Reservoir water budget estimation using satellite and ground measurement data

Rising air temperatures, increased rainy runoff, reduced dry season runoff, and severe weather conditions have intensified floods and droughts, significantly affecting the reservoir water supply. The accuracy of reservoir water balance is crucial for meeting water needs. The study compares satellite data and ground measurements to analyze the water budget of Sutami Reservoir in Indonesia. Satellite data collected included precipitation (Tropical Rainfall Measuring Mission-TRMM) and evaporation (Global Land Data Assimilation System-GLDAS). The water balance approach was utilized to analyze the water budget. The suitability tests used were Root Mean Square Error (RMSE), Nash-Sutcliffe Efficiency (NSE), Correlation Coefficient (CC), and Relative Error (RE). The study revealed that the data from TRMM and GLDAS satellites closely resembled ground measurements. The reservoir water balance analysis revealed that satellite data aligns with ground measurements, indicating water shortages in the dry season and excess water in the rainy season. Satellite data is particularly beneficial for watershed management in areas lacking ground measurement equipment, as it can be analyzed for various purposes.


Introduction
Reservoirs are constructed for various purposes, including water balance, agriculture, and flood mitigation [1].They store drinking water, transport, generation of hydropower.Additionally, aquatic and production of fish, as well as recreational activities, are accommodated.Reservoirs regulate regional climate, balance ecosystems, and protect biodiversity.However, global change has led to changes in lakes and reservoirs, affecting their morphodynamic characteristics and influencing the global hydrological cycle [2].Seasonal change in lake water storage (LWSsc) is essential for estimating terrestrial water availability and influencing the hydrologic process.The global estimation of Land Surface Water Storage change (LWSsc) presents significant challenges, mostly coming from the limited availability of in-situ measurements, particularly in developing nations [3].A study on Qinghai Lake in Western China revealed that anthropogenic activities have not affected the lake's water levels.The observed decline in lake levels can be attributed to the underlying arid climatic conditions [4].This study emphasizes the significance of understanding the water balances in reservoirs and the underlying 2 factors causing the decrease in lake levels.
Accurate and precise evaluations and regional as well as global precipitation estimations are important in scientific inquiries and practical applications.Satellite rainfall estimation has evolved significantly, with various products such as Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks (PERSIANN) [5], Climate Prediction Center morphing method (CMORPH) [6], TRMM Multi-satellite Precipitation Analysis (TMPA) [7] and The Global Precipitation Measurement Mission (GPM) [8] is currently undergoing development to offer essential rainfall data to the academic community, specifically in areas where there is a scarcity of sufficient rain gauges or radar systems [9].However, satellite-based precipitation estimates have advantages and disadvantages, highlighting the need for further improvement.The TRMM satellite, launched in 1997, has been indispensable in advancing and improving precipitation estimation at high resolution, real-time, and on a quasi-global scale by utilizing TRMM data.The latest TMPA 3B42V7 has shown improved accuracy and gains many hydrological applications [10].Parameters for analyzing water balance that are easy to access using satellite data are rainfall and evaporation.
This study divides the evaporation data into measurement data and Global Land Data Assimilation System (GLDAS) 2.1 satellite data.Estimating evapotranspiration in areas without measuring gauges is accomplished by utilizing The Moderate Resolution Imaging Spectroradiometer (MODIS) products and data.It compares instantaneous latent heat flux and Bowen Ratio measurements, demonstrating its operationality and applicability without ground observations [11].
In recent years, hydrologists and engineers utilizing satellite data for various objectives have extensively examined the water balance of reservoirs.For example, lake water balance using remotely sensed data in Iran by [5].This study found no differences between rainfall from the remote sensing source and ground measurements, representing a lesser effect of meteorological sources in the water balance simulation on the water balance was also conducted for the croplands of Kansas, USA.The findings revealed the spatial distribution of crop water utilization and equilibrium in Kansas, offering valuable insights for effectively managing the state's irrigation agriculture and water resources [6].
This research needs to be done because the study site is a very important reservoir in Java because of the multipurpose reservoir.This research will be useful for areas that do not yet have a measurement station because they can directly use satellite data without calibrating.The benefit for stakeholders is that it can be used for decision support about what to do in operating the reservoir.The aims and novelty of this research lie in its use of satellite data to estimate the reservoir water budget compared to ground measurements.This approach offers cost-effective and efficient monitoring, improved decision support, and fills the data gap in areas without measurement stations.

