Application of TRMM for Spatio-Temporal Analysis of Precipitation in the Taiwan Strait and Its Adjacent Regions

Abstract

Precipitation patterns are highly valued in the fields of weather forecasting, water resource management, and estuary environment research. In this study, daily and monthly precipitation TRMM data from 1998 to 2019 were selected, and EOF analysis was employed to analyze the precipitation patterns of the Taiwan Strait and its neighboring regions. We obtained the following results: (1) The rainy season (May–June) is the main contributor to precipitation in the study area. The EOF first mode reflected the overall consistency of the precipitation spatial distribution. However, within each river basin, the magnitude of precipitation variation is spatially different. The magnitude of precipitation variation is significant in the northwestern part of the Minjiang River basin, the southwestern part of the Jiulong River basin, and the southwestern corner of the Hanjiang River basin. These areas happen to correspond to the mountain areas, revealing that topographic precipitation plays a role in the spatial distribution of precipitation in the three river basins. (2) The spatial distributions of the EOF first mode and of precipitation during El Niño in the Minjiang River basin are consistent. This reveals that ENSO is probably the dominant factor in precipitation in the Minjiang River basin. The significant increase in precipitation during El Niño compared with a normal year in the Minjiang River basin confirms this point. (3) In all three strong El Niño years, 1998, 2010, and 2016, the Minjiang River basin experienced significant heavy precipitation in the fall and winter, whereas the Jiulong River and Hanjiang River basins did not (except in 2016). In other words, the Minjiang River basin is more affected by ENSO, while the Jiulong River and Hanjiang River basins are only limitedly impacted by ENSO.

1. Introduction

Precipitation is highly valued in the fields of weather forecasting and water resource management and in studies of river basins and estuarine environments [1,2,3]. The accuracy and applicability of the TRMM precipitation products have been evaluated and corroborated in many river basins, such as the Andean–Amazon River basins [4], the Yangtze River basin [5], the Dongjiang River basin [6], the Huaihe River basin [7], the Ganjiang River basin [8], the Mekong River basin [9], and the Hunza River basin [10]. There are also many studies on the spatial and temporal distribution, vertical structure, regulatory mechanisms, and even watershed quality assessment of precipitation based on TRMM data for medium and large river basins around the world [2,5,7,11,12,13]. The Minjiang River, the Jiulong River, and the Hanjiang River on the west coast of the Taiwan Strait bring large quantities of fresh water and terrestrial material to the Taiwan Strait every year. With the economic development and urbanization of the west coast of the Taiwan Strait, the ecological environment of the local watershed has deteriorated, and the terrestrial pollutants transported by runoff have further exacerbated the environmental problems (e.g., algal blooms) of the offshore ecosystems. The precipitation pattern in a river basin plays an extremely critical role in the transport of nutrient fluxes from the river, offshore marine ecosystems, and freshwater–seawater interactions [1].

Studies on precipitation and heavy precipitation in the Taiwan Strait and its neighboring regions have mostly been based on the analysis of long time series of meteorological station observations [14,15] as well as the statistical analysis of the long-term Doppler weather radar data [13]. The west coast of the Taiwan Strait is mostly hilly and mountainous, and meteorological stations are sparsely distributed in this region. The quantity of data acquired through meteorological stations has been quite limited. TRMM remotely sensed precipitation data have the advantages of large coverage, huge quantities of data, and continuous time series. In addition, their spatial measurements are not restricted by geography, endowing the TRMM data with high reliability in spatial analysis. Therefore, they are an important supplement to monitoring data from meteorological stations.

Empirical orthogonal function (EOF) analysis has been applied to the discussion of precipitation mechanisms. Lyons [16] analyzed monthly mean rainfall data encompassing a 37-year period from 63 stations in Hawaii and showed that the first to third modes were trade wind precipitation modes. Singh [17] used EOF analysis to investigate the intraseasonal and interannual variability of monsoon rainfall. In this study, we intended to select daily and monthly precipitation TRMM data for the period 1998–2019 and took the Taiwan Strait and its neighboring regions as the study area. On the basis of ensuring that this dataset truly reflected the trend of precipitation distribution in the region, EOF analysis was employed to study the spatio-temporal patterns of precipitation in the study area.

