Sensor-Driven Preventive Preservation of Underground Heritage: A Case Study of the Wine Cellars of Baltanás

Abstract

Underground wine cellars represent a fragile form of cultural heritage, where long-term microclimatic imbalance can lead to material degradation, structural instability, and internal collapses. High humidity, limited ventilation, and the difficulty of access complicate both diagnosis and conservation. This study presents preliminary results from a preventive monitoring strategy applied to the underground wine cellars of Baltanás (Palencia, Spain), focusing on temperature, relative humidity, wall moisture content, and ventilation as key drivers of deterioration. A wireless network of commercial temperature–humidity sensors, wall moisture probes, and airflow sensors was deployed in four sections of a representative cellar over a monitoring period exceeding two years. In addition, mobile monitoring was performed using a quadruped robot equipped with a rotating environmental sensing module, enabling measurements in confined and unstable areas. Results reveal strong thermal inertia, persistently high relative humidity frequently approaching saturation, low and intermittent natural ventilation, and sustained internal wall moisture. These conditions are consistent with observed material decay and internal landslides. The monitoring with quadruped robot proved particularly valuable for identifying localized humidity pockets and stagnant air zones beyond the reach of fixed sensors. The study demonstrates how different solutions for monitoring can support preventive conservation strategies for subterranean heritage, providing a scalable framework for early risk detection and informed management decisions.

1. Introduction

Cultural heritage, as an irreplaceable record and legacy of human civilization, is increasingly endangered by complex and interacting threats that accelerate material degradation and compromise structural stability. For much of the twentieth century, conservation strategies were predominantly reactive, relying on passive restoration after visible damage had already occurred. While this approach has proven effective in certain contexts, it is often costly, delayed, and irreversible when degradation processes advance beyond critical thresholds. In recent decades, accelerating climate change, fluctuating humidity regimes, atmospheric pollution, and intensified human activity have exposed heritage assets to cumulative and synergistic stressors that traditional inspection-based approaches struggle to detect and interpret [1,2,3].

In response to these challenges, a growing international consensus has incorporated preventive conservation, a preservation paradigm that seeks to reduce risks at their source through continuous monitoring, early detection, and proactive management rather than post-damage intervention [4]. Preventive conservation depends on reliable, high-frequency environmental data capable of capturing long-term trends, seasonal cycles, and anomalous events. However, many heritage sites, particularly underground structures, are characterized by limited accessibility, fragile conditions, and restricted infrastructure, rendering manual inspections and sporadic measurements insufficient [5,6].

Advances in the Internet of Things (IoT) and wireless sensor networks have enabled the deployment of low-power, non-invasive monitoring systems capable of continuously recording microclimatic parameters in challenging heritage environments. These technologies have been successfully applied in museums, historic buildings, archaeological sites, and subterranean spaces, demonstrating their capacity to support preventive conservation through data-driven assessment of environmental risks [7,8,9]. When combined with artificial intelligence (AI) techniques, long-term sensor data can further support anomaly detection, predictive modeling, and the development of digital twins for heritage management.

AI-driven methods are increasingly demonstrating their value across both tangible and intangible heritage domains, from high-accuracy cultural pattern recognition, deep-learning restoration, and (Natural Language Processing) NLP-based oral-history preservation [10], to sustainability-oriented governance frameworks that guide the responsible and culturally aligned adoption of AI technologies in heritage practice [11]. In parallel, neural-network models have shown strong performance in short-term microclimate forecasting for museum environments, supporting preventive conservation through low-error environmental prediction [12], while real-time AI-enabled control systems are emerging as effective tools for mitigating environmental hazards such as radon exposure in subterranean heritage spaces [13]. AI-supported mobile inspection is also advancing, with recent work demonstrating that mobile platforms, e.g., drone-based environmental monitoring, can be improved through machine learning calibration models that correct flow field induced distortions in CO₂ measurements, enabling more reliable data acquisition in hard to reach heritage environments [14].

Within this context, the EU-funded ARGUS project [15] advances a structured threat-to-sensor mapping methodology that systematically links identified degradation mechanisms with appropriate sensing technologies [16]. This approach enables tailored monitoring strategies adapted to the specific vulnerabilities of each heritage site, supporting scalable and evidence-based preventive conservation across diverse cultural contexts.