Site Description
The investigated geographical region is the Sutami Reservoir, situated in the Eastern part of Java, Indonesia.The reservoir is situated within the confines of the Brantas River Basin, encompassing a catchment area of approximately 12,000 square kilometers, which corresponds to approximately 35% of the total land area of the East Java Province (Figure 1).The overall distance of the primary river amounts to 320 km, wherein the Brantas River stands as the second most extensive river in the Java region.It is situated 110°30'-112°55' South latitude and 70°1'-8°15' East longitude.The Sutami Reservoir is managed directly by the Jasa Tirta 1 Public Corporation.The Sutami Reservoir was built in 1972 to irrigate 34,000 ha of irrigation and 488 million kilowatts of hydropower annually.The type of Sutami dam is a rockfill with a soil core and a height of 97.50 m.

Data
The data was obtained from Jasa Tirta 1 Public Corporation, the Meteorology-Climatology and Geophysics Agency, and through global data TRMM and GLDAS 2.1 [12].
To construct the water balance simulation model spanning from 2002 to 2019, the acquisition of the subsequent data is essential: a. Inflow and rainfall data from ground measurement.b.Reservoir release data.c.Rainfall data from TRMM and evaporation data from GLDAS 2.1.

Gridded Rainfall Data
Along with observed data, two satellite-based (TRMM and GLDAS 2.1) gridded rainfall and evaporation products were derived using Google Earth Engine (GEE).The TRMM (http://trmm.gsfc.nasa.gov/)The mission concluded in 2015, marking the cessation of processing the final TMPA data ended in 2019, enhancing our understanding of tropical precipitation distribution and variability.GLDAS (https://ldas.gsfc.nasa.gov/gldas/) is an innovative tool that combines satellite and ground-based technology to generate a comprehensive representation of fluxes and land surface states through advanced land surface modeling and data assimilation methodologies.By employing a range of land surface models, such as Noah, Mosaic, Community Land Model, and Variable Infiltration Capacity, GLDAS can provide global fluxes at varying spatial and temporal resolutions.NASA's Hydrology Data and Information Services Center describes GLDAS models and forcing datasets, including the evaporation dataset from the GLDAS 2.1 National Oceanic and Atmospheric Administration (NOAH) model.

Figure 1. Map of the Study Location
Monthly rainfall data and evaporation were collected from 2002 to 2019 according to the availability of observational data in the field to estimate long-term trends.Subsequently, a comparison was conducted between the data above and satellite data that was readily accessible.This comparison was performed by aligning the data according to the station's geographical coordinates, considering both high spatial resolution and comparable observation periods.

The Utilization of Data Handling and Statistical Applications in the GEE Platform
The gridded rainfall products monthly were acquired from the Google Earth Engine (GEE) platform.The ImageCollection algorithm and the filter command are components utilized in image processing and analysis.The study utilized high-resolution modeled datasets, which were obtained and subjected Catchment Area of Sutami Reservoir to analysis.Bilinear interpolation resampling was employed to ensure spatial similarity during the analysis process.Nearest point-pixel rainfall values were extracted for station localities.Using satellite data presents a significant challenge to the scientific community due to its inherent limitations.Certain statistical techniques were employed to assess the data's distribution and compare it to the grid data set to address this issue.This analysis aimed to ascertain the reliability and accuracy of the satellite-derived rainfall and evaporation data concerning the observed data collected at the station under investigation.Furthermore, several statistical tests that have been carried out are (a) the Rainfall data consistency test using the Multiple Mass Curve and Rescaled Adjusted Partial Sums (RAPS) test method, (b) Evaporation data consistency test using RAPS test method, (c) Rainfall data quality test and evaporation (Stationary) with the F test and T-test to determine the stability of the value data variance and determine the stability of the average value of the tested data and determine the nature of the data, The RMSE, NSE, CC, and Relative Error are used to assess the average error, Nash-Sutcliffe efficiency, correlation coefficient, and relative error in comparing satellite data to measurement data.