2. Study Area, Data, and Methodology

2.1. Study Area

The west coast of the Taiwan Strait consists mainly of small and medium-sized watersheds of the Minjiang, Jiulong, and Hanjiang Rivers (Figure 1). This region features a typical subtropical maritime monsoon climate, with rainfall mainly concentrated in the wet period (April–September); in spring and summer, the river basins predominantly receive frontal precipitation, while during the summer and fall seasons, they are frequently affected by tropical cyclones, often resulting in heavy precipitation. The Minjiang River is the largest river in Fujian Province, originating from the southern foot of Wuyi–Shanling, with a total length of 514 km. The Minjiang River basin encompasses a longitude of 116°23′–119°43′ E and a latitude of 25°23′–28°19′ N latitude and is located in the north of Fujian, with a watershed area of 60,992 km², accounting for 62.5% of the province’s area. Its average annual runoff is around 53.66 billion cubic meters (1950–2020). The Jiulong River is the second largest river in Fujian Province, with a total length of 285 km and an annual runoff of 12.35 billion cubic meters (1997–2020). The Jiulong River basin is located in the southern part of Fujian Province, with longitudes of 116°50′–118°02′ E and latitudes of 24°12′–25°44′ N. The river basin covers an area of 14,741 km², accounting for about 12% of Fujian Province’s land area. The main stream of the Hanjiang River is 470 km long (from Heyuan to Dongxikou), with an average annual runoff of 24.89 billion cubic meters (1951–2020). The Hanjiang River basin is located in the east of Guangdong and southwest of Fujian at a longitude of 115°13′–117°09′ E and a latitude of 23°17′–26°05′ N. It is the second largest river basin in Guangdong Province after the Pearl River basin, with a river basin area of 30,112 km², of which Guangdong accounts for 60.4% and Fujian accounts for 39.6%.

2.2. Data and Methodology

The precipitation data used in this study include TRMM 3B42 v7 daily precipitation grid data and 3B43 monthly precipitation grid data for the 1998–2019 period, with a spatial resolution of 25 × 25 km and a coverage of 180° W–180° E, 50° S–50° N (https://disc.gsfc.nasa.gov/datasets/TRMM_3B42_Daily_7/summary (accessed on 12 October 2023)). The 3B42 data incorporates precipitation data from several microwave remote sensing measurements. The 3B42 V6 product was generated in three steps: firstly, revising the TRMM/TMI data and integrating the SSMI/I, AMSR-E, and AMSU-B data; secondly, revising the microwave precipitation using the infrared precipitation valuation from the Global Precipitation Climate Program (GPCP); and lastly, integrating the ground-based rain gauge data for the joint valuation of the microwave and infrared data. The 3B42V6 product provides an optimal valuation of precipitation for each grid with a temporal resolution of up to 3 h, and its daily data are averaged over the 3 h data. Compared to 3B42V6, V7 has a more updated algorithm. The major revisions to the algorithm involved the radar reflectivity–rainfall rate relationship, surface clutter detection over high terrain, a new reference database for the passive microwave algorithm, and a higher-quality gauge analysis product for monthly bias correction [4]. The 3B42V7 product has been applied to precipitation and runoff estimation [19], hydrological modeling [8], drought monitoring [5], and other fields. The 3B43 dataset is a globally averaged, month-by-month, gridded global dataset that is a composite of the 3B42 data product, the NOAA Climate Prediction Center’s Climate Anomaly Monitoring System (CAMS), and rain gauge information from the Global Precipitation Climate Center (GPCC).

In addition to the TRMM data, the daily precipitation data from meteorological stations shared by the World Meteorological Organization (WMO) (https://www.ncei.noaa.gov/data/global-summary-of-the-day/archive/ (accessed on 15 October 2023)) were also used in this study, including data from SHAOWU, WUYISHAN, PUCHENG, NANPING, FUZHOU, CHANGTING, YONGAN, ZHANGPING, JIUXIANSHAN, MEIXIAN, GAOQI, and SHANTOU stations, as well as the MEI [20,21] data (https://psl.noaa.gov/enso/mei/ (accessed on 15 October 2023)).

2.3. Validation of Data

The daily precipitation data (for the 1998–2019 period) of 12 meteorological stations within the three major river basins on the west coast of the Taiwan Strait were employed to validate the TRMM data in the study area. The cumulative probability distribution curves of meteorological station precipitation and TRMM precipitation for 1998–2019 are shown in Figure 2a. In the figure, the blue solid line indicates the precipitation data from the meteorological station, and the red dashed line indicates the TRMM data. The cumulative distribution curves of these two types of data show different degrees of agreement between different stations. In general, meteorological stations located in coastal areas or cities usually have a higher degree of agreement; examples of these stations include GAOQI, SHANTOU, NANPING, and so on. For meteorological stations located in mountainous areas or counties, such as SHAOWU, WUYISHAN, CHANGTING, and so on, the degree of agreement is low. The TRMM satellite has a 720 km scanning width, and its data quality should be consistent over the study area. However, meteorological stations in mountainous areas or counties may have unstable data quality due to the weakness of observational equipment or technical capabilities, presenting a lower degree of agreement in the cumulative distribution curve of precipitation.

