Rainfall/runoff hydrological modeling using satellite precipitation information

Appropriate water-related management tools are required for suitable decision-making in a region or basin. The hydrological modeling of the rainfall/runoff process, incorporating a geographic information system, is a very useful tool for comprehensive hydrological basin management. These models aim at representing an actual hydrological system and thereby forecast hydrographs and basin outlet flows.

The construction of a hydrological model requires general as well as very specific precipitation information, such as geographic, physiographic, and hydrometric information. In Mexico, as in many other countries, the availability of sufficient, reliable, official meteorological information is a challenge owing to the difficulty of finding constant and updated records. In some regions, an adequate amount of information is available, but it is of poor quality; while in other areas, few weather stations are established but have only been getting information for a short period of time. On the other hand, with the tools of the 4.0 industrial revolution, satellite products can be provided with information that is a good alternative for these regions and, in areas without a network of weather stations, the lacking information might become available. Thus, satellite precipitation information is available to almost everyone and has a consistent spatial and temporal resolution to be considered in a hydrological model (Balzcázar et al. 2019).

However, the quality and usefulness of satellite information must still be assessed. For these purposes, a comparison exercise between information sources may prove extremely useful and necessary. An alternative to obtaining up-to-date satellite precipitation time series is the database Climate Hazards Group InfraRed Precipitation with Station data (CHIRPS) from the CLIMATESERV project included in the SERVIR GLOBAL product catalog, available at https://climateserv.servirglobal.net/. This base is a quasi-global rainfall data set, since it has coverage in all areas between latitudes of 50° S to 50° N. It has over 35 years of continuous records (1981-present), global, spatial, and temporal consisting of daily and monthly records. CHIRPS combines satellite data sources and terrestrial observations, that is, high-resolution (0.05°) satellite images and on-site station data to create precipitation time series in a mesh (Funk et al. 2015).

The objective of this work is to present the comparative analysis and the results obtained from the hydrological simulation of the daily flows of the Conchos River at the hydrometric stations: La Boquilla, Fco. I. Madero and Ojinaga, located in the upper, middle and lower parts of the basin, respectively. The Hydrologic Modeling System program of the Hydrological Engineering Center (HEC-HMS) of the US Army Corps of Engineers was used for this work. For the generation of the precipitation time series, the official information of the Rapid Climate Information Extractor (ERIC III) of the National Water Commission of Mexico (CONAGUA) and the estimated information from the CHIRPS database were used.

The chosen model simulation year was 1981, since full runoff information from the hydrometric stations was available for that year. In addition, that year presented a runoff anomaly of interest in the month of October (Hurricane Norma category 2 as per the Saffir-Simpson scale (CENAPRED 2014)). Therefore, that year was selected to calibrate the model. For the validation of the simulation model, the year 1991 was considered since a tropical storm was recorded. Additionally, the 1992–2001 period was also considered, in which a severe drought occurred in the region.

The Rio Conchos basin is located in northern Mexico and is part of the hydrological region No. 24 (HR-24) in the Mexican classification. It is the most important tributary area of the hydrological region, which reaches the Rio Grande in Ojinaga, flowing towards east/northwest until the Gulf of Mexico. The Conchos River is born in the eastern part of the Sierra Madre Occidental, which is called Sierra Tarahumara, at more than 2,600 meters above sea level. The main tributaries of the Conchos River are the Nonoava, Balleza, Florido, Parral, San Pedro, Satevó, Santa Isabel, Chuvíscar, and Sacramento rivers. Due to its economic and social impact in the region, the most important aquifers that interrelate with the main river or its tributaries are the Jiménez–Camargo, Meoqui-Delicias, Alto San Pedro River, Chihuahua-Sacramento, and Bajo Conchos River aquifers. The largest capacity dams within the basin are La Boquilla (Toronto Lake), Fco. I Madero (Las Vírgenes), Luis L. León (El Granero), and Mexican Federalism (San Gabriel) (Kelly 2001; CONAGUA 2011, 2019; Ingol-Blanco & McKinney 2011).

