Background & Summary

The northern coastline regions of Java have been soliciting the attention of many studies because a large portion of land in at least ten cities is subsiding1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18 (Fig. 1). The subsiding land has been triggered by a wide range of natural and anthropogenic activities, such as the compaction of sediments in Pekalongan, Semarang, and Demak19,20, gas extraction in Sidoarjo16, and structural loadings in Jakarta1. In addition, excessive groundwater extraction is the most significant triggering factor due to the increasing demand and need for residential and industrial water supply2,16,21,22,23,24. In these cities, the impacts of land subsidence such as widespread coastal inundation and structural damage to buildings, have been significantly reducing the quality of the living environment1,2,5,14,17,20,25,26,27. In Jakarta, the capital city of Indonesia, land subsidence is so severely affecting the city’s future urban viability28 that government authorities are planning to move the capital to Borneo29.

Fig. 1
figure 1

Administrative boundaries of coastal cities along the northern coastline regions of Java that are known to experience land subsidence1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18.

Full size image

Monitoring efforts to study the spatial extent of land subsidence and its rates in these cities have been continuously made using land-based and space-borne techniques1,2,3,4,6,7,8,9,10,11,15,16,27,30. Out of the ten cities, land subsidence in Jakarta and Semarang has been the most intensively studied with a long monitoring history. In Jakarta, levelling surveys and campaign GNSS measurements between 1982 and 2010 estimate that the rates are from 1 to 28 cm/year1. A recent study using Sentinel-1 InSAR data between 2014 and 2020 estimates that the rates are from 1 to ~11 cm/year7,8. Different rates between the past and recent monitoring efforts have also been observed in Semarang: GNSS measurements between 1999 and 2011 estimate that the rates are from 14 to 19 cm/year2, while a recent study using Sentinel-1 InSAR data between 2015 and 2020 reveals that the rates are from 2 to 3 cm/year8.

Besides Jakarta and Semarang, land subsidence monitoring efforts using InSAR data in the remaining cities have been rapidly growing since 20133,4,6,9,11,16,31,32,33 thanks to the availability of open access SAR data from the Copernicus Sentinel-1 satellites operated by the European Space Agency. On the contrary, since 2013, the monitoring efforts using campaign GNSS measurements have been lacking; the latest measurements were in 2010 for Jakarta1, in 2017 for Semarang12, and in 2018 for Demak10. Even worse, no study reported GNSS-based monitoring efforts in other cities that are known to experience land subsidence, such as Bekasi, Subang, Pekalongan, and Surabaya. This lack of GNSS-based monitoring efforts is worrying because the GNSS data is still needed for several reasons.

First, although InSAR observations provide all-weather and day-night monitoring capacity at high spatial coverage and resolution34,35, InSAR accuracy may still be degraded by various noise sources such as atmospheric phase delays, satellite orbit uncertainty, and unwrapping errors36,37. The degraded accuracy may mislead the interpretation of the subsidence rate. Incorporating independent data from GNSS measurements can help mitigate the false interpretation38,39. Second, InSAR velocity maps are relative to a reference point within the SAR data footprints. In areas where GNSS data is not available, a common approach to select a reference point is by assuming a certain area to be stable. However, this approach is subjective and may result in varying InSAR velocity maps across different studies. For example, Tay et al.7 showed that a location on the northern coastline of Jakarta subsides ~7x faster than that reported by Wu et al.8. One possible explanation for this discrepancy is the use of different reference points. Therefore, GNSS data is necessary to provide a priori information for selecting a stable reference point. Third, InSAR velocity maps are 1D measurements of surface deformations in the radar line-of-sight direction of SAR satellites34,35. In the case of land subsidence monitoring where vertical motions are of interest, other data sets such as GNSS observations are needed to isolate the vertical motions precisely (e.g.40,41,42,43). Fourth, Shirzaei et al.44 suggest the need for incorporating geocentric global reference frame vertical land motion (VLM) into global mean sea level (GMSL) studies. Geocentric is the natural for a global frame. Therefore, GMSL studies relative to this frame will allow us to determine whether a given location is rising or falling relative to the centre of the Earth. The InSAR-based VLM measurements are ideal for this purpose because InSAR data provide global coverage observations. However, the main challenge is that InSAR results are provided in a local reference frame. Thus, establishing worldwide InSAR-based VLM measurements needs GNSS data to tie the VLM measurements into a global reference frame44. In this study, we publish a time series of 3D displacements observed at twenty continuous GNSS stations between 2010 and 2021 along the northern coastline regions of Java (Fig. 1). The data may potentially be used for all the purposes mentioned above.

Observation specifications

We obtain the Receiver Independent Exchange (RINEX) GNSS data from the Geospatial Information Agency of Indonesia (BIG) which has been establishing and maintaining continuous GNSS stations in the country since 1996. Most stations are located on the national telecommunication company network. The stations use different monument types (Fig. 2) and record data continuously at one sample per second using high-precision L1/L2 geodetic type receivers and standard Choke Ring antennas (Table 1). In addition, the stations also have meteorological instrument systems, an automatic battery charger that connects to the national power network, and a cell modem (Fig. 3) that will stream the recorded raw data via a secure TCP/IP connection to BIG’s data processing centre in Cibinong, West Java up to one-hour latency.

