Monitoring meteorological drought in Hoa Vang district, Da Nang City, using Google Earth Engine and satellite-derived precipitation data combined with ground-based weather stations

Drought is a typical natural disaster that causes damage to agriculture and the environment and dramatically impacts a region's economy (Samarah 2005). After a prolonged dry period, droughts become noticeable, but it can be challenging to determine their start, duration, and end. Droughts can be complex in duration, magnitude, and intensity. To simplify, we can define drought as a persistent water shortage caused by a continuous lack of rainfall. Droughts can take various forms, such as agricultural, hydrological, and meteorological. Meteorological droughts are the most common type (Ray et al. 2015), and they occur when there is insufficient rainfall on average. Since it frequently serves as the initiation to other types of drought, it is the most researched type of drought (Ahmed et al. 2019). The frequency of meteorological droughts does not depend on the amount of precipitation in an area but instead depends on its variability (Pande et al. 2022). Drought is a continuous climatic phenomenon that cannot be avoided. However, efficient administration and observation of a drought can help minimize its impact on socioeconomic sectors, especially agricultural production (Patel & Yadav 2015).

A variety of drought indicators have been developed to monitor droughts (Chen et al. 2017). These indicators allow researchers to quantitatively assess climatic anomalies in terms of time, intensity, and spatial extent (Mishra & Singh 2010). The most common and detailed drought indicator that was widely used is the standardized precipitation index (SPI). It is simple, straightforward, and independent of climatic factors (Kushwaha et al. 2022). Precipitation-based drought indicators, including SPI, are based on two assumptions: (1) precipitation variability is significantly higher than other parameters, such as temperature and potential evapotranspiration, and (2) other variables are constant over time. In this case, the importance of these other factors is minimal, and variations in rainfall over time lead to drought (Vicente-Serrano et al. 2010). SPI has been applied to study various aspects of drought, including frequency analysis (Guo et al. 2017), drought forecasting (Mishra et al. 2007), spatiotemporal analysis, and climate impact studies (Ashraf & Routray 2015).

Vietnam is significantly impacted by natural disasters, including storms, floods, and droughts. The World Bank has assessed Vietnam as a country that has experienced numerous natural disasters (World Bank 2018). Central Vietnam is frequently affected by droughts. This is primarily due to a lack of rainfall and the limited water storage capacity of local reservoirs, exacerbated by prolonged hot and dry climatic conditions. Similar to the risks posed by storms, the impact of droughts often focuses on specific geographical regions, depending on the meteorological and topographical characteristics of each area (Lan Huong et al. 2022). In 2016, Vietnam experienced its most severe drought in a century, partly attributed to the effects of El Niño. In the 20 most affected provinces, approximately 2 million people received relief assistance, including 500,000 residents in the coastal provinces of central Vietnam (UNDP 2015; World Bank & GFDRR 2018). Da Nang City, situated in an area greatly influenced by climate change, frequently encounters droughts, particularly in the Hoa Vang district. Hoa Vang district exhibits diverse and complex terrain, spanning mountainous, midland, and plain regions, divided from northwest to southeast. With an average annual rainfall of approximately 1,873 mm, heavy rainfall predominantly occurs in September, October, and November, leading to flooding in lowland areas. Conversely, periods of low rainfall result in droughts, adversely impacting agricultural activities in the district. Consequently, it is essential to assess and monitor droughts in the Hoa Vang district, with a particular focus on promptly evaluating rainfall patterns, as they are the primary factor influencing drought conditions in the area.

