ORIGINAL RESEARCH article

Front. Clim., 15 October 2026

Sec. Predictions and Projections

Volume 8 - 2026 | https://doi.org/10.3389/fclim.2026.1937736

Characterization and regionalization of dry spells across Java-Indonesia using relative frequency and fractional contribution to total duration

  • 1. South Sumatera Climatological Station, Indonesia Agency for Meteorology, Climatology and Geophysics (BMKG), Palembang, Indonesia

  • 2. Directorate of Climate Change, Indonesia Agency for Meteorology, Climatology and Geophysics (BMKG), Jakarta, Indonesia

  • 3. Department of Meteorology and Geophysics, IPB University, Bogor, Indonesia

Abstract

Dry spells represent an important dimension of rainfall variability because rainfall accumulation alone may not adequately capture the temporal distribution of rainfall and its associated impacts. This study characterizes and regionalizes dry spells across Java, Indonesia, using relative frequency (RF) and fractional contribution to total duration (FCTD). Daily rainfall data from 771 BMKG stations across Java for 2010–2019 were analyzed to characterize dry-spell duration. RF and FCTD were calculated for different dry-spell durations. Principal component analysis (PCA) and K-means clustering were subsequently applied to regionalize dry-spell characteristics. Short-duration dry spells (≤10 days) exhibited the highest RF throughout Java, whereas FCTD revealed pronounced spatial variability. Western Java was characterized by a greater contribution from short-duration dry spells, while eastern Java showed a greater contribution from long-duration dry spells (>10 days). A distinct north–south gradient was also identified, reflecting the influence of local topography, with long-duration dry spells contributing more substantially in northern regions. Dry-spell characteristics closely corresponded to the spatial distribution of mean annual rainfall, with wetter regions dominated by short-duration dry spells and drier regions by long-duration dry spells. Mean annual rainfall showed a moderately strong positive correlation with the FCTD of short-duration dry spells (Pearson’s r = 0.657). Regionalization identified four clusters representing distinct dry-spell regimes across Java. The findings demonstrate that combining RF, FCTD, and multivariate clustering provides a physically meaningful framework for characterizing the spatial variability of dry spells across Java, capturing both the west–east rainfall contrast and the north–south topographic gradient.

1 Introduction

Dry spell represents the day-to-day variability of rainfall based on consecutive non-rainy periods (Ratan and Venugopal, 2013; Li et al., 2016). As part of rainfall characteristics, dry spells provide important information on the temporal distribution of rainfall that accumulation-based metrics alone cannot capture. Trenberth (2011) demonstrated that similar monthly rainfall totals can mask substantially different rainfall regimes, as a month dominated by a prolonged dry spell followed by intense rainfall may pose greater risks than one with more evenly distributed precipitation. Thus, rainfall accumulation alone may underestimate climate-related impacts associated with the temporal distribution of rainfall (Ferijal et al., 2021; Bal et al., 2022).

This distributional dimension of rainfall is becoming increasingly consequential under a warming climate, as global and regional climate models project changes in rainfall distribution and dry-spell characteristics (Pendergrass et al., 2017; Intergovernmental Panel on Climate Change (IPCC), 2023). Although projected changes vary spatially, several studies indicate an increase in dry spell duration, particularly during dry periods (Sushama et al., 2010; Bouagila and Sushama, 2013; Supari et al., 2020; Marinović et al., 2021; Wainwright et al., 2021). In Indonesia, projected changes are reflected in an increase in the maximum duration of consecutive dry days (CDD), with a stronger tendency during the dry season (JJA and SON) (Supari et al., 2020). Longer dry spells during already dry periods can compound drought severity, as rainfall deficits coincide with the high evaporative demand typical of the dry season in the region (Breinl et al., 2020).

In addition to long-term climate change, dry-spell variability over Java is influenced by regional and large-scale climate processes, including monsoon variability, the El Niño–Southern Oscillation (ENSO), and the Indian Ocean Dipole (IOD). The monsoon system plays an important role in determining the seasonal distribution of rainfall over Java, while its interaction with the island’s complex topography contributes to spatial differences in rainfall between windward and leeward regions (Tjasyono, 2004; Qian et al., 2010). ENSO can further modify monsoon circulation and rainfall over Java, with El Niño conditions generally associated with drier conditions, particularly during the dry and transition seasons. However, the rainfall response to ENSO can vary spatially because of interactions between large-scale circulation and local land–sea and mountain–valley circulations (Qian et al., 2010). The IOD also contributes to rainfall variability over Indonesia, particularly during JJA and SON, with its influence varying spatially across the Indonesian region (As-syakur et al., 2014). The combined influence of these large-scale climate modes and monsoon variability provides an important climatological context for understanding the spatial variability of dry-spell characteristics across Java.

