Refining Rainfall Derived from Satellite Radar for Estimating Inflows at Lam Pao Dam, Thailand

Abstract

This project aimed to evaluate the use of meteorological satellite-derived rainfall data to estimate water inflows to dams. In this study, the Lam Pao Dam in the Chi Basin, Thailand, was used as a case study. Rainfall data were obtained using the PERSIANN technique. To improve accuracy, satellite-derived rainfall estimates were adjusted using ground-based rainfall measurements from stations located near and within the catchment area, applying the 1-DVAR method. The Kriging method was employed to estimate the spatial distribution of rainfall over the catchment area. This approach resulted in a Probability of Detection (POD) of 0.92 and a Threat Score (TS) of 0.72 for rainfall estimates in the Chi Basin. Rainfall data from the Weather Research and Forecasting (WRF) numerical models were used as inputs for the HEC-HMS model to simulate water inflows into the dam. To refine rainfall estimates, various microphysics schemes were tested, including WSM3, WSM5, WSM6, Thompson, and Thompson Aerosol-Aware. Among these, the Thomson Aerosol-Aware scheme demonstrated the highest accuracy, achieving an average POD of 0.96, indicating highly reliable rainfall predictions for the Lam Pao Dam catchment. The findings underscore the potential benefits of using satellite-derived meteorological data for rainfall estimation, particularly where installing and maintaining ground-based measurement stations is difficult, e.g., forests/mountainous areas. This research contributes to a better understanding of satellite-derived rainfall patterns and their influence on catchment hydrology for enhanced water resource analysis.

1. Introduction

Hydrological modeling has been widely applied in Thailand using tools such as SWAT and HEC-HMS. These models rely heavily on accurate data inputs to understand catchment hydrodynamics and to develop effective strategies for water resource management, flood mitigation, drainage planning, water supply, and irrigation systems. This research focuses on simulating and predicting inflows to dams for the purposes of improved water resource investigations. The Lam Pao Dam, a key agricultural reservoir in northeastern Thailand, plays a critical role in supporting irrigation, particularly during dry seasons. Constructed in 1968, the dam has an estimated storage capacity of 1980 million cubic meters. It supplies water to approximately 54,080 hectares during the rainy season and 4240 hectares during the dry season [1].

The study incorporates radar-derived rainfall data to a HEC-HMS hydrological model of the Lam Pao Basin to evaluate its effectiveness for water resource purposes. Variability in rainfall significantly influences river levels and the overall water balance. Additionally, changes in runoff behavior have been exacerbated by deforestation within the catchment area. Research conducted in the Chi River Basin from 1951 to 2003 reported a 20% reduction in forest cover over a span of 52 years [2]. To assess water availability across different sub-regions of the Chi Basin—such as the Lumsapung Basin and Mae Sruai Dam—hydrological models employing net water balance methods have been utilized. These models typically account for inflows (Qin), precipitation (Pt), evaporation (Et), and outflows (Qout) to estimate variations in reservoir storage (ΔS). Additional studies [3,4,5] have sought to improve hydrological modeling techniques to enhance water resource management practices. The integration of hydrological and meteorological data has proven effective in improving streamflow estimation across numerous river basins and catchments in Thailand [6,7,8].

Radar systems operate by emitting electromagnetic waves—similar to those used in mobile phones and wireless networks—which are transmitted in brief pulses. These pulses reflect off objects in their path, with a portion of the signal returning to the radar. When the pulses encounter precipitation, part of the energy is scattered back, allowing the radar to detect and estimate rainfall intensity [9]. While rain gauges offer precise measurements of rainfall at specific locations, they are limited in their ability to capture spatial variability. In contrast, radar provides broad spatial coverage but measures precipitation within the atmosphere rather than the exact amount that reaches the ground [9].