Currently, the calculation of the SPI primarily relies on actual meteorological station data and is conducted over a wide area (Livada & Assimakopoulos 2007; Ionita et al. 2016; Mahajan & Dodamani 2016; Rahman & Lateh 2016; Dabanlı et al. 2017; Soydan Oksal 2023). However, for small areas with limited numbers of stations, accurately assessing SPI values and drought characteristics becomes challenging. The Hoa Vang district is a typical example, with a small area and limited meteorological stations. Therefore, assessing drought characteristics based on these stations will not ensure accuracy regarding the spatial distribution of drought levels. However, in recent years, the significant development of rain-measuring satellites has opened up opportunities to generate drought maps for regions that lack monitoring data. These satellites are capable of providing information about rainfall worldwide, with relatively good spatial and temporal resolution (Serrat-Capdevila et al. 2014). Satellite precipitation integrated with ground-based station data, such as Climate Hazards Group InfraRed Precipitation with Station data (CHIRPS), possesses a comparatively high spatial resolution (0.05 × 0.05°) and has been consistently collected from 1981 to the present. This comprehensive dataset can facilitate detailed spatiotemporal analysis of drought variations in localized areas, exemplified by Hoa Vang district in Da Nang City. Furthermore, the study compares SPI calculations from CHIRPS with SPI results from similar satellite and ground station datasets like the Center for Hydrometeorology and Remote Sensing (CHRS) and field observations to evaluate the reliability of the research findings. Consequently, this method can be applied to assess drought characteristics in areas with similar conditions. This study was conducted with the following primary objectives: (1) to calculate SPI indices based on CHIRPS data; (2) to select the best SPI index based on the Mann–Kendall (MK) test and Sen's slope analysis; (3) to evaluate the accuracy of SPI calculated from CHIRPS compared to other precipitation data sources such as CHRS and ground observations through statistical parameters such as Pearson's linear correlation coefficient (r), root mean squared error (RMSE), mean absolute error (MAE), relative bias (Bias), and Kling–Gupta efficiency (KGE); and (4) to generate drought maps based on the SPI index, including monthly drought frequency, drought intensity, and extreme drought events from 1981 to 2022 in the Hoa Vang district.

In this study, we used the Google Earth Engine (GEE) cloud computing system to calculate the SPI based on CHIRPS data for the period from 1981 to 2022 in the Hoa Vang district located in Da Nang City, Vietnam. In the field of environmental monitoring, GEE is a cloud-supported system that is designed to store and analyze massive datasets (Kumar & Mutanga 2018). GEE contains a large collection of satellite imagery from various sources and different types of geoprocessing tools (Khan & Gilani 2021). The current archival dataset includes data from other satellites, as well as vector datasets based on geographic information systems (GIS), social, population, weather, digital elevation models, and climate data layers (Kumar & Mutanga 2018). The main advantages of GEE are that it is time-saving and has fast processing. This study takes advantage of GEE and satellite-derived precipitation data integrated with ground station data using different modeling methods in GIS to assess the drought hazard via the SPI index in the Hoa Vang district, Da Nang City.

The average annual temperature in the Hoa Vang district is 26.7 °C, with the highest usually occurring in May, June, July, and August, with temperature in the range of 29–30 °C, and the lowest in the months of December, January, and February, with average temperature of 21–24 °C, especially the mountainous forest area of Ba Na, with an altitude of nearly 1,500 m, usually with an average temperature of about 20 °C. The average annual sunshine hours in the district are 2,438 h, with the most solar radiation in May and June, averaging from 262 to 282 h per month, while December and January usually have little sunshine, with an average of 136–152 h per month (An et al. 2021).

The Hoa Vang district has the Cu De River system in the north and rivers such as the Túy Loan River, the Yen River in the south, downstream of the Ai Nghia River, and the Vu Gia River. The hydro conservation of these rivers varies seasonally, leading to flooding in the rainy season and water shortage in the dry season, causing drought conditions in the Hoa Vang district.

In this study, we used CHIRPS data and the GEE cloud computing platform to calculate the SPI value for the Hoa Vang district, Da Nang City. The main types of data used in the study are shown in Table 1.

Estimating variations in precipitation spatially and temporally is a key aspect of droughts and environmental monitoring. However, estimating precipitation based on satellite data can be challenging due to the influence of complex terrain, which can lead to the underestimation of intense rainfall. In addition, the network of rainfall monitoring stations is often sparsely distributed in a given area. To address these limitations, researchers have used the CHIRPS dataset, which was developed in collaboration with scientists at the USGS Earth Resources Science and Observation Center (EROS). CHIRPS aims to provide comprehensive, reliable, and regularly updated datasets for various early warning objectives, including trend analysis and drought monitoring (University of California & Santa Barbara 2023).

