Scheme 3 versus Scheme 1 are presented in Figure 6. For the Xinanjiang model, the magnitude of the NSE

_{Q}increase exceeds that of the NSE

_{Q}decline in all cases, i.e., the performance improvement dominates for both Scheme 2 and Scheme 3. For SIMHYD and GR4J, the improvement dominates in all cases for Scheme 3, but the net ΔNSE

_{Q}is relatively marginal. For all three RR models, the net performance improvement of Scheme 3 is more significant than that of Scheme 2, demonstrating the superiority of Scheme 3.

_{E}of Scheme 1 exceeds the respective NSE

_{thr}. Accordingly, Scheme 3 using the Xinanjiang, SIMHYD, and GR4J models are compared with Scheme 1 over 169, 83, and 62 MOPEX basins, respectively. Results show that for all these models, Scheme 3 surpasses Scheme 2 for both simulation and regionalization. Note that Scheme 3 outperforms Scheme 1 in 91.1%, 59.0%, and 53.2% basins in regionalization for the Xinanjiang, SIMHYD and GR4J models, respectively, which indicates that using RS-ET data as partial direct input can improve streamflow prediction in most ungauged basins.

_{E}of Scheme 1 is above the NSE

_{thr}) for streamflow prediction, Figure 7 compares simulated streamflow hydrographs between Scheme 1 and Scheme 2 in regionalization for the representative basins with NSE

_{E}> NSE

_{thr}and NSE

_{E}< NSE

_{thr}. It can be seen that for all three models, Scheme 2 simulates streamflow better than Scheme 1 in the basins with NSE

_{E}> NSE

_{thr}, but performs worse than Scheme 1 in the basins with NSE

_{E}< NSE

_{thr}. Specifically, in the basins with NSE

_{E}> NSE

_{th}

_{r}, both high flows and low flows are matched better in Scheme 2 than in Scheme 1. In the basins with NSE

_{E}< 0.3 for the Xinanjiang model, these two schemes perform similarly for high flows, but low flows are more overestimated in Scheme 2. In the basin with NSE

_{E}< 0.25 for the SIMHYD model, high flows are extremely underestimated in Scheme 2. In the basins with NSE

_{E}< 0.1 for the GR4J model, high flows are underestimated overall and are matched better in Scheme 1 than in Scheme 2. Therefore, Scheme 3, which uses RS-ET data provided that the NSE

_{E}of Scheme 1 exceeds the respective NSE

_{thr}, outperforms Scheme 1.

#### 3.4. Water Balance Condition

_{E}of Scheme 1 is above the respective threshold). Figure 9a analyzes the relationship between WBI of Scheme 2 and NSE

_{E}of Scheme 1 in the validation period. Results demonstrate that the water balance based on RS-ET data is relatively better with the higher NSE

_{E}of Scheme 1, which explains the better WBI results of Scheme 3 compared to Scheme 2 in Figure 8. Furthermore, as inferred from the relationship between WBI and NSE

_{Q}results of Scheme 2 in Figure 9b, the NSE

_{Q}values of Scheme 2 are relatively higher, with better water balance results, which is consistent with the better streamflow performance of Scheme 3 compared to that of Scheme 2 in Figure 4.

#### 3.5. Regional Spatial Patterns

_{Q}being below 0.45 for simulation and below 0.35 in regionalization, and the efficacy of using RS-ET is actually marginal for the GR4J model. Considering the heterogeneity in hydro-meteorological characteristics and evapotranspiration magnitudes over 208 study basins, the modeling results of the Xinanjiang model are analyzed from a geographical standpoint to reveal the regional spatial patterns.

_{Q}results of Scheme 3 with the average annual precipitation (mm) over 208 basins (Figure 1), there is a correlation between low precipitation values and poor model performances. The regional spatial pattern of Scheme 3 can be connected with the spatial pattern of the original Scheme 1 and the geographical characteristic of performance improvements by using RS-ET data. Figure 10 shows NSE

_{Q}in Scheme 3 and Scheme 1 versus the average annual precipitation (mm). It can be seen that there is a definite low-performance zone (NSE

_{Q}is below 0.5 in validation and below 0.4 in regionalization for Scheme 3, and NSE

_{Q}is below 0.4 in validation and below 0.3 in regionalization for Scheme 1) with annual precipitation below 1200 mm.