To overcome the limitations of both methods, rainfall accumulation data from radar are typically presented as images that represent estimated precipitation totals over defined time intervals—such as every five minutes or hourly. These images are created by combining the high spatial resolution of radar with the point-specific accuracy of rain gauge data. Rain gauges are used to calibrate radar outputs, improving rainfall estimation in areas between gauge locations [10]. This combined approach yields more accurate precipitation estimates than using either radar or gauges independently. However, the quality of these accumulation images depends on accurate radar reflectivity, reliable rain gauge measurements, and effective correction of errors and biases [9,10].

This study incorporates satellite-based rainfall estimates using the PERSIANN (Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks) approach. Specifically, infrared-based PERSIANN rainfall data were applied to the Chi River Basin and integrated with the HEC-HMS model to evaluate hydrological conditions at the Lam Pao Dam. While ground-based rain gauge data were utilized, their spatial coverage is limited due to the Chi Basin’s diverse and complex terrain. Comparing gridded satellite precipitation data with point-based gauge observations within the HEC-HMS modeling framework provides valuable insights into the influence of data source variability on hydrological model outputs.

In addition to radar and rain gauge data, rainfall information can also be derived from satellite observations and outputs from numerical weather prediction (NWP) models. One such model is the Weather Research and Forecasting (WRF) model—a state-of-the-art mesoscale NWP system designed for both research and operational flow estimation. WRF supports meteorological simulations across a wide range of spatial scales, from tens of meters to several thousand kilometers. It generates precipitation accumulation outputs based on real-time radar observations, which are calibrated with rain gauge data to reduce uncertainties. The WRF modeling framework has been applied in various regions worldwide [11,12,13,14]. In this study, WRF-derived data are integrated with the HEC-HMS hydrological model to simulate streamflow in the Lam Pao Basin.

This study aims to evaluate the effectiveness of integrating rainfall data from the WRF model and the PERSIANN algorithm with the HEC-HMS model for water resource and irrigation purposes. The primary objective is to simulate streamflow within the Lam Pao Basin using this combined modeling approach. The HEC-HMS model was employed to estimate the Lam Pao Dam’s storage levels over annual periods, facilitating an analysis of long-term hydrological trends. The simulation period of up to one year was selected to assess the model’s performance and consistency, with the potential for extension to longer-term applications in future research. Overall, this approach supports the evaluation of the model’s reliability and its potential use as a decision-support tool for managing water resources at Lam Pao Dam.

2. Data and Methods

This study is organized into three core components. First, rainfall estimates across the catchment were obtained using the PERSIANN satellite-based technique. To improve the precision of these estimates, the 1-DVAR method was applied, incorporating observational data from local rain gauge stations. Spatial analysis of rainfall over the reservoir area was further conducted using the kriging interpolation method.

In the second phase, rainfall estimates for the catchment were produced using the WRF atmospheric model, employing five different microphysics parameterization schemes. The performance of each scheme was assessed through statistical evaluation, using observed rainfall data from ground stations for validation.

Lastly, the satellite-derived PERSIANN rainfall data were fed into the HEC-HMS hydrological model to simulate inflow into Lam Pao Dam. These simulated values were compared with observed discharge data to assess accuracy. In parallel, rainfall estimates from the WRF model were also used as input for hydrological simulations, and the outputs were evaluated using statistical validation techniques.

2.1. Study Area

The Chi Basin, where PERSIANN satellite rainfall data were collected and WRF estimated rainfall was estimated, is depicted in Figure 1. Located in northeastern Thailand, the basin spans an area of 49,131.92 km² and covers 14 provinces, including Chaiyaphum, Khon Kaen, Nong Bua Lamphu, Udon Thani, Maha Sarakham, Nakhon Ratchasima, Loei, Phetchabun, Kalasin, Roi Et, Yasothon, Ubon Ratchathani, Si Sa Ket, and Mukdahan.

The Lam Pao Dam, located in northern Kalasin Province, was constructed between 1963 and 1968. It has a storage capacity of 1430 million cubic meters and plays a vital role in flood prevention and agricultural water supply. The dam’s catchment area extends into a large part of Udon Thani Province. Downstream, the Lam Pao River flows through Kalasin Province before joining the Chi River at the boundary between Kalasin and Roi Et Provinces.