Recently, advancements in satellite observation data have allowed for the development of high-resolution climate maps of precipitation, utilizing satellite-based estimates from organizations like NASA and NOAA. These maps have a resolution of 0.05 × 0.05° and are valuable in eliminating systematic biases when applied to satellite-based rainfall estimates, which is a crucial step in creating datasets like CHIRPS. The CHIRPS dataset has been constructed from 1981 to the present, and it has been instrumental in supporting drought monitoring initiatives of the United States Agency for International Development (USAID) (University of California & Santa Barbara 2023).

In this case study, we have taken into account the capabilities of GEE to collect CHIRPS data and calculate SPI values for the period spanning from 1981 to 2022. The study focused on the Hoa Vang district in Da Nang City, allowing for a detailed characterization of rainfall patterns and drought conditions in that specific area.

This study used the precipitation data obtained from the PERSIANN-CCS system, which is a high-resolution, real-time global satellite rain monitoring system (0.04 × 0.04° or 4 × 4 km) developed by the CHRS at the University of California, Irvine (UCI). The system uses cloud classification techniques to estimate precipitation from satellite data. This allows information on rain to be collected globally quickly and reliably. The rain data from PERSIANN-CCS provides good resolution and gives us a detailed view of the rainfall distribution in the study area (Center for Hydrometeorology and Remote Sensing 2023).

By using rainfall data from the PERSIANN-CCS system, it is possible to make accurate judgments and analyses of rainfall over time and space. From there, it helps us better understand climate and weather characteristics in the study area (Center for Hydrometeorology and Remote Sensing 2023).

In this study, we extracted monthly precipitation data from the study area and conducted SPI value calculations based on SPIGenerator software (National Drought Mitigation Center 2018).

To assess the accuracy of satellite-derived rainfall rain data and SPI value calculation results, this study used daily rain gauge data from 1981 to 2020 in the study area. These data were collected at the Central Hydrometeorological Center, Vietnam. From this actual rain data, the study also processed and calculated SPI values on SPIGenerator software.

The drought index is an important parameter to characterize various aspects of droughts, including their termination, duration, affected area, and severity at different scales (Piechota & Dracup 1996). Numerous drought indicators have been developed by meteorologists and climatologists worldwide (Dai 2011). Among these many indicators, the most popular is the SPI (McKee et al. 1993). SPI has been recognized globally and widely applied due to its robustness, efficiency, flexibility, and standardization (Tsakiris & Vangelis 2004). As a result, the SPI model, which relies on rainfall data, is often the preferred choice for assessing drought conditions. It offers advantages such as minimal data requirements and straightforward calculations (Viste et al. 2013). One of the advantages of the SPI is its ability to facilitate direct comparisons of SPI values between different regions, even with contrasting climates. This index is effective in tracking the onset, intensity, and duration of drought events (Viste et al. 2013).

The SPI can be calculated for various timescales, such as 1, 2, 3, 6, 9, 12, 24, and 72 months (Stagge et al. 2015). The selection of the timescale depends on the specific application and the characteristics of the phenomenon being studied (Duc et al. 2022). From a statistical perspective, the range of 1–24 months is generally considered to provide the most reliable and informative SPI values (Guttman 1999). This range covers both short-term and long-term drought conditions, enabling a comprehensive assessment of drought severity and duration. Shorter timescales, such as 1, 3, and 6 months, are particularly relevant for measuring the impact of drought on agriculture (Morid et al. 2006). These shorter-term SPI values capture more immediate and localized drought effects on crops and vegetation, providing valuable insights for agricultural planning and management. It is important to note that the selection of the appropriate timescale for calculating SPI should be based on the specific objectives of the study and the temporal characteristics of the phenomenon being analyzed.