This paper applies the ARGUS framework to the Baltanás underground wine cellars (Palencia, Spain), an ensemble of subterranean heritage structures threatened by persistently high humidity, inadequate ventilation, and internal landslides. Although three cellars are currently being monitored within the broader research project, Cellar 88 was selected as the primary case study because it has experienced several collapses in recent years and has not undergone major renovations, making it the most representative and diagnostically valuable example for understanding active deterioration processes. Focusing on temperature and humidity monitoring, ventilation, wall moisture characterization, and targeted mobile monitoring, the study aims to:

• Characterize the microclimatic behavior of a representative underground cellar;
• Identify environmental conditions associated with observed deterioration phenomena;
• Demonstrate how sensor-based monitoring supports preventive conservation strategies in fragile subterranean heritage contexts.

The work presented here reports preliminary results from a more extensive ongoing investigation aimed at fully characterizing the environmental and structural dynamics of the Baltanás cellar network.

2. Materials and Methods

2.1. Context of Case Study Site

2.1.1. Historical Background and Cultural Value

The underground wine cellars of Baltanás constitute an exceptional example of vernacular subterranean architecture associated with traditional viticulture. The complex comprises 374 excavated cellars distributed across a hillside, forming a dense underground network that has been continuously used and adapted since at least the mid-sixteenth century. The earliest documented reference to the Baltanás estate dates to 1543, when ownership by Don Pedro de Zúñiga was formally recorded [17].

Beyond their architectural significance, the cellars embody a rich intangible heritage linked to local winemaking practices, oral traditions, and social customs. Together with associated tools, storage systems, and spatial organization, they represent a living ethnographic landscape of high cultural and historical value. In recognition of this importance, the site has benefited in recent years from coordinated preservation and promotion initiatives, including legal protection as a cultural interest site, 3D documentation campaigns, and the development of a preventive conservation plan [18,19].

2.1.2. Description of the Cellar Complex and Selection of Cellar 88

The Baltanás cellar complex is excavated into a hillside composed of easily excavated geological materials, resulting in a labyrinthine system of galleries, chambers, chimneys, and unloading ducts (Figure 1a). The cellars are configured in a multi-level system, with the cells constructed at varying depths. This configuration ensures that each cellar is surrounded by cells on both sides, as well as cells situated above and below. Figure 1a provides an aerial view of part of the Baltanás cellar town, while Figure 1b shows the point cloud of the studied cellar and its surrounding structures.

For the purposes of this study, Cellar 88 was selected as a representative case study. This cellar has remained within the same family for three generations and underwent a documented restoration in 1927, followed by minor structural reinforcements in recent years [20]. Its geometry, ventilation layout, and state of conservation are characteristic of many cellars within the complex, making it a suitable testbed for sensor-based preventive monitoring. Figure 1a shows an aerial view of the Baltanas Wine Cellar Town and Figure 1b shows the case study used for this paper, cellar 88.

Importantly, Cellar 88 exhibits different types of damage while remaining accessible for instrumentation and inspection. This balance between risk exposure and operational feasibility motivated its selection for detailed monitoring and experimental deployment within the broader cellar system.

2.1.3. Suspected Issues Related to Collapses

Like many subterranean heritage structures, the Baltanás cellars are affected by conditions that increase the risk of structural collapse. The primary contributing factors are persistently high humidity, insufficient ventilation, and ground vibrations or movements.

Elevated internal temperature and air humidity, driven by soil moisture, limited air exchange, and strong thermal inertia, promotes material weakening through condensation, salt crystallization, and biological growth. Ventilation within the cellar network is often inadequate and spatially heterogeneous, depending on chimney geometry, door opening events, and external wind conditions. Poor ventilation favors the accumulation of moisture and gases, further exacerbating material degradation and reducing overall structural resilience.

In addition, the cellars are excavated within a hillside that is now affected by surface traffic, paved roads, and frequent human activity, all of which introduce ground vibrations and dynamic loads. Pavement and road construction also alter the natural infiltration and flow of surface water, increasing the likelihood of moisture accumulation within the subsurface. These factors, combined with moisture-induced weakening of the geological materials, have contributed to localized internal landslides and collapses in recent years, including within parts of Cellar 88. Such unstable areas are often unsafe for direct human access, underscoring the need for alternative monitoring and inspection approaches.

2.2. Wireless Sensors Networks

Different systems based on wireless sensor networks (WSNs) were deployed in Cellar 88 (Figure 2) to enable continuous, non-invasive monitoring of structural and environmental parameters. Owing to the complex geometry and underground setting of the cellar network, a hybrid communication architecture was implemented. This architecture integrates LoRaWAN (Long Range Wide Area Network) sensors, providing long-range data transmission and connectivity with other monitored cellars, with Wi-Fi-based sensors used for higher-frequency measurements and local data access.