Water usage from the dam is distributed across various sectors: irrigation (87.1%), aquaculture (9.5%), domestic consumption (1.8%), industry (1.6%), and livestock (0.1%). Water shortages typically occur during the dry season, and demand during drought conditions—particularly for aquaculture and irrigation—places significant pressure on water allocation.

The irrigated area includes approximately 50,000 hectares during the wet season and around half that in the dry season. Rice is the predominant crop, occupying about 50% of the farmland, followed by cassava, groundnuts, sugarcane, and kenaf.

Aquaculture is mainly practiced in major water bodies, including the Lam Pao River and Lam Pao Reservoir. In 1998, total aquaculture production reached nearly 251 tons. Additionally, commercial-scale shrimp farming within the Lam Pao Irrigation Project in Kalasin Province produced about 700 tons in the same year.

2.2. Rainfall Data

The rainfall data used in this study were supplied by the Thailand Deptartment of Meteorology. Rainfall estimation in this study made use of the PERSIANN dataset, which is based on satellite remote sensing data. This method relies on infrared (IR) brightness temperature observations from NCEP, collected at 15 min intervals between 2009 and 2019 [15]. In addition, Tropical Rainfall Measuring Mission (TRMM) Multi-Satellite Precipitation Analysis (TMPA) was employed, which combines data from several satellites along with ground-based rain gauge observations [16,17]. The TRMM dataset covers the latitudinal range of 50° S to 50° N and spans the years 2009 to 2018. For this study, the TRMM 3B42 Version 7 (TRMM-3B42V7) algorithm was used, offering high-resolution data at 0.25° × 0.25° spatial and 6 h temporal scales for detailed rainfall analysis.

Statistical Verification and Evaluation Methods to Compare Satellite with Rain Gauge Stations

The 1DVAR technique was applied to generate an optimal estimate of the vertical atmospheric state profile, represented as x, by integrating observational data (y_o) with a background state estimate (x_b). This method takes into account the associated error covariance matrices for both the observational inputs and the background information. The 1DVAR approach is mathematically framed as the minimization of the following function [1]:

$$J\left(x\right)={\displaystyle \frac{1}{2}}{\left(x-x_b\right)}^T{B}^{-1}\left(x-x_b\right)+{\displaystyle \frac{1}{2}}{\left(y_0-H⌊x⌋\right)}^T{O}^{-1}\left(y_0-H⌊x⌋\right)$$

(1)

where y_o = elevation of observation (m); T = temperature in K; O = observation error covariance matrix (m, m); B = background error covariance matrix (n, n); and H(x) = converts the n dimensional state vector defined on the 1D-Var levels and surface into a corresponding profile of refractivity given on the m refractivity levels.

To evaluate the model’s performance, the coefficient of determination (R²) was utilized. The correlation coefficient (CC) was also used to quantify the strength and direction of the relationship between two variables—typically either two observed data series or components of a multivariate dataset. The CC ranges from −1.0 to 1.0, with values approaching 1 indicating a strong positive correlation, meaning the estimated values closely align with the observed trends [18].

$$CC={\displaystyle \frac{{\sum }_{i=1}^N\left(Y_i-\overline{Y}\right)\left(O_i-\overline{O}\right)}{\sqrt{{\sum }_{i=1}^N(Y_i-{\left.\overline{Y}\right)}^2{\sum }_{i=1}^N\left(O_i-{\left.\overline{O}\right)}^2\right.}}}$$

(2)

where Y_i = gauge rainfall measurement, $\overline{Y}$ = average gauge rainfall measurement, O_i = estimated rainfall, and $\overline{O}$ = estimated rainfall average.

Table 1. Binary contingency table for whether there is rain.

