Estimation of Runoff Under Changed Climatic Scenario of a Meso Scale River by Neural Network Based Gridded Model Approach

Climate change is linked with change in precipitation, evapotranspiration, and other climatological parameters, and therefore the runoff of a river basin will be affected. The Gomati River basin is the largest in Tripura. The increased settlement in the Gomati River basin and climate change may threaten the natural flow patterns that enable its diversity. This study assesses the impact of climate change on total flow from a catchment in northeast India (the Gomati River catchment). For this assessment, the Group Method of Data Handling (GMDH) model was used to simulate the rainfall–runoff relationship in the catchment with respect to the observed data during 2008–2009. The statistically downscaled outputs of the Hadley Centre Global Environment Model version 2 (HadGEM2-ES) general circulation model scenario was used to assess the impacts of climate change on the Gomati River basin. Future projections were developed for the 2030s, 2040s, and 2050s. The results of this study may contribute to the development of adaptive strategies and future policies for the sustainable management of water resources in northeast Tripura.


Introduction
Global climate is changing due to an increased greenhouse effect, which will affect temperature distribution in time and space. Since the 1920s, the temperature of the planet has shown an increasing trend. Recently, the number of warm years observed exceeds the expected range of natural variability (Abram et al. 2021;Oscar-Junior 2021). The expected increase in mean surface temperature is expected to be 0.3-0.6 °C per decade in the current century (Rustad 2001). India's average temperature increased by around 0.7 °C between 1901 and 2018, and it is projected to increase by approximately another 4.4 °C by the end of this century (Krishnan et al. 2020). Key hydrological processes in the water cycle, such as precipitation, are directly impacted by this change in temperature. Additionally, over areas of central India, such as the southwest coast, the southern peninsula, and northeast India, have experienced two droughts per decade on average over the last few decades (Krishnan et al. 2020). This has changed the spatiotemporal variation in water resources in river basins (Strohmeier et al. 2020;Shin et al. 2021). Therefore, river discharges are likely to change in volume and distribution over the course of the year. Quantitative assessment of these changes is required to set long-term policies for future river management.
Thus, a future projection of river discharge is needed to deal with water-related disasters caused by climate change, such as floods, droughts, and water scarcity. Hydrological and flow routing models can play a key role in converting climate model outputs into river discharge. It is possible to evaluate future changes in the availability of water resources, flood discharge, and drought frequency, and identify possible future hotspots for water-related disasters only using projections of river discharge information. Runoff models with a 1° spatial resolution and general circulation model (GCM) outputs have been used to study variation in the future threats of floods and droughts on a global scale (Rodriguez et al. 2020;Mehdi et al. 2021). While a runoff model with a 1° spatial resolution is sufficient for analyzing river discharge variation on a global scale, it is not suitable for analyzing variation at the regional or country scale. Instead, detailed, high-resolution hydrological models are used to study the impact of climate change at the basin scale (Mali et al. 2021). It is difficult to use high-resolution complex hydrological models to identify hotspots at the regional or country scale because of the high degree of computational power required to compute hydrological models and complications in identifying the model parameters.
Climate change is leading to changes in temperature, precipitation, Evapotranspiration, and other climatological parameters. Therefore, it is necessary to identify changes in flow patterns and the availability of water resources for different regions, as this depends directly on all climatological parameters of each given region.

Objectives and Novelty of the Study
The present study assesses the impact of climate change on rainfall runoff relationship of the Gomati River basin in northeast India. In this regard the following sub objectives have been adopted: (i) Conduct a rainfall-runoff simulation of the Gomati River basin. (ii) Apply a gridded approach to the behavior of the hydrological system of the Gomati River basin.
(iii) Assess the impact of climate change on the rainfall-runoff relationship.
The novelty of the present study is the Representative Elementary Area concept is used for watershed modeling with the help of polynomial neural networks. This paper highlights the first instance of the application of GMDH in the prediction of discharge by the REA concept (Jana and Majumder 2010). As the model is distributed this can be considered as the first instance of the application of GMDH in distributed modeling approach.