Dry-spell variability has direct implications for agricultural production, which is strongly influenced by both the seasonal availability and temporal distribution of rainfall. Rain-fed agricultural systems are especially sensitive to prolonged dry spells, while irrigated systems may also experience water stress where irrigation infrastructure or water distribution is insufficient (Sivakumar, 1992; Surmaini and Syahbuddin, 2016). The timing and duration of dry spells can be particularly consequential for crops because rainfall deficits occurring during sensitive growth stages may cause greater stress than comparable anomalies in total seasonal rainfall(Kumar et al., 2006; Bichet and Diedhiou, 2018; Gobin and Van De Vyver, 2021; Thoithi et al., 2021). Prolonged dry spells may also affect water resources through reduced surface runoff and increased evaporative losses, potentially placing additional pressure on reservoirs and other water-storage systems (Hettiarachchi et al., 2022). These conditions are of particular concern in Java, as the major rice-producing region of Indonesia (Prasada et al., 2022), and contains extensive agricultural areas that depend on both rainfall and irrigation. Consequently, detailed information on dry-spell characteristics is relevant for agricultural production, water-resource management, and the assessment of climate-related risks across the island.

Given these implications, characterizing the frequency and duration structure of dry spells is essential for understanding their potential impacts. Dry spell characteristics are commonly described using statistical indicators such as event frequency, mean duration, and maximum duration (Huang et al., 2015). Studies from different regions have shown that the occurrence and contribution of dry spells vary substantially across duration classes. For example, previous studies in West Africa, Brazil, and Singapore have found a predominance of short-duration dry spells, although the duration classes contributing most to the total number of dry days differ among regions (Li et al., 2016; Froidurot and Diedhiou, 2017; Oliveira et al., 2021). In Indonesia, however, dry spell studies have focused predominantly on annual maximum duration, particularly through the CDD index (Supari et al., 2018; Ramadhan et al., 2022). While maximum duration provides information on the most extreme dry-spell event, it does not describe the occurrence or contribution of shorter and more frequent events. Dry spells shorter than the annual maximum may nevertheless be important for assessing rainfall-related water stress, particularly when changes in total rainfall accumulation are not apparent (Nieuwolt, 1989; Gong et al., 2005). Moreover, existing studies in Indonesia have generally been conducted over relatively limited areas or as part of global analyses using coarser-resolution datasets, constraining the characterization of spatial variability in dry spell regimes (Ratan and Venugopal, 2013; Mandapaka et al., 2016). A more detailed characterization across a range of dry spell durations and at a finer spatial scale is therefore needed, particularly for Java.

To address this gap, this study provides a duration-based characterization of dry-spell regimes across Java using two complementary metrics: relative frequency (RF) and fractional contribution to total duration (FCTD). RF quantifies the proportion of dry-spell events occurring at a given duration, whereas FCTD quantifies the contribution of each duration class to the total number of dry days. Their combined use enables both the frequency structure and cumulative contribution of different dry-spell durations to be evaluated, providing information that is not captured by maximum-duration indices alone. Unlike drought indices such as the Standardized Precipitation Index (SPI), which quantify precipitation anomalies over specified accumulation periods, RF and FCTD directly describe the occurrence and duration structure of consecutive dry periods. Because the objective of this study is to characterize the frequency structure and cumulative contribution of dry spells across a wide range of durations, RF and FCTD provide a suitable framework for describing dry-spell regimes rather than drought severity alone. The study further examines the relationship between dry-spell characteristics and regional rainfall climatology and regionalizes rainfall stations according to their dominant dry-spell characteristics using principal component analysis (PCA) and K-means clustering. This integrated approach provides a spatially explicit characterization of dry-spell regimes and their regional climatic context across Java.

2 Study area and dataset

2.1 Study area

The study area covers Java Island, located between 5°52′34′′–8°46′49′′ S and 105°05′59′′–114°36′18′′ E. Geographically, Java Island is bordered by the Java Sea to the north, Bali Strait to the east, Indian Ocean to the south and the Sunda Strait to the west. The Landscape of Java Island includes mountainous areas, highlands, lowlands, and coastal regions.

Based on the mean annual rainfall cycle, Java Island is classified as a monsoonal rainfall type, characterized by a single wet-season and a single dry-season, influenced by the West Monsoon (November to March), and the East Monsoon (May to September) (Aldrian and Susanto, 2003). In addition to this regional monsoonal influence, the topographic conditions of Java Island also act as a local factor affecting the daily patterns of wind and rainfall. This local influence is strengthened during periods of weak monsoon activity, such as during an active El Niño phase of ENSO. Under these conditions, rainfall patterns across Java become more heterogeneous, producing wet anomalies in the mountainous southern region and dry anomalies in the north (Qian et al., 2010).

2.2 Dataset

This study employed daily rainfall data obtained from a surface-based rainfall observation network operated by Indonesia’s Agency for Meteorology, Climatology, and Geophysics (BMKG) across Java Island (Figure 1). The initial dataset comprised 966 rainfall stations, which were subjected to a quality control procedure following the methodology described by Wibawanty et al. (2025). This process flagged and removed stations based on four criteria: data duplication over 365-day and 28-day windows, zero rainfall recorded throughout the DJF season, and unrealistically prolonged dry spells. After QC, 771 stations (79.8% of the initial network) were retained for analysis, covering the period 2010–2019. The spatial distribution of the retained stations was not perfectly uniform: post-QC station density was higher in eastern Java, with comparatively sparser coverage in the western and central parts of the island (Figure 1).