Estimate	Observed		Total
	Yes	No
Yes	a (hit)	b (false alarm)	a + b
No	c (miss)	d (correct rejection)	c + d
Total	a + c	b + d	a + b + c + d = n

illustrates how the relationship between satellite data (estimate) and rain gauge station data (observed) is classified. In this study, “Yes” refers to rainfall amounts greater than or equal to 0.1 mm, while “No” refers to rainfall amounts less than 0.1 mm. When comparing rainfall quantities from the rain gauge station (observed) and those estimated from satellite data (estimate), the variables in Table 1 are classified based on the following criteria:

• Hit = the rainfall quantity of the satellite (estimate) and rain gauge station (observed) ≥ 0.1;
• False alarm = the rainfall quantity of the satellite (estimate) ≥ 0.1 and rain gauge station (observed) < 0.1;
• Miss = the rainfall quantity of the satellite (estimate) < 0.1 and the rate of the rain gauge station (observed) > 0.1;
• Correct rejection = the satellite (estimate) and rain gauge station (observed) < 0.1.

The scores in Table 1 are divided into four indices, as outlined in Equations (3)–(6). The Probability of Detection (POD) is the ratio of hits (a) to the total number of observed events (a + c). The False Alarm Ratio (FAR) represents the number of false alarms (b) compared to the total number of “yes” estimates (a + b). The Probability of False Detection (POFD) is the ratio of false alarms (b) to the total number of “no” estimates (b + d). The Threat Score (TS) is the ratio of hits (a) to the total number of observed “yes” (rain) events plus false alarms (a + b + c).

1.

Probability of Detection (POD) [19]

$$POD={\displaystyle \frac{a}{a+c}}$$

(3)

2.

False Alarm Ratio (FAR) [20]

$$FAR={\displaystyle \frac{b}{a+b}}$$

(4)

3.

Probability of False Detection (POFD) [19]

$$POFD={\displaystyle \frac{b}{b+d}}$$

(5)

4.

Threat Score (TS) [21]

$$TS={\displaystyle \frac{b}{a+b+c}}$$

(6)

The perfect estimate has a POD equal to 1. Similarly, a perfect estimate occurs where the FAR and POFD equal to 0. The TS gives the rate of all well-predicted locations with respect to the reference, where the reference or the simulation are above the threshold. The POD, FAR, and TS generally need to be examined together with MAE and RMSE goodness of fit criteria.

2.3. WRF Model Rainfall Estimates

Rainfall across the Chi Basin in northeastern Thailand was monitored using meteorological satellite data. The Thai Meteorological Department (TMD) utilized the WRF model, driven by the National Center for Environmental Prediction (NCEP) Global model, to generate rainfall estimates. For calibration purposes, rainfall data from 10 nearby stations (listed in

Table 2. Rainfall station used in this study.

Code	Station	Latitude	Longitude
48354	Udon Thani	17.383	102.800
48356	Sakon Nakhon	17.150	104.133
48355	Sakon Nakhon (1)	17.125	104.061
48360	Nong Bua Lumphu	17.233	102.429
48381	Khon Kaen	16.461	102.790
48384	Tha Pra (1)	16.333	102.817
48382	Kosum Phisai (2)	16.247	103.068
48390	Kamalasai (2)	16.333	103.588
48405	Roi Et	16.050	103.683
48404	Roi Et (1)	16.073	103.608

) were collected over a 10-year span from 2008 to 2017. A statistical summary of the data is given in Table S3. The WRF-simulated rainfall was then compared to observed values from these rain gauge stations and adjusted accordingly to improve model accuracy.

Among the various factors influencing numerical weather prediction models, cloud microphysics plays a critical role as a major source of uncertainty. Five different WRF microphysics schemes used in this study are listed in

Table 3. Microphysics (MP) schemes used in WRF model.

MP Scheme	Scheme	Reference
3	WSM3	[22,23]
4	WSM5	[23]
6	WSM6	[24]
8	Thompson	[25]
28	Thompson Aerosol-Aware	[25,26,27]

.