_{Q}

_{3-1}= NSE

_{Q}

_{3}− NSE

_{Q}

_{1}) versus the average annual precipitation (mm). Positive ΔNSE

_{Q}

_{3-1}indicates that Scheme 3 outperforms Scheme 1, and negative values indicate the opposite. It is noteworthy that higher ΔNSE

_{Q}

_{3-1}values appear in basins that receive less than 1200 mm of annual precipitation. Results indicate that the performance-improved zone shown in Figure 11 coincides with the low-performance zone inferred from Figure 10. Accordingly, the correlation between the relative performances and the original performance of Scheme 1 is also analyzed in Figure 11, which demonstrates that the performance improvement is more significant in the originally poorly simulated basins.

## 4. Conclusions

## Author Contributions

## Funding

## Conflicts of Interest

## References

- Sivapalan, M.; Takeuchi, K.; Franks, S.W.; Zehe, E. IAHS Decade on Predictions in Ungauged Basins (PUB), 2003–2012: Shaping an exciting future for the hydrological sciences. Hydrol. Sci. J.
**2003**, 48, 857–880. [Google Scholar] [CrossRef] - Zhang, Y.; Chiew, F.H.S. Relative merits of different methods for runoff predictions in ungauged catchments. Water Resour. Res.
**2009**, 45, W07412. [Google Scholar] [CrossRef] - Guo, Y.; Zhang, Y.; Zhang, L.; Wang, Z. Regionalization of hydrological modeling for predicting streamflow in ungauged catchments: A comprehensive review. Wiley Interdiscip. Rev. Water
**2020**, 8, e1487. [Google Scholar] [CrossRef] - Yang, X.; Li, F.; Qi, W.; Zhang, M.; Xu, C.Y. Regionalization methods for PUB: A comprehensive review of progress after the PUB decade. Hydrol. Res.
**2023**, 54, 885–900. [Google Scholar] [CrossRef] - Lakshmi, V. The role of satellite remote sensing in the Prediction of Ungauged Basins. Hydrol. Process.
**2004**, 18, 1029–1034. [Google Scholar] [CrossRef] - Yoon, H.N.; Marshall, L.; Sharma, A.; Kim, S. Bayesian Model Calibration Using Surrogate Streamflow in Ungauged Catchments. Water Resour. Res.
**2022**, 58, e2021WR031287. [Google Scholar] [CrossRef] - Hrachowitz, M.; Savenije, H.H.G.; Blschl, G.; Mcdonnell, J.J.; Cudennec, C. A decade of Predictions in Ungauged Basins (PUB)—A review. Hydrol. Sci. J.
**2013**, 58, 1198–1255. [Google Scholar] [CrossRef] - Dembélé, M.; Ceperley, N.; Zwart, S.J.; Salvadore, E.; Mariethoz, G.; Schaefli, B. Potential of satellite and reanalysis evaporation datasets for hydrological modelling under various model calibration strategies. Adv. Water. Res.
**2020**, 143, 103667. [Google Scholar] [CrossRef] - Guo, D.; Westra, S.; Maier, H.R. Impact of evapotranspiration process representation on runoff projections from conceptual rainfall-runoff models. Water Resour. Res.
**2017**, 53, 435–454. [Google Scholar] [CrossRef] - Xu, Z.; Wu, Z.; Shao, Q.; He, H.; Guo, X. A two-step calibration framework for hydrological parameter regionalization based on streamflow and remote sensing evapotranspiration. J. Hydrol.
**2022**, 613, 128320. [Google Scholar] [CrossRef] - Zhang, Y.; Chiew, F.H.S.; Zhang, L.; Li, H. Use of Remotely Sensed Actual Evapotranspiration to Improve Rainfall–Runoff Modeling in Southeast Australia. J. Hydrometeorol.
**2009**, 10, 969–980. [Google Scholar] [CrossRef] - Vervoort, R.W.; Miechels, S.F.; Ogtrop, F.F.V.; Guillaume, J.H.A. Remotely sensed evapotranspiration to calibrate a lumped conceptual model: Pitfalls and opportunities. J. Hydrol.
**2014**, 519, 3223–3236. [Google Scholar] [CrossRef] - Kunnath-Poovakka, A.; Ryu, D.; Renzullo, L.J.; George, B. The efficacy of calibrating hydrologic model using remotely sensed evapotranspiration and soil moisture for streamflow prediction. J. Hydrol.