Therefore, we selected the daily precipitation data from the GAOQI meteorological station located in Xiamen city, a special economic zone, for validation of the TRMM daily precipitation data. During the 1998–2019 period, the GAOQI station had the highest precipitation during the wet period (April–September) in 2006 and the second highest in 2016. Therefore, the daily precipitation data during the wet periods of these two years were selected for comparison and validation, and the results, as shown in Figure 2b, present a good agreement. The correlation coefficients reached 0.86 (2006) and 0.78 (2016), implying that the uncertainty of the TRMM remote sensing data is acceptable. In other words, the TRMM products generally agreed with the weather station data in reflecting the trend and intensity of daily heavy precipitation events. The differences in several dates might have been due to inconsistencies in the spatial and temporal scales of observations between the two datasets. The meteorological stations measured hour-by-hour precipitation data at a particular point, while a grid in the TRMM data represents a spatial extent of 25 × 25 km with a raw time resolution of only 3 h. Considering the instantaneous characteristics of real precipitation, remote sensing data might not be able to reach the accuracy of station data for localized small-scale precipitation observation, but due to the advantages of remote sensing data in terms of properties such as their large area, quasi-real-time sensitivity, high resolution, and time continuity, they still serve as useful datasets with respect to reflecting the spatial structure of and temporal variations in precipitation at large scales. In conclusion, the TRMM data are suitable for the statistical analysis of precipitation on spatial and temporal scales and could enable spatial optimization in the statistical analysis process by virtue of their high spatial resolution and uniform distribution of station-like grid data.

In addition, in this study, we confirmed that the annual precipitation calculated via TRMM agreed quite well with the annual precipitation from the Fujian Provincial Climate Bulletin in terms of precipitation trend and intensity on an annual scale, with a correlation coefficient as high as 0.98 (p ≤ 0.001, n = 23). The TRMM data could reflect the characteristics of precipitation variations at large spatial and temporal scales in the Taiwan Strait and its neighboring regions well, although some slight biases were observed. Several studies have demonstrated the validity and high reliability of TRMM data. Ali et al. [10] evaluated the performance of three TRMM precipitation products (3B42V6, 3B42RT, and 3B42V7) with respect to rain gauge observations in the Hunza River basin, Northern Pakistan. The results showed that the monthly and annual performance of 3B42V7 could be used as an acceptable station precipitation substitute for applications in the study region. Chen et al. [5] demonstrated that the TRMM 3B43 precipitation product provided a new data source for reliable drought monitoring in the Yangtze River basin. Collischonn et al. [22] pointed out that TRMM-based calculated hydrographs are comparable with those obtained using rain gauge data according to the results of hydrological modeling.

3. Annual and Interannual Variability of Precipitation in the Taiwan Strait and Its Neighboring Regions

3.1. Annual Variability of Precipitation

Figure 3a,b represent the spatial distribution and temporal variation histograms of monthly precipitation in the Taiwan Strait and its neighboring regions for the multi-year averages from 1998 to 2019, respectively. It is evident that the precipitation in the Taiwan Strait and its neighboring regions is mainly concentrated in the April–September period, with an average monthly precipitation of >150 mm, accounting for 72.44% of the annual precipitation. In the April–September period, the most precipitation occurred in June, with an average monthly precipitation amount of 296.68 mm, accounting for 16.94% of the year. The main precipitation areas were located in the northern part of Fujian Province, the border between Guangdong and Fujian, and the southwestern part of Taiwan (Figure 3a), where the monthly amount of precipitation could be greater than 300 mm. August was the second-highest precipitation month, and the average monthly precipitation was 234.50 mm. The area with the greatest precipitation was located in Taiwan and its southern sea area, where monthly precipitation could also be greater than 300 mm. Monthly precipitation was lowest in the winter (December, January, and February), accounting for only 11.29% of the yearly total precipitation, and was more evenly distributed across the study area, with most regions receiving no more than 100 mm of monthly precipitation.

