A long-term, high-density, minute-resolution rainfall dataset from the Da-Tun Mountain of Northern Taiwan during 2009–2023 Authors: Lin-Wen Cheng, Cheng-Ku Yu, Che-Yu Lin, Shabeh ul Hasson Published: 2026 ABSTRACT Here we present a 15-year record (2009–2023) of quality-controlled, one-minute rainfall observations from the Da-Tun rain gauge network (DTRGN), deployed to capture fine-scale precipitation variability over Da-Tun Mountain (DT). DT, situated immediately inland of northern Taiwan’s coast, is an isolated three-dimensional volcanic massif ~15 km wide and ~1 km high, and represents a key climatological center of heavy orographic rainfall because tropical cyclones and northeasterly monsoonal flow frequently bring abundant moisture and precipitation to this mountain barrier. The DTRGN comprises 25 tipping-bucket rain gauges measuring rainfall across major mountain peaks, slopes, and valleys that are potential hotspots of orographically enhanced precipitation. We additionally provide temporal aggregations of one-minute rainfall observations at 10-minute, hourly, daily, and monthly intervals to support the research community. As a high-density, station-based rainfall dataset, DTRGN serves as an essential benchmark and resource for a broad spectrum of research and operational applications, including studies of orographic precipitation processes, tropical cyclone rainfall characterization, radar-derived precipitation correction, hydrological modeling, and assessments of long-term precipitation trends in complex terrain. Background & Summary Long-term, high-frequency precipitation observations over steep coastal mountainous regions remain a critical missing component of the global Earth system observing network, particularly in regions frequently affected by tropical cyclones and monsoonal circulations. The Da-Tun Mountain (DT), located immediately adjacent to the northern coast of Taiwan (Fig. 1a), exemplifies this class of under-observed mountainous environments. The DT features complex volcanic terrain with sharp elevation gradients, rising steeply to a maximum elevation of 1120 m above mean sea level (MSL). Owing to its geographic setting, the DT region frequently receives substantial precipitation in different seasons and various synoptic conditions. Notably, it has been identified as one of the climatological maxima of heavy rainfall during the influence of tropical cyclones and northeasterly monsoonal flow1–6. The DT region thus serves as an exceptional natural laboratory for investigating orographic precipitation processes and related hydrometeorological phenomena. Besides such importance of the DT region, it was being monitored by a sparse distribution of surface stations maintained by the Taiwan Central Weather Administration (CWA) (Fig. 1b). Most CWA stations in and around the DT region are located below 200 m MSL, over surrounding lowland and foothill areas, with only four stations above 500 m MSL. Among these, one is located on the slope area and three near local mountain crests, including one situated at a relatively high elevation and two at low crests within the DT region. However, their limited spatial representation fails to capture the detailed variations and intensities of rainfall over the DT region. In addition, the DT region provides a compact yet topographically complex environment that includes ridges, slopes, and valleys within a limited spatial extent, making it scientifically suitable for high-density observations aimed at resolving fine-scale orographic precipitation processes. Relatively better accessibility of the DT region compared to other high-elevation areas of Taiwan (e.g., Central Mountain Range) also facilitates long-term station deployment and maintenance.Fig. 1(a) Terrain elevation map of Taiwan, with the black rectangle in northern Taiwan indicating the region of the Da-Tun Mountain (DT). The location of the Wu-Fen-San (WFS) Doppler radar is denoted by a red triangle. (b) Enlarged view of the DT region indicated by the black rectangle in (a), showing the terrain height and the spatial distribution of surface weather stations operated by the Central Weather Administration (CWA). Squares and circles indicate the locations of manual and automatic CWA stations, respectively. (c) Same as (b), but also showing the 25 rain gauge stations (stars with station ID numbers) deployed as part of the Da-Tun Rain Gauge Network (DTRGN). The color shading in all panels represents terrain elevation. Station ID numbers correspond to those listed in Table 2. (a) Terrain elevation map of Taiwan, with the black rectangle in northern Taiwan indicating the region of the Da-Tun Mountain (DT). The location of the Wu-Fen-San (WFS) Doppler radar is denoted by a red triangle. (b) Enlarged view of the DT region indicated by the black rectangle in (a), showing the terrain height and the spatial distribution of surface weather stations operated by the Central Weather Administration (CWA). Squares and circles indicate the locations of manual and automatic CWA stations, respectively. (c) Same as (b), but also showing the 25 rain gauge stations (stars with station ID numbers) deployed as part of the Da-Tun Rain Gauge Network (DTRGN). The color shading in all panels represents terrain elevation. Station ID numbers correspond to those listed in Table 2. To address this limitation, a dense and high-resolution surface rainfall observation network