Abstract
Fine-temporal resolution rainfall is a prerequisite for flood risk-related studies. In recent years, satellite precipitation products (SPPs) have gained some attention in the hydro-meteorology field, but the application of these products for hydrological studies is still uncertain. This study discusses the methodology for temporal disaggregation of daily rain gauge (DRG) data using SPP for improved accuracy in hydrologic simulation. The objectives of this research article are (i) to temporally dis-aggregate the DRG data using 3-h SPP TRMM_3B42RT; (ii) to suggest the best approach in terms of hydrologic simulation capabilities using the compound factor (CF) technique. The hydrologic simulation in HEC-HMS was performed for various temporally disaggregated rain gauge (TDRG) data and results were analyzed using 10 efficiency criteria. The best approach was selected using the CF technique. The lowest CF of 2.4 reveals that peak discharge can be computed with less than 8% deviation using the TDRG data, which utilizes the least error tile/grid/pixel from rain gauge rainfall value out of four nearby tile/grid/pixel. This study concluded that flood volume can be computed with less than 5% deviation using DRG data but it failed to reproduce the flood peak. This approach has overcome the insufficiency of sub-daily rainfall observation.
HIGHLIGHTS
Evaluation of SPP TRMM_3B42RT for a major flood event using statistical measurements like RB (%), CC, and RMSE (mm/day).
Temporal disaggregation of the DRG data using 3-h SPP TRMM_3B42RT.
Identification of the best approach in terms of hydrologic simulation by using 10 efficiency criteria and the use of the CF technique.
INTRODUCTION
Rainfall is the primary variable required to drive the hydrologic model. Rainfall has a strong influence on the hydrologic response (Hou et al. 2020) due to its ample spatial-temporal variability (Cho 2020). Hence, it becomes essential to capture the spatial-temporal variability of rainfall for accurate runoff prediction. Generally, the conversion of daily rainfall to sub-daily rainfall relies on the characteristics of observed sub-daily rainfall in the past. There are some temporal disaggregation approaches like multiplicative cascade models, multivariate techniques and simple methods of fragments that preserve the rainfall characteristics like rainfall intermittency, rainfall asymmetry at finer temporal resolution and autocorrelation. The availability of historically recorded sub-daily rainfall of sufficient length is still very limited in most of the regions.
In recent years, the application of satellite precipitation products (SPPs) has grown due to their fine spatial-temporal resolutions. Microwave- and infrared sensor-based satellites for precipitation retrievals are available with global coverage (Ghimire et al. 2022). The error produced in SPPs subjected to the interpolation technique used (Nerini et al. 2015; Chen et al. 2016). In addition, SPP is an alternative or supplementary data source to ground-based rain gauge data to drive the hydrologic model. The SPPs have gained significant momentum in hydrologic modeling due to their high spatial-temporal coverage. Furthermore, the SPPs are helpful for the temporal disaggregation of daily rainfall to sub-daily rainfall (Yaswanth et al. 2023).
There are studies (Ryo et al. 2014; Shrestha et al. 2019; Acharya et al. 2022) related to improving the temporal resolution of ground-based rain gauge rainfall for hydrologic simulation. The findings of these studies suggested that temporally disaggregated rain gauge (TDRG) data increases the simulation performance of flood peaks.
Multiple hydrological applications like flood forecasting, rainfall–runoff modeling, water balance modeling and computer-based models for transport of pollutants require high temporal resolution rainfall data (Güntner et al. 2001). The hydrological model's application for practical purposes is often limited by the temporal resolution of rainfall data (Aronica et al. 2005). The accuracy of hydrological models is largely influenced by the spatial resolution of input data (Yang et al. 2001). The temporal disaggregation approaches have appreciable benefits. They can increase the resolutions of hydrological processes (Koutsoyiannis et al. 2003).
Hazarika et al. (2007) compared the observed point gauge rainfall data with TRMM_3B42RT rainfall data for real-time flood forecasting in the Bagmati River in Nepal. The pixel-based comparison of TRMM_3B42RT rainfall and observed rainfall showed that TRMM_3B42RT rain fields closely correspond with aerial average rainfall. It was also found that TRMM_3B42RT underestimated the rain for peak monsoon period for the wet rain regimes whereas it overestimated the rain for dry rain regimes. Dönmez & Tekeli (2017) tested TRMM_3B42RT rainfall data for Gaziantep, a city located in a semi-arid region of Turkey for the flood predictions. Their research demonstrated that TRMM_3B42RT rainfall data indicated high rainfall rates during the flood events. Yuan et al. (2019) proposed a framework for statistical and hydrological assessment of the TRMM and GPM SPPs in both near and post-real-time versions at sub-daily temporal scales in a poorly gauged watershed in Chindwin River basin, Myanmar. Statistical assessment at grid and basin scales shows that TRMM_3B42RT rainfall data generally presents higher quality among various SPPs. The TRMM_3B42RT obtains satisfactory hydrological performance in a 3-h flood simulation, with a Nash–Sutcliffe model efficiency of 0.868.