Figure 1

3 Methods

The analysis consists of three steps: (1) identification of dry spell characteristics based on RF and FCTD, (2) clustering of dry spells, and (3) transect analysis of rainfall and topography.

3.1 Dry spell characteristics

In this study, a daily rainfall threshold of 1 mm is used to define a rainy day. This threshold has been widely adopted in previous studies and is recommended by the Expert Team on Climate Change Detection and Indices (ETCCDI) for analyses related to extreme conditions (Omay et al., 2024). A dry spell was therefore defined as a sequence of consecutive days with daily rainfall below 1 mm. Dry-spell characteristics were quantified using relative frequency (RF) and fractional contribution to total duration (FCTD) metrics (Li et al., 2016). The calculation of RF and FCTD are given in Equations 1 and 2, where represents the frequency of dry spells with duration , while denotes the maximum dry-spell duration. RF and FCTD were first calculated annually for each station, and the annual values were subsequently averaged over the 2010–2019 period to obtain the mean annual RF and FCTD for each station. These mean values were then used for spatial analysis and boxplot visualization.

In addition, dry spells were classified into two duration categories (≤10 days and >10 days) to facilitate the interpretation of the relative contribution of shorter and longer dry-spell events. The 10-day threshold was adopted based on the reported potential effects of dry periods exceeding 10 days on the productivity of shallow-rooted, rain-fed crops under tropical conditions (Nieuwolt, 1989). No sensitivity analysis using alternative thresholds was conducted in this study; therefore, the 10-day threshold should be interpreted as an operational classification rather than a universal agro-climatic threshold.

To characterize the dominant dry-spell duration at each station, FCTD values were cumulatively summed from the shortest to the longest dry-spell duration. The shortest duration at which cumulative FCTD reached 60% was identified as the representative dominant duration. The 60% threshold was adopted as an operational criterion to identify the duration range accounting for a clear majority of the total dry-spell contribution. This threshold was selected for comparative and interpretive purposes and was not intended to represent a universal physical or agricultural threshold. Therefore, the resulting duration should be interpreted as an indicator of the dominant cumulative contribution of dry-spell durations rather than as a critical threshold.

3.2 Dry spell clustering

The dry-spell clustering procedure begins with principal component analysis (PCA), followed by non-hierarchical clustering using the K-means algorithm.

3.2.1 Principal component analysis (PCA)

PCA is a multivariate data analysis method that aims to extract significant information from the data and represent it in a new set of orthogonal coordinates, known as principal components (Abdi and Williams, 2010). The use of PCA for regionalization of dry-spell characteristics has also been demonstrated using station-based rainfall observations. Serra et al. (2014), for example, applied PCA to four L-moments describing dry-spell-length distributions at 267 European rain gauges before regionalizing the stations into homogeneous groups. In their analysis, PCA reduced redundant information among the original dry-spell descriptors before the clustering procedure.

In general, PCA involves the following steps: (1) standardizing the variables and estimating the covariance matrix, (2) computing the loading matrix, and (3) identifying significant principal components (Azam et al., 2017). Several rules of thumb are commonly used to determine the number of significant principal components, including cumulative variance analysis and eigenvalue-based criteria. Cumulative variance analysis is applied using an 80% threshold to determine the number of significant principal components (Jolliffe, 2002). The eigenvalue-based criterion retains principal components with eigenvalues greater than 1, as such components explain more variance than a single original variable (Kaiser, 1960). In addition to these criteria, the number of principal components is also determined using parallel analysis, which retains components with eigenvalues greater than those obtained from random data (Franklin et al., 1995).

In this study, PCA was applied to three annual dry-spell characteristics at each of the 771 rainfall stations: FCTD for dry spells with durations ≤10 days, FCTD for durations >10 days, and the dry-spell duration required to account for at least 60% of cumulative FCTD. These variables represent complementary aspects of the duration structure of dry spells but may contain overlapping information. PCA was therefore used to identify their dominant combinations and minimize the influence of multicollinearity before clustering. Prior to PCA, the variables were standardized to zero mean and unit variance so that differences in measurement scale did not affect the resulting components.

3.2.2 Clustering method

The principal components obtained from PCA are subsequently used as input variables for clustering using the K-means algorithm. K-means is an unsupervised, partition-based clustering method that directly divides the data into a predefined number of groups by optimizing a specified objective function (Halkidi et al., 2001). An unsupervised approach is appropriate because homogenous dry spell regimes across Java were not known a priori. The objective of K-means is to identify groups of rainfall stations with similar dry-spell characteristics.

K-means clustering was implemented using the k-means++ initialization method. To reduce the influence of centroid initialization on the final solution, 30 independent initializations were performed for each specified number of clusters. For each initialization, a maximum of 500 iterations was allowed, with the algorithm terminating when the convergence criterion was satisfied or the maximum number of iterations was reached. The random seed was fixed at 42 to ensure reproducibility. The solution with the lowest WSS among the 30 initializations was retained.