Root Mean Square Error
The root mean square error (RMSE), akin to the mean absolute error (MAE), emphasizes the computation of the average error size in satellite and in situ data [13].

Nash-Sutcliffe Efficiency (NSE)
Nash-Sutcliffe efficiency (NSE) evaluates hydrologic model performance, assessing bias and goodness of fit based on satellite rainfall data [14]: The variable denoted as Qo,i represents the measured streamflow data obtained from the hydrologic station, Qs,i is simulated.The NSE efficiency indicates the similarity of simulated discharge to observed discharge, with a value close to 1 indicating identical or nearly identical discharge.A negative number signifies that the mean value of the observed data is superior to that of the model.

Correlation Coefficient
The variables RSi and ROi represent the satellite-derived and observed rainfall values, respectively.The variable N represents the total number of observations in the dataset.

Reservoir Water Budget Estimation
The change in storage in a reservoir can be described as the residual of the water balance [4].
Where ΔS is the reservoir water budget change, P is precipitation, E is total evaporation from the reservoir water bodies, I is inflow, and O is the outflow for irrigation.

Analysis of Rainfall and Evaporation Data Quality A. Data Consistency Test (Multiple Mass Curve Method and RAPS)
Data quality is essential for analyzing statistical data presented in periodic series, including data related to rainfall and evaporation.There are six rainfall stations and one Meteorology, Climatology, and Geophysical Agency (MCGA) evaporation station in the area of the Sutami reservoir, as well as available TRMM and GLDAS.The Multiple Mass Curve method consistency test was applied to the rainfall station for the measurements.In contrast, for the MCGA evaporation station, TRMM, and GLDAS, the Ranking Alternatives by Perimeter Similarity (RAPS) method was applied.All data that have passed the consistency test calculations had consistent results and could be utilized in the next stage of the process.

B. Stationary Test (T Test and F Test)
Based on the findings of the Stationary Test (t-test and F-Test), the MCGA evaporation station and TRMM were stationary, as the mean and variance values of the data were stable or of the same type.Whereas for GLDAS, the results were non-stationary, which was indicated by the variance value of the data being unstable.

C. Average Rainfall
The TRMM data acquired was in the format of rainfall data averaged across a specific region.Hence, the data for the six-rainfall station first had to be converted into regional average rainfall data to compare them with the TRMM data, which is the regional average rainfall data.The regional average rainfall analysis was performed to obtain the amount that can represent the amount for the entire Sutami Reservoir Watershed.The method for calculating the regional average rainfall in the Sutami Reservoir was the Thiessen Polygon Method with the ArcGIS 10.3 software.1, the value of Ac was obtained for each rainfall station.The extent of influence of each rainfall station in the Sutami Reservoir influences the value of Ac.The calculation results for the Ac value were subsequently utilized to determine the average regional precipitation using the Thiessen Polygon Method.

D. Uncorrected Data Conformity Test
Data Suitability Test is an evaluation of the model to get an idea of the level of uncertainty that a model has in predicting hydrological processes.This study used a data suitability test to compare measurement data (rainfall and evaporation) and satellite data (TRMM and GLDAS).Satellite data is considered simulation data, while measurement data is considered observation data (data considered according to the conditions it should be).The highest value of RMSE is observed during the ten days, as depicted in Table 2.The value in the NSE method shows the results of " Unsatisfactory" for the evaporation data for all periods.The classification of the model's predictive efficiency is determined [16], where it is considered good when the Nash Sutcliffe Efficiency (NSE) is greater than 0.75, satisfactory when NSE ranges from 0.36 to 0.75, and unsatisfactory when NSE is less than 0.36.The results of this interpretation show that the NSE value <0.36 is included in the unsatisfactory category or does not meet the expected standard statistically.As for the rainfall data for all periods, the results show " Satisfactory " because the value is 0.36<NSE<0.75.
The relative error (RE) value in the rainfall data for all periods has the same value of 0.05 or 5%.The relative error of TRMM rainfall data compared to post-station rainfall can be considered low.Meanwhile, the relative error in evaporation data has negative results.The correlation coefficient value manifests "Very Strong" results exclusively during the monthly rainfall data duration due to a correlation coefficient value greater than 0.70.In contrast, for the monthly period on evaporation data, the 10-days period on rainfall and evaporation data, the result is "moderate up to very strong" due to the correlation coefficient of 0.60-0.79.
Table 2 shows that although the Correlation Coefficient (R) of measured and uncorrected satellite data shows a strong relationship, the overall results are "Strong" except for the monthly period for measured rainfall and TRMM.However, the NSE value does not meet the evaporation data, and the relative error value is negative, while the 10-day rainfall data shows an NSE value of <0.5.So, it is necessary to make corrections to the Evaporation and Rainfall data.