Study Area
Tripura is a tiny state in northeast India, with a geographical area of 10,486 km 2 (Forest Survey of India 2017; Mathur and Bhattacharya 2022), which is bordered by Bangladesh on three sides. The Gomati River basin is located in the lower middle part of Tripura and spreads from the eastern to western boundaries of the state. The catchment area of the Gomati River is 2205.04 km 2 within the boundary of India. It has the largest basin of any river in Tripura. Its catchment area is nearly 21.03% of the total geographic area of Tripura. It also hosts the one and only hydropower project in Tripura, at Dumboor. Thus, this river basin is by far the most important of all of the river basins of this state. In recent years, due to the influence of climate change, the water resource availability of this basin has changed drastically, resulting in important effects on the life of the residents of the basin, directly or indirectly. In addition, changes in water resource availability are reducing the output of the hydropower project at Dumboor. Therefore, it is very important to understand the impact of climate change on the water resource availability of the Gomati River basin (Fig. 1).

Data Compilation
In this present study, computed and observed data were used to develop the model. The detailed data compilation procedure has been discussed in the Sects. 2.1.1 and 2.1.2 respectively.

Computed Data
The catchment boundary and watershed area of the Gomati River basin were delineated and computed using Google Earth Pro tool. The watershed area was divided into 19 grids of 15 × 15 km (Fig. 2).
Land use land cover data were extracted using Geomatica FreeView software with Google Earth imagery of the Gomati River basin. Images of all of the 19 grids were taken, and land use data for each grid were compute.
The composite runoff coefficient was computed by taking the runoff coefficient for agricultural land, settlement, and forest land as 0.28, 0.45, and 0.35 (Subramanya 2013). Equation (1) was used for computing composite runoff coefficient. (1) Gomati River basin
Climate change data for the Gomati basin (2010-2050) were compiled using the HadGEM2-ES climate model for scenario RCP-2.60.

Modeling Techniques
For this study, eight modeling techniques were compared to compute runoff for the study basin: Hydrologic Engineering Centre-Hydrologic Modelling System (HEC-HMS) (Zotou et al. 2020;Kumar et al. 2021 (Freedman 2009). The climate model HadGEM2-ES (Collins et al. 2011;Gyamfi et al. 2021) was used to analyze the impact of climate change in rainfall-runoff. After developing this model using random values, its accuracy was analyzed in terms of root mean square error (RMSE), correlation (R), coefficient of determination (R 2 ), mean absolute error (MAE), NES, and PBIAS (Moriasi et al. 2007).

Computation of Grid Wise Rainfall
Station-wise observed rainfall data for 2008-2018 (collected from IMD) were converted into grid-wise rainfall data (calculated in Sect. 3.6) using the Thiessen polygon method (Subramanya 2013).

Selection of Best Model for Total Flow
This study discusses the application of different approaches to the computation of total runoff in the form of a gridded approach, a station-wise approach, and a lumped approach using the D8 method. After the computation of total runoff through these different approaches, the best approach is determined, and base flow of the river is added to the runoff, which provides the total flow.
Grid-wise downscaled GCM data from 2010 to 2018 were compared to grid-wise observed rainfall data (i.e., computed using Thiessen polygons), and grid-wise absolute error was computed. Finally the weighted average of the grid-wise absolute error was computed. The prediction of total flow for the Gomati River basin was performed using the best approach for total flow computation, as identified earlier in the study.

Downscaling
Climate change data were acquired from the climate model HadGEM2-ES using climate scenario as RCP 2.6 for 19 grids for the period of 2010 to 2050. However, such GCM data are at a coarse scale, so they need to be converted to a local scale. The delta method (Dessu and Melesse 2013;Wootten et al. 2021) was used to downscale the climate data, as shown in Eq. (2).
where P Delta SD is the downscaled data of precipitation, P Obs refers to the average observed, and P GCMhist represents the GCM mean simulation historical data of precipitation. The subscript GCM rcp represents the GCM's RCP outputs over the future period, and the subscript Obs represents the observation values.
The monthly factor of P Obs P GCMhist.

Climate Change Scenario
To evaluate the impact of future climate change on the flow of the Gomati River basin, future climate data (including precipitation) were acquired from the climate model HadGEM2-ES, one of the most reliable climate models of recent times. Using this model, climate data for the 2030s, 2040s, and 2050s were acquired for use in the GMDH model.