As shown in Figure 3b, the annual trends are generally consistent across the region as well as across the three river basins. The average precipitation in the Minjiang River basin was greater than that in the Jiulong River and Hanjiang River basins, except during the July–October period. The annual variability of precipitation in each river basin is strongly characterized by the influence of the East Asian monsoon. In the winter, it was dry, there was less rain under the influence of the northeast monsoon, and dry and cold air flowed in from the northern continent. In the spring, the cold air weakened and retreated northward, and warm and humid air from the oceans strengthened and shifted northward, especially during the turn of the spring and summer in May–June, when the fronts of the two air masses hovered over Fujian Province, producing a large amount of precipitation. This is the main period of precipitation in Fujian Province, especially in the northern part of the Minjiang River basin. In June, the Pearl River Estuary and the southern part of the Taiwan Strait became the precipitation centers. The return of the South China Sea summer winds in May and June not only brought abundant precipitation to the northern part of the South China Sea but also affected areas in the southern China area and generated monsoon convective precipitation [23].

In the summer, all the regions are mainly governed by subtropical high pressure in the northwestern Pacific Ocean and are frequently affected by tropical cyclones (typhoons) that form in the equatorial Pacific Ocean. This makes the precipitation in southeastern Taiwan, Eastern and Southern Fujian, and the coastal zone tend to show the characteristics of being heavy but short in duration, similar to typhoon precipitation. In addition, topography also has a strong influence on the spatial distribution of precipitation. Fujian Province is characterized by hilly terrain, with the Wuyi mountain range towering over the Fujian–Jiangxi border and the Minzhong mountain system (Jiufeng, Daiyun, and Bopingling Mountains) in the central part of the province, with both areas often being the locations of high-precipitation areas. As for Taiwan, precipitation is mainly concentrated in the southern mountainous areas of the island due to the influence of the Central Mountain Range [24].

3.2. Interannual Variation in Precipitation

The variation in annual precipitation in the Taiwan Strait and its neighboring regions from 1998 to 2019 was quite significant, and the interannual trends were not quite the same in the different river basins (Figure 4a,b). Generally, the annual precipitation in 1998, 2006, 2010, 2012, and 2016 was more significant during the 22-year period, with an average annual precipitation of more than 2000 mm for the whole region, which is 17% higher than the total 22-year average precipitation, while the lowest annual precipitation was recorded in 2003, with only 1237 mm. Based on the spatial distribution of annual precipitation (Figure 4b), the Minjiang River basin was generally a strong precipitation area, especially in 2010, 2012, 2015, and 2016, when the amount of precipitation was higher than 2300 mm, and in 1998, 2001, 2002, 2005, and 2006, when it exceeded 2000 mm. In contrast, the frequency of strong precipitation in the Jiulong River and Hanjiang River basins is much lower, with annual precipitation exceeding 2000 mm only in 2006 and 2016. The annual precipitation levels of these two river basins were significantly lower than those of the Minjiang River basin for the vast majority of the 22-year period. In 2016, heavy precipitation occurred in all river basins, with annual precipitation exceeding 2500 mm in the Hanjiang River basin and the Minjiang River basin. In addition, the southern part of Taiwan and its surrounding waters experienced heavy precipitation in several years, including 1998, 2010, 2012, and 2016.

The heavy precipitation that occurred in 1998, 2010, and 2016 was probably related to the El Niño phenomenon in those years. A larger positive MEI value indicated a stronger warm event in the eastern equatorial Pacific. Gong et al. [25] pointed out that the precipitation in the Southern China area would be high in the winter and fall of El Niño years. Our analysis also showed (Figure 5a) that there were significant positive anomalies of precipitation in the winter and fall in these years. El Niño might generate precipitation anomalies in the Fujian region by affecting the location and intensity of subtropical high pressure in the western Pacific Ocean [26]. In 2010, precipitation in the Minjiang River basin significantly increased due to the strong cold air moving southward, and there was a rare heavy rainstorm in February of that year. In the fall, under the combined influence of atmospheric low-level shear and a southwestern warm and humid airflow, heavy rain was frequent and precipitation was unusually high in Fujian, constituting the highest level for the same period since 1961 [27]. In 2016, precipitation peaked in all the river basins, all of whose values were the highest on record since 1961. During this year, there was extraordinarily high precipitation during the winter, the rainy season (May–June), and the typhoon season (July–September). The super El Niño event, which began in September 2014 and ended in May 2016, had the longest duration, highest cumulative intensity, and strongest peak intensity since 1951. There was widespread heavy precipitation over the upper Minjiang River at the beginning of May 2016 (5–10 May), and the last heavy rainfall of the rainy season lasted 7 days (12–18 June). After entering the typhoon season, nine typhoons occurred consecutively. Typhoon Nepartak mainly affected the middle and lower reaches of the Minjiang River, leading to a record-breaking 217 mm of daily precipitation in Minqing. Super Typhoons Moranti and Megi brought extensive heavy rainfall to the Jiulong River basin and Hanjiang River basin.