E. Suitability Test of Corrected Data
The corrected data suitability test was performed on ten days and monthly periods.In each of these periods, the utilized data was also with rainfall values.Before performing the suitability test for the data, correction was made with a simple regression equation to determine the equation correction factor to be utilized by examining the coefficient of determination (R 2 ) value for each equation correction factor.

F. Regression Analysis
The correction factors for TRMM satellite data and evaporation data are used to determine the magnitude of the x and y parameters as the correction factors for the line equation between observation data and satellite data.The shape of the utilized regression equation was determined by looking at the pattern of the measured rainfall data and evaporation data series with the TRMM and GLDAS, as well as the highest generated coefficient of determination (R 2 ) for each equation.The correction factor was then utilized to correct TRMM and GLDAS data by entering TRMM data and GLDAS data into the resulting correction equations to select corrected TRMM and GLDAS data.
Based on the evaporation regression analysis of measurement data and GLDAS, several possible regression equations can be obtained between the relationship between the two data.The equation results obtained via scatterplot as follows: a. Exponential Equation : Y = 1.69 e 0.1503X R 2 = 0.56 b.Linear Equation : Y = 0.89X R 2 = 0.15 c.Logarithmic Equation : Y = 1.49ln(X) + 1.07 R 2 = 0.55 d.Polynomial Equation : Y = -0.099X 2 + 1.17X R 2 = 0.51 e.Power Equation : Y = 1.52 X 0.512 R 2 = 0.54 Selection of the equation used to correct evaporation data by referring to the R 2 value produced in each regression equation.The highest determined value, or R 2 will be chosen as the equation for correcting the evaporation data.Based on the results obtained for the correction of monthly period evaporation data, the equation chosen is the exponential equation with an R 2 value of 0.56 because it is the highest R 2 value among the other equations.Based on the result presented in Table 3, it can be observed that the ten-day period yielded the highest value for the Root Mean Square Error (RMSE) for each dataset.The value for the NSE method shows "Satisfied" results for all periods of evaporation data and ten daily periods for rainfall data.Table 3 shows ten days is the best duration for using rainfall data.In contrast, monthly data is suitable for evaporation, so the corrected data can be used for further analysis.

