The magnetic observatory on Tatuoca , Belém , Brazil : history and recent developments

The Tatuoca magnetic observatory (IAGA code: TTB) is located on a small island in the Amazonian delta in the state of Pará, Brazil. Its location close to the geomagnetic equator and within the South Atlantic Anomaly offers a high scientific return of the observatory’s data. A joint effort by the National Observatory of Brazil (ON) and the GFZ German Research Centre for Geosciences (GFZ) was undertaken, starting from 2015 in order to modernise the observatory with the goal of joining the INTERMAGNET network and to provide real-time data access. In this paper, we will describe the history of the observatory, recent improvements, and plans for the near future. In addition, we will give some comments on absolute observations of the geomagnetic field near the geomagnetic equator.

. Satellite image of the Tatuoca island (©Google 2016). The main buildings and infrastructure of the observatory are marked and annotated.
The LEMI system is powered by a 45 Ah lead-acid battery which is charged by a dedicated 30 W solar panel on the roof of the variometer house. Further, a POS-2 proton gradiometer is located in the southeastern corner of the variometer house (white circle). This instrument was never in operation and has been removed in October 2016 (see below).
The absolute house has an approximate size of 4.8 by 8.0 m, and is roughly oriented in N-S direction. It houses ten pillars, four of which are located at the northern end, including the main pillar, and six of which are located at the southern end.

5
The latter pillars carry several historic instruments, including the Ruska theodolite donated by UNESCO. The main pillar is equipped with a ZEISS 020B theodolite in degree-scale to which a Canadian EDA fluxgate magnetometer has been attached.
The EDA fluxgate has an analogue current reading, and therefore the absolute measurements had to be performed with the zero residual method (Newitt et al., 1996, p. 43ff). As described in Sec. 4, the fluxgate has been replaced with a digital instrument during our first trip in November 2015. For absolute measurements, an azimuth mark is located at a distance of 150 m to the 10 4 Geosci. Instrum. Method. Data Syst. Discuss., doi:10.5194/gi-2017-19, 2017 Manuscript under review for journal Geosci. Instrum. Method. Data Syst. Discussion started: 9 March 2017 c Author(s) 2017. CC-BY 3.0 License. southwest. Further, a GEM System GSM-19 proton overhauser magnetometer is available for measuring the magnetic field intensity. Until recently, the time of the absolute measurements was taken from an analogue wall clock which is regularly set according to the GPS time of the LEMI electronics in the variometer house.
In total, there are three observers and one cook who swap shifts in teams of two each week. Therefore, the observatory is usually occupied by two persons who do two consecutive absolute measurements on three days each week. In addition, the 5 head of the observatory and one technician and are both located in Belém, and frequently visit the island. For this purpose, and for transporting goods and fuel to the island, the observatory owns a small motorboat.
Concerning power supply, the observatory is equipped with recently upgraded solar panels of nominal 324 W total, charging eight 165 Ah lead-acid batteries, i.e. 1320 Ah. In addition, there exist two diesel generators of 5 and 6 kW at 120 V which also can be used to charge the batteries. The diesel generator directly powers the lights in the variometer and absolute houses via 10 a dedicated electric cable system. The batteries provide energy mainly for the accomodation building via a 127 V inverter. In parallel, the batteries power the recently installed equipment (Sec. 4).

Recent improvements
With the intention to prepare the Tatuoca Observatory to join the INTERMAGNET network, a team of ON and GFZ visited the observatory for two weeks from November 17 th , 2015 to November 27 th , 2015. During this time, new instruments were 15 installed and new methods for absolute measurements were introduced. During a follow-up visit from October 24 th , 2016 to October 28 th , some further improvements to the instrumentation and absolute measurements were made, as described below.

Variometer House
A Technical University of Denmark (DTU) FGE fluxgate variometer was installed in the variometer house on November, 21th, 2015 (Pedersen and Merenyi, 2016;Rasmussen and Lauridsen, 1990), and baselines are available for this variometer since November 22, 2015. As shown in Fig. 5 and 6, the FGE was installed on the existing western socket, at a distance of about 2.2 m from the LEMI-417 sensor. For testing purposes and as a backup system, the LEMI was kept in operation. The FGE 5 was oriented to magnetic north (HDZ) by minimizing the output of its unbiased Y-sensor while an appropriate bias field was chosen for the X (horizontal north) and Z (vertical down) channels in order to extend the dynamic range of the readings to 6 Geosci. Instrum. Method. Data Syst. Discuss., doi:10.5194/gi-2017-19, 2017 Manuscript under review for journal Geosci. Instrum. Method. Data Syst. Discussion started: 9 March 2017 c Author(s) 2017. CC-BY 3.0 License.