2.4. HEC-HMS Model of Lam Pao Dam

The Lam Pao Dam, located in the Chi Basin in northern Kalasin Province, was constructed between 1963 and 1968. Its primary function is agricultural irrigation. The Lam Pao Reservoir effectively divides the catchment into two sections (see Figure S1). Most of the catchment area upstream of the dam is mountainous. To the north-northeast, the Phu Phan Mountain Range extends, with part of it designated as a national park, and beyond that lies Sakhon Nakhon Province. To the northwest of the catchment is Udon Thani. The river upstream of the dam originates from Nong Han Kumphawapi Lake in Udon Thani Province, flowing through Kalasin Province before converging with the Chi River at the boundary between Kalasin and Roi Et Provinces.

For this study, the HEC-HMS model was employed to simulate specific rainfall events and assess inflows into the Lam Pao Dam. Flow data measured at the dam were used to validate the model’s output. These flow data were supplied by the Thailand Department of Irrigation. Information regarding the model setup, description, and calibration (see Figure S2) can be found in the Supplementary Information (SI).

3. Results and Discussion

3.1. Comparison and Validation of PERSIANNs Satellite Data

This section presents a comparison between rainfall data from ground-based rain gauge stations and satellite-derived meteorological data from PERSIANN, along with the application of bias correction using the 1-DVAR method. The PERSIANN data from 2009 to 2018, covering the rainy season months of June, July, August, and September, were selected for analysis. The correlation coefficient (R²) values, both pre- and post-bias correction, for the ten-year period are shown in

Table 4. Correlation coefficient during June–September (R²) of rainfall gauge data compared with PERSIANN satellite.

Years	June		July		August		September
	R2 +	R2 *	R2 +	R2 *	R2 +	R2 *	R2 +	R2 *
2009	0.04	0.79	0.23	0.72	0.12	0.69	0.10	0.60
2010	0.35	0.19	0.19	0.49	0.01	0.27	0.24	0.05
2011	0.73	0.94	0.40	0.73	0.33	0.83	0.01	0.49
2012	0.34	0.87	0.56	0.78	0.44	0.76	0.01	0.25
2013	0.21	0.52	0.41	0.78	0.27	0.70	0.21	0.58
2014	0.48	0.69	0.46	0.70	0.00	0.82	0.65	0.90
2015	0.02	0.78	0.54	0.83	0.18	0.58	0.05	0.48
2016	0.01	0.37	0.30	0.66	0.44	0.81	0.63	0.83
2017	0.41	0.82	0.58	0.80	0.38	0.73	0.18	0.63
2018	0.28	0.86	0.52	0.81	0.63	0.88	0.04	0.32
Avg	0.28	0.68	0.42	0.73	0.28	0.71	0.21	0.51

⁺ before bias correction; * after bias correction.

.

Satellite imagery, including data from rain gauge stations, PERSIANN satellite observations, and the images after data assimilation with the 1-DVAR technique, are presented in Figure S3. In July 2018, rainfall recorded by the rain gauge stations ranged from 200 to 340 mm, with the heaviest precipitation occurring in the central, eastern, and northeastern parts of the Chi Basin. The PERSIANN data estimated rainfall between 240 and 400 mm, covering areas from the north to the east and northeast of the basin. The correlation coefficient with the rain gauge data was R² = 0.54. Following data assimilation with 1-DVAR, the satellite data showed improved agreement with the rain gauge data, resulting in a higher R² of 0.83. In August 2018, rain gauge data varied from 200 to 300 mm, while the satellite data ranged from 220 to 340 mm, yielding an initial R² of 0.33. After applying 1-DVAR assimilation, the satellite rainfall data aligned much more closely with the ground measurements, improving the R² to 0.83. For September 2018 (see Figure S3), the rain gauge data ranged from 200 to 420 mm, whereas the satellite estimates ranged from 200 to 320 mm, with an R² of 0.66. After 1-DVAR assimilation, the satellite data adjusted to a range of 220 to 340 mm, resulting in an R² of 0.90. September showed the most significant improvement, with the best correlation between the rain gauge data and the post-assimilation satellite estimates.