This research discusses the different approaches for the conversion of daily to sub-daily rain gauge (DRG) rainfall values and implications related to hydrologic simulation. An effort is devoted to improving the hydrologic simulation capabilities by using TDRG data by SPP. The objectives of this research article are (i) to evaluate SPP TRMM_3B42RT for a major flood event using statistical measurements; (ii) to temporally disaggregate the DRG data using 3-h SPP TRMM_3B42RT; and (iii) to suggest the best approach in terms of hydrologic simulation capabilities using compound factor (CF) technique.
STUDY AREA AND DATA COLLECTION
The weblinks to access the DRG rainfall, SPP TRMM_3B42RT, digital elevation model, land use land cover, and hydrologic soil group data and related information are provided in the ‘Data availability statement’. The 2-h discharge data at the stream gauge station was collected from the Bisalpur dam authority, which is located downstream of the stream gauge station.
METHODOLOGY
Approach-1 (Figure 2(a)): The ground-based rain gauge station is located in tile-2. So, tile-2 is utilized for the extraction of rainfall values. This scenario is abbreviated as S-1 in the later part of the article.
Approach-3 (Figure 2(a)): The R1 denotes DRG rainfall. P1, P2, P3, and P4 are TRMM_3B42RT aggregated daily rainfall values. A tile that shows the minimum deviation from R1 is utilized for the extraction of rainfall values. So, the tile which represents the minimum of {|R1–P1|, |R1–P2|, |R1–P3|, |R1–P4|} is utilized for extraction of rainfall values. This scenario is abbreviated as S-3 in the later part of the article.
Approach-4 (Figure 2(b)): In this approach the dry period was identified from the rainfall series generated from S-1. After, the identification of the dry period the rainfall was uniformly distributed in the wet period. The dry period and wet period were identified using the threshold of 0 and >0 mm/h rainfall, respectively. This scenario is abbreviated as S-4 in the later part of the article.
Model . | Method . | Parameter . |
---|---|---|
Loss | SCS curve number | Initial abstraction Curve number |
Transform | SCS unit hydrograph | Lag time |
Routing | Muskingum | Storage-time constant (K) Weighting factor (X) |
Baseflow | Recession | Initial discharge Recession constant Ratio to peak |
Rain gauge data | Thiessen polygon | Thiessen weight |
Model . | Method . | Parameter . |
---|---|---|
Loss | SCS curve number | Initial abstraction Curve number |
Transform | SCS unit hydrograph | Lag time |
Routing | Muskingum | Storage-time constant (K) Weighting factor (X) |
Baseflow | Recession | Initial discharge Recession constant Ratio to peak |
Rain gauge data | Thiessen polygon | Thiessen weight |
To strengthen the persuasiveness of the results an additional event from 04 August 2019 to 21 August 2021 with an observed peak discharge of 2,309 m3/s was also selected. The hydrologic simulation was performed on a daily scale (as per the availability of daily runoff data during this period) which shows an NSE and R2 value of 0.81 and 0.90, respectively.
Furthermore, the sensitivity analysis was also performed to assess the model parameter's sensitivity to the results. Sensitivity analysis provides a comprehensive understanding of physical processes within the complex hydrologic systems. The one value at a time, i.e., local sensitive analysis was performed with parameters variation from +30 to −30% at an increment of 10%. The results of sensitivity analysis revealed that curve number and storage-time constant have the highest and ratio to peak has the lowest sensitivity for peak discharge. Whereas, curve number and recession constant have the highest and weighting factor has the lowest sensitivity for flood volume.
To check the efficiency of various TDRG data to reproduce the flood hydrograph, a total of ten efficiency criteria were tested as listed in Table 2.