**2016**, 535, 509–524. [Google Scholar] [CrossRef] - Gui, Z.; Liu, P.; Cheng, L.; Guo, S.; Wang, H.; Zhang, L. Improving runoffff prediction using remotely sensed actual evapotranspiration during rainless periods. J. Hydrol. Eng.
**2019**, 24, 04019050.1–04019050.12. [Google Scholar] [CrossRef] - Herman, M.R.; Nejadhashemi, A.P.; Abouali, M.; Hernandez-Suarez, J.S.; Daneshvar, F.; Zhang, Z.; Anderson, M.C.; Sadeghi, A.M.; Hain, C.R.; Sharifi, A. Evaluating the role of evapotranspiration remote sensing data in improving hydrological modeling predictability. J. Hydrol.
**2018**, 556, 39–49. [Google Scholar] [CrossRef] - Yang, Y.; Guan, K.; Peng, B.; Pan, M.; Franz, T.E. High-resolution spatially explicit land surface model calibration using field-scale satellite-based daily evapotranspiration product. J. Hydrol.
**2020**, 596, 125730. [Google Scholar] [CrossRef] - Zhang, L.; Zhao, Y.; Ma, Q.; Wang, P.; Ge, Y.; Yu, W. A parallel computing-based and spatially stepwise strategy for constraining a semi-distributed hydrological model with streamflow observations and satellite-based evapotranspiration. J. Hydrol.
**2021**, 599, 126359. [Google Scholar] [CrossRef] - Li, H.; Zhang, Y.; Chiew, F.H.S.; Xu, S. Predicting runoff in ungauged catchments by using Xinanjiang model with MODIS leaf area index. J. Hydrol.
**2009**, 370, 155–162. [Google Scholar] [CrossRef] - Taconet, O.; Bernard, R.; Vidal-Madjar, D. Evapotranspiration over an Agricultural Region Using a Surface Flux/Temperature Model Based on NOAA-AVHRR Data. J. Clim. Appl. Meteorol.
**1986**, 25, 284–307. [Google Scholar] [CrossRef] - Nemani, R.R.; Running, S.W. Estimation of regional surface resistance to evapotranspiration from NDVI and Thermal-IR AVHRR data. J. Appl. Meteorol.
**1989**, 28, 276–284. [Google Scholar] [CrossRef] - Zhang, K.; Kimball, J.S.; Nemani, R.R.; Running, S.W. A continuous satellite-derived global record of land surface evapotranspiration from 1983 to 2006. Water Resour. Res.
**2010**, 46, 109–118. [Google Scholar] [CrossRef] - Cleugh, H.A.; Leuning, R.; Mu, Q.; Running, S.W. Regional evaporation estimates from flux tower and MODIS satellite data. Remote Sens. Environ.
**2007**, 106, 285–304. [Google Scholar] [CrossRef] - Leuning, R.; Zhang, Y.Q.; Rajaud, A.; Cleugh, H.; Tu, K. A simple surface conductance model to estimate regional evaporation using MODIS leaf area index and the Penman-Monteith equation. Water Resour. Res.
**2008**, 44, 652–655. [Google Scholar] [CrossRef] - Zhang, Y.Q.; Chiew, F.H.S.; Zhang, L.; Leuning, R.; Cleugh, H.A. Estimating catchment evaporation and runoff using MODIS leaf area index and the Penman-Monteith equation. Water Resour. Res.
**2008**, 44, 2183–2188. [Google Scholar] [CrossRef] - Guerschman, J.P.; Van Dijk, A.I.J.M.; Mattersdorf, G.; Beringer, J.; Hutley, L.B.; Leuning, R.; Pipunic, R.C.; Sherman, B.S. Scaling of potential evapotranspiration with MODIS data reproduces flux observations and catchment water balance observations across Australia. J. Hydrol.
**2009**, 369, 107–119. [Google Scholar] [CrossRef] - Zhou, Y.; Zhang, Y.; Vaze, J.; Lane, P.; Xu, S. Improving runoff estimates using remote sensing vegetation data for bushfire impacted catchments. Agr. Forest. Meteorol.
**2013**, 182–183, 332–341. [Google Scholar] [CrossRef] - Zhang, Y.Q.; Vaze, J.; Chiew, F.H.S.; Liu, Y. Incorporating vegetation time series to improve rainfall-runoff model predictions in gauged and ungauged catchments. In Proceedings of the 19th International Congress on Modelling and Simulation, Perth, Australia, 12–16 December 2011. [Google Scholar]
- Szilagyi, J. Application of MODIS-Based Monthly Evapotranspiration Rates in Runoff Modeling: A Case Study in Nebraska, USA. Open J. Mod. Hydrol.