Given x_j and x_k as the time series values at times j and k, respectively (where j > k), the median of these N values of S_i represents Sen's slope estimator. A positive value of this slope indicates an upward trend, whereas negative values indicate downward trends, with the magnitude of the slope reflecting the rate of change.

The value of r ranges from −1 to 1. For the SPI study case, the closer the value is to 1, the more optimal it is.

If these results show no correlation, it means that the data source used to assess drought does not meet the requirements, and further testing or searching for alternative data sources is needed. If there is a correlation among the calculated SPI values, we can use a satellite data source that provides a better value to evaluate drought in the study area. In this study, we compared the results of constructing the SPI index based on CHIRPS data on the GEE platform with the results of SPI based on CHRS data and observed data on the SPIgenerator application. The SPIgenerator application is used to generate SPI data and illustrate how to interact with the SPI Dynamic Link Library (DLL). This application allows reading rainfall data and supports different timescales and data types (weekly, monthly). It provides SPI data, drought frequency, and different drought durations.

All three indices, RMSE, MAE, and Bias, are more optimal when their values reach 0.

The methodology for assessing drought conditions in Hoa Vang District utilizes a systematic approach integrating satellite-derived and observational data. This process begins with data collection from three primary sources: CHIRPS spanning 1981–2022, CHRS covering 2003–2022, and observational data spanning 1981–2020 from meteorological stations within the study area. The district is divided into 28 regions, each defined by a spatial resolution of 0.05 × 0.05° to facilitate precise analysis (Figure 3).

Based on the statistical analysis of SPI values for 1-, 3-, 6-, 9-, and 12-month periods, the study conducted a frequency analysis of drought occurrences in the Hoa Vang district over the 12 months during the period 1981–2022. The results of this analysis are presented in Table 3.

Table 3 presents the frequency and percentage of drought occurrence for different SPI indices (1, 3, 6, 9, and 12 months) during the period 1981–2022 in the Hoa Vang district. For SPI 1, the month with the highest drought frequency is May (18.5%), while the lowest is January (11.1%). Regarding SPI 3, the months with the highest drought frequency are February, July, and August (around 18%), while the lowest is November (13.2%). For the 6-month SPI, June has the highest drought frequency (19.6%), while January has the lowest (15%). In cases of 9- and 12-month SPI, the drought frequencies are relatively consistent, ranging from 11 to 18%.

In general, drought tends to occur more frequently during the dry season (May to June), which aligns with the rainfall patterns in the study area (September to December). The analysis of the data table demonstrates seasonal variations in the frequency of drought occurrence based on different SPI indices. The results of the analysis align with the climate cycles of the region.

From Figure 5, we can observe that the frequency of moderate drought occurrences is generally higher compared to severe drought and extreme drought. However, within each SPI value, the frequency of occurrence differs. For the periods 1981–2000, concerning moderate drought, SPI values of 9 and 12 months tend to have higher frequencies compared to SPI values of 1, 3, and 6 months. The frequency of severe drought does not exhibit significant differences among the SPI values during the 1981–2000 period. In 2001–2022, the frequency of moderate drought occurrences shows a decreasing trend with increasing SPI values. The frequencies of severe and extreme drought are relatively low during the 2001–2022 period.

This study has conducted the MK test and Sen's slope analysis in the study to assess the correlation of SPI values for 1, 3, 6, 9, and 12 months on a monthly basis. The results are presented in Table 4.

Based on Table 4, we can observe that the MK values increase with the SPI index. For the SPI 1 index, only the MK values for January and August are statistically significant (p < 0.05). For the SPI 3 index, there are 6 months with statistically significant MK values (months 1, 2, 3, 9, 10, and 11). The MK values for the SPI indices of 6, 9, and 12 are statistically significant for 9, 11, and 12 months, respectively. Therefore, we utilize the SPI 12 values calculated from CHIRPS data to assess the correlation with the SPI 12 results calculated from CHRS data and measurements.