Statistic . | Equation . | Optimal value . |
---|---|---|
Percent difference in runoff volume | 0 | |
Percent difference in peak flow | 0 | |
Difference in flood peak time (h) | 0 | |
Coefficient of determination | 1 | |
Nash–Sutcliffe efficiency | 1 | |
Index of agreement | 1 | |
Kling-Gupta efficiency | 1 | |
Root mean square error | 0 | |
Normalized root mean square error | 0 | |
Mean absolute error | 0 |
Statistic . | Equation . | Optimal value . |
---|---|---|
Percent difference in runoff volume | 0 | |
Percent difference in peak flow | 0 | |
Difference in flood peak time (h) | 0 | |
Coefficient of determination | 1 | |
Nash–Sutcliffe efficiency | 1 | |
Index of agreement | 1 | |
Kling-Gupta efficiency | 1 | |
Root mean square error | 0 | |
Normalized root mean square error | 0 | |
Mean absolute error | 0 |
Vs is simulated runoff volume (mm), Vo is observed runoff volume (mm), Qs is simulated discharge (m3/s), Qo is observed discharge (m3/s), Qsp is simulated peak discharge (m3/s), Qop is observed peak discharge (m3/s), Qoa is average observed discharge (m3/s), Qsa is average simulated discharge (m3/s), TPS is simulated flood peak time, TPO is observed flood peak time, cc is correlation coefficient, σs is the standard deviation of simulated discharge (m3/s), and σo is the standard deviation of observed discharge (m3/s).
RESULTS AND DISCUSSION
The evaluation of TRMM_3B42RT for the 2016 flood event revealed that biases are present. The total rainfall depth was underestimated and overestimated for higher rainfall and lower rainfall, respectively. Hence, the direct use of TRMM_3B42RT data for hydrologic simulation can produce misleading results. So, temporal disaggregation of existing rain gauge data is adopted as discussed in methodology sections by using Equation (5). TDRG data were prepared considering S-1, S-2, S-3, S-4, and S-5 approach.
The S-1 approach has given the maximum 3-h rainfall at RGS7 with a magnitude of 111.41 mm, which is 299.78% higher than the DRG rainfall. There was a significant difference in the maximum 3-h rainfall in all approaches. So, the method of rainfall extraction at any point location using four nearby tiles plays a significant role.
Rainfall . | Flood volume (mm) . | Peak discharge (m3/s) . | Time of peak (dd-mm-yyyy hh:mm) . | |||
---|---|---|---|---|---|---|
Observed . | Simulated . | Observed . | Simulated . | Observed . | Simulated . | |
DRG | 87.81 | 91.69 | 6,500 | 5,366.7 | 8-8-2016 | 8-8-2016 |
18:00 | 12:00 | |||||
S-1 | 87.81 | 115.55 | 6,500 | 8,062.5 | 8-8-2016 | 8-8-2016 |
18:00 | 16:00 | |||||
S-2 | 87.81 | 93.22 | 6,500 | 5,464.4 | 8-8-2016 | 9-8-2016 |
18:00 | 12:00 | |||||
S-3 | 87.81 | 107.04 | 6,500 | 7,017.9 | 8-8-2016 | 9-8-2016 |
18:00 | 7:00 | |||||
S-4 | 87.81 | 53.37 | 6,500 | 3,555.3 | 8-8-2016 | 9-8-2016 |
18:00 | 7:00 | |||||
S-5 | 87.81 | 102.95 | 6,500 | 8,472.8 | 8-8-2016 | 9-8-2016 |
18:00 | 3:00 |
Rainfall . | Flood volume (mm) . | Peak discharge (m3/s) . | Time of peak (dd-mm-yyyy hh:mm) . | |||
---|---|---|---|---|---|---|
Observed . | Simulated . | Observed . | Simulated . | Observed . | Simulated . | |
DRG | 87.81 | 91.69 | 6,500 | 5,366.7 | 8-8-2016 | 8-8-2016 |
18:00 | 12:00 | |||||
S-1 | 87.81 | 115.55 | 6,500 | 8,062.5 | 8-8-2016 | 8-8-2016 |
18:00 | 16:00 | |||||
S-2 | 87.81 | 93.22 | 6,500 | 5,464.4 | 8-8-2016 | 9-8-2016 |
18:00 | 12:00 | |||||
S-3 | 87.81 | 107.04 | 6,500 | 7,017.9 | 8-8-2016 | 9-8-2016 |
18:00 | 7:00 | |||||
S-4 | 87.81 | 53.37 | 6,500 | 3,555.3 | 8-8-2016 | 9-8-2016 |
18:00 | 7:00 | |||||
S-5 | 87.81 | 102.95 | 6,500 | 8,472.8 | 8-8-2016 | 9-8-2016 |
18:00 | 3:00 |
The efficiency criteria were evaluated as represented in Table 2. The efficiency criteria values for DRG, S-1, S-2, S-3, S-4, and S-5 are listed in Table 4. DRG rainfall outperformed in terms of a percent difference in runoff volume criteria. S-1 rainfall outperformed in terms of the difference in flood peak time, coefficient of determination, Nash–Sutcliffe efficiency and index of agreement criteria. S-2 rainfall outperformed in terms of Kling-Gupta efficiency and RMSE criteria. S-3 rainfall outperformed in terms of a percent difference in peak flow and normalized RMSE criteria. S-4 rainfall fails to outperform in selected efficiency criteria. S-5 rainfall outperformed in terms of mean absolute error criteria.