**2013**, 3, 172–178. [Google Scholar] [CrossRef] - Roy, T.; Gupta, H.V.; Serrat-Capdevila, A.; Valdes, J.B. Using satellite-based evapotranspiration estimates to improve the structure of a simple conceptual rainfall and -runoff model. Hydrol. Earth Syst. Sci.
**2017**, 21, 879–896. [Google Scholar] [CrossRef] - Duan, Q.; Schaake, J.; Andréassian, V.; Franks, S.; Goteti, G.; Gupta, H.V.; Gusev, Y.M.; Habets, F.; Hall, A.; Hay, L. Model parameter estimation experiment (MOPEX): An overview of science strategy and major results from the second and third workshops. J. Hydrol.
**2006**, 320, 3–17. [Google Scholar] [CrossRef] - Jung, M.; Reichstein, M.; Ciais, P.; Seneviratne, S.I.; Sheffield, J.; Goulden, M.L.; Bonan, G.; Cescatti, A.; Chen, J.; de Jeu, R.; et al. Recent decline in the global land evapotranspiration trend due to limited moisture supply. Nature
**2010**, 467, 951–954. [Google Scholar] [CrossRef] - Miralles, D.G.; Holmes, T.R.H.; De Jeu, R.A.M.; Gash, J.H.; Meesters, A.G.C.A.; Dolman, A.J. Global land-surface evaporation estimated from satellite-based observations. Hydrol. Earth Syst. Sci.
**2010**, 7, 453–469. [Google Scholar] [CrossRef] - Vicente-Serrano, S.M.; Beguería, S.; López-Moreno, J.I. Comment on “Characteristics and trends in various forms of the Palmer Drought Severity Index (PDSI) during 1900–2008” by Aiguo Dai. J. Geophys. Res.
**2011**, 116, D19112. [Google Scholar] [CrossRef] - Mátyás, C.; Sun, G. Forests in a water limited world under climate change. Environ. Res. Lett.
**2014**, 9, 085001. [Google Scholar] [CrossRef] - Wang, D.; Alimohammadi, N. Responses of annual runoff, evaporation, and storage change to climate variability at the watershed scale. Water Resour. Res.
**2012**, 48, W05546. [Google Scholar] [CrossRef] - Chen, X.; Alimohammadi, N.; Wang, D. Modeling interannual variability of seasonal evaporation and storage change based on the extended Budyko framework. Water Resour. Res.
**2013**, 49, 6067–6078. [Google Scholar] [CrossRef] - Merz, R.; Blöschl, G. Regionalisation of catchment model parameters. J. Hydrol.
**2004**, 287, 95–123. [Google Scholar] [CrossRef] - Oudin, L.; Andréassian, V.; Perrin, C.; Michel, C.; Le Moine, N. Spatial proximity, physical similarity, regression and ungaged catchments: A comparison of regionalization approaches based on 913 French catchments. Water Resour. Res.
**2008**, 44, W03413. [Google Scholar] [CrossRef] - Zhao, R.J.; Liu, X.R.; Singh, V.P. The Xinanjiang model. In Computer Models of Watershed Hydrology; Singh, V.P., Ed.; Water Resources Publications: Littleton, CO, USA, 1995; pp. 371–381. [Google Scholar]
- Chiew, F.H.S.; Peel, M.C.; Western, A.W.; Singh, V.; Frevert, D. Application and testing of the simple rainfall-runoff model SIMHYD. In Mathematical Models of Small Watershed Hydrology and Applications; Water Resources Publication: Littleton, CO, USA, 2002; pp. 335–367. [Google Scholar]
- Perrin, C.; Michel, C.; Andréassian, V. Improvement of a parsimonious model for streamflow simulation. J. Hydrol.
**2003**, 279, 275–289. [Google Scholar] [CrossRef] - Zhao, R.J. The Xinanjiang model applied in China. J. Hydrol.
**1992**, 135, 371–381. [Google Scholar] [CrossRef] - Edijatno; Nascimento, N.D.O.; Yang, X.; Makhlouf, Z.; Michel, C. GR3J: A daily watershed model with three free parameters. Hydrol. Sci. J.
**1999**, 44, 263–277. [Google Scholar] [CrossRef] - Goldberg, D.E. Genetic Algorithms in Search, Optimization, and Machine Learning; Addison-Wesley: Reading, MA, USA, 1989. [Google Scholar]
- Rosenbrock, H.H. An Automatic Method for Finding the Greatest or Least Value of a Function. Comput. J.