Based on Figure 6, it is evident that drought can occur in all months during the period 1981–2022. The dry season (June to August) tends to exhibit stronger drought intensity compared to other periods. Specifically, there are 8 months with severe drought intensity recorded in September 1995 (−1.75), October 1998 (−1.73), August 1998 (−1.71), October 1995 (−1.64), July 2019 (−1.58), December 1982 (−1.55), April 1983 (−1.54), and May 1983 (−1.54). Conversely, there are 64 months with moderate drought intensity. In general, moderate drought occurs in almost all months. When categorized by periods, from 1981 to 1990, there are 28 months with moderate drought, accounting for 43.8%. From 1991 to 2000, there are 30 months, accounting for 46.9%. From 2001 to 2010, there are only 2 months (June and July 2005), accounting for 3.1%, and from 2011 to 2022, there are 4 months, accounting for 6.2%. Therefore, we can observe a trend of increasing rainfall in the periods from 20011 to 2022. However, drought can still occur in specific time periods.

According to Figures 7–9, the average correlation (r) between CHIRPS data and the observed data is 0.85, indicating a strong positive relationship. In addition, the average correlation between CHIRPS data and CHRS data is 0.77, indicating a moderately strong positive relationship. The average RMSE of SPI 12 calculated from CHIRPS data compared to CHRS data is 0.70, while compared to observed data, it is 0.56. This suggests that the error in SPI 12 calculated from CHIRPS compared to CHRS is relatively higher than that compared to observed data. For MAE, the value for CHIRPS compared to CHRS is 0.57, which is also higher than that for observed data (0.37). According to the Bias index, the SPI 12 value calculated from CHIRPS tends to be higher compared to SPI 12 calculated from CHRS (increasing by 18%) and from observed data (increasing by 7%). However, these RMSE, MAE, and Bias values are all at relatively low levels (optimal value). Finally, the KGE index is used to evaluate the performance of SPI 12 calculated from CHIRPS compared to CHRS and observed data. The KGE index for observed data is high (0.80), indicating high accuracy between SPI 12 calculated from CHIRPS and observed data. In contrast, the KGE index, when compared with CHRS is moderate (0.67). Overall, SPI 12 values from CHIRPS data show high accuracy with observed data across all indices, while with CHRS data, accuracy ranges from moderate to high. This suggests that CHIRPS data is a reliable source for estimating SPI 12 values and can be used as a substitute for observed data in SPI analysis.

Based on the results of the MK test, Sen's slope analysis, and the correlation assessment between the SPI values derived from CHIRPS, CHRS, and the observed data, we find that the SPI 12 values have significant MK test results (12 out of 12 months with p < 0.05), and the Sen's slope (0.04019) has statistical significance. The indices r, RMSE, MAE, Bias, and KGE all demonstrate favorable values. This indicates that drought mapping based on the SPI 12 index derived from CHIRPS data exhibits high reliability. Therefore, in this study, we establish a frequency map of drought occurrence and extract the time periods with drought events based on the SPI 12 index during the study period to create a drought risk zoning map in the Hoa Vang district.

The study extracted times when severe droughts occurred in the period from 1981 to 2022. Accordingly, we extracted SPI 12 values at the times of May 1983 (period 1981–1990), September 1995 (period 1991–2000), July 2005 (period 2001–2010), and July 2019 (period 2011–2022). On that basis, we simulate this value in space using the IDW interpolation method. The results are shown in Figure 13.

This study evaluated the meteorological drought conditions in the Hoa Vang district, Da Nang City, using satellite-based rainfall data integrated with ground station data (CHIRPS) and the cloud computing platform GEE. The SPI was calculated for different timescales (1, 3, 6, 9, and 12 months) to assess the characteristics of drought in the study area.

Based on the statistical results of the SPI index (1, 3, 6, 9, and 12 months), the study determined the frequency of drought occurrence for the period 1981–2022 in Hoa Vang district. With the 1-month SPI, the month with the highest frequency of drought is May (18.5%), while January has the lowest frequency (11.1%). With the 3-month SPI, the highest frequency of drought occurs in February, July, and August (about 18%), and November has the lowest frequency (13.2%). The 6-month SPI shows that June has the highest frequency of drought (19.6%), while January has the lowest frequency (15%). The 9- and 12-month SPIs show relatively uniform drought frequency, ranging from 11 to 18%. In general, droughts often occur more often in the dry season, especially in May and June.