The basin is mostly located within the state of Chihuahua (93.1%) and a small percentage in the state of Durango (6.9%), representing almost 30% of the HR-24. It has a drainage area of 66,682 km², and it comprises the following basins: Rio Florido, Rio Conchos–Presa de la Colina, Rio San Pedro, Rio Conchos–Presa el Granero, Rio Conchos–Ojinaga. According to the official division of the National Water Commission (CONAGUA 2011), these basins are divided into 11 sub-basins (Figure 1). For the construction of the hydrological model, each of these divisions was considered, taking into account hydrometric and weather stations as well as the hydraulic infrastructure with sufficient information for its calibration and validation. The main river of the Rio Conchos basin is approximately 750 km long.

The Rio Conchos basin is characterized by its desertic and dry climate, particularly in the middle and lower basins. Warm and semi-humid climates predominate in the upper part, with seasonal rains during the year. The highest temperatures are recorded in the summer (from June to August), and the lowest temperatures are recorded in winter (from November to February). Throughout the basin, the average annual temperature is 17.5 °C. However, the maximum annual temperatures are 35 °C, 27 °C, and up to 20 °C in the lower, middle, and upper basins, respectively. Regarding minimum temperatures, they range from 15 °C in the lower parts to 5 °C in the other regions (IMTA 2009). The annual average rainfall is 700, 500, and 260 mm in the upper, middle, and lower basins, respectively (CONAGUA 2021b).

The geographical information of the study basin was largely obtained in 1:50,000 and 1:250,000 scales, which are the most common scales supported by the official agencies in Mexico. This information is available as .shp (Shapefile) and raster files for Digital Elevation Models from the Lambert Conformal Conic projection coordinate system (CCL_ITRF_1992), which uses data from the 1992 International Terrestrial Reference Frame (D_ITRF_1992). Some of this information had to be reprojected owing to its lack of an established coordinate system. For this study, this information was processed through the ArcGIS program (version 10.4). The most relevant information found, that supported the hydrological model was: Hydrological regions, basins, sub-basins, hydrometric stations, weather stations, main rivers, bodies of water, land use, irrigation districts, and main cities. This information was used to create a relational data model that establishes relationships between georeferenced information and climate information.

For the generation of the precipitation time series, the ERIC III and the CHIRPS databases were used for the Conchos River basin. The first one contains information from the National History Data Bank of CONAGUA and the National Meteorological Service (SMN) (IMTA 2009). The latter contains information from multiple sources of satellite data and terrestrial observations and combines this information to create precipitation time-series meshes (Funk et al. 2015).

Average daily rainfall information was downloaded from the CHIRPS database using an input file with the GeoJSON extension, which is a geospatial data exchange format based on the JavaScript object notation (Butler et al. 2016). The download file has a csv extension, which may be viewed and manipulated in MS Excel®. Figure 3 displays the entry of the GeoJSON file, which contains geospatial data for the Rio Conchos basin, thus delimiting the geographical area from which the precipitation time-series meshes are required.

Figure 4 shows average-daily rainfall for 1981 as obtained from the two databases previously mentioned, which served as the main input parameter for the hydrological model. It is important to note that the difference in millimeters of rainfall of the time series of the two databases is 16.40 mm, which is equivalent to 2.85% difference.

Additionally, the hydrometric information was obtained from the National Surface Water Data Bank of CONAGUA (BANDAS). Naturalized flow information from the Texas Commission on Environmental Quality (TCEQ) was used, which used a basic procedure applied to the analysis of water availability in the Rio Conchos basin. This analysis involved knowing naturalized flows through historical hydrological information. With this information and with the Water Rights Analysis Package (Wurbs 2005), the water availability of water in the region was simulated over a period of 61 years (1940–2000) considering the water rights of each user and authorized diversion and quantities of storage according to with the TCEQ (R. J. Brandes Company 2003).

Also, there are 31 hydrometric stations (HS) within the study area. However, only those stations that presented constant and reliable information were used in the calibration and validation of the hydrological model. Thus, after the analysis of the stations within the basin, 8 HS located in the main sub-basin riverbed, 6 HS located in bypass channels and 4 HS at dam entries were used. Table 1 describes the HS located in the Rio Conchos tributary riverbeds.

Figure 5 depicts the spatial location of the hydrometric stations used throughout the study area while Figure 6 shows the 1981 average daily-runoff observed for the 24230 HS-Ojinaga gauge station, which is located at the outlet of the basin and was the determining HS for the calibration and validation of the hydrological model.