The K-means clustering algorithm is generally implemented through the following procedure: (i) specification of the number of clusters; (ii) random initialization of cluster centroids and assignment of each observation to the nearest centroid; (iii) updating of centroid locations by computing the mean of observations within each cluster; (iv) reassignment of observations based on the updated centroids; and (v) iterative repetition of steps (ii)–(iv) until convergence is achieved, as indicated by no further change in centroid positions (Gao et al., 2023).

Among available clustering approaches, K-means was selected because it provides a direct partitioning of observations into mutually exclusive groups based on their distance from cluster centroids. This characteristic is suitable for the present objective, in which each rainfall station is assigned to one dominant dry-spell regime. K-means also allows observations to be reassigned during successive iterations as cluster centroids are updated, allowing the partition to be optimized rather than fixing cluster membership at an earlier stage. This flexibility is an advantage over hierarchical methods, in which observations assigned to a branch cannot subsequently be reassigned. Zhang et al. (2016) highlighted this distinction in hydroclimatic regionalization and found that K-means produced stable cluster patterns and relatively high within-cluster similarity in their precipitation regionalization analysis.

The number of clusters was evaluated using the within-cluster sum of squares (WSS), which measures the total squared distance between observations and the centroid of their respective clusters. The WSS is expressed in Equation 3, where k is the number of clusters, g represents an observation within cluster j, tg is the value of observation g, and tj is the centroid of cluster j.

The elbow method was used to identify the candidate number of clusters by examining the point at which increasing the number of clusters resulted in progressively smaller reductions in WSS. Because identification of the elbow may involve visual interpretation, the silhouette coefficient was additionally calculated to evaluate the quality of each cluster solution. The silhouette coefficient measures the similarity of an observation to other observations within its assigned cluster relative to its similarity to observations in the nearest neighboring cluster. It ranges from −1 to 1, with higher values indicating greater within-cluster cohesion and between-cluster separation. The silhouette coefficient was used as an additional measure of cluster quality and to assess the robustness of the WSS-based selection.

3.3 Transect analysis of rainfall and topography

To investigate the possible influence of topography on the north–south spatial pattern of dry-spell characteristics, three north–south transects representing western, central, and eastern Java were examined. The transects followed those used by Tjasyono (2004). Rainfall, elevation, and the number of rainy days were analyzed along each transect during the southeast monsoon period (June–August) for 2010–2019. A rainy day was defined as a day with rainfall ≥1 mm.

4 Result and discussion

4.1 Dry spell distribution

4.1.1 Annual distribution of RF and FCTD

Figure 2 presents boxplots of the mean annual RF across the 771 rainfall stations in Java Island for the period 2010–2019. The mean annual RF decreased with increasing dry-spell duration, indicating that shorter-duration dry spells occurred more frequently than longer-duration dry spells. One dry day exhibited the highest mean annual RF, ranging from 15.0 to 52.0% across the observation stations, with an average of 37.2%. This was followed by two-day and three-day dry spells, with mean annual RF ranging from 11.0 to 28.2% (average: 19.4%) and from 6.0 to 18.6% (average: 11.5%), respectively. These findings are consistent with those reported by Li et al. (2016) in Singapore and Mandapaka et al. (2016) for Jakarta, who likewise found that one- to three-day dry spells dominated the dry spell frequency distribution. A similar predominance of short-duration dry spells has also been reported in the Mediteranian (Caloiero and Coscarelli, 2020).

Figure 2

Figure 3 presents boxplots of the mean annual FCTD across rainfall stations in Java Island. The mean annual FCTD decreased rapidly with increasing dry-spell duration up to approximately 23 days. Beyond this duration, the FCTD values generally increased with dry-spell duration among the stations that experienced such prolonged events. This result indicates that, although long-duration dry spells occurred at relatively few stations, they contributed substantially to the total dry-spell duration at those locations. Overall, the results highlight the spatial heterogeneity of dry-spell characteristics across Java Island.

Figure 3

Overall, the greatest contribution to the total dry-spell duration was associated with two-day dry spells, with mean annual FCTD ranging from 2.2 to 19.6% (average: 7.7%). This was followed by one dry day, with FCTD ranging from 1.6 to 20.7% (average: 7.6%), and three-day dry spells, with values ranging from 2.0 to 14.2% (average: 6.7%). Two- and three-day dry spells accounted for the highest FCTD at 42.9% of the observation stations, indicating that short-duration dry spells contributed most to the total dry-spell duration across the majority of Java Island. Although the mean FCTD increased for dry-spell durations exceeding approximately 23 days, these high values were confined to only a few stations. The highest mean FCTD values were observed for dry spells lasting 234, 270, and 262 days, with mean FCTD values of 14.6%, 8.0%, and 7.9%, respectively. These extreme-duration events occurred at only one or two stations and were concentrated mainly in eastern Java, where mean annual rainfall is generally lower. Some events coincided with the El Niño years of 2015 and 2019, suggesting that large-scale climate variability may have contributed to their persistence. However, this relationship was not explicitly tested in the present study and therefore should not be interpreted as a causal effect of El Niño.