There was no obvious El Niño signal in 2006 (Figure 5b), but there were also significant heavy rainfall events in this year. There are two main reasons for the high annual precipitation in 2006. The first reason is the intersection of a weak, cold airflow and a southwestward, warm, and humid airflow. In this year, the rainy season arrived earlier, between April and June, with frequent and long-duration heavy rainfall (>100 mm) [27]. The second reason is related to typhoons. For example, in May, the severe typhoon Chanchu’s precipitation intensity was the strongest in history in the same month. In July, the severe tropical storm Bilis, which had the strongest intensity and widest range of precipitation among tropical cyclones that struck Fujian in July since 1956, made landfall [27]. It should be noted that the total annual typhoon precipitation in 2003 was the lowest since 1998 [15], which might be a major reason for the 22-year minimum annual precipitation that occurred in this year.

During the pre-flood season in the Southern China area (February–June), the Minjiang River basin was subjected to frontal precipitation generated by the combined action of dry and cold air from the north and warm and humid air from the tropical oceans. Under the background of the development of the East Asian blocking high-pressure system, the shift of the subtropical high-pressure system in the western Pacific Ocean caused a north–south shift of the rainy zone in the middle and lower reaches of the Yangtze River basin [28], which might be one of the reasons for the interannual variability of precipitation in the Minjiang River basin. Another impact factor might be the South China Sea’s summer winds. There is a significant positive correlation between the intensity index of the South China Sea’s summer winds and precipitation in the Jiangnan region and the later flood season in Southern China [29]. The Jiulong River basin and Hanjiang River basin, in contrast, are mainly affected by tropical weather systems, and most of the heavy precipitation occurs in the later flood season, whose interannual variability in precipitation is likely due to differences in typhoon precipitation [15].

4. EOF Analysis

EOF analysis is a powerful tool in oceanography, providing a robust method for extracting and interpreting patterns of variability within complex oceanic datasets [30,31]. EOF analysis begins with the construction of a covariance matrix, representing the statistical relationships between different spatial points within a dataset. The second step involves the diagonalization of the covariance matrix through eigenvalue decomposition. This process yields a set of eigenvectors and corresponding eigenvalues. The eigenvectors, also known as EOF modes, represent the spatial patterns of variability within the dataset, while the eigenvalues quantify the amount of variance associated with each mode. The first mode captures the largest amount of variance, and subsequent modes explain progressively smaller portions. Once the EOF modes are determined, they can be used to reconstruct the original data by multiplying the corresponding spatial patterns by the associated temporal coefficients. The significance of EOF analysis in oceanography lies in its ability to identify and characterize the major modes of variability within a dataset, effectively reducing its dimensionality while retaining essential information.

EOF decomposition analysis of the TRMM monthly precipitation anomaly was conducted for the Minjiang River basin, the Jiulong River basin, and the Hanjiang River basin. Then, the typical spatial distribution field and the corresponding time coefficients of each river basin were obtained. The first three modes of each river basin passed the 95% significance test, and their variance contributions and cumulative variance contributions are shown in

Table 1. The variance contribution and cumulative variance contribution of the first three modes obtained from the EOF decomposition of precipitation in three river basins.

Name of River Basin	Mode of EOF	Variance Contribution	Cumulative Variance Contribution
Minjiang R.	1	90.80%	90.80%
	2	6.12%	96.93%
	3	2.00%	98.93%
Jiulong R.	1	94.40%	94.40%
	2	4.53%	98.93%
	3	0.47%	99.41%
Hanjiang R.	1	93.35%	93.35%
	2	4.50%	97.85%
	3	1.09%	98.94%

.

4.1. The Minjiang River Basin

The contribution of the first EOF mode in the Minjiang River basin reached 90.80%, which is much higher than that of the other two modes. As shown in Figure 6a, the first mode is positive throughout the Minjiang River basin, with the high-value center located in the Wuyi–Shanling area in the northwestern part of the river basin, and the low-value area located in the southeastern part of the river basin. This typical phenomenon reflects the overall consistency of the spatial distribution of precipitation in the Minjiang River basin, i.e., more rainfall or less rainfall over the whole area, while the variation magnitude was the largest in the northwest zone. The corresponding time coefficients of this mode (Figure 6b) indicated that the precipitation in the Minjiang River basin was mainly produced during the wet season (April–September) and the dry season (the other months). In the rainy season, the variation during May–June was the strongest and fluctuated greatly from year to year, which was significant in 1998 and 2010 and weak in 2003 and 2011. The time coefficients during the winter dry period were negative, with slight interannual variation. During the twenty-two-year period (1998–2019), the variation patterns of precipitation in the Minjiang River basin showed single-peak (2003, 2005, 2006, 2010, 2019, etc.), double-peak (1998, 1999, 2012, 2019, etc.), and multi-peak (2000, 2007, 2014, 2016, etc.) structures.