Among the hardware used to capture and transmit sensor data, the system includes commercial Arduino MKR WAN 1310-based LoRaWAN microcontrollers. In addition, more robust devices such as Piconodes (1 channel) and Dataloggers (up to 4 channels) Worldsensing are used specifically for the LoRaWAN sensors. The Arduino units have shown communication issues under the cellar’s high humidity conditions, as they are not designed for such harsh underground environments.

Data streams were routed through two cloud-based platforms: ThingSpeak of MathWorks [21] and CMT Cloud of Worldsensing [22]. These platforms support real-time data acquisition, long-term data storage, alert configuration, and advanced analytics for structural and environmental monitoring.

Wireless monitoring was implemented in Cellar 88 to characterize the microclimatic conditions driving material degradation and instability. First sensors installed included temperature and humidity sensors. Given the suspected issues of excessive humidity, limited ventilation, and ground vibrations, the monitoring strategy further focused on those variables. All sensors were installed in a non-invasive manner and operated continuously to capture both short-term fluctuations and long-term trends. Figure 3 shows the location of sixteen deployed sensors and

Table 1. Sensor deployment and measurement types.

Place	Place Characteristics	Sensor Name	Sensor Information
Outside	Accelerometer buried next to the door.	Acc1	Vibrations on the vehicle path outside the cellar
	Temperature and humidity sensor placed on top of door.	TH4	External temperature and air humidity
Entrance	Entrance of the cellar.	AF0	Airflow at entrance
Kitchen	High humidity zone	TH1	Temperature and air humidity
		AF1	Airflow at kitchen chimney
		IH3	Internal humidity in kitchen wall
Wine Press Room	Collapse zone	TH2	Temperature and air humidity
		AF2	Airflow at collapse zone
Gallery 2		TH3	Temperature and air humidity
	Safe zone	WP1	Water potential in healthy wall
		IH1	Internal humidity in healthy wall
Passage 1	Collapse zone	WP2	Water potential in collapsed wall
		IH2	Internal humidity in collapsed wall
Gallery 3	Chimney	AF3	Airflow at cellar middle point
	Wall	Acc2	Vibrations on the cellar
Gallery 4	Chimney	AF4	Airflow at cellar end

summarizes their functions and measurement parameters.

2.2.1. Temperature and Humidity

Ambient temperature and relative humidity were monitored using commercial Shelly wave humidity and temperature (H&T) sensors (Shelly Europe Ltd., Sofia, Bulgaria, Figure 2a,b), selected for their low power consumption, compact form factor, and suitability for long-term deployment in confined environments. Sensors were installed at four strategically selected locations within Cellar 88, chosen to represent areas with differing depths, proximity to the entrance, and ventilation conditions. One sensor was placed outside the cellar to provide a reference for ambient atmospheric conditions and to allow comparison with nearby public weather stations. A second sensor was installed at the base entrance stairs, representing areas most directly influenced by external air exchange. A third sensor was positioned in the kitchen area, a zone known for consistently high humidity but without documented structural collapses. The fourth sensor was installed in the wine press room (lagar), the section of the cellar exhibiting the highest incidence of internal landslides and structural instability. The locations of all sensors are shown in Figure 3.

Temperature and relative humidity were recorded since April 2023, covering approximately 2.5 years. This long-term record enables the identification of seasonal cycles and serves as a foundation for understanding the microclimate of the cellar. The Shelly sensor was intentionally selected for its low-power operation and adaptive logging capabilities, which are particularly suited for long-term monitoring in locations with limited access and constrained power availability. The data are not acquired at a constant sampling frequency because the Shelly sensor employs an adaptive logging strategy inherent to its factory configuration, which is intended to minimize power consumption. In our setup, the device records at least one measurement per day, with additional records generated whenever changes in temperature exceed 1 °C or changes in relative humidity exceed 5%. These thresholds were selected from the available configuration ranges (0.5–5 °C for temperature and 0.5–50% for relative humidity), as variations below 1 °C and 5% were considered not significant for the present analysis and would unnecessarily increase power consumption and data volume. As a result, a given day may contain a single measurement or multiple measurements, depending on the temporal variability of the monitored conditions. Temperature and humidity are reported in Celsius degrees and percentual humidity.