To further examine the contribution of short- and long-duration dry spells, FCTD was analyzed by dividing dry-spell duration into two categories: ≤10 days and >10 days (Figure 4A). Dry spells with durations of ≤10 days exhibited FCTD values ranging from 0.3 to 19.6%, with a mean of 4.4%. In contrast, dry spells lasting >10 days had FCTD values ranging from 0.4 to 14.6%, with a mean of 2.3%. Nevertheless, at 25.4% of the observation stations, dry spells longer than 10 days contributed the largest proportion of the total dry-spell duration, as indicated by the highest FCTD values. Among these stations, the dominant long-duration dry spells ranged from 30 to 270 days, with a mean duration of 191 days.

Figure 4

Dry spells lasting more than 10 days were further classified into five percentile-based groups according to the distribution of dry-spell duration (>10 days): <25th, 25th–50th, 50th–75th, 75th–90th, and >90th percentiles (Figure 4B). The distribution of FCTD was then examined for each duration group. The results show that the highest FCTD values were predominantly associated with the upper percentile groups of long-duration dry spells. This finding suggests that the extended duration of these dry spells compensates for their relatively low occurrence frequency, resulting in a substantial contribution to the total dry-spell duration. A similar pattern was reported by Li et al. (2016) in Singapore, where dry spells lasting more than 6 days contributed approximately 30% of the total dry-spell duration, exceeding the contributions of 1-day (15%) and two-day (16%) dry spells. Nevertheless, these long-duration dry spells accounted for less than 10% of all dry-spell events, as indicated by their low RF.

4.1.2 Dry spell spatial distribution

The spatial distribution of FCTD for dry spells with durations ≤10 days and >10 days is shown in Figure 5. The two duration classes exhibit complementary spatial patterns. Stations dominated by short-duration dry spells (≤10 days) are characterized by high FCTD values for the ≤10-day class and correspondingly low FCTD values for the >10-day class. Conversely, stations dominated by long-duration dry spells (>10 days) exhibit the opposite pattern. Since FCTD represents the fractional contribution of each duration class to the total dry-spell duration, the FCTD values of the two classes sum to 100% at each station.

Figure 5

Overall, western Java was characterized by a greater contribution from short-duration dry spells (≤10 days), particularly across its central region. A north–south spatial gradient was also evident, with the northern coastal areas exhibiting lower FCTD values for dry spells of ≤10 days than the southern areas. In contrast, eastern Java was generally characterized by a greater contribution from long-duration dry spells (>10 days). A similar north–south gradient was observed in this region, where the northern coastal areas exhibited higher FCTD values for dry spells lasting more than 10 days than the southern areas. The central part of Java exhibited transitional characteristics, with short-duration dry spells (≤10 days) still contributing the largest proportion of the total dry-spell duration. However, this pattern should be interpreted with caution because the relatively limited number of observation stations may not adequately represent the spatial variability across central Java.

The cumulative FCTD was further analyzed to identify the dry-spell duration required to account for 60% of the total dry-spell duration at each observation station (Figure 6). Across much of western Java, particularly its central region, dry spells lasting up to 10 days accounted for 60% of the total dry-spell duration. In the remaining parts of western Java, the cumulative FCTD generally reached 60% within dry-spell durations of up to 20 days. Only a few locations in the northern and southern parts of western Java required dry spells exceeding 20 days, with the maximum duration reaching 39 days. The central part of Java was generally characterized by dry spells of up to 20 days accounting for 60% of the total dry-spell duration. In contrast, eastern Java required substantially longer dry-spell durations to reach the same cumulative contribution, with durations exceeding 20 days dominating most stations. Along the northern coast of eastern Java, dry spells of up to 197 days were required to account for 60% of the total dry-spell duration. This tendency toward longer dry spells in eastern Java is consistent with historical observations during 1991–2020, which recorded a maximum dry-spell duration of 298 days in the region. According to BMKG (2022), this exceptionally long dry spell was associated with the 2009 El Niño event.

Figure 6

4.1.3 Regional rainfall patterns and dry-spell characteristics

The spatial distribution of dry-spell characteristics across Java Island showed a spatial correspondence with regional rainfall climatology. Figure 7 presents the spatial distribution of mean annual rainfall during the 1991–2020 climatological period. Overall, the spatial pattern of mean annual rainfall corresponded to the spatial distribution of mean annual FCTD, particularly the north–south gradient observed in western and eastern Java. The northern coastal areas of western and eastern Java generally received lower annual rainfall than the surrounding regions and were characterized by higher FCTD contributions from dry spells lasting >10 days.