The intra-annual distribution of precipitation reflected in the first modes is closely related to the topography and monsoon climate. The middle and upper reaches of the Minjiang River covered the northwestern part of Fujian, located between the Wuyi and Jiufeng–Daiyun Mountain Ranges. The orientations of these two mountain ranges are nearly parallel to the coastline, and their windward slopes can lift the warm and humid air from the Pacific Ocean, leading to heavy topographic rain. After the spring, the warm and humid airflow brought by the monsoon converges with the cold airflow from the north in the south of the Wuyi–Shanling area, forming frontal precipitation. This frontal precipitation is abundant, long-lasting, and covers a large area, and the daily precipitation level can be as high as 200–400 mm. The combined effect of topographic precipitation and frontal precipitation resulted in significant high values in the first mode in the Wuyi–Shanling area in the northwest of the Minjiang River basin.

The variance contribution of the second mode was 6.12%, and the northwest–southeast inversion pattern is presented with the Minzhong mountain system serving as the center of spatial precipitation distribution (Figure 6c), i.e., the trends of precipitation changes are opposite in the northwestern and southeastern areas of the river basin. What is more, the variability is the greatest at the northwestern and southeastern ends of the river basin. Considered together with the corresponding time coefficients (Figure 6d), the precipitation pattern in this mode is still generally dominated by seasonal variation. During the spring and winter, the northwestern part of the river basin is rainy, while the southeastern part is in a state of low rainfall; the reverse is true during the summer and fall. In terms of interannual variability, the strong precipitation signal in the northwestern part of the river basin mainly occurred in 1998, which might be related to the strong El Niño phenomenon that occurred in the same year. No significant precipitation change signal was observed in the other years. The interannual variability of precipitation in the summer and fall seasons was significant, most notably in the summer of 2013, followed by the summer of 2006, and this behavior is probably correlated with typhoon precipitation.

The second mode mainly reflects the differences in the spatial distribution of precipitation during the spring, rainy season, and the summer (the later flood season). In the spring, the precipitation in the Minjiang River basin was predominantly frontal precipitation. In the summer typhoon season, tropical cyclones formed in the western Pacific Ocean often made landfall along the coastal areas of Fujian, such as Ningde, Fuzhou, Putian, and so on, bringing rainfall with large quantities and short durations [15]. The difference in precipitation causes is the main reason for the spatial and temporal variance of the second mode. The variance contribution of the third mode was further reduced to 2.00%.; we will not continue the analysis and discussion of this topic in this study.

4.2. The Jiulong River Basin

The first mode obtained from the EOF analysis of precipitation in the Jiulong River basin has a high contribution rate of 94.40%. Like the Minjiang River basin, the first mode exhibited an overall consistent spatial distribution of precipitation (Figure 7a), with the high-value area located in the center of the river basin, i.e., the south area of Bopingling in the southern part of the Minzhong mountain system. In the time dimension, it also has obvious seasonal variations (Figure 7b). As shown, the positive signals during the rainy season and their interannual variations are significant, such as those in 2000, 2006, 2010, and 2017. In addition, the time coefficient curve of the first mode also reflected that there are various types of precipitation structures in the Jiulong River basin, such as single-peak (2002, 2008, 2017, etc.), double-peak (2004, 2006, 2010, 2013, etc.), and multi-peak (2000, 2007, 2019, etc.).

The contribution of the second mode is 4.53%, and the spatial distribution of precipitation is also characterized by a northwest–southeast inversion pattern (Figure 7c). By matching the second mode time coefficient (Figure 7d), we observed that the precipitation in the river basin was concentrated in the southeastern part during the rainy season. In 2005, the time coefficient curve reached its strongest negative signal, which indicated that the precipitation in the northwestern part of the river basin was extremely significant. This corroborated the heavy precipitation event occurring in this area in the specified year. Moreover, the time coefficient curve reached the highest positive signal in 2006, coinciding with the once-in-a-decade typhoon with heavy precipitation along the coast of southeastern Fujian that occurred in the same year. The variance contribution of the third mode is further reduced to 0.47% and is considered to be negligible.