In July 2018, the Probability of Detection (POD) reached its highest value of 0.924, while the False Alarm Ratio (FAR) was 0.15, as shown in Figure S3. In August 2018, the POD remained high at 0.892, which also corresponds to the data presented in Figure S3. For September 2018, the POD score was 0.75, and the FAR was 0.37, demonstrating the best match between the rain gauge data and the post-assimilation satellite data using the 1-DVAR technique, as illustrated in Figure S3.

3.2. WRF-Derived Rainfall

The results from the Numerical Weather Prediction (NWP) approach using the WRF model are presented. The WRF rainfall estimates for the period from 2013 to 2018 were utilized to predict rainfall in the Lam Pao Dam catchment.

An analysis of the data in

Table 5. The event of hits, false alarms, missing events, correct negatives, PODs, FARs, POFDs, and TSs for June 2013, June 2014, and June 2018. Values with the highest POD and TS and lowest FAR and POFD are highlighted.

June 2013
Scheme	3	4	6	8	28	Avg
Hits	275	283	300	304	309
False Alarm	75	73	60	48	50
Missing Event	26	14	21	13	22
Correct negative	44	50	39	55	39
POD	0.914	0.953	0.935	0.959	0.934	0.939
FAR	0.214	0.205	0.167	0.136	0.139	0.172
POFD	0.630	0.593	0.606	0.466	0.562	0.572
TS	0.731	0.765	0.787	0.833	0.811	0.786
June 2014
Scheme	3	4	6	8	28	Avg
Hits	199	249	253	256	258
False Alarm	63	104	104	106	103
Missing Event	56	34	30	27	25
Correct negative	32	19	19	17	20
POD	0.780	0.880	0.894	0.905	0.912	0.874
FAR	0.240	0.295	0.291	0.293	0.285	0.281
POFD	0.663	0.846	0.846	0.862	0.837	0.811
TS	0.626	0.643	0.654	0.658	0.668	0.650
June 2018
Scheme	3	4	6	8	28	Avg
Hits	194	245	241	238	243
False Alarm	79	110	112	104	105
Missing Event	36	21	22	24	24
Correct negative	41	44	45	54	48
POD	0.843	0.921	0.916	0.908	0.910	0.900
FAR	0.298	0.310	0.317	0.304	0.304	0.304
POFD	0.658	0.714	0.713	0.658	0.686	0.686
TS	0.628	0.652	0.643	0.653	0.653	0.645

for June 2013 indicates that Scheme 8 outperformed the other four schemes, showing the best fit overall. Scheme 8 achieved the highest values for Probability of Detection (POD) and Threat Score (TS), and the lowest values for False Alarm Ratio (FAR) and Probability of False Detection (POFD), as shown in Table 5, although Scheme 28 performed similarly, though slightly less effectively.

In June 2014, Scheme 28 achieved the highest POD value of 0.912 and the highest TS value of 0.668. Meanwhile, Scheme 3 had the lowest FAR value of 0.285 and the lowest POFD value of 0.837.

For June 2018, results varied across the different schemes. Scheme 4 had the highest POD value, while Scheme 3 exhibited the lowest FAR. Schemes 8 and 28 showed slightly higher FAR values. Both Scheme 3 and Scheme 20 had the lowest POFD values, while Scheme 28 attained the highest TS value.

Overall, Scheme 28 demonstrated the most robust performance, as indicated by its high POD, FAR, POFD, and TS values, highlighted in Table 5. Schemes 3 and 8 also performed well but slightly behind Scheme 28 in terms of effectiveness. Figure 2 illustrates this by showing the spatial distribution of rainfall derived from rain gauge data (Figure 2a) and the mapping results from kriging spatial methods for Schemes 3, 8, and 28 (Figure 2b–d). Scheme 28 (Figure 2d) most closely mirrored the spatial map from the rain gauge stations, especially in terms of rainfall intensity variation from west to east, ranging from 200 to 320 mm.

3.3. Hydrology Modeling Coupling with PERSIANN Satellite and WRF Data

The PERSIANN satellite-derived rainfall data were integrated into the HEC-HMS model, and the resulting simulations were compared with observed inflow data from the Lam Pao Dam (Figure 3). For comparison, a separate model using rain gauge data was also developed and evaluated.