**1960**, 3, 175–184. [Google Scholar] [CrossRef] - Nelder, J.A.; Mead, R. A simplex method for function minimization. Comput. J.
**1965**, 7, 308–313. [Google Scholar] [CrossRef] - Duan, Q.; Sorooshian, S.; Gupta, V. Effective and efficient global optimization for conceptual rainfall-runoff models. Water Resour. Res.
**1992**, 28, 1015–1031. [Google Scholar] [CrossRef] - Li, Z.; Liu, P.; Deng, C.; Guo, S.; He, P.; Wang, C. Evaluation of Estimation of Distribution Algorithm to Calibrate Computationally Intensive Hydrologic Model. J. Hydrol. Eng.
**2016**, 21, 04016012. [Google Scholar] [CrossRef] - Kittel, C.M.M.; Arildsen, A.L.; Dybkjr, S.; Hansen, E.R.; Bauer-Gottwein, P. Informing hydrological models of poorly gauged river catchments—A parameter regionalization and calibration approach. J. Hydrol.
**2020**, 587, 124999. [Google Scholar] [CrossRef] - Yang, X.; Magnusson, J.; Huang, S.; Beldring, S.; Xu, C.Y. Dependence of regionalization methods on the complexity of hydrological models in multiple climatic regions. J. Hydrol.
**2019**, 582, 124357. [Google Scholar] [CrossRef] - Yang, X.; Magnusson, J.; Xu, C.Y. Transferability of regionalization methods under changing climate. J. Hydrol.
**2018**, 568, 67–81. [Google Scholar] [CrossRef] - Arsenault, R.; Breton-Dufour, M.; Poulin, A.; Dallaire, G.; Romero-Lopez, R. Streamflow prediction in ungauged basins: Analysis of regionalization methods in a hydrologically heterogeneous region of Mexico. Hydrol. Sci. J. J. Des Sci. Hydrol.
**2019**, 64, 1297–1311. [Google Scholar] [CrossRef] - Beck, H.E.; Pan, M.; Lin, P.; Seibert, J.; Dijk, A.I.J.M.V.; Wood, E.F. Global Fully Distributed Parameter Regionalization Based on Observed Streamflow From 4229 Headwater Catchments. J. Geophys. Res. Atmos.
**2020**, 125, e2019JD031485. [Google Scholar] [CrossRef] - Young, A.R. Stream flow simulation within UK ungauged catchments using a daily rainfall-runoff model. J. Hydrol.
**2006**, 320, 155–172. [Google Scholar] [CrossRef] - Nash, J.E.; Sutcliffe, J.V. River flow forecasting through conceptual models part I—A discussion of principles. J. Hydrol.
**1970**, 10, 282–290. [Google Scholar] [CrossRef] - Legates, D.R.; McCabe, G.J. Evaluating the use of “goodness-of-fit” Measures in hydrologic and hydroclimatic model validation. Water Resour. Res.
**1999**, 35, 233–241. [Google Scholar] [CrossRef] - Li, H.; Zhang, Y.; Zhou, X. Predicting Surface Runoff from Catchment to Large Region. Adv. Meteorol.
**2015**, 2015, 720967. [Google Scholar] [CrossRef] - Rientjes, T.H.M.; Muthuwatta, L.P.; Bos, M.G.; Booij, M.J.; Bhatti, H.A. Multi-variable calibration of a semi-distributed hydrological model using streamflow data and satellite-based evapotranspiration. J. Hydrol.
**2013**, 505, 276–290. [Google Scholar] [CrossRef] - Budyko, M.I. Climate and Life; Academic Press: Cambridge, MA, USA, 1974. [Google Scholar]
- Madadgar, S.; Moradkhani, H. Improved Bayesian multimodeling: Integration of copulas and Bayesian model averaging. Water Resour. Res.
**2014**, 50, 9586–9603. [Google Scholar] [CrossRef] - Budyko, M.I. The Heat Balance of the Earth’s Surface. Eurasian Geogr. Econ.
**1961**, 2, 3–13. [Google Scholar] [CrossRef] - Liu, W.; Wang, L.; Zhou, J.; Li, Y.; Sun, F.; Fu, G.; Li, X.; Sang, Y.F. A worldwide evaluation of basin-scale evapotranspiration estimates against the water balance method. J. Hydrol.
**2016**, 538, 82–95. [Google Scholar] [CrossRef]