4.3. The Hanjiang River Basin

The first mode obtained from the EOF analysis of precipitation in the Hanjiang River basin has a high contribution rate of 93.35%. Like the other two river basins, the first mode exhibited an overall consistent spatial distribution of precipitation (Figure 8a). As illustrated in the figure, the magnitude of precipitation changes in the Hanjiang River basin gradually increased from north to south and showed some latitudinal changes. In the time dimension, precipitation in the river basin was also predominantly governed by seasonal variations (Figure 8b), i.e., precipitation was concentrated in the rainy season, and this phenomenon was most obvious in 2005, 2007, 2008, and 2017. The second mode (Figure 8c), which contributed 4.50% of the variance, exhibited significant latitudinal differences. According to the time coefficient curve (Figure 8d), this mode’s seasonal variation was slightly different from that of the Minjiang River and the Jiulong River. In the spring, there was a brief but obvious negative signal, at which point frontal precipitation occurred in the northern part of the Hanjiang River basin. After the onset of the rainy season, the negative signal quickly jumped to a positive signal, and the precipitation area shifted to the south. The arrival of the South China Sea summer winds brought a large amount of monsoon precipitation to the Southern China area in May–June of each year [29]. Therefore, the South China Sea summer winds had an earlier impact on the Hanjiang River basin compared to the Minjiang River and Jiulong River basins. These South China Sea summer winds not only affected summer precipitation but also had some influence on rainy-season precipitation, leading to a shift of the precipitation center of the Hanjiang River basin from north to south. The variance contribution of the third mode was further reduced to 1.09%; this finding will not be analyzed or discussed further in this study.

5. Discussion

Due to the combined effects of atmospheric circulation, topography, and other factors, there are north–south differences and east–west (coastal and inland) differences in the spatial distributions of precipitation in all three river basins. The high precipitation areas of the three river basins are generally located in mountainous areas. In the time dimension, there are also significant seasonal differences in precipitation in all three river basins. When the spatial and temporal dimensions are combined, the spatial distribution of precipitation in the spring, the rainy season, and the summer in different river basins exhibits some special features. Figure 9 reflects the multi-year average spatial distribution of precipitation in the spring (March–April), the rainy season (May–June), the summer (July–September), and the fall/winter (October–February) from 1998 to 2019. In the spring (Figure 9a), frontal precipitation and topographic precipitation were predominant, and the high-value areas were mainly located in the northwestern parts of the river basins, especially in the Wuyi–Shanling area. The rainy season accounted for the greatest share of annual precipitation in terms of intensity and breadth (Figure 9b). The spatial distribution and precipitation formation mechanisms in the rainy season are similar to those in the spring. A significantly high amount of precipitation (more than 300 mm) covered the Wuyi–Shanling area as well as the lowlands in the middle and upper reaches of the Minjiang River valley. At the same time, the southwest part of the Hanjiang River basin also experienced high precipitation due to the influence of warm and humid water vapor brought about by the arrival of the South China Sea summer winds. In summer (Figure 9c), the precipitation center of the river basin shifted from the northwest inland to the southeast coast, and the average monthly precipitation was more than 200 mm. Precipitation in the fall and winter (Figure 9d), which accounted for a low percentage of the annual precipitation, was generally evenly spatially distributed. A comparison with the results of the EOF analysis of precipitation in each river basin revealed that the spatial distribution of the first mode in the three river basins is very similar to that of rainy-season precipitation. In other words, the first mode mainly reflected the distribution characteristics of the rainy season.

Since 2004, the changes in annual precipitation in the river basins along the west coast of the Taiwan Strait have become increasingly erratic, and the interannual fluctuations are becoming increasingly pronounced. This confirms the intensification of atmospheric circulation anomalies and climate instability in the region over the past decade. In terms of interannual changes, precipitation in the three river basins fluctuated, but the overall trend did not change much in the spring and fall/winter. Meanwhile, summer precipitation showed a stable and weakly decreasing trend. The interannual variability of rainy season-precipitation in the three river basins is significant, with excessive precipitation occurring in the rainy seasons of 2005, 2006, 2010, and 2016. Rainy-season precipitation in the Minjiang River basin was as high as 481 mm/month in 2005, the highest level in 22 years. The reason for the significant fluctuations in the interannual variability of rainy-season precipitation can be traced back to the north–south shift of the rainy zone in the middle and lower reaches of the Yangtze River basin caused by the annual offset of the subtropical high-pressure system in the western Pacific Ocean [28], which, in turn, affected rainy-season precipitation in the major river basins on the west coast of the Taiwan Strait as well as the north–south distribution of rainy-season precipitation in Fujian [26]. The strength of the South China Sea summer winds also affected summer precipitation in the river basins along the west coast of the Taiwan Strait. The results of correlation analyses (not shown) indicate that summer precipitation in all three river basins was quite consistent with the changes in the index of the South China Sea summer winds.