The Rio Conchos basin has a drainage area of 66,682 km², which was obtained through the ArcGIS program using the .shp hydrological basins file from the National Water Information System as a base layer (CONAGUA 2020b). It has a watershed perimeter length of less than 15,000 km and the length of the main riverbed is estimated at 750 km.

The Time of Concentration (T_c) is the time required for the runoff to travel from the furthest point of the basin to its outlet. T_c is only calculated for surface runoff and depends on the slope and length of the main riverbed. T_c can be calculated through various methods; however, in this work, the Kirpich equation was used (Aparicio-Mijares 1992). The slope was calculated through the Taylor–Swarts method. Lag time (T_p) is the period from the start of the runoff produced by a certain rainfall until it reaches its peak value. Conceptually, T_p is considered as a weighted T_c. The United States Department of Agriculture (USDA) established that the relationship between T_p and the T_c as T_p=0.6T_c (USDA 2010).

The determination of a Curve Number (CN) is necessary to simulate the behavior of the rainfall/runoff process by the Soil Conservation Service (SCS) method. This CN is defined by the Hydrological Soil Groups, soil cover type, soil treatment and the Antecedent Moisture Conditions (AMC) (USDA 1986).

To establish the hydrological groups of soil, the INEGI edaphological vector data set scale 1:250,000 Series II or updated is used, which contains information on the different types of soil in Mexico obtained by using the World Reference Base of Soil Resources International System (WRB) for its classification (INEGI 2013). To determine land use and vegetation, the 1:250,000 scale vector data set V or present series is used, which contains information on land use and vegetation in Mexico obtained based on photo-interpretation techniques of satellite images and supported by field work (INEGI 2014). If different hydrological groups and land uses are found in the analyzed watershed, the value of the CN parameter is calculated by subdividing the watershed and obtaining the weighted average of each of them.

Based on the collected and analyzed information, the hydrological model was developed to simulate the behavior of the rainfall/runoff process at the Rio Conchos basin using the HEC-HMS program (version 4.2) and the transformation method of the SCS unit hydrograph. This model was fed by the two sources of rainfall data mentioned above and with the physiographic basin information, variables for the determination of precipitation losses, as well as variables used to transform excess rainfall into runoff (Figure 7). This information was entered into the program while taking into consideration each of the main components: Basin model, meteorological model, control specifications, and time series data (US Army Corps of Engineers 2016). The simulation used a period of 365 days (from January 1 to December 31, 1981), establishing a daily interval. The time series of the different databases, specifying increases in millimeters (mm), were captured manually. Table 2 shows the input parameters for each of the sub-basins considered within the Rio Conchos Basin hydrological model.

The model calibration required a quantitative assessment of the hydrological response of the sub-basin. For these purposes, both hydrograph charts (observed and simulated) were compared. This is fundamental for the assessment of the model because it supports data distribution and variation comparisons (Legates & MacCabe 1999). The observed hydrographs versus the model output hydrographs using the time series from the ERIC III and CLIMATESERV databases for the calibration period (1981) are shown in Figure 8 for three of the stations located across the basin: 8085-Boquilla, 8202- Fco. I Madero, 24230-Ojinaga. The Boquilla station is located upstream from the outlet of the R. Conchos–Presa de la Colina sub-basin, which is the largest in the basin (upper part). The second station is located at the outlet of the San Pedro River sub-basin, upstream of the Fco. I Madero dam (middle part). Finally, Ojinaga HS is located 1 km upstream of the convergence between Rio Conchos and Rio Bravo (lower part). The main parameters for the model calibration were the lag time, antecedent moisture conditions, the use of land and the type of soil.

In general, the model, despite the information used, tends to reproduce maximum flows with a greater adjustment than the minimum flows. The simulated daily flow with satellite rainfall information accurately represents the hydrological response of the basin in the three stations.

Model validation is defined as ‘the process of demonstrating that a specific model for a given site is capable of making accurate predictions for periods outside the calibration period’ (Refsgaard & Knudsen 1996). Thus, a model is deemed as validated if its predictive accuracy and capacity are within acceptable limits or errors.

Therefore, after the model calibration process, two independent simulations were performed using the precipitation time series from the CHIRPS database. The first simulation was performed across of a period of one year (1991), when the tropical storm Ignacio passed through the region. The second simulation was performed across a period of 10 years (1992–2001). A period of drought was recorded in the basin during that decade (Ortega-Gaucin 2013). One of the objectives of selecting a drought period for validation is that this hydrological model is intended to be used to assess the impact of climate change in the basin, where drought periods are expected to increase in the following years.