was established beginning in 2009. The resulting dataset is referred to as the Da-Tun rain gauge network (DTRGN). By 2011, the network had already achieved reasonably good spatial coverage across the DT region, enabling reliable monitoring of rainfall variability over the complex terrain. The spatial distribution of DTRGN stations is shown in Fig. 1c. The network comprises 25 tipping-bucket rain gauges measuring rainfall at one-minute temporal resolution across the DT region and its vicinity. As the locations of hotspots for orographically enhanced precipitation are not known a priori, stations were deployed to sample a range of terrain settings, with particular attention to major mountain peaks and windward-facing slopes and valleys where orographic enhancement is more likely to occur. At the same time, spatial coverage was also considered to maintain a relatively uniform distribution of stations over the DT region. These topographically informed deployment strategies enable a more comprehensive representation of precipitation characteristics across the complex terrain. Within the DT region (elevation > 100 m), covering approximately 225 km², the CWA network includes 9 stations and the DTRGN includes 18 stations, corresponding to approximately 25 km² per station for the CWA network and 12.5 km² per station for the DTRGN. When combined, the integrated network achieves approximately 8.3 km² per station. These values are substantially higher than the WMO-recommended minimum station density for mountainous regions (250–2500 km² per station, depending on station type)7, indicating that the DTRGN provides high spatial resolution for precipitation observations. Some precipitation measurements from the DTRGN dataset have been used earlier in research related to orographic precipitation. For example, Cheng and Yu3 analyzed DTRGN data in combination with radar observations and upslope modeling to investigate precipitation distribution and intensity over DT under a wintertime northeasterly monsoonal environment. Cheng et al.4 identified key mechanisms of tropical cyclone generated orographic precipitation in DT region. Additional studies have employed DTRGN data for hydrological and geochemical analysis in volcanic watersheds8, for validating and improving satellite-based precipitation radar retrievals9, and for diagnosing valley rainfall enhancement and terrain–airflow interactions5. More recent work has further utilized integrated observations from multiple measurement platforms—including Doppler and polarimetric radars, disdrometers, and the DTRGN—to document the kinematic and microphysical characteristics of orographic precipitation over DT during Typhoon Chanthu (2021)6. However, these previous studies focused primarily on specific physical processes or applications and did not provide a comprehensive description of the DTRGN itself. Detailed information on the observational network design, deployed instrumentation, maintenance practices, data quality control procedures, and the overall strengths and limitations of the dataset is therefore presented in this Data Descriptor. With its high spatial density, long-term continuity, and fine temporal resolution, the DTRGN dataset provides a valuable foundation for advancing research in orographic precipitation processes, tropical cyclone rainfall, radar-derived rainfall correction, hydrological modeling, and the assessment of long-term precipitation trends in complex terrain. The release of this dataset is expected to significantly complement the existing CWA surface network, fostering interdisciplinary investigations into meteorological, hydrological, and environmental sciences, and advancing understanding of precipitation processes and related land–atmosphere interactions in complex mountainous regions. Methods In this section, we describe the key components of the data acquisition process for the DTRGN, including hardware specifications, site selection, sensor maintenance, data transmission, quality control (QC), and temporal aggregation. The procedures were informed by international best practices as outlined in the World Meteorological Organization (WMO) guidelines10–13. Sensors and device hardware Each station of the DTRGN collects precipitation data at 1-minute intervals, enabling high-frequency monitoring of rainfall variability. The rain gauges are equipped with 200-mm-diameter collector orifices and are manufactured by three companies: Nippon Electric Instrument, Inc. (model N-68), Takeda Meteorological Instrument Co., Ltd. (model TK-1), and Climatec, Inc. (model CTKF-1). The measurement accuracy of these gauges is ±0.5 mm for rainfall intensities less than 20 mm h−1. Wind shields were not installed at the DTRGN stations, following the deployment configuration commonly used in operational rain gauge networks in Taiwan. This may introduce wind-induced measurement errors, particularly under strong wind conditions12. Each station is equipped with a data logger for data storage, and a remote transmission module for data retrieval. The data logger clocks were configured with an automated clock check function that performs daily synchronization with the data server located at the Department of Atmospheric Sciences, National Taiwan University, Taipei. If the time deviation exceeds a predefined threshold (on the order of 1 second), the system automatically corrects the station clock. The internal memory