Rainfall . | Dv (%) . | Dp (%) . | Tp (h) . | R2 . | NSE . | d . | KGE . | RMSE (m3/s) . | NRMSE . | MAE (m3/s) . |
---|---|---|---|---|---|---|---|---|---|---|
DRG | 4.419 | –17.435 | –6.00 | 0.604 | 0.526 | 0.874 | 0.767 | 1,024.940 | 0.720 | 603.811 |
S-1 | 31.591 | 24.038 | –2.00 | 0.776 | 0.722 | 0.910 | 0.668 | 1,031.142 | 0.575 | 581.713 |
S-2 | 6.161 | –15.932 | 18.00 | 0.673 | 0.609 | 0.900 | 0.800 | 919.808 | 0.635 | 533.661 |
S-3 | 21.900 | 7.968 | 13.00 | 0.717 | 0.677 | 0.908 | 0.755 | 952.817 | 0.573 | 576.304 |
S-4 | –39.221 | –45.303 | 13.00 | 0.482 | –1.549 | 0.680 | –0.192 | 1,282.580 | 1.547 | 749.029 |
S-5 | 17.242 | 30.351 | 9.00 | 0.696 | 0.676 | 0.901 | 0.751 | 1,008.664 | 0.631 | 523.767 |
Rainfall . | Dv (%) . | Dp (%) . | Tp (h) . | R2 . | NSE . | d . | KGE . | RMSE (m3/s) . | NRMSE . | MAE (m3/s) . |
---|---|---|---|---|---|---|---|---|---|---|
DRG | 4.419 | –17.435 | –6.00 | 0.604 | 0.526 | 0.874 | 0.767 | 1,024.940 | 0.720 | 603.811 |
S-1 | 31.591 | 24.038 | –2.00 | 0.776 | 0.722 | 0.910 | 0.668 | 1,031.142 | 0.575 | 581.713 |
S-2 | 6.161 | –15.932 | 18.00 | 0.673 | 0.609 | 0.900 | 0.800 | 919.808 | 0.635 | 533.661 |
S-3 | 21.900 | 7.968 | 13.00 | 0.717 | 0.677 | 0.908 | 0.755 | 952.817 | 0.573 | 576.304 |
S-4 | –39.221 | –45.303 | 13.00 | 0.482 | –1.549 | 0.680 | –0.192 | 1,282.580 | 1.547 | 749.029 |
S-5 | 17.242 | 30.351 | 9.00 | 0.696 | 0.676 | 0.901 | 0.751 | 1,008.664 | 0.631 | 523.767 |
Bold values indicate the best results.
To suggest the best approach, we have used the multi-criteria decision-making-based CF technique. A study by Shaikh et al. (2022) also used the CF technique for the ranking of synthetic unit hydrograph methods. This method converts the qualitative understanding of a phenomenon into a quantitative one (Hembram & Saha 2020). However, the CF technique gives lumped value by considering equal weightage to selected efficiency criteria, resulting in the exaggeration of the final output. The ranks were assigned based on the optimal value of efficiency criteria. A lower rank is assigned to the best-performing scenario, i.e., scenario which shows the closest value to the optimal value of efficiency criteria.
Assigned rank and CF values (Equation (9)) are listed in Table 5. The lowest value of CF was 2.4 for approach S-3. So, the first rank was assigned to approach S-3. The second rank was assigned to approaches S-1 and S-2 due to the tie between CF values.