**Figure 1.**Spatial distribution of 208 MOPEX basins used in this study and their average annual precipitation (mm).

**Figure 4.**Relationship of the relative performances (ΔNSE

_{Q}

_{2−1}= NSE

_{Q}

_{2}− NSE

_{Q}

_{1}) between Scheme 2 and Scheme 1 versus NSE

_{E}of Scheme 1 in the calibration period. The percent below the vertical dashed line is the different NSE

_{thr}value, and the number in the bracket is the number of basins with the corresponding NSE

_{thr}exceeded. The percent above the vertical dashed line is the basin percent of Scheme 2 surpassing Scheme 1 when the corresponding NSE

_{thr}is exceeded.

**Figure 5.**Comparison of the NSE

_{Q}values among the three schemes for three RR models computed on 208 basins. The red line in the boxplots represents the median value, the ends of the boxes represent the 1st and 3rd quartiles, the whiskers represent the values at 1.5 standard deviations, and outliers (more than 1.5 standard deviations from the mean) are shown as red crosses.

**Figure 6.**Comparison of the relative performances (ΔNSE

_{Q}) of Scheme 2 and Scheme 3 versus Scheme 1 for three RR models computed on 208 basins. The orange bar represents the average value of increased NSE

_{Q}(positive ΔNSE

_{Q}), the blue bar represents the average value of decreased NSE

_{Q}(negative ΔNSE

_{Q}), the bar filled with diagonals represents the average net ΔNSE

_{Q}value, and the three RR models (Xinanjiang, SIMHYD, and GR4J) are referred to as ‘X’, ‘S’ and ‘G’.

**Figure 7.**Comparison of simulated streamflow hydrographs between Scheme 1 and Scheme 2 in regionalization for the representative basins with NSE

_{E}> NSE

_{thr}and NSE

_{E}< NSE

_{thr}for three RR models. Only a representative segment from the whole simulation period is shown here.

**Figure 8.**Comparison of WBI values among the three schemes for the three RR models computed on 208 basins. The red line in the boxplots represents the median value, the ends of the boxes represent the 1st and 3rd quartiles, the whiskers represent the values at 1.5 standard deviations, and outliers (more than 1.5 standard deviations from the mean) are shown as red crosses.

**Figure 9.**Relationship between the │WBI│ values of Scheme 2 versus (

**a**) NSE

_{E}of Scheme 1 and (

**b**) NSE

_{Q}of Scheme 2 in the validation period.

**Figure 10.**Relationship between (

**a**) the NSE

_{Q}values of Scheme 1 and (

**b**) the NSEQ values of Scheme 3 versus the average annual precipitation (mm) in validation and regionalization.

**Figure 11.**Relationship between ΔNSE

_{Q3}

_{−1}(NSE

_{Q}

_{3}− NSE

_{Q}

_{1}) versus the (

**a**) average annual precipitation (mm) and (

**b**) NSE

_{Q}of Scheme 1 in validation and regionalization.

**Table 1.**Summary of relevant literature on RR modeling using RS-ET as direct input (the current paper is added for completeness). Not all papers determined under what circumstances RS-ET can be used to improve runoff prediction in RR models and performed geographical analyses, and these are denoted with a N/A ‘not applicable’ in the relevant part of the ‘Key results’ column. In the ‘Key results’ column, the three components are identified by the code: (1) assess the relative performance of runoff prediction versus the conventional method; (2) determine the circumstances under which RS-ET can be used as direct input in RR models; and (3) analyze the model performances from a geographical standpoint to reveal the regional pattern. Studies are ordered chronologically then alphabetically.