Data transmission
In the same building as the batteries are located (labelled "electronics" in Fig. 3), a netbook and a 3G router were installed. The netbook can connect to remote servers using a reverse SSH tunnel via the 3G network. Indeed, increasing coverage by mobile telecommunication network makes data transmission easy and cheap even in more remote places where expensive solutions (satellites, direct link, dedicated landlines) would have been the only alternative before. However, the SIM card that was used 5 to transfer the variometer data stopped working from February, 4 th to July 20 th , 2016. Since then, data transfer has been reliable thanks to a new SIM card. The laptop is also used as a backup for the variometer data and displays a daily magnetogram for the local staff to check the correct operation of the system. Since October 2016, the absolute measurements are also manually stored in the netbook and transmitted to a remote server. In this way, quasi-definitive data can be produced with reduced latency.
Due to initial problems with a fibre-optical link between the variometer house and the electronics house, the data are manually 10 downloaded from the RaspberryPi datalogger in the variometer house on a daily basis via an ethernet link. Since February 2017, the fibre optics link is fully operational and data is continuously transmitted to a remote server.

Absolute Measurements
In the absolute house, changes were kept at a minimum level while making some significant improvements: First, the EDA fluxgate (E.D.A. electronics Ltd., Ottawa, CA) was replaced by a DTU model G fluxgate and electronics (serial number 15 0151, sensor PIL 7451) on November, 24th, 2015 after eighteen absolute measurements to determine baselines for the FGE variometer were made. Second, the absolute house was cleared from a number of magnetic and nonmagnetic objects on November, 26th, 2015. As a result, potential future movement of magnetic objects and associated changes in the level of the observatory (showing up as apparent changes in the baselines) can be avoided. Also, a clean absolute house makes it easier to identify new and potentially magnetic objects that have accidentally been forgotten. Before and after removing these objects, 20 five absolute measurements were done, respectively. These ten absolute measurements revealed a difference in the absolute level of the observatory of +1.5 nT in the horizontal component (H) and +0.4' (≈3 nT) in declination (D) after the removal of these objects while no difference was found for the vertical component. We note that any change in the absolute level should not exceed one nT in order to preserve the accuracy of the secular variation data from TTB (Matzka et al., 2010). This could have been achieved by correcting all future or past data with an appropriate constant offset. However, there are strong indications 25 that the absolute level of the observatory was not stable to better than 3 nT in the previous periods, and therefore the previous data have not been corrected for this relatively low change in the observatory's absolute level. Instead, the baseline was adopted by introducing a baseline jump corresponding to the jump in the measured absolute values.
In consequence of installing the model G fluxgate, the residual method of absolute measurements was introduced (Jankowsky and Sucksdorff, 1996, p. 89;Worthington and Matzka, 2017, this issue). In this way, the accuracy of the available ZEISS wall clock set according to the LEMI GPS in the variometer house. However, this clock is magnetic and had to be located far enough from the observer, making it hard to read. Therefore, it was replaced by an almost nonmagnetic stopwatch in October 2016. This stopwatch allows to easily read the time with one second accuracy and is set according to the system time of the netbook in the electronics house. In turn, the netbook's system time is synchronized via NTP with its GPS and several remote NTP servers. Mainly, these challenges result from the trivial fact that inclination is close to zero near the magnetic equator.
A first problem arises as the telescope is nearly vertical during inclination measurements and a zenith ocular is needed to read the vertical circle for positions where the telescope points upwards (for an alternative method, see Brunke and Matzka (2017)).

10
This situation is even complicated by the fact that the widely used Zeiss Theo 020B has no degree numbers on the vertical circle from 162 • to 179 • and from 181 • to 198 • . Thus, only the minute marks can be read from the vertical circle if the telescope is pointing down. A slow and cumbersome remedy is to count the number of degree-marks between the closest numbered mark and the desired telescope postion. Another method is to assume a feasible degree number (e.g. the same one as with the last absolute measurement) and to compare the results of the absolute measurement (baselines, sensor offset, collimation angles) 15 with the previous absolute measurements. In this way, a wrong reading will lead to non-consistent absolute measurements and can easily be identified. Then, the corresponding erroneous reading of the vertical circle must be corrected by a full degree or even multiples of it, and the correct absolute value can be calculated.
Another problem near the magnetic equator arises as formulas to calculate inclination from DI-flux measurements differ in sign for the northern and southern hemisphere (note that Eq. 5.4 of Jankowsky and Sucksdorff (1996, p. 95) has the wrong 20 sign for the southern hemisphere as well as other sign errors (Matzka and Hansen, 2007)). When the geomagnetic equator is passing the observatory location due to secular variation, it may even happen that an observatory changes its magnetic hemisphere during a single absolute measurement due to the additional daily variation.
Further, telescope positions during inclination measurements are typically denoted 'Sensor up, telescope North' and so on.
If the inclination is very shallow, however, it is not easy to identify if the telescope actually points south or north, and if the sensor is positioned up or down relative to the telescope. Here, a simple rule can help to find the correct position: on 5 the northern hemisphere, the north-pointing telescope will always point upwards, and on the southern hemisphere, the northpointing telescope will always point downwards (Fig. 8). This still may lead to some confusion if an observatory is changing magnetic hemispheres due to the movement of the magnetic equator. Then, certain positions, e.g. 'Sensor up, telescope North' will instantaneously be rotated by 180 degrees (see Fig. 8. However, observers might not realize this situation immediately due to the slow change in inclination and they might report readings in mixed up positions.