Figure 7

Regions with lower mean annual rainfall generally exhibit higher FCTD values associated with dry spells lasting >10 days, whereas regions with moderate to high mean annual rainfall are characterized by higher FCTD values associated with dry spells lasting ≤10 days. This spatial correspondence suggests a close relationship between mean annual rainfall and the dominant dry-spell characteristics across Java Island. This spatial correspondence is further supported by a moderately strong positive Pearson correlation (r = 0.657, p < 0.001, n = 771; Supplementary Figure 4) between mean annual rainfall class and the FCTD of dry spells lasting ≤10 days.

The relationship between FCTD and annual rainfall varied across Java Island. In western Java, dry spells lasting >10 days were primarily associated with areas of low annual rainfall, whereas dry spells lasting ≤10 days contributed more substantially in areas with moderate to high annual rainfall. In central Java, greater variability in FCTD was observed within the moderate rainfall category, as several locations exhibited relatively high contributions from dry spells lasting >10 days. In eastern Java, dry spells lasting >10 days contributed substantially to the total dry-spell duration in both low- and moderate-rainfall regions. These results indicate that dry-spell characteristics can vary even among areas receiving similar amounts of annual rainfall. Moreover, a clear west-to-east gradient was evident, with the contribution of long-duration dry spells increasing progressively toward eastern Java.

The observed relationship between rainfall climatology and dry-spell characteristics is consistent with findings from previous studies. Gong et al. (2005) reported that short-duration dry spells occur more frequently during wetter seasons, whereas long-duration dry spells are less common during the same period. Similarly, Caloiero and Coscarelli (2020) and Ratan and Venugopal (2013) demonstrated that regional rainfall regimes influence the contribution of different dry-spell durations to total dry-spell duration. These findings support the pattern observed in Java, where regions with relatively higher rainfall generally showed a greater contribution from short-duration dry spells, while drier regions tended to exhibit a greater contribution from long-duration dry spells. A similar west–east increase in dry-spell variability has also been reported for the Wei River Basin in China, where spatial variations in dry-spell characteristics were associated with regional rainfall patterns (She et al., 2016).

The north–south gradient observed in the FCTD distribution is likely associated with the influence of orographic rainfall. Java Island is characterized by a mountain range extending from west to east across its central and southern regions. This asymmetric topography interacts with the southeast monsoon, forcing moist air to ascend over the southern mountain slopes (windward side), where precipitation is enhanced through orographic lifting. As the air descends toward the northern side (leeward side), it becomes warmer and drier, resulting in lower annual rainfall and a greater contribution of long-duration dry spells.

The influence of topography on rainfall distribution and rainy days across Java Island was previously investigated by Tjasyono (2004) using three north–south transects located in western, central, and eastern Java. The analysis was based on rainfall observations collected during the southeast monsoon (June–August) in 1979. In the present study, a comparable north–south transect framework was applied to rainfall data from 2010–2019, with observation stations selected independently for the western, central, and eastern sections of Java. This approach was used to assess whether the topographic influence identified by Tjasyono (2004) remains evident under more recent climatic conditions. Figure 8 shows the locations of the three transects, where Transects I, II, and III represent eastern, central, and western Java, respectively. For each transect, elevation, rainfall, and the number of rainy days were analyzed, as presented in Figure 9.

Figure 8

Figure 9

Figure 9 presents the north–south variation in elevation, rainfall, and the number of rainy days along Transect I, extending from Parigi (Pangandaran) in southern Java to Sukra (Indramayu) on the northern coast. During the southeast monsoon period (June–August), rainfall reached 867 mm in the southern region and gradually decreased northward to a minimum of 72.5 mm (Supplementary Table 1). A similar pattern was observed for the number of rainy days, which declined from 25 days in the south to only 6 days in the northern coastal region. This pattern reflects the influence of the southeast monsoon interacting with the mountainous terrain of southern Java. Moist air is forced to ascend over the windward slopes, enhancing orographic rainfall in the southern region. As the air descends toward the lower-elevation northern plains (the leeward side), precipitation decreases, resulting in lower rainfall totals and fewer rainy days. A similar topographic influence was also observed in eastern Java along Transect III (Supplementary Figure 2).

Unlike the western and eastern parts of Java, a distinct north–south gradient in dry-spell characteristics was not observed along the central Java transect (Supplementary Figure 1). Rainfall decreased from the southern coast toward the central part of the transect, reaching a minimum of 90.8 mm, before increasing again to 277.4 mm toward the northern coast (Supplementary Table 2). This pattern is associated with the presence of mountainous terrain close to the northern coast, particularly Mount Muria (Figure 8), which is located near Transect II. The proximity of the mountain enhances local atmospheric circulation, including sea-breeze and valley-breeze systems, which promote moisture convergence and orographic uplift during the southeast monsoon. As a result, cloud development and rainfall are enhanced over the northern part of the transect despite the prevailing dry-season conditions.

4.2 Identifying dry spell spatial patterns

The number of principal components (PCs) retained for the regionalization analysis was determined using three commonly adopted criteria: cumulative explained variance, the Kaiser criterion (eigenvalue >1), and parallel analysis. The cumulative explained variance reached the 80% threshold with the first principal component (PC1). Both the Kaiser criterion and parallel analysis likewise indicated that a single principal component was sufficient to represent the variability in the dataset. The consistency among the three selection criteria supported the retention of one principal component for the subsequent regionalization analysis.