Study | RS-ET or Vegetation Data/RR Model Used | Location/Climate/Number of Catchments/Size Range of Catchment/Length of Time Series/Time Step | RR Modeling Using RS-ET As Direct Input/Regionalization Method | Key Results |
---|---|---|---|---|

1. Zhang et al. [24] | ET estimated from MODIS LAI with the Penman–Monteith (PM) equation/SIMHYD | The Murray-Darling Basin in Australia/N/A/120/50–2000 km^{2}/5 years/annual | Deriving discharge from water balance estimates (R_{RS} = P − E_{RS})/the spatial proximity method | (1) The simulated R_{RS} in gauged/ungauged basins had an accuracy similar to that of SIMHYD in the gauged catchments.(2) N/A (3) RS-ET was successfully used to estimate long-term runoff in semi-humid and humid regions. |

2. Li et al. [18] | ET estimated from MODIS LAI with the PM equation/Xinanjiang model | Southeast Australia/semi-arid and semi-humid/ 210/50–2000 km^{2}/7 years/ daily | Modification of the Xinanjiang model to use MODIS LAI directly/the spatial proximity method and the physical similarity method | (1) Incorporation of MODIS LAI into Xinanjiang model improved both the model calibration and prediction of runoff in ungauged catchments. (2) N/A (3) N/A |

3. Zhang et al. [11] | ET estimated from MODIS LAI with the PM equation/SIMHYD | Southeast Australia/N/A/120/50–2000 km^{2}/5 years/daily | Modification of SIMHYD to use MODIS-LAI directly/the spatial proximity method | (1) The runoff simulation results were reduced, while the regionalization results were improved significantly. (2) N/A (3) N/A |

4. Zhang and Chiew [2] | ET estimated from MODIS LAI with the PM equation/Xinanjiang and SIMHYD | Southeast Australia /relatively unimpacted/210/50–2000 km^{2}/13 years/daily | The two models are revised to incorporate RS-LAI/the spatial proximity method, the physical similarity method, and the integrated similarity method | (1) The revised models generally perform better than the original RR models, but the improvements are marginal. (2) N/A (3) The revised models give significantly better results in the poorer modeled ungauged catchments. |

5. Zhang et al. [27] | LAI from AVHRR/SIMHYD | Australia/N/A/470/50–5000 km^{2}/26 years/daily | Modification of SIMHYD to incorporate RS-LAI directly (SIMHYD-ET)/the spatial proximity method | (1) For both model calibration and regionalization, the runoff modeled by the SIMHYD-ET model are similar to (or only very marginally better than) those simulated by the original SIMHYD model. (2) N/A (3) The SIMHYD-ET outperformed SIMHYD especially for poorly simulated catchments with low NSE of daily runoff and high water balance errors. |

6. Szilagyi [28] | RS-ET estimated with CREMAP method/a lumped conceptual model of Jakeman and Hornberger (JH model) | The Little Nemaha River in Nebraska, USA/continental/1/2051 km^{2}/6 years/daily | Modification of the JH model to incorporate CREAMP-ET directly/N/A | (1) The accuracy of runoff simulation remained practically unchanged. (2) N/A (3) N/A |

7. Zhou et al. [26] | ET estimated from MODIS LAI with the PM equation/Xinanjiang | Southeast Australian/bushfire impacted/4/360–900 km^{2}/28 years/daily | Modification of Xinanjiang model to incorporate RS-LAI directly (Xinanjiang-ET model)/N/A | (1) Inclusion of RS-LAI resulted in a slight improvement of runoff simulation and noticeable decrease in water balance errors. (2) N/A (3) Use of RS-LAI can improve runoff simulation in three wetter catchments, not in a dry catchment. |

8. Willem Vervoort et al. [12] | MODIS-ET from Montana University/IHACRES | New South Wales, Australia/semi-arid/4/146–2184 km^{2}/12 years/daily | To replace the ET sub-module directly with RS-ET data/the spatial proximity method | (1) Using RS-ET reduced runoff simulation performance. (2) N/A (3) N/A |

9. Roy et al. [29] | ET from GLEAM/HYMOD | The Nyangores River basin in Kenya and Tanzania/N/A/1/697 km^{2}/7.5 years/daily | Modification of HYMOD to simulate ET as by GLEAM/N/A | (1) The modified model can provide improved simulations of streamflow. (2) N/A (3) N/A |