In 2010 (Figure 3a,b), the model utilizing satellite-derived rainfall closely aligned with the results from the rain gauge-based model, indicating good agreement between the two data sources. Table S4 shows that the inflow to Lam Pao Dam was within 10%. However, in 2011 (Figure 3c,d), a noticeable discrepancy emerged between the results obtained from satellite data and those based on rain gauge observations. Both models, however, failed to fully replicate the observed inflows, particularly during September, suggesting limitations in capturing peak flow events. The estimated rainfall over the Lam Pao Dam catchment area in 2010 was slightly underestimated compared to the measured rainfall, highlighting the need for further refinement (Table S4).

These findings underscore the potential for improving the simulation of catchment flows into the Lam Pao Dam. A more detailed assessment of satellite rainfall data accuracy is required, along with appropriate corrections to better align with ground-based measurements. Moreover, the acquisition of higher-resolution Digital Terrain Model (DTM) data is critical for enhancing the spatial accuracy of hydrological modeling. By employing advanced HEC-HMS capabilities and integrating both radar and satellite-derived rainfall data, the overall precision and reliability of the hydrological simulations can be significantly improved. A sensitivity analysis on input parameters (satellite and WRF) could be carried out to estimate the uncertainty in rainfall inputs and assess how this affects HEC-HMS outputs and, by extension, their application in water management planning.

WRF data for the years 2013 and 2014 were available, and the comparison, shown in Figure S4, focuses specifically on the months of June to September, where WRF data were available. No comparison with the HEC-HMS based on rain gauge data was possible due to lack of data in those years. The limited comparison shows that the rain estimates in the water catchment area of the Lam Pao Dam provided higher estimates than observed data.

This comparison provides insight into the potential of satellite data to serve as a viable alternative to ground-based rain gauge measurements. A robust conclusion on the possibility of replacing ground-based gauges with satellite data would require long-term data across multiple locations. This is necessary to capture the diversity of rainfall mechanisms that occur over time and across different geographical settings. Reaching such a conclusion would require the accumulation of evidence from many studies conducted over time and in varied regions [28,29]. Accordingly, our study is a contribution toward that broader objective.

4. Conclusions

In the initial phase of the study, the focus was on validating PERSIANN rainfall estimates within the Chi Basin. Validation was carried out using bias correction threshold scores and the 1D-VAR data assimilation technique. During June 2011, PERSIANN satellite-derived estimates indicated higher intensities than rain gauge data and a correlation coefficient (R²) of 0.73. The 1D-VAR assimilation method resulted in a significantly improved fit with an R² of 0.94, closely aligning with the ground-based measurements. Rainfall validation POD, TS, etc., demonstrated a close agreement between the corrected PERSIANN estimates and observed rainfall data. The PERSIANN-derived rainfall data were subsequently integrated into the HEC-HMS model and gave adequate simulated estimates of water inflows into the Lam Pao Dam.

Rainfall estimation was carried out using the WRF atmospheric model, which employed five different microphysics schemes. The estimated rainfall outputs were statistically validated by comparing them with observed rainfall data from measurement stations across the Chi Basin. The performance of the microphysics schemes, reflected by an average Probability of Detection (POD) value of 0.90, indicated that the WRF model provided consistent and reasonably accurate rainfall predictions for the Lam Pao Dam catchment area when benchmarked against ground-based observations. These validated estimates, used with the HEC-HMS hydrological model, were able to simulate water inflow into the Lam Pao Dam, albeit in a limited comparison.

The results of this study provide insight into the potential of PERSIANN satellite-based rainfall estimates, particularly in regions where the installation and maintenance of ground-based measurement stations are impractical—such as forested areas or rugged terrain. Furthermore, the integration of satellite-derived rainfall data with other observational and modeling resources offers significant potential to enhance the ability to provide flow estimates for water resource studies. This approach can help mitigate economic losses, improve operational planning, and ultimately strengthen community resilience.