Results and Discussion
The detailed results from the study have been discussed step by step in the Sects. 3.1 to 3.12.

Image Processing
The highest percentage of settlement is found at grid number 7, which includes Udaipur, the largest and most populated city in the Gomati River basin. Grid number 15, which contains Dumboor Lake, has the highest percentage of water surface.

Grid-wise Runoff Coefficient
The composite runoff coefficient is computed by taking the runoff coefficient for agricultural land, settlement, and forest land as 0.28, 0.45, and 0.35.

Development of the Model
For the purposes of this study, four main types of models were developed: • a hydrological model (HEC-HMS); • a polynomial neural network (GMDH); • an artificial neural network (quick propagation, conjugate gradient descent, Levenberg-Marquardt, incremental back Propagation, and batch back propagation); and • Multi-Linear Regression (MLR).
After development, the model accuracy was analyzed using RMSE, R, R 2 , MAE, NES, and PBIAS. The values of model accuracy for a given dataset are shown in Table 1.
A GMDH model was developed that takes into account the runoff coefficient, rainfall intensity, and land area as inputs and the runoff as output. The model was exported into a spreadsheet to predict the grid-wise normalized runoff of the Gomati River basin. The model is described in Eq. (3).

Thiessen Polygon Method
Station-wise yearly rainfall data from IMD (2008 to 2018) were converted into grid-wise yearly rainfall data using the Thiessen polygon method.

Computation of Discharge (from 2008-2009) and Selection of Best Approach
In this study, the base flow value was added to the surface runoff by observing the secondary data collected from CWC. The total flow was computed using three different approaches: the gridded approach, the sub-basin approach, and the lumped approach.

Gridded Approach
After calculating grid-wise total flow for 2008 and 2009, the total flow of the Gomati River basin was computed with the help of the D8 method. The error computation of the computed total flow was performed using the observed total flow for 2008 and 2009 taken from the CWC, as shown in Fig. 4.
The average yearly absolute error for the gridded approach was found to be 1.84%.

Sub-Basin Approach
After calculating station-wise total flow for the years 2008 and 2009, the total flow of the basin was computed using the D8 method. The error computation of the computed total flow was performed using observed total flow for 2008 and 2009 drawn from the CWC, as shown in Fig. 5. The average yearly absolute error for the sub-basin approach was computed to be 2.09%.

Lumped Approach
After total flow for 2008 and 2009 was computed, the error computation for total flow was performed using observed total flow for 2008 and 2009 drawn from the CWC, as shown in Fig. 6. The average yearly absolute error for the lumped approach was found to be 3.46%. A comparison of all three approaches indicated that the gridded approach was the best for the computation of total flow of the Gomati River basin, as it had the lowest percentage of error.

Downscaling and Error Computation of GCM Data (2010-2018)
Climate change data were collected from the climate model HadGEM2-ES using the RCP 2.6 climate scenario for the 19 grids for the period of 2010 to 2050. However, the GCM data show huge variation with respect to observed data (2010-2018) from IMD (which has been converted into grid-wise rainfall data with Thiessen polygons).
The average yearly absolute error was 35.47%, much higher than the threshold of 20%. The GCM data were downscaled so that the average yearly absolute error could be minimized reasonably.
Using the delta method of statistical downscaling, the average yearly absolute error was found to be 13.86%, which is less than the threshold of 20%.

Downscaling of GCM Data (2021-2050) and Prediction of Total Flow for the 2030s, 2040s, and 2050s
This monthly factor of P Obs P GCMhist.

Downscaled GCM Data and Total Flow for the 2030s, 2040s and 2050s
To predict the total flow for the 2030s, the best approach, i.e., the gridded approach was used. The average monthly flow and downscaled GCM data for the entire Gomati basin for 2021-2030, 2031-40, 2041-50 are shown in Fig. 7.

Changes in Flow by Decade
This study found that the average decade-wise change in flow in the Gomati River basin in the 2020s (computed using converted IMD rainfall data with the help of Thiessen polygons) and predicted flow from the 2030s, 2040s, and 2050s decreased drastically with respect to the 2010s. This change in flow is shown graphically in Fig. 8.