10. This study | ET estimated from AVHRR NDVI/Xinanjiang, GR4J and SIMHYD | The MOPEX basins in the continental United States/highly diverse/401/67–10,329 km^{2}/21 years/daily | (i) Scheme 2: Using RS-ET as direct input; (ii) Scheme 3: Using RS-ET as partial direct input/the spatial proximity method | (1) Using RS-ET as direct input improved model performances for the Xinanjiang model, but worsened runoff prediction for SIMHYD and GR4J in most cases; using RS-ET as partial direct input improved runoff prediction in 91.1%, 59.0%, and 53.2% basins for Xinanjiang, SIMHYD, and GR4J, respectively. (2) If the simulated ET from a particular hydrological model matches the RS-ET data well, then the RS-ET data may be used as direct input in this model. (3) The efficacy of using RS-ET is superior for relatively arid and originally poorly simulated basins. |

**Table 2.**Statistical summary (percentiles) of streamflow and ET simulation and prediction performances for Scheme 1 over 208 MOPEX basins.

Model | Model Performance (%) | Calibration | Validation | Regionalization | ||||||
---|---|---|---|---|---|---|---|---|---|---|

25 | 50 | 75 | 25 | 50 | 75 | 25 | 50 | 75 | ||

Xinanjiang | NSE_{Q} | 52.22 | 64.67 | 74.05 | 46.85 | 59.39 | 70.15 | 29.01 | 45.84 | 59.27 |

NSE_{E} | 36.59 | 50.86 | 56.92 | 39.87 | 51.30 | 57.69 | 33.19 | 48.21 | 56.38 | |

SIMHYD | NSE_{Q} | 32.70 | 42.01 | 49.64 | 28.55 | 39.93 | 47.81 | 18.27 | 32.96 | 41.90 |

NSE_{E} | 20.18 | 28.30 | 37.43 | 16.12 | 23.40 | 30.92 | 15.51 | 26.25 | 33.27 | |

GR4J | NSE_{Q} | 57.22 | 65.79 | 72.68 | 50.73 | 63.44 | 73.22 | 25.91 | 46.53 | 62.06 |

NSE_{E} | −10.61 | 2.69 | 14.65 | −12.47 | 4.88 | 16.67 | −9.50 | 3.12 | 14.13 |

**Table 3.**Comparison among the streamflow performance of Scheme 2 and Scheme 3 versus Scheme 1 by basin percent.

Model | Periods | Percent (Basin Numbers) | |
---|---|---|---|

Scheme 2 > Scheme 1 | Scheme 3 > Scheme 1 | ||

Xinanjiang model | Calibration | 85.1% (177/208) | 92.3% (156/169) |

Validation | 86.1% (179/208) | 93.5% (158/169) | |

Regionalization | 85.1% (177/208) | 91.1% (154/169) | |

SIMHYD | Calibration | 51.9% (108/208) | 67.5% (56/83) |

Validation | 40.4% (84/208) | 56.6% (47/83) | |

Regionalization | 50.5% (105/208) | 59.0% (49/83) | |

GR4J | Calibration | 42.3% (88/208) | 67.7% (42/62) |

Validation | 32.2% (67/208) | 45.2% (28/62) | |

Regionalization | 38.0% (79/208) | 53.2% (33/62) |

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. |

© 2023 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).

## Share and Cite

**MDPI and ACS Style**

Gui, Z.; Zhang, F.; Chang, D.; Xie, A.; Yue, K.; Wang, H.
A General Method to Improve Runoff Prediction in Ungauged Basins Based on Remotely Sensed Actual Evapotranspiration Data. *Water* **2023**, *15*, 3307.
https://doi.org/10.3390/w15183307

**AMA Style**

Gui Z, Zhang F, Chang D, Xie A, Yue K, Wang H.
A General Method to Improve Runoff Prediction in Ungauged Basins Based on Remotely Sensed Actual Evapotranspiration Data. *Water*. 2023; 15(18):3307.
https://doi.org/10.3390/w15183307

**Chicago/Turabian Style**

Gui, Ziling, Feng Zhang, Da Chang, Aili Xie, Kedong Yue, and Hao Wang.
2023. "A General Method to Improve Runoff Prediction in Ungauged Basins Based on Remotely Sensed Actual Evapotranspiration Data" *Water* 15, no. 18: 3307.
https://doi.org/10.3390/w15183307