of the logger can store approximately 25 days of 1-minute precipitation records. Power is supplied by a solar panel in combination with two sealed lead-acid batteries, enabling continuous operation under varying weather conditions. All components are housed in a weatherproof enclosure designed to withstand high humidity and the harsh conditions typical of mountainous environment. A summary of the key specifications for sensors, loggers, and power supply components is provided in Table 1.Table 1Overview of sensors and associated hardware components used in the observation network.ItemComponentType/Model1Tipping Bucket Rain GaugeN-68 by Nippon Electric Instrument, Inc.TK-1 by Takeda Meteorological Instrument Co., Ltd.CTKF-1 Climatec, Inc,2Data loggerCampbell Scientific CR200/3003Wireless ModuleKepro MA8-9L4Antenna4 G (900/1800/2600 MHz)5Solar Panel20 Watt (16.8 V)6Rechargeable BatteriesLead-Acid Battery (12 V, 14 Ah) × 2 Table 1Overview of sensors and associated hardware components used in the observation network.ItemComponentType/Model1Tipping Bucket Rain GaugeN-68 by Nippon Electric Instrument, Inc.TK-1 by Takeda Meteorological Instrument Co., Ltd.CTKF-1 Climatec, Inc,2Data loggerCampbell Scientific CR200/3003Wireless ModuleKepro MA8-9L4Antenna4 G (900/1800/2600 MHz)5Solar Panel20 Watt (16.8 V)6Rechargeable BatteriesLead-Acid Battery (12 V, 14 Ah) × 2 Overview of sensors and associated hardware components used in the observation network. Data transmission is carried out through a 4 G cellular network. Communication between the data server and each rain gauge site is configured using LoggerNet, a data acquisition and communication software developed by Campbell Scientific. The base station regularly attempts to retrieve data from each gauge site on an hourly basis. In the event of a failed transmission due to unstable signal conditions, the system automatically retries the connection every 10 minutes until successful. This design ensures timely and reliable data acquisition while accounting for the communication challenges in complex terrain environment. Sensor network deployment and maintenance As shown in Fig. 1b, the CWA surface stations are located mostly in the elevation range of 7-188 m MSL, and only four stations are deployed above 500 m MSL. In addition, the temporal resolution of CWA surface observations varies by station type: manual weather stations record rainfall at 1-minute intervals, while automatic weather stations operate at a 10-minute resolution. Given the highly heterogeneous nature of orographic precipitation, the sparse distribution of the CWA stations over the DT region emphasizes the need for a denser observational network. The DTRGN addresses this gap by deploying additional rain gauge stations with the elevation range of 36-1011 m MSL, enabling more accurate characterization of terrain-induced precipitation variability. To fulfill the scientific objective of capturing spatial variability in orographic precipitation, the station placement prioritized coverage across key topographic features of ridgelines, slopes, and valleys, known to considerably modulate rainfall distribution and intensity. This terrain-informed station deployment strategy is aligned with international recommendations for achieving adequate spatial representation in complex terrain11. This terrain-driven deployment strategy, while essential for capturing fine-scale precipitation information under diverse local conditions, also posed practical challenges. Specifically, the rugged landscape and limited accessibility in certain areas made it difficult to fully adhere to the siting guidelines recommended by the WMO12. As a result, some gauges were installed in locations surrounded by natural vegetation, such as grasses or shrubs, which may introduce minor wind-related biases—an issue commonly noted in studies of point-rainfall measurement uncertainty14. These siting constraints are also acknowledged in international observing guidelines, which note that ideal exposure conditions are rarely attainable in real-world environments and that practical compromises are often necessary12. In a few cases, gauge stations were placed on the rooftops of low-rise buildings when suitable open ground was unavailable. Although these settings do not fully conform to the ideal WMO standards, care was taken to minimize potential obstructions and to ensure that the measurements remain representative and reliable for long-term monitoring over this mountain barrier. Photographs of all DTRGN stations, each annotated with its station ID, are presented in Fig. 2 to illustrate the diverse siting environments across the network. These images provide visual confirmation of the installation settings, including the instrument mounting structures, the presence or absence of nearby obstructions, ground conditions, and basic site accessibility. The photographs also help illustrate variations in station exposure and maintenance conditions across the network. Metadata for each station, including geographic coordinates, elevation, and installation date, are summarized in Table 2.Fig. 2Photographs of the 25 DTRGN stations. Each image is annotated with the corresponding station ID listed in Table 2.Table 2Details of deployed DTRGN stations.IDStation NameLatitudeLongitudeElevation (m MSL)Deploy Date (dd-mm-yy)1Xiao You Keng (XYK)25.1781121.547781223-06-092Lu Jiao Keng (LJK)25.1917121.567934323-06-093Jie