2.2.2. Internal Humidity and Water Potential

To complement ambient measurements, commercial capacitive soil moisture sensors (model SEN0308 of DFRobot, Zhiwei Robotics Corp., Shanghai, China) and soil water potential sensors (model TEROS 32 of METER Group, München, Germany) were installed to assess humidity and water potential within the cellar walls (Figure 2e,f). The internal humidity sensors consist of elongated probes inserted into the wall, allowing direct measurement of internal humidity conditions at approximately 5–10 cm depth rather than surface-level or ambient air values. The water potential sensors, which quantify the energy state of water within porous materials, operate by measuring the tension required to equilibrate water between a porous ceramic matrix and the surrounding material, providing a direct indication of how strongly water is retained or attracted within the masonry. Water potential is expressed in pressure units (kPa), with lower values indicating stronger water attraction and higher moisture-driven transport potential. The water potential sensors were installed deeper within the walls (40 cm approximately) than the internal humidity probes, allowing differentiation between near-surface moisture conditions and deeper water migration processes. Internal humidity values are reported as percentages, while water potential measurements are reported in kilopascals (kPa).

While ambient humidity reflects short-term ventilation and occupancy effects, internal moisture content integrates longer-term processes such as capillary rise, infiltration, and delayed evaporation. These processes are strongly influenced by the fine-grained nature of the surrounding silty–clayey soils [23], characterized by a high proportion of limos and clays and low to medium plasticity, which favor moisture retention and capillary transport. Monitoring both parameters provides a more complete understanding of moisture accumulation mechanisms that could contribute to plaster detachment, salt crystallization, and loss of mechanical strength in earthen and masonry structures [24]. Water potential monitoring helps understanding the water behavior within the cellar walls. Low water potential regions attract water, while water moves away from high water potential regions. By measuring the water potential, regions that show a higher capacity for water absorption, can be identified and studied in more depth.

Three locations were selected to measure internal humidity, and water potential was evaluated in two of the same zones. The first is a region that shows no structural humidity problems. The second one is a region in which several collapses occurred in a span of two years. The third is an area that shows high internal humidity but with no structural problem for now. In the first and second location internal humidity and water potential are measured while in the third only internal humidity is recorded. Figure 3 shows the location of the internal humidity and water potential sensors.

2.2.3. Ventilation

Natural ventilation plays a key role in the environmental behavior of underground constructions. As summarized by [23], passive airflow in subterranean spaces is strongly influenced by ground temperature, outdoor conditions, and the geometry of elements such as tunnels and chimneys. Previous studies have focused on cellar ventilation as it is a complex process that governs cellar behavior through the interaction between external atmospheric conditions and the internal microclimate [23,24,25,26,27].

Ventilation conditions within Cellar 88 were monitored using MEMS (Micro-Electro-Mechanical Systems) airflow sensors (model DF6V, Omron Corporation, Kyoto, Japan) installed at the five air entrances: the main door, the hatch to the dining room, the central air vent and the two chimneys (Figure 2g,h). These openings constitute the primary natural ventilation paths of the cellar. These sensors measure air velocity, and include directional information, with positive and negative signs indicating whether the airflow is directed toward or away from the sensor, respectively. They enable the identification of dominant air exchange patterns and periods of stagnation. Figure 4a illustrates the seasonal ventilation behavior of a single-chimney configuration presented by [28] showing the difference between warm-season and cold-season. In contrast, Figure 4b shows the monitored cellar in this study, highlighting its more complex configuration with multiple air inlets and outlets.

2.2.4. Vibrations

Vibrations were monitored using two accelerometer Worldsensing Vibration Meter (model LSG7ACL-BILH-VIB) sensors configured to measure Peak Particle Velocity (PPV), Figure 2c,d. PPV represents the maximum instantaneous velocity of a particle as a vibration wave passes through the ground. It is calculated as the square root of the sum of the squared PPV components in the x, y, and z directions.

One of the accelerometers was placed inside the cellar while the other was placed outside. This decision was made as the cellar is subterranean and a path that facilitates the passage of vehicles is placed above. Heavy vehicles and diverse human activity were suspected to affect structural stability of the cellar. Hence one of the objectives of this work is to analyze whether vibration produced on the outside or surface level were being transmitted to the cellar structure. The exact location of these sensors is shown in Figure 3.