Figure 9 shows the simulated daily hydrographs against those observed for the validation periods (1991, 1992–2001 respectively) at the outlet of the Rio Conchos basin in Ojinaga, showing a good performance of the model according to the variation and distribution of the simulated flows against the observed flows for the validation periods. The model tends to reproduce better maximum flows, such as those produced in September 1991, January and June 1992, July 1993, May 1994, October 1996, and May 1997.

The RSR concentrates the benefits of the error index and contains a normalization factor so that the resulting statistics and reported values can be applied to several components. The RSR range starts at an optimal value of 0, which indicates that there is no residual variation; therefore, the simulated model is perfect up to positive values. The lower the RSR, the smaller the RMSE and the better the performance of the simulated model (Moriasi et al. 2007).

This efficiency ranges from −∞ to 1.0, NSE=1.0 being the optimum value. The values between 0.0 and 1.0 are considered acceptable levels of performance, whereas levels less than 0 indicate that the observed value is a better predictor than the simulated value, indicating unacceptable performance (Moriasi et al. 2007).

This index has a range between 0.0 and 1.0; the value of 1.0 reveals a perfect relationship between the observed and simulated data. On the contrary, a value of 0.0 reflects a total disagreement (Willmott 1981). Table 3 shows the classification of the model performance evaluation based on what was considered and recommended by Moriasi et al. (2007; 2015).

The distribution and variation of the simulated flows for the calibration period show a good performance of the model, since it tends to reproduce the observed flows, especially in the months of August, September, and October (see Figure 8). On the other hand, the results of the statistical evaluation of the simulations using the two precipitation databases in comparison with the information observed in the EH Boquilla, Fco. I. Madero, and Ojinaga are shown in Table 4.

These results show r² equal to 0.66, 0.77, and 0.60 with information from ERIC III, so the performance is classified as Satisfactory for the upper and lower part of the basin, and Good for the middle part. Considering the CHIRPS information, the results of r² were 0.67, 0.53, and 0.60, so that its performance was Satisfactory for the upper and lower basin and Not Satisfactory for the middle basin.

The performance of the model using the two databases was Good for the middle-upper part and Very Good for the lower part of the basin considering the RSR (0.54, 0.58, 0.02 for ERIC III and 0.59, 0.12, 0.02 for CHIRPS). With respect to the NSE, the results simulated with ERIC III showed a Good performance (0.71) for the upper basin, Satisfactory (0.67) for the middle zone, and Very Good (1.00) for the lower part of the basin. On the other hand, with CHIRPS data the performance of the model was Satisfactory (0.65) for the upper zone and Very Good (0.99 and 1.00) for the rest of the basin. Meanwhile, the adjustment index d shows a Very Good performance using the two precipitation databases with values from 0.95 to 1.00.

The results of the MAE, considering ERIC III and CHIRPS, indicated an error of 36.36 m³/ s and 39.79 m³/s in the upper basin, and an error of 16.57 m³/s and 3.32 m³/s in the middle basin. An error of 0.81 m³/s (both) occurred at the outlet of the basin, which represents an error of less than 1% of m³/s for the calibration period.

Figure 9 shows the good performance of the model according to the variation and distribution of the simulated flows against the observed flows for the validation periods. The model tends to reproduce better maximum flows such as those produced in September 1991, January and June 1992, July 1993, May 1994, October 1996, and May 1997. Table 5 shows all statistical results of the model in the validation periods.

The Conchos River basin has been simulated in different years and with different hydrological models, but there are few studies that consider the simulation of the basin with a daily time interval, since they have focused generally on monthly flows. You have, for example, Gómez-Martinez et al. (2005), who tested two different distributed models, on the one case with HEC-HMS, and on the other the CEQUEAU. These models were intended to simulate the monthly flow rates in different periods of time (1980–1999) of each of the sub-basins of the Conchos River. With the available information of the different time periods, it was not possible to calibrate or validate with the HEC-HMS model. With the CEQUEAU model, only the middle-upper part of the basin was calibrated due to the lack of available hydrometric information. According to the authorś conclusion, they comment: ‘if you want to calibrate the entire basin to the Pegüis HS, it is necessary to have the complete information of all the dams’ (Gómez-Martinez et al. 2005).