The study was conducted using MK and Sen's slope test methods to evaluate the reliability of SPI values. Accordingly, the SPI 12 value has the highest reliability, with 12/12 having a p-value <0.05 (statistically significant).

The study also evaluated the accuracy assessment between the results of calculating the SPI 12 index based on rain data from the CHIRPS on the GEE platform and the results from CHRS data along with actual measured rainfall data in the study area. The correlation coefficient (r) was 0.77 with CHRS data and 0.85 with observed data. RMSE was 0.7 with CHRS data and 0.56 with observed data, while the MAE was 0.57 with CHRS data and 0.37 with observed data. The Bias was 0.18 with CHRS data and 0.07 with observed data, and the KGE was 0.67 with CHRS data and 0.8 with observed data. The results indicate that the SPI 12 calculated from CHIRPS data is highly correlated with observed data and moderately correlated with CHRS data. Results from this study indicate that CHIRPS data-derived precipitation could provide a high potential application in drought hazard assessment, especially for areas with limitations in field rainfall data measurement.

The study used the IDW method to create maps showing the frequency of drought occurrence by month in the Hoa Vang district. Drought can occur in almost all areas of the district, but the frequency of occurrence varies over time. The frequency of drought occurrence in the period 1981–1990 and 1991–2000 is often quite high, accounting for 19.17–32.5% in the period 1981–1990 and 20–36.67% in the period 1991–2000. However, in the period 2001–2010, the frequency of drought occurrence was very low, only from 0.83 to 2.5%. In the period from 2011 to 2022, the frequency of drought occurrence increased compared to the previous period, from 2.08 to 11.8%.

In Hoa Vang district, moderate drought is the most common intensity, occurring in 9.72% to 14.48% of the months in the study period. Severe drought is infrequently observed, appearing in only 1.19% to 3.77% of the months. Extreme severe drought is rarely seen in the district, primarily limited to a small area in the Northwest. The spatial distribution of drought levels indicates that moderate drought is concentrated in the Northwest and Southern regions, while severe drought is primarily found in the Southeast. Additionally, the study identified periods of significant drought from 1981 to 2022 and simulated drought maps for those times. Overall, the findings highlight that moderate and severe drought levels are the main concerns in the study area.

The study also extracted the times with the greatest drought level in the period from 1981 to 2022 and simulated drought maps at these times. In general, drought focuses on two levels: medium and severe drought in the study area.

In conclusion, the research results provide a better understanding of the impacts of drought on the natural environment and socioeconomics, especially in the agricultural sector of the Hoa Vang district, Da Nang City. This, in turn, provides a scientific basis for decision-making processes related to environmental protection policies while ensuring the efficiency of local agricultural production. Scientifically, the study proposed a research framework that includes appropriate integrated methods (the SPI using CHIRPS data on the GEE cloud computing platform in combination with the MK test and Sen's slope method to evaluate the reliability of the SPI indices) to achieve the most accurate and comprehensive assessment of drought impacts on the agricultural sector in the study area. This study is a new advancement in the application of the CHIRPS dataset for estimating spatial and temporal changes in rainfall. This application has overcome potential difficulties caused by complex terrain, which can lead to the underestimation of heavy rainfall. In addition, it addresses the challenge in data collection in areas where the network of rain gauge stations is often sparsely distributed. However, it is important to note that this study has geographical limitations. Since the results of the study are based on a pilot experiment in Hoa Vang district, Da Nang City, they can only serve the management and policy-making needs of the local area. Future studies can apply similar methods in different regions. This will not only help assess the accuracy and refinement of the research framework we propose but also provide a scientific basis for finding management solutions for similar issues in other localities.

Le Ngoc Hanh was funded by the Master, PhD Scholarship Programme of Vingroup Innovation Foundation (VINIF), code VINIF.2023.TS.033.

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

The authors declare there is no conflict.