Dry-spell regionalization was performed using an unsupervised K-means clustering algorithm to group rainfall stations with similar dry-spell characteristics. The number of clusters was evaluated using both the within-cluster sum of squares (WSS) and the silhouette coefficient. The elbow analysis of WSS indicated four clusters as a parsimonious solution, beyond which the reduction in within-cluster variation became progressively smaller (Figure 10). The silhouette analysis, however, produced its highest mean coefficient for seven clusters (Supplementary Figure 3), indicating somewhat stronger internal cluster separation at this level. Although the seven-cluster solution yielded a higher silhouette coefficient, its finer spatial partitioning was not accompanied by a similarly clear large-scale climatic interpretation, whereas the four-cluster solution produced broader and more spatially coherent regions. Therefore, four clusters were retained to balance within-cluster similarity with the interpretability and parsimony of the resulting regionalization.

Figure 10

The spatial distribution of the K-means clustering results and the corresponding FCTD distributions for each cluster are presented in Figures 11 and 12, respectively. Cluster 1, which covers most of western Java, was characterized by the highest contribution from short-duration dry spells (≤10 days). This pattern is consistent with the spatial distribution of mean annual rainfall shown in Figure 7, where relatively high annual rainfall is observed in several areas of western Java. Cluster 2 contained the largest number of observation stations and was characterized by a relatively balanced contribution from dry spells lasting ≤10 days and >10 days. This cluster includes the northern part of western Java, central Java, and portions of eastern Java. Cluster 3 exhibited slightly greater contributions from dry spells lasting >10 days than from those lasting ≤10 days and was primarily distributed across the northern coast of western Java and several areas of eastern Java. Cluster 4 was characterized by the greatest contribution from long-duration dry spells (>10 days), indicating a stronger influence of prolonged dry spells on the total dry-spell duration. This cluster was mainly located along the northern coast of eastern Java, which is also characterized by relatively low mean annual rainfall.

Figure 11

Figure 12

The differences in dry-spell characteristics among clusters were further assessed using the Kruskal–Wallis test. The median FCTD contribution from dry spells exceeding 10 days increased progressively from 38.28% in K1 to 69.85% in K4, while the median of FCTD 60% increased from 10 days in K1 to 124 days in K4. Significant differences were identified among the four clusters for both FCTD contribution from dry spells exceeding 10 days (H(3) = 680.79, p < 0.001, ε2 = 0.884) and FCTD 60% (H(3) = 646.47, p < 0.001, ε2 = 0.839). Dunn’s post-hoc test with Holm adjustment further indicated significant differences for all pairwise comparisons among clusters (p < 0.001). A summary of the Kruskal–Wallis and Dunn’s post-hoc test results is provided in Supplementary Table 3. These results provide statistical evidence that the four clusters represent distinct dry-spell characteristics.

The spatial distribution of the clusters (Figure 11) was consistent with the observed dry-spell characteristics and the underlying physical mechanisms discussed in the previous sections. External validation showed a significant association between cluster membership and mean annual rainfall category during 1991–2020 (χ2(6) = 239.37, p < 0.001; Cramer’s V = 0.394). Cluster 1 was strongly associated with the ≥3,000 mm rainfall category, whereas Cluster 4 was predominantly associated with rainfall below 1,500 mm. Clusters 2 and 3 were mainly characterized by rainfall between 1,500 and 3,000 mm (Supplementary Figure 5). Interpreting clustering results in the context of underlying physical processes is essential for establishing the climatological meaning of the identified clusters (Badr et al., 2015; Zhang et al., 2016). This is particularly important because clustering is an unsupervised learning technique that groups observations based solely on data similarity, without prior knowledge of the true class labels.

5 Concluding remarks

Dry-spell characteristics across Java Island are dominated by frequent short-duration events, as indicated by the high relative frequency (RF) of dry spells lasting 1 to 3 days. However, the fractional contribution to total duration (FCTD) revealed substantial spatial variability, with western Java primarily characterized by short-duration dry spells (≤10 days) and eastern Java by long-duration dry spells (>10 days). The spatial distribution of FCTD also captured a distinct north–south gradient, reflecting the influence of local topography and orographic processes.

The spatial variability of dry-spell characteristics was closely associated with the rainfall climatology of Java Island. Regions with lower mean annual rainfall generally exhibited greater contributions from long-duration dry spells, whereas regions with higher annual rainfall were dominated by short-duration dry spells. This relationship was supported by a moderately strong positive correlation (r = 0.657) between mean annual rainfall class and the FCTD of short-duration dry spells, indicating that regional rainfall regimes play an important role in shaping dry-spell characteristics.