For their part, Amato et al. (2006) applied the WEAP hydrology model to the Conchos river basin using a monthly time interval for testing and tried to compare it with the model of Gómez-Martínez et al. of the year 2005, but this comparison could not be made because of the lack of results of the latter model. Amato et al. calibrated their model in 1980 with naturalized flows of TCEQ obtaining good approximations from the simulated against the monthly flows.

Ingol-Blanco & McKinney (2009, 2013) used the WEAP hydrology model of Amato et al. (2006) and extended the calibration period to 10 years (1980–1989) and added a validation period from 1990 to 1999 using a monthly time interval. The model reproduced accurately and with good performance (NSE = 0.81 and 0.84, in the upper and lower basin, respectively) simulated flows for the calibration and validation periods in the upper and lower parts of the basin.

Considering models with a daily time interval we have, on the one hand, the final report New technologies for flood prevention and analysis of the Mexican Institute of Water Technology (IMTA 2012) established a methodology for the hydrological modeling of the flows of the Conchos river basin on the Hydrotel platform and the Physitel application (Fortin et al. 1995).The simulated results with respect to the observed flows showed a performance of the Non-Satisfactory model in a large part of the basin (NSE = −0.69, −1.46 and −0.09, in the upper, medium, and lower basin, respectively). The model could only reproduce the flows of the Florido river sub-basin with a good (0.73) performance in the simulation in the period 1975–1985.

On the other hand, Hernández-Romero & Patiño-Gómez (2018), Hernández-Romero et al. (2019) used punctual and satellite precipitation information to model the flows on a daily time scale of the Conchos–P. de la Colina sub-watershed, corresponding to the Upper Conchos River basin. The NSE efficiency (0.74 and 0.65) of the model showed Satisfactory performance using satellite information and Good using punctual data for the year of calibration (1981).

The importance of analyzing in detail the process of rainfall-runoff with a time interval lies in the fact that it is possible to evaluate the availability, the management of the resource, the expectations of the fulfillment of the 1944 binational agreement, and the impact of the current drought that plagues the region every day or week, thus subtracting the risk of bias that may occur when obtaining monthly average flows due to the climatic variability that is so evident in this unique and complex basin. In addition, when considering the analysis of extreme events, such as hurricanes, the simulation with a daily time interval can be a tool to help in early warning systems in the vulnerable regions of the basin.

The hydrological model simulated the rainfall/runoff process of the Rio Conchos basin through the HEC-HMS program, which was fed with satellite and timely rainfall information as well as with other physical parameters. The quality and usefulness of satellite rainfall information was assessed through an inter-comparison exercise between the time series, obtaining a difference of less than 5% between the two sources.

Based on this result, the hydrological response of the 1981 basin was simulated. The model was calibrated with three stations along the basin (upper, middle, and lower) using the time series mentioned above. After the calibration process, two independent periods were simulated with information from the CHIRPS database to validate the model: 1991 and 1992–2000.

Since the adjustment of the distribution and variation of the output flow was good, the analysis indicated good behavior of the model against the historical information observed. In addition, a statistical assessment of the model was performed for the calibration and validation periods. In general, the model tends to reproduce maximum flows with a greater adjustment than minimum flows. Using satellite rainfall information, the daily flow of the basin's hydrological response may be represented well.

In fact, the statistical assessment (r, r², MAE, RSR) of the model indicates a good and very good representation in the calibration and validation periods, respectively. The goodness of fit of the model performance (NSE and d) indicates a good goodness of fit for the calibration period and very good for the validation period.

With these results, it can be concluded that the satellite records are a good alternative to reinforce or supply the lack of information available in Mexican weather stations. These records can be used for hydrological modeling in basins with little or no information, such as the Conchos River basin analyzed in this research. Once the quality and usefulness of the satellite information have been evaluated, the lack of climatological information can be filled with confidence and thus updated hydrological models for better decision making on issues related to water availability at the watershed level.

The authors thank the University of the Universidad de las Américas Puebla and the UNESCO-UDLAP Chair in Hydrometeorological Risk for all the support and facilities provided for carrying out this work.

All relevant data are included in the paper or its Supplementary Information.