ENSO is an essential factor influencing the spatio-temporal distribution and interannual variability of precipitation in the analyzed river basins. Figure 10a,b illustrate the spatial distributions of precipitation during El Niño events (MEI ≥ 1) and La Nina events (MEI ≤ −1). During La Nina events, all river basins on the west coast of the Taiwan Strait experienced a lack of precipitation. During El Niño events, precipitation in the Minjiang River basin was the heaviest, with a significant increase over normal years (Figure 10c). However, the southern part of the Hanjiang River basin and the Jiulong River basin exhibited insignificant changes in precipitation. The spatial distributions of EOF first mode (Figure 6a) and of precipitation during El Niño (Figure 10a) in the Minjiang River basin are consistent. Cai et al. [26] pointed out that El Niño may generate precipitation anomalies in the Fujian region by affecting the location and intensity of the subtropical high-pressure system in the western Pacific Ocean. Stephan et al. [32] validated the high positive correlations between ENSO and precipitation in Eastern China in the winter, along the Yangtze River in the summer, and in southeast China during the spring. Similar results were found in this study. In all three strong El Niño years, namely, 1998, 2010, and 2016 (Figure 5b), the Minjiang River basin experienced significant heavy precipitation in the fall and winter, but the Jiulong River basin and the Hanjiang River basin did not (except in 2016). The heavy precipitation in the Jiulong River basin and the Hanjiang River basin in 2016 was mainly due to typhoon precipitation: nine typhoons occurred successively in this year, the most prominent of which were Super Typhoons Moranti and Megi, which brought extensive amounts of extremely heavy rainfall. In conclusion, the Minjiang River basin is more affected by ENSO, while the Jiulong River basin and the Hanjiang River basin are limitedly impacted by ENSO.

6. Conclusions

In this study, daily and monthly precipitation TRMM data from 1998 to 2019 were selected to analyze the precipitation patterns of the Minjiang River basin, the Jiulong River basin, and the Hanjiang River basin on the west coast of the Taiwan Strait. The five important points of this study are as follows:

(1)

This study expands the application of TRMM remote sensing data in mesoscale watershed precipitation studies. The EOF analysis further highlights the advantages of TRMM’s large coverage, huge quantities of data, and continuous time series.

(2)

Our analysis of precipitation annual variation shows that the rainy season (May–June) is the main contributor to precipitation in the Taiwan Strait and its neighboring regions. The analysis of precipitation interannual variability revealed a correlation between regional precipitation and ENSO.

(3)

The EOF analysis of precipitation in the three river basins revealed that the variance contribution of the first mode reached more than 90% and that the values of the first mode were all positive, reflecting the spatial synchronization of precipitation changes in the three river basins. Among them, the spatial distributions of EOF first mode (Figure 6a) and of precipitation during El Niño (Figure 10a) in the Minjiang River basin are consistent, and those in the Jiulong River and Hanjiang River basins are inconsistent. This finding reflects the fact that ENSO is probably the dominant factor of precipitation in the Minjiang River basin but not in the Jiulong River basin or the Hanjiang River basin. The significant increase in precipitation during El Niño compared with normal years in the Minjiang River basin (Figure 10a) and the insignificant increase in precipitation in the Jiulong River and the Hanjiang River basins confirm this point.

(4)

The magnitude of precipitation variation is spatially different in each river basin. The magnitude of precipitation variation is significant in the northwestern part of the Minjiang River basin (Figure 6a), the southwestern part of the Jiulong River basin (Figure 7a), and the southwestern corner of the Hanjiang River basin (Figure 8a). These areas happen to correspond to the Wuyi–Shanling Mountain area, the Bopingling Mountain area, and the Lianhua Mountain area, respectively (Figure 1). This finding reveals that topographic precipitation plays a reinforcing role in the spatial distribution of precipitation in the Minjiang River basin, while it probably plays a dominant role in the Jiulong River basin and the Hanjiang River basin.

(5)

The MEI indicated three strong El Niño years: 1998, 2010, and 2016 (Figure 5b). By incorporating the EOF first mode time series of precipitation in the Minjiang River basin (Figure 6b), we found that the Minjiang River basin experienced significant heavy precipitation in the fall and winter in all three strong El Niño years, whereas the Jiulong River basin and the Hanjiang River basin did not (except in 2016). In conclusion, the Minjiang River basin is more affected by ENSO, while the Jiulong River basin and the Hanjiang River basin are only limitedly impacted by ENSO.