Water Balance Analysis
Water balance analysis calculates the relationship between total water input and total water output in a watershed or reservoir where there is river flow discharge, rainfall, evaporation, etc. Water balance analysis is a way to get answers to problems, such as in the quantitative evaluation of regional water resources, changes due to interventions of human activities, and a comparison of the potential of water resources of an area with other regions.The calculation of water balance analysis in this research uses 2 data, such as measurement data (rainfall and evaporation) and satellite data (TRMM and GLDAS).What will be compared is the results or value of changes in water reserves between analysis of water balance using measurement data vs satellite data.Satellite data can be employed as a substitute dataset for calculating water availability in a system to determine the differences in the value of changes in water reserves.According to the findings in Figure 2, the outcomes of the water balance analysis for the monthly period using ground measurement data and satellite data have a difference in values that is not too far or can be said to be close.The results show that satellite data's water balance analysis is 17% lower than ground measurement data.The various elements contributing to the disparity are outlined below: (1) measurement accuracy is crucial when utilizing remote sensing technologies to measure rainfall and evaporation through satellite data.Despite the rapid development of this technology, the potential for measuring inaccuracies remains.Differences in measurement results might arise due to factors such as the inherent inaccuracies of data processing algorithms or the defects present in remote sensing instruments; (2) The spatial resolution of satellite data is comparatively poorer than that of ground measurement data.Satellite data encompasses a broader spatial extent inside a singular pixel, but ground measurement data offers more intricate insights into localized conditions at a specific geographical point.The disparity in resolution may lead to variations in the outcomes of water balance analysis; (3) Data Unavailability: There may be a lack of ground measurement data for specific places or periods.In this scenario, utilizing satellite data as a substitute may prove more advantageous than without data.Nevertheless, disparities in these data sources may result in variations in the outcomes of water balance analyses; (4) Regional Attributes: Each geographical area possesses distinct qualities that can impact the disparities observed between satellite-derived data and ground-based measurement data.Various factors such as terrain, vegetation, and land use conditions have the potential to exert an influence on the patterns of rainfall and evaporation within a given geographical region.Hence, variations in these specific regional attributes may give rise to disparities in the outcomes of water balance analyses; (5) How data is processed plays a crucial role in the disparities between satellite data and ground measurement data when analyzing the water balance.It should be noted that even slight variations in the results may arise due to disparities in the approaches or assumptions employed during the analysis.It can stated that the satellite data used in the form of rainfall and evaporation data can be used as an alternative in using water balance analysis calculations.

Conclusions
The analysis results show that the suitability test of TRMM rainfall corrected data to ground measurement rainfall data shows satisfactory results.The similarity or coherence in the patterns observed in both data types is indicated by the similar patterns observed in ground measurement data and satellite data.Although discrepancies exist in the numerical values between ground measurement data and satellite data, comparable patterns suggest a shared correlation or association between the variables under observation.Within the analytical framework, equivalent trends in field measurement data and satellite data may be related to the observation of consistent rainfall or evaporation patterns across both data types.For example, suppose that the ground measurement data shows a consistent pattern of intense rainfall in a specific rainy season.In this particular case, it is worth mentioning that satellite data can demonstrate a corresponding pattern characterized by higher levels of precipitation during that particular timeframe.The test results for the suitability of GLDAS corrected data to ground measurement evaporation data also show satisfactory results.The results of the water balance analysis for all periods show a series pattern that is almost the same or close to the ground measurement data and satellite data.However, satellite data's water balance analysis value is generally 17% lower than the measurement data.Measurement accuracy, spatial resolution, data unavailability, local characteristics, and data processing methods are some factors that may give rise to disparities between satellite data and ground measurement data in water balance analysis.When utilizing satellite data as a substitute for water balance analysis, it is essential to consider these factors.So, it is concluded that satellite data can be used as an alternative in calculating water balance analysis.

Figure 2 .
Figure 2. Water balance assessment using ground measurement data and satellite data

Table 1 .
Summary of the influence area coefficient of the rainfall station

Table 2 .
This study's data suitability test stage was divided into the initial data suitability test phase and the subsequent data suitability test phase, which involved corrections.The uncorrected suitability test compares measured rainfall or evaporation data and TRMM rainfall data or satellite evaporation for 18 years.Meanwhile, the corrected data suitability test compares satellite data that has undergone correction with measurement data.The study involved the application of various regression equations, namely exponential regression, linear regression, logarithmic regression, results if there is data that has a value of "zero".The methods employed for assessing the suitability of the data include the RMSE, NSE, CC, and RE Test.The data range used is 18 years (2002-2019).Summary of Uncorrected Data Conformity Test Results polynomial regression, and rank regression, to correct the satellite data.This study aims to enhance the coherence between hydrological responses by optimizing parameter values.The uncorrected data suitability test is calculated using ten days and monthly periods.In each of these periods, data is used with rainfall values (data with "zero" rainfall values are not used) in post-station and TRMM rainfall data.It is because some of the regression equations used, such as logarithmic and rank, cannot bring up 6 the equation

Table 3 .
Summary of Corrected Data Conformity Test Results