Regionalization using principal component analysis and K-means clustering identified four distinct dry-spell regions across Java. The resulting clusters successfully represented the major spatial patterns of dry-spell characteristics, including the west–east contrast associated with regional rainfall climatology and the north–south gradient related to topographic influences. The four clusters represent a gradual transition from regions dominated by short-duration dry spells (≤10 days) in western Java, through intermediate regions with relatively balanced contributions from short- and long-duration dry spells, to regions characterized by a greater contribution from long-duration dry spells (>10 days) in eastern Java. Cluster 4 was characterized by the greatest contribution of long-duration dry spells and was primarily distributed along the northern coast of eastern Java, indicating a greater potential for prolonged rainfall interruptions in this region. The agreement between the statistical clustering results and the underlying climatic and topographic controls demonstrates that the proposed regionalization provides a physically meaningful representation of dry-spell variability across Java.

The practical value of the regionalization lies in providing a spatially explicit classification of areas with different dominant dry-spell characteristics. This classification can support the prioritization of drought monitoring and climate-risk assessment, particularly in regions characterized by a greater contribution from long-duration dry spells, where prolonged rainfall interruptions may pose greater risks to agricultural water availability. The identified dry-spell regimes may also provide a climate-based screening framework to support irrigation planning, planting-season decisions, and water-resource management by identifying areas with differing levels and characteristics of dry-spell risk. Thus, the regionalization can complement existing drought monitoring and agricultural and water-resource management practices by providing spatial information on the dominant dry-spell regime.

These findings contribute to a better understanding of dry-spell variability at both regional and local scales. At the regional scale, the identified clusters demonstrate that dry-spell characteristics vary systematically across Java in association with the spatial pattern of rainfall climatology, particularly the west–east transition from relatively wetter to drier conditions. At the local scale, the north–south patterns identified in western and eastern Java highlight the role of topography and orographic processes in modifying rainfall and dry-spell characteristics within relatively short geographical distances. The transect analysis further illustrates how variations in elevation, rainfall, and rainy days can contribute to local differences in dry-season rainfall conditions.

Several limitations should be considered when interpreting these results. First, the 10-day duration threshold was adopted based on reported impacts of prolonged dry spells on unirrigated shallow-rooted crops in tropical regions, but its relevance may vary across the different agro-climatic and agricultural conditions of Java. Second, the 60% cumulative FCTD threshold was adopted as an operational criterion rather than being derived from an independent physical threshold or empirical optimization. Therefore, both thresholds should be interpreted as methodological criteria rather than universal critical values. Third, the four-cluster solution was selected primarily using the WSS elbow method and supported by the spatial and climatic interpretability of the resulting clusters; other validation indices may favor a different number of clusters. Finally, the relationships between dry-spell characteristics, rainfall, and topography should not be interpreted as causal, as ENSO, IOD, monsoon variability, and other atmospheric processes were not explicitly incorporated into the analysis.

Future studies could evaluate the sensitivity of dry-spell patterns and regionalization to alternative duration and cumulative FCTD thresholds, as well as alternative cluster-validation methods. Incorporating ENSO, IOD, monsoon indices, seasonal rainfall, topography, atmospheric circulation, and moisture transport could further clarify the climatic mechanisms underlying the identified dry-spell regimes. Longer observational records and independent datasets could also be used to assess the temporal stability and robustness of the regionalization across different climatic conditions.

Statements

Data availability statement

The datasets presented in this article are not readily available because access to the data is subject to BMKG data-sharing policies. Requests to access the datasets should be directed to www.bmkg.go.id.

Author contributions

DW: Conceptualization, Data curation, Methodology, Writing – review & editing, Investigation, Writing – original draft, Formal analysis, Visualization. Supari: Conceptualization, Validation, Formal analysis, Supervision, Writing – review & editing. IS: Formal analysis, Writing – review & editing, Validation, Conceptualization, Supervision.

Funding

The author(s) declared that financial support was received for this work and/or its publication. The first author was supported by a master’s scholarship from the Indonesia Endowment Fund for Education (LPDP), Ministry of Finance of the Republic of Indonesia.

Acknowledgments

The author gratefully acknowledges the Indonesia Agency for Meteorology, Climatology, and Geophysics (BMKG) for providing the rainfall observation data used in this study.

Conflict of interest

The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Generative AI statement

The author(s) declared that Generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fclim.2026.1937736/full#supplementary-material

References

Summary

Keywords

clustering, consecutive, daily rainfall, dry spell, duration

Citation

Wibawanty DR, Supari and Santikayasa IP (2026) Characterization and regionalization of dry spells across Java-Indonesia using relative frequency and fractional contribution to total duration. Front. Clim. 8:1937736. doi: 10.3389/fclim.2026.1937736

Received

14 July 2026

Revised

02 September 2026

Accepted

30 September 2026

Published

15 October 2026

Volume

8 - 2026

Edited by

Hindumathi Palanisamy, World Meteorological Organization, Switzerland

Reviewed by

Heri Mulyanti, Universitas Bojonegoro, Indonesia

Shaidatul Azdawiyah Bt Abdul Talib, Malaysian Agricultural Research and Development Institute (MARDI), Malaysia

Updates

Copyright

*Correspondence: Supari,

Disclaimer

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.

Outline

Figures

Cite article

Copy to clipboard


Export citation file


Share article

Article metrics