Bei (JB)25.1985121.594521823-06-094San Zhi (SZ)25.2252121.536041423-06-095Huang Zui Shan (HZS)25.1761121.603890123-06-096Xing Hua (XH)25.2037121.501427323-06-097Da Tun (DT)25.1555121.504032823-06-098Ping Deng (PD)25.1295121.576743513-01-109Zhu Zi Shan (ZZS)25.2153121.5611101113-01-1010Zhu Gao Shan (ZGS)25.1582121.572982913-01-1011Xi Di (XD)25.1445121.606737317-06-1012Ye Liu (YL)25.2059121.68963617-06-1013Zhong Tai (ZT)25.2244121.484518018-06-1014Ma Ling (ML)25.1098121.68995117-06-1015Jin Shan (JS)25.2629121.617724129-06-1016Song Shan (SS)25.2007121.552585629-06-1017Meng Hu (MH)25.1283121.636031927-12-1018Ding Nei (DN)25.0980121.75378130-12-1019Bei Gang (BG)25.0891121.64454130-12-1020Long Sheng (LS)25.1564121.69399628-12-1021Zhang Xing (ZX)25.0922121.71375628-12-1022Peng Lai (PL)25.2048121.616332231-12-1023Jian Shan Hu (JSH)25.2410121.563029918-02-1424Fei Cui (FC)25.1795121.674338220-02-1425Zhong Guang (ZG)25.2162121.60395819-11-18 Photographs of the 25 DTRGN stations. Each image is annotated with the corresponding station ID listed in Table 2. Table 2Details of deployed DTRGN stations.IDStation NameLatitudeLongitudeElevation (m MSL)Deploy Date (dd-mm-yy)1Xiao You Keng (XYK)25.1781121.547781223-06-092Lu Jiao Keng (LJK)25.1917121.567934323-06-093Jie Bei (JB)25.1985121.594521823-06-094San Zhi (SZ)25.2252121.536041423-06-095Huang Zui Shan (HZS)25.1761121.603890123-06-096Xing Hua (XH)25.2037121.501427323-06-097Da Tun (DT)25.1555121.504032823-06-098Ping Deng (PD)25.1295121.576743513-01-109Zhu Zi Shan (ZZS)25.2153121.5611101113-01-1010Zhu Gao Shan (ZGS)25.1582121.572982913-01-1011Xi Di (XD)25.1445121.606737317-06-1012Ye Liu (YL)25.2059121.68963617-06-1013Zhong Tai (ZT)25.2244121.484518018-06-1014Ma Ling (ML)25.1098121.68995117-06-1015Jin Shan (JS)25.2629121.617724129-06-1016Song Shan (SS)25.2007121.552585629-06-1017Meng Hu (MH)25.1283121.636031927-12-1018Ding Nei (DN)25.0980121.75378130-12-1019Bei Gang (BG)25.0891121.64454130-12-1020Long Sheng (LS)25.1564121.69399628-12-1021Zhang Xing (ZX)25.0922121.71375628-12-1022Peng Lai (PL)25.2048121.616332231-12-1023Jian Shan Hu (JSH)25.2410121.563029918-02-1424Fei Cui (FC)25.1795121.674338220-02-1425Zhong Guang (ZG)25.2162121.60395819-11-18 Details of deployed DTRGN stations. Routine maintenance was performed annually at each station, while additional unscheduled site visits were conducted when data issues were identified. Routine maintenance primarily included cleaning of funnels and filters, and vegetation control at selected stations (e.g., ID1, 2, 3, 9, 11, 16, 17, and 23). Unscheduled site visits were typically associated with site-specific environmental factors. Under persistent northeasterly monsoonal conditions, orographically forced lifting usually results in prolonged cloud cover over the windward slopes of DT. This substantially reduces the solar panel charging efficiency and, in some cases, causes battery levels to drop below operational thresholds, necessitating manual battery replacement at some stations (e.g., ID1, 2, 3, 9, 11, 16, 17, 22, and 24). In addition to power-related issues, certain stations were affected by environmental interference. For example, birds were observed dropping small fruits and seed detritus onto the mesh filter of the tipping-bucket gauges (Fig. 3a). This often led to clogging of the filter, preventing rainwater from entering the measurement funnel and resulting in missed rainfall detection at several stations (e.g., ID8, 12, 14, 18, 19, and 20). Such cases required unscheduled site visits for maintenance activities (e.g., battery replacement, filter cleaning, and functional checks). These maintenance activities typically required less than 30 minutes for a given station and therefore had only very minor impacts on the continuity of observations.Fig. 3Representative issues encountered during long-term operation and maintenance of the DTRGN. (a) Seed detritus dropped by birds onto the mesh filter of a tipping-bucket gauge, leading to clogging and potential underestimation of rainfall. (b) Access to one station was blocked by discarded branches and vegetation dumped at the entrance. (c) A beehive was discovered beneath the solar panel at the MH station (Table 2), presenting a safety hazard. The right panel shows a magnified view of the beehive. Representative issues encountered during long-term operation and maintenance of the DTRGN. (a) Seed detritus dropped by birds onto the mesh filter of a tipping-bucket gauge, leading to clogging and potential underestimation of rainfall. (b) Access to one station was blocked by discarded branches and vegetation dumped at the entrance. (c) A beehive was discovered beneath the solar panel at the MH station (Table 2), presenting a safety hazard. The right panel shows a magnified view of the beehive. Accessibility also posed logistical challenges. While most stations are reachable by vehicle, a few stations, such as HZS, ZGS, and MH (Table 2), require a significant hike. The HZS station is particularly demanding, as access requires a strenuous 6 km round-trip hike carrying tools and replacement batteries, with certain sections lacking well-defined trails. Each maintenance visit typically takes up to 6 hours, including rest breaks. Moreover, unexpected hazards encountered during maintenance have further complicated field operations. These include access routes obstructed by dumped vegetation (Fig. 3b) and the formation of a