Conclusion
The overall conclusion, limitations and future scope of the study have been discussed in Sects. 4.1 and 4.2.

Overall Conclusion
The watershed areas in this study were delineated using Google Earth Pro, and the area of the Gomati River basin was obtained as 2205.04 km 2 . The watershed as a whole was divided into 19 grids of 15 × 15 km. Grid-wise land use areas were calculated using Geomatica FreeView. Areas of settlement, forest, water body, agricultural land, and barren land were set at 1.47%, 62.22%, 4.82%, 30.54%, and 0.99% of the total geographic area of the Gomati River basin. Using this grid-wise land use data, a grid-wise runoff coefficient was computed, and the composite runoff coefficient for the Gomati basin was identified at 0.328. A GMDH model was developed that took into account the runoff coefficient, rainfall intensity, and area as inputs and runoff as output, and the GMDH model was exported into a spreadsheet to enable the prediction of the grid-wise runoff of the Gomati River basin. This study was conducted to identify whether climate change in this region had any impact on the total flow in the Gomati River basin, and if so, how much. The study was conducted by combining the GMDH model and the HadGEM2-ES climate change model. The average total flow of the 2020s and the future decades of the 2030s, 2040s, and 2050s will decrease 27.76%, 26.50%, 25.10%, and 23.62%, respectively, with the 2010s as a baseline, due to the fall in the volume of precipitation in present and future decades. This indicates that climate change can cause a large reduction in precipitation, which will further contribute to a reduced total flow in the Gomati River basin.

Limitations and Scope for Future Research
The findings of this study refer to the future water resource availability of the Gomati River basin, in the context of possible climate change scenarios, while acknowledging the high degree of uncertainty present in any climate change projection. Future climate change is the most important factor in water resource availability in this basin. In the downscaling method it was assumed that when the value of the GCM data approached zero, then the downscaled value of the GCM data would also approach zero. Future land use is a major factor in the prediction of total flow, but it is almost impossible to predict possible land use changes.
GMDH was used to create a hydrological model in this study, and other popular tools, such as SWAT, were not used, although they may be suitable for this river basin; their use, therefore, may lead to a higher accuracy in model calibration and validation. The REW approach, for instance, similar to other innovative ungauged catchment methods, can be used to find a same result and observe differences in the present method and new methods. It is also possible by this means to identify the water scarcity or availability of this basin if the water demand data of this basin is obtained. Finally, if another appropriate base flow calculation method that is appropriate for an ungauged catchment is used, a different result may be found.

Authors Contribution
The study, concept and design, material preparation, data collection, analysis, writing of manuscript were performed by Debajit Das, Tilottama Chakraborty, Mrinmoy Majumder and Tarun Kanti Bandyopadhyay. All authors read and approved the final manuscript.

Availability of Data and Materials
Source of data has been clearly mentioned in the manuscript.

Ethical Approval
The manuscript should not be submitted to more than one journal for simultaneous consideration. Answer: Not submitted. The submitted work should be original and should not have been published elsewhere in any form or language (partially or in full), unless the new work concerns an expansion of previous work. (Please provide transparency on the re-use of material to avoid the concerns about text-recycling ('self-plagiarism'). Answer: The submitted work is original and the work not has been published elsewhere. A single study should not be split up into several parts to increase the quantity of submissions and submitted to various journals or to one journal over time (i.e. 'salami-slicing/publishing'). Answer: Not done. Concurrent or secondary publication is sometimes justifiable, provided certain conditions are met. Examples include: translations or a manuscript that is intended for a different group of readers. Answer: Not applicable. Results should be presented clearly, honestly, and without fabrication, falsification or inappropriate data manipulation (including image based manipulation). Authors should adhere to discipline-specific rules for acquiring, selecting and processing data. Answer: Results has been presented clearly, honestly, and without fabrication, falsification or inappropriate data manipulation (including image based manipulation). No data, text, or theories by others are presented as if they were the author's own ('plagiarism'). Proper acknowledgements to other works must be given (this includes material that is closely copied (near verbatim),