Rainfall . | Dv (%) . | Dp (%) . | Tp (h) . | R2 . | NSE . | d . | KGE . | RMSE (m3/s) . | NRMSE . | MAE (m3/s) . | CF . | Overall rank . |
---|---|---|---|---|---|---|---|---|---|---|---|---|
DRG | 1 | 3 | 2 | 5 | 5 | 5 | 2 | 4 | 5 | 5 | 3.7 | 4 |
S-1 | 5 | 4 | 1 | 1 | 1 | 1 | 5 | 5 | 2 | 4 | 2.9 | 2 |
S-2 | 2 | 2 | 5 | 4 | 4 | 4 | 1 | 1 | 4 | 2 | 2.9 | 2 |
S-3 | 4 | 1 | 4 | 2 | 2 | 2 | 3 | 2 | 1 | 3 | 2.4 | 1 |
S-4 | 6 | 6 | 4 | 6 | 6 | 6 | 6 | 6 | 6 | 6 | 5.8 | 5 |
S-5 | 3 | 5 | 3 | 3 | 3 | 3 | 4 | 3 | 3 | 1 | 3.1 | 3 |
Rainfall . | Dv (%) . | Dp (%) . | Tp (h) . | R2 . | NSE . | d . | KGE . | RMSE (m3/s) . | NRMSE . | MAE (m3/s) . | CF . | Overall rank . |
---|---|---|---|---|---|---|---|---|---|---|---|---|
DRG | 1 | 3 | 2 | 5 | 5 | 5 | 2 | 4 | 5 | 5 | 3.7 | 4 |
S-1 | 5 | 4 | 1 | 1 | 1 | 1 | 5 | 5 | 2 | 4 | 2.9 | 2 |
S-2 | 2 | 2 | 5 | 4 | 4 | 4 | 1 | 1 | 4 | 2 | 2.9 | 2 |
S-3 | 4 | 1 | 4 | 2 | 2 | 2 | 3 | 2 | 1 | 3 | 2.4 | 1 |
S-4 | 6 | 6 | 4 | 6 | 6 | 6 | 6 | 6 | 6 | 6 | 5.8 | 5 |
S-5 | 3 | 5 | 3 | 3 | 3 | 3 | 4 | 3 | 3 | 1 | 3.1 | 3 |
CONCLUSIONS
The following conclusions can be drawn from this research article.
- 1.
Higher rainfall values were underestimated and lower rainfall values were overestimated by SPP TRMM_3B42RT.
- 2.
The rainfall results of different scenarios suggest that satellite precipitation tile/grid/pixel to point estimate using four nearby tiles/grids/pixels is sensitive to the extraction technique used.
- 3.
The CF technique revealed that the approach S-3 performed excellently. For the temporal disaggregation of rain gauge rainfall, approach S-3 utilizes a tile/grid/pixel representing the least deviation from observed rainfall at RGS out of four nearby tiles/grids/pixels.
- 4.
The DRG data were found to be capable of capturing the runoff volume with less than a 5% deviation in runoff volume.
- 5.
The approach S-3 was found to be capable of capturing the peak discharge with less than 8% deviation in peak discharge.
- 6.
It can be concluded that flood volume can be computed accurately using DRG data but it failed to reproduce the flood peak. The peak discharge can be computed accurately using the S-3 approach.
These results were derived using the 2016 flood event, which is characterized by a bimodal hydrograph. This study strengthens the findings of Yaswanth et al. (2023), in which improved hydrologic simulation results were reported by the use of TDRG data using an approach similar to S-3. This approach will be useful in remote areas where rain gauge is limited in number and available at daily temporal resolution. Real-time TRMM_3B42 data are utilized, which in the application can simulate the real-time stream flow. In the future, this methodology can be verified and modified with an enhanced understanding of different SPPs.
ACKNOWLEDGEMENT
The authors acknowledge the Bisalpur dam authority for providing the discharge data at the Bigod stream gauging station.
DATA AVAILABILITY STATEMENT
Daily rain gauge data can be downloaded from http://www.water.rajasthan.gov.in/wrd. 3-Hourly TRMM rainfall data can be downloaded from https://disc.gsfc.nasa.gov/datasets/TRMM_3B42RT_7/summary?keywords=TRMM_3B42RT_7. Digital elevation model can be downloaded from https://portal.opentopography.org/raster?opentopoID=OTSRTM.082015.4326.1. Land use land cover can be downloaded from https://esri.maps.arcgis.com/apps/instant/media/index.html?appid=fc92d38533d440078f17678ebc20e8e2. Hydrologic soil group can be downloaded from https://daac.ornl.gov/Soils/guides/Global_Hydrologic_Soil_Group.html. The discharge data collected from Bisalpur dam authority cannot be made publicly available.
CONFLICT OF INTEREST
The authors declare there is no conflict.