beehive beneath the solar panel at the gauge station (Fig. 3c), both of which represented additional maintenance challenges and safety concerns for field personnel. Data collection and processing All 1-minute rainfall data transmitted from the field stations are automatically uploaded to a centralized server for QC and further processing. Measurements are systematically checked for consistency, and the missing or anomalous values are flagged using a designated missing data code (−99.9), allowing for robust onward analysis. The overall workflow for automated data acquisition and centralized processing is consistent with internationally recommended operational practices for data management10. To ensure data integrity, a multi-step QC procedure was applied. First, raw data were screened for physically implausible values. Any 1-minute rainfall rate exceeding 10 mm was automatically excluded, as such intensities are considered unrealistic for tipping-bucket gauges in this region and are typically indicative of mechanical malfunction. A post hoc examination of these extreme values indicates that such cases were rare, with only four events identified over the entire observation period. Subsequent inspection of maintenance records and field conditions confirmed that these events were associated with instrument malfunctions, supporting their exclusion. Rainfall rates between 5 and 10 mm min−1 were subjected to manual inspection, which involved cross-referencing radar reflectivity data from the Wu-Fen-San (WFS) Doppler radar (location indicated in Fig. 1a) to identify the potential presence of precipitation systems, evaluating temporal consistency within the station’s own records, and comparing observations with nearby gauges15,16. Radar information was used as qualitative supporting evidence rather than a strict filtering criterion. Spatial consistency was used as the primary criterion for data retention. High-intensity rainfall observations were retained when supported by clear and consistent rainfall signals at nearby stations, indicating spatial coherence in the precipitation field. In cases where spatial consistency was absent, both temporal continuity and radar support were then required for data retention. As a result, isolated high-intensity rainfall observations at a single station without radar support were removed, even if they exhibited short-term temporal continuity. Such manual verification is consistent with the broader QC requirements emphasized for hydrometeorological observations in typhoon-affected regions17. This process helped differentiate genuine high-intensity rainfall from spurious spikes. These procedures follow widely adopted QC principles, including physical screening, temporal consistency checks, and cross-station comparisons13. In addition, daily accumulated rainfall values were regularly assessed to identify possible gauge blockages. Figure 4 illustrates an example of this procedure. The spatial rainfall field on 27 September 2019 (Fig. 4a), objectively analyzed using the Cressman weighting scheme18, shows that the rain gauge at Station ID 10 (indicated by red arrow) reported anomalously low rainfall compared with surrounding sites during a widespread precipitation event, indicating likely partial blockage of the funnel. A subsequent site visit on 2 October 2019 confirmed that the funnel was obstructed by seed debris. As shown in Fig. 4b, this station typically recorded rainfall amounts comparable to neighboring gauges (e.g., 4–5 September), providing a reference for determining the onset of underreporting. After the blockage was verified, all measurements from 20 September to 2 October—the interval during which the gauge underestimated rainfall—were retrospectively flagged and replaced with the designated missing value (−99.9), as highlighted by the black box in Fig. 4b. Similar cases were identified but occurred infrequently across the network, roughly only two times per station over the observation period. This procedure ensures that gauge obstructions—caused by accumulated leaves, small fruits, or seed debris—do not compromise the accuracy of the dataset.Fig. 4(a) Daily accumulated rainfall on 27 September 2019 derived from CWA and uncorrected DTRGN observations. The red arrow marks the rain gauge at Station ID 10, which exhibited abnormally low rainfall compared with surrounding stations and was later confirmed to be partially blocked. (b) Time–station diagram of daily rainfall for September–October 2019, illustrating that Station ID 10 generally recorded rainfall comparable to nearby gauges during normal operation, as highlighted by the red box (4–5 September). The black box indicates the period of underestimation due to blockage, which was retrospectively replaced with missing values during QC. (a) Daily accumulated rainfall on 27 September 2019 derived from CWA and uncorrected DTRGN observations. The red arrow marks the rain gauge at Station ID 10, which exhibited abnormally low rainfall compared with surrounding stations and was later confirmed to be partially blocked. (b) Time–station diagram of daily rainfall for September–October 2019, illustrating that Station ID 10 generally recorded rainfall comparable to nearby gauges during normal operation, as highlighted by the red box (4–5 September). The black box indicates the period of underestimation due to blockage, which was retrospectively replaced with missing values