2.3. Mobile Monitoring with Quadruped Robot

A new mobile inspection system was developed specifically for this study to enable dynamic, mobile environmental monitoring inside underground wine cellars. The system consists of an in-house-designed air-quality “sniffer” module mounted on a Unitree Go2 quadruped robot. This custom module is built around a rotating sensing platform equipped with gas-sensing and microclimate instruments, allowing the robot to sample the surrounding air while moving through the cellar. This represents the first application of such a mobile gas-monitoring system in this underground heritage context, addressing limitations of traditional fixed-sensor networks.

Mobile monitoring was performed along the longitudinal axis of the cellar, covering the accessible length from the entrance (after the stairs) toward the deepest section. Due to the complex geometry of the cellar and the underground environment, wireless communication with the robot was intermittently lost near the terminal section (Gallery 4) of the cellar, limiting real-time data transmission in that area. For this reason, mobile measurements were organized around four predefined locations distributed along the cellar. At each location, the robot briefly paused to ensure sufficient data acquisition, while measurements were continuously recorded during both movement and stopping phases. The values reported in this study therefore correspond to averaged measurements obtained in the vicinity of each marked location rather than to static point measurements.

Single-point monitoring may miss localized variations in microclimate or structural stress. Preservation efforts can be further enhanced by integrating fixed-sensor monitoring with mobile monitoring, which provide complementary insights into site-specific processes, improving the detection of potential hazards and informing targeted conservation interventions. Mobile measurements allow the identification of spatial heterogeneities in microclimatic conditions, such as zones of elevated CO₂, localized humidity accumulation near walls or ceilings, and temperature variations associated with airflow patterns or proximity to entrances. These spatial differences are particularly relevant in underground cellars, where complex geometries, variable cross-sections, and constrained air exchange can lead to stagnant zones and uneven environmental conditions. Moreover, mobile sensing enables the detection of transient or localized phenomena, such as short-term gas accumulation. Overall, the mobile monitoring system complements static monitoring by providing spatially resolved environmental information essential for a comprehensive assessment of cellar conditions and associated hazards.

The mobile monitoring system is composed of an Arduino MKR 1010 Wi-Fi (Arduino S.r.l., Ivrea, Italy) and a set of commercial sensors integrated to measure temperature, humidity, CO₂, CO, SO₂, NO₂, NH₃, VOCs and PM_2.5. The sensing suite includes a Sensirion SCD41 (Sensirion AG, Stäfa, Switzerland) for temperature, humidity and CO₂; a MICS-6814 MOX sensor for CO, NH₃ and NO₂; a DFRobot Gravity SO₂ electrochemical sensor (Zhiwei Robotics Corp., Shanghai, China) for SO₂; and a Sensirion SEN55 (Sensirion AG, Stäfa, Switzerland), which provides PM_2.5 measurements together with a VOC index and NOx index. To characterize airflow conditions, an RK120-09 ultrasonic wind speed and direction sensor (Rika Electronic Tech Co., Ltd., Changsha, China) was incorporated. Mechanical positioning of the rotating sensing head is achieved using a Parallax Inc. (Rocklin, CA, USA) servomotor (model 900-00008). The rotation logic is governed by the wind direction: the servomotor orients the sensing platform toward the incoming airflow to maximize the representativeness of gas measurements. A magnetic end-stop provides an absolute reference position. During initialization, the platform rotates until the end-stop is detected, defining the zero position; from this reference, the system rotates until the detected wind direction is reached. Figure 5a illustrates the mobile sensing system mounted on the robot, while Figure 5b shows the robot inspecting the antechamber leading to the wine press room and Passage 1 in Cellar 88. The points of the cellar that were monitored with the mobile system are shown in Figure 3.

2.4. Data Processing and Analysis

All data processing was performed using Python (version 3.11.9), and all time-series figures were generated using the Matplotlib library (version 3.11) [31]. The original sampling frequency of each variable was preserved during preprocessing to avoid introducing artificial temporal patterns. Internal humidity, water potential, accelerometer, and airflow sensor data were recorded at a constant temporal resolution of 5 min. In contrast, Shelly sensor measurements were acquired at a non-uniform frequency, with at least one measurement per day (n ≥ 1), and additional records logged when variations in the measured variables exceeded predefined threshold values.

For long-term analysis and visualization of the approximately 2.5-year dataset, a median filter with a kernel size of k = 11 was applied to the time-series data. The kernel size was selected empirically: smaller values resulted in noisier signals, while larger values led to excessive smoothing and loss of information. The median filter was applied exclusively for visualization purposes, in order to reduce short-term fluctuations and enhance interpretability, while preserving the original trends. The filtered data were not used for quantitative analysis