Fig. 2
figure 2

Monument types of the BIG GNSS stations used in this study.

Full size image
Table 1 GNSS station specifications.
Full size table
Fig. 3
figure 3

Components of the BIG GNSS stations.

Full size image

Methods

We processed the RINEX GNSS data and obtained a time series of GNSS station coordinates using the GPS at MIT/Global Kalman filtering (GAMIT/GLOBK) software package version 10.7145,46,47. Our GPS processing consisted of two steps48,49. In the first step, we used double-differencing methods in the GAMIT software to estimate daily station positions, atmospheric parameters, satellite orbits, and earth orientation parameters from ionosphere-free linear combination GPS phase observations. During this step, we fixed the satellite orbit parameters to the IGS final orbits and applied a second-order ionospheric correction using IGS final ionospheric products. We set the computation parameters to the default GAMIT setting, except for the atmospheric delay parameters, which were modeled and estimated every hour using the Vienna Mapping Function50. We corrected the station displacements due to ocean tides using the most recent global ocean tide model, Finite Element Solution 200451. To adjust the effect of solar and solid-earth tides, we applied the International Earth Rotation and Reference System Service 2010 standard model52 and the atmospheric pressure loading model corrections53. Finally, we included GPS data from 12 International GNSS Services (IGS) stations (ALIC, BAKO, COCO, DARW, DGAR, GUAM, HYDE, IISC, LHAZ, PIMO, XMIS, YARR) in our daily processing to integrate our local network into the ITRF2014 reference frame54.

In the second step, we used the GLOBK software to combine our daily solutions with the global GPS solutions provided by the MIT analysis centre. During this step, we aligned our combined solutions with the ITRF2014 reference frame54 by minimising the position differences of eight selected sites55, using a priori values defined by the IGb14 realisation of ITRF201454. To accomplish this position difference minimation, we calculated six Helmert transformation parameters (three translations and three rotations) of eight selected reference sites: YARR in Australia, MAW1 and DAV1 in Antarctica, STJO and FLIN in North America, WSRT, ONSA, and NOT1 in Europe55. These sites were selected because they are less affected by earthquake deformations and hydrological loading55. Lastly, we generated daily time series coordinates for all the GNSS stations with respect to IGb14 realisation of ITRF201454.

Data Records

The processing results are a time series of 3D displacements from 2010 to 2021, relative to the ITRF2014. Most stations record negative velocities in the vertical component and are likely related to land subsidence (Table 2 and Fig. 4). The time series of the 3D displacements that include horizontal motions can be found in this repository: https://doi.org/10.5281/zenodo.777501656.

Table 2 Vertical velocity recorded by the BIG GNSS stations along the northern coastline regions of Java.
Full size table
Fig. 4
figure 4

Vertical component of the twenty BIG GNSS stations. (a) Negative vertical velocities are likely related to land subsidence. (b) Daily time series of the GNSS vertical component from 2010 to 2021.

Full size image

Technical Validation

Bad environments (e.g., buildings and trees) that degrade the sky view of the GNSS antenna will reflect and refract satellite signals before arriving at the antenna57. The reflected and refracted signals are called multipath signals which will introduce carrier phase measurement errors and subsequently lead to a positioning error58. A simple approach to inspect the environments surrounding the antenna is by plotting the signal-to-noise ratio (SNR) values measured by the GNSS receivers59. The SNR, like the carrier phase measurement, is also impacted directly by multipath signals and hence can therefore be used as a proxy to assess the environments surrounding the GNSS antenna59,60. Low SNR values indicate a large tracking error59, meaning that multipath objects are present. We plot the SNR values using the L1 data recordings only. We do not use the L2 data due to its encrypted C/A code and the lack of civilian access to the P-code which affects the L2 SNR reliability59. We plot the SNR values as a function of azimuth and elevation angle, both in the time series and sky plot (Figure S1). The SNR plot shows that all the stations have SNR values greater than 30 decibels, indicating good environments surrounding the GNSS stations hence the negative velocities in the vertical component are robust.

In addition to SNR analysis, we use independent observation measured by a deep pile benchmark to validate the negative velocity at the GNSS CPKL station in Pekalongan (Table 2 and Fig. 4). The benchmark (Fig. 5) was installed by the centre for groundwater and environmental geology, Indonesia’s geological agency, on 17 March 2020 ~500 m northeast of the GNSS CPKL station. The benchmark measurements between 06 April 2021 and 02 October 2022 estimate that ~80 ± 1 mm land subsidence occurs within ~1.5 years (Fig. 5c,d). Unfortunately, we cannot make a one-to-one comparison between this result and the amplitude of land subsidence measured at the GNSS CPKL station due to two reasons: 1) the benchmark and the GNSS CPKL station are ~500 m apart, 2) our GNSS data ended in 2021. Nevertheless, the benchmark measurements are still important in the sense that Pekalongan city is experiencing severe land subsidence.

Fig. 5
figure 5

A deep pile benchmark to measure land subsidence in Pekalongan. (a) A schematic design of the deep pile benchmark. (b) The newly installed benchmark. (c) The benchmark measures 60 ± 1 mm land subsidence by 06 April 2021. (d) The benchmark measures 140 ± 1 mm land subsidence by 02 October 2022, meaning that 80 ± 1 mm land subsidence occurs within ~1.5 years.

Full size image