during QC. By combining automated filtering with targeted manual verification, this QC strategy maintains the high temporal resolution of the original records while ensuring their physical consistency across both spatial and temporal domains. This two-step approach also minimizes false detections arising from instrument malfunctions or environmental contamination, thereby preserving the integrity of the long-term precipitation time series. As a result, the QC-processed dataset provides a reliable foundation for subsequent analyses of orographic precipitation variability in the DTRGN region. Data upscaling Following the QC procedures described in the previous section, we performed temporal upscaling to produce 10-minute, hourly, daily, and monthly rainfall datasets from the original 1-minute resolution measurements. This approach allows the dataset to support a broader range of applications, including hydrological modeling, trend analysis, and comparison with other observational and model datasets that operate on coarser time scales. To ensure consistency across time aggregations, we applied a data availability threshold approach, as demonstrated in previous hydrometeorological network studies19. To minimize the impact of short-term data gaps on aggregated rainfall estimates, we applied an availability-based threshold approach that preserves the temporal characteristics of the original observations. While our method does not perform imputation, the general principle of maintaining temporal completeness and reducing biases associated with missing data is consistent with practices emphasized in large-scale climate data reconstructions20. For each 10-minute interval, the availability of the 1-minute rainfall data was initially assessed. If at least 80% of the 1-minute records (i.e., a minimum of 8 out of 10 valid values) were present, the mean rainfall rate (mm min−1) was first computed from the accumulated rainfall divided by the number of valid data. Assuming that rainfall rates during missing time steps were comparable to the mean rate of valid observations, the 10-minute rainfall rates (i.e., rainfall accumulation within the 10-minute interval) were then estimated by multiplying the mean rainfall rate by the total number of time steps (10). Intervals failing to meet the availability threshold were flagged as missing and excluded from further aggregation. For hourly rainfall, a similar 80% availability criterion was applied to the 10-minute data: at least 5 valid 10-minute values (out of 6) were required to compute the hourly total. If this condition was satisfied, the hourly rainfall rate (mm h−1) was calculated as the mean value of the valid 10-minute rainfall rates multiplied by six to estimate the full hourly accumulation. Otherwise, the hourly value was marked as missing. Daily rainfall accumulations were computed from at least 80% of valid hourly observations, which correspond to 20 valid hourly observations within 24 hours of that day. Similarly, monthly rainfall was calculated when 24 valid daily rainfall observations within a 30-day month were available. This rate-based upscaling method adopted herein minimizes underestimation of total rainfall due to short-term missing observations, ensuring reliability across multiple temporal scales. Data records The dataset is available at Figshare21 and is provided as plain CSV files: DTRGN_Quality_Controlled_1min_Rainfall.csv, DTRGN_Upscaled_10min_Rainfall.csv, DTRGN_Upscaled_Hourly_Rainfall.csv, DTRGN_Upscaled_Daily_Rainfall.csv, and DTRGN_Upscaled_Monthly_Rainfall.csv. These five files include different temporal scales of rainfall measurements collected from 25 DTRGN stations, covering the period from June 2009 to December 2023. Users can access the dataset at multiple temporal resolutions according to their analytical needs. The file, DTRGN_Quality_Controlled_1min_Rainfall.csv, contains the quality-controlled 1-minute rainfall data. The file follows a consistent tabular structure in which the first row lists the station names as column headers; the second and third rows indicate the latitude and longitude of each station, respectively; and the fourth row specifies the unit of measurement (mm). The remaining rows contain the 1-minute precipitation values, with the first column recording the timestamps in YYYY-MM-DD HH:MM format. Each timestamp represents the ending time of the 1-minute accumulation interval. Subsequent columns correspond to different stations, ordered by station ID. The upscaled files provide temporally aggregated rainfall data at 10-minute, hourly, daily, and monthly resolutions, respectively. Each file follows the same format as the 1-minute data file. Timestamps are placed in the first column, and subsequent columns correspond to different stations. The timestamp format varies depending on the resolution: YYYY-MM-DD HH:MM for 10-minute and hourly data, YYYY-MM-DD for daily data, and YYYY-MM for monthly data. For sub-daily datasets (10-minute and hourly), each timestamp represents the ending time of the accumulation interval. For daily and monthly datasets, the timestamp indicates the corresponding calendar day or month over which precipitation is accumulated. Due to file size, some files may not be directly viewable in standard spreadsheet software such as Microsoft Excel. Technical Validation To ensure the reliability and completeness of the dataset, a multi-step validation