10
Moreover, sun observations are potentially necessary to determine geographic north. In this case, the standard methods that involve the leading and trailing limb of the sun are not practicable as the sun is moving nearly vertically. Special considerations on sun observations are detailed in Wienert (1970, p. 136).
Still, absolute observations near the magnetic equator do not only make the measurement process more complicated. Since the vertical component is close to zero, the leveling of the telescope is not very critical for declination measurements at the 15 magnetic equator. On the other hand, leveling errors can cause significant problems for observatories at mid-to high-latitudes, and usually happen due to inexperienced or careless observers.

Data
All available digital variometer data of Tatuoca have been processed along with the available absolute measurements. These   than for the LEMI sensor. Here, the vertical red lines indicate the period for which direct data transmission from Tatuoca was not possible, and the vertical black lines indicate the beginning and end of the visits to Tatuoca. Period IV is the time after the second visit until a lightning strike occurred (2016-10-29 to 2016-12-30).
The variances of the preliminary base values were estimated by first linearly detrending the data. This detrending was done separately for periods when the base values changed abruptly. Then, the standard deviation was calculed for different periods and for the horizontal field, the declination, and the vertical field. In Tab. 1, the resulting standard deviations are summarized.
Overall, the base values are stable to within 3 nT, but very large outliers occur frequently. For the period before our first visit (period I: 2008-06-02 -2015-11-17), only LEMI data are available, and standard deviations are a bit higher as compared 5 to period II, which spans the time from after the first visit to before the internet connection was lost (2015-11-28 to 2016-02-03). Also, the variances of the base values as derived from the FGE sensor are slightly smaller than those of the LEMI sensor, confirming the quality of the FGE instrument. When Internet connection was lost until the second visit (period III: 2015-02-05 to 2016-10-23), the variances of the base values significantly increased in all three components to a level that would be problematic for an Intermagnet observatory. Mainly, the reason is that no immediate feedback could be given to 10 the observatory staff doing the absolute measurements, underlining the importance of regular data transmission for ensuring data quality. After our most recent visit in October 2016 (period IV: 2016-10-29 to 2016-12-30), the quality of absolute measurements has improved, although significant scatter still occurs in D 0 and Z 0 . However, it is expected that a more detailed investigation and correction or removal of misreadings in the absolute measurements in the course of preparing the definitve data 2016 will lead to a significantly better standard deviation.

7 Summary and Outlook
Since 2015, the Observatorio Nacional (ON) of Brazil and the German Research Centre for Geosciences (GFZ) have collaborated in preparing the Tatuoca magnetic observatory to become a member of the Intermagnet network of magnetic observatories. Intermagnet has defined criteria for quality control and data checking, and provides centralized infrastructure for data Figure 11. The power cable between the variometer house and the electronics house was damaged a few meters from the variometer house due to currents induced by a lightning strike close to Tatuoca observatory on December 31 st , 2016.
distribution (Love and Chulliat, 2013). Thus, our efforts will add an observatory adhering to high data quality standards at an insteresting location within the magnetic equator and the south atlantic magnetic anomaly.
As of the end of 2016, a new DTU suspended variometer is installed on Tatuoca along with a modern datalogging system and a GemSystems GSM-90F1 scalar magnetometer. Further, a 3G modem is used to transmit the data to central servers on a daily basis. As well, the EDA fluxgate magnetometer on the ZEISS 020B theodolite in the absolute house was replaced with 5 a DTU fluxgate model G. This latter change allowed to introduce the residual method of absolute measurements, increasing the accuracy of absolute measurements. Definitive base values have been calculated for the period from 2008 to 2015, and preliminary baselines are available for 2016. Most of the time, the base values are stable to within 3 nT, but very large outliers exist. Also, we experienced that it is important to provide immediate feedback to the observatory staff in order to assure high quality absolute measurements. This is particularly important, as a variety of peculiarities complicate absolute measurements 10 near the equator. For example, missing degree marks at some vertical telescope positions make the readings prone to errors.
On the 31 st of December 2016, a lightning strike hit the island of Tatuoca. In consequence, severe currents were induced in the power cable between the variometer house and the electronics house (Fig. 11). Apparently, the lightning protection in the variometer house was protecting the sensitive GSM and FGE electronics, but further testing is required. However, the 10 m ethernet cable attached to the datalogging system was unprotected, and induced currents destroyed the RaspberryPi. Also, the 15 inverter, the netbook, and solar charge controllers were destroyed, probably due to induced currents in the cables leading to the batteries and solar panels. This event underlines the importance of lightning protection at magnetic observatories. We intend to fully repair the damage in February and March 2017.
Although the observatory is in a promising state, further improvements are needed to become a reliable member of the Intermagnet network. First, the stability of the baseline still can be improved. Second, the power supply chain for data recording 20 should become independent of the power supply chain that is available for living and housing of the observatory staff.
In addition to these major tasks, there exists various smaller improvements that we may consider in the future. For example, the temperature of the FGE sensors electronics is changing by approximately four degrees per day. Such a temperature change