process was implemented. During routine field maintenance, additional checks were conducted to verify the proper functioning of the rain gauges. These included cleaning the internal components of the tipping-bucket mechanism and manually triggering the tipping lever, which corresponds to a fixed incremental rainfall amount, to confirm that the recorded signals aligned with the expected values. This procedure helped validate the performance of both the sensor and the data logging system. On-site inspections were essential for identifying and addressing issues such as clogging, mechanical faults, or data transmission problems. We also assessed data completeness and availability across all stations over the observation period. The monthly percentage of data availability was calculated for each of the 25 stations using the 1-minute rainfall dataset spanning from 2009 to 2023 (Fig. 5a). Most stations consistently maintained high data availability throughout the years. However, occasional gaps were identified, typically occurring on average 4 times per station per year. These gaps were primarily caused by temporary power outages, transmission interruptions, or the manual removal of spurious records due to debris-induced blockage of the tipping buckets.Fig. 5(a) Monthly data availability of 1-minute rainfall records at each of the 25 DTRGN stations from 2009 to 2023. Each row represents a station (identified by its station ID), and each column corresponds to a calendar month. Color shading indicates the percentage of valid 1-minute data available in each month. White areas represent 0% data availability, while gray areas indicate periods before station deployment. (b) Time series of monthly accumulated precipitation at all 25 stations from June 2009 to December 2023, derived from the quality-controlled daily rainfall dataset. Each row represents a station (identified by its station ID), and each column corresponds to a calendar month. Color shading indicates the monthly accumulated precipitation (mm). White areas denote missing data, and gray areas indicate periods before station deployment, consistent with panel (a). (a) Monthly data availability of 1-minute rainfall records at each of the 25 DTRGN stations from 2009 to 2023. Each row represents a station (identified by its station ID), and each column corresponds to a calendar month. Color shading indicates the percentage of valid 1-minute data available in each month. White areas represent 0% data availability, while gray areas indicate periods before station deployment. (b) Time series of monthly accumulated precipitation at all 25 stations from June 2009 to December 2023, derived from the quality-controlled daily rainfall dataset. Each row represents a station (identified by its station ID), and each column corresponds to a calendar month. Color shading indicates the monthly accumulated precipitation (mm). White areas denote missing data, and gray areas indicate periods before station deployment, consistent with panel (a). A notable period of data loss occurred between April and May 2017, during which nearly all stations experienced a complete interruption in data reception. This gap was caused by the local telecom providers’ discontinuation of 3 G mobile data services. As a result, the existing 3G-based communication modules became obsolete, and real-time data transmission failed across the network. Full data reception was restored only after on-site replacement of all communication modules with new 4G-compatible units. During the interruption period (April–May 2017), the data availability across most stations ranged between approximately 30% and 60% (Fig. 5a). Because the missing data during this period were continuous across all temporal resolutions, the same gap is reflected in the 10-minute, hourly, daily, and monthly aggregated datasets. To further evaluate the quality of the rainfall records, we examined the spatiotemporal variability of monthly accumulated precipitation, as shown in Fig. 5b. The dataset reveals pronounced variability across stations and over time, reflecting the influence of complex terrain and site exposure. Stations located over windward ridges and slopes tend to exhibit stronger variability, whereas those at lower elevations or in leeward areas show weaker signals. In addition, several stations (e.g., stations ID2, ID5, ID9 and ID16) display identifiable seasonal patterns, with enhanced accumulated precipitation during the winter months (December to February) at high-elevation or windward locations, consistent with the influence of the northeasterly monsoon. However, these seasonal features are not uniformly observed across the network, underscoring the strong spatial variability in precipitation over the DT region. These results highlight the importance of the high-density network in capturing terrain-related rainfall variability. Furthermore, to demonstrate the improvement in spatial precipitation representation provided by the DTRGN, Fig. 6 compares the annual accumulated rainfall field for 2020 derived from the CWA network alone (Fig. 6a) and from the combined CWA–DTRGN observations (Fig. 6b). Both fields were objectively analyzed using the same Cressman interpolation scheme18 to ensure methodological consistency. In the CWA-only analysis (Fig. 6a), the primary rainfall maximum appears along the eastern slope and extends eastward toward the foothills and northeastern coastal areas. After incorporating DTRGN data, the major rainfall core shifts westward toward the mountain crest, better aligning with the expected orographic lifting region. In addition, an independent local maximum of rainfall becomes evident along the northern ridge of DT, which was not captured in the CWA-only analysis, as shown in Fig. 6b. The inclusion of DTRGN gauges substantially refines the spatial structure of the analyzed field, particularly along steep slopes and ridge where the CWA network coverage is sparse. It is noteworthy that, although ridge regions frequently exhibit strong orographic enhancement of precipitation, the location of rainfall maxima over DT can vary considerably depending on the prevailing wind conditions. Rainfall maxima may also occur over slopes or valley regions due to the complicated interaction between prevailing wind and the terrain3,5. These important features are better resolved by the high-density DTRGN observations.Fig. 6(a) Spatial distribution of annual accumulated precipitation (mm) for 2020 derived solely from the CWA rain gauge network. (b) Same as (a), but including observations from the DTRGN. (c) Same as (b), but excluding observations from the HZS station. Color shading represents annual rainfall accumulation, while black contours indicate terrain elevation at 200 m intervals from 100 m to 900 m MSL. Squares and circles denote manual and automatic CWA stations, respectively, and stars indicate DTRGN rain gauges. (a) Spatial distribution of annual accumulated precipitation (mm) for 2020 derived solely from the CWA rain gauge network. (b) Same as (a), but including observations from the DTRGN. (c) Same as (b), but excluding observations from the HZS station. Color shading represents annual rainfall accumulation, while black contours indicate terrain elevation at 200 m intervals from 100 m to 900 m MSL. Squares and circles denote manual and automatic CWA stations, respectively, and stars indicate DTRGN rain gauges. To further assess the role of high-elevation gauges in resolving localized precipitation structure, a sensitivity test was performed by excluding the HZS station from the interpolation analysis. When HZS data were excluded, the eastern local rainfall maximum became weaker and less distinct, and the spatial pattern of rainfall shifted toward a single dominant maximum (Fig. 6c). The peak annual accumulated rainfall was reduced by approximately 3000 mm. These results indicate that this particular high-elevation station provides critical rainfall information that cannot be appropriately reconstructed from surrounding stations. These results highlight the capability of the DTRGN to provide a more realistic and topographically consistent precipitation field than what can be obtained from the CWA network alone, particularly in regions characterized by sharp terrain gradients and slope-dependent rainfall variability. Beyond the spatial improvements demonstrated in Fig. 6, the overall performance of the DTRGN was further verified through a series of quality assurance procedures. Hardware inspections, manual calibration checks, and spatiotemporal consistency analyses collectively confirm that the DTRGN offers a reliable and high-quality dataset for investigating orographic precipitation and associated hydrometeorological processes in complex mountainous environments. These validation steps demonstrate that the dataset preserves both the minute-scale variability and the long-term precipitation characteristics essential for research applications ranging from event-based analyses to climatological assessments. Usage Notes The DTRGN precipitation dataset provides high-temporal-resolution rainfall records collected across a mountainous region in northern Taiwan. Users should be aware of several considerations when utilizing this dataset for scientific or operational applications. First, although all data underwent rigorous QC—including the removal of physically implausible values and periods affected by gauge blockage—some minor errors may persist due to environmental factors or unexpected hardware behavior. Users are advised to perform additional screening or sensitivity testing, especially when analyzing extreme rainfall events. Second, gaps in the dataset occur due to power shortages, communication failures, or sensor maintenance. Missing or invalid data points are clearly marked with a placeholder value of −99.9. Researchers focusing on long-term trends or statistical summaries should account for these gaps appropriately, using imputation, temporal smoothing, or exclusion based on specific study objectives. Third, due to the limited available space for instrument installation, some gauges are located near natural vegetation or on rooftops, which may introduce minor wind-related or site-specific measurement biases. Users conducting inter-station comparisons or high-precision hydrological modeling should consult Table 2 and Fig. 2 for information on siting environments and consider incorporating uncertainty estimates into their analysis. Lastly, the dataset includes multiple temporal resolutions (1-minute to monthly) to accommodate different research needs. While upscaling was performed with an 80% data availability threshold at each temporal aggregation level, users requiring stricter completeness may choose to apply custom thresholds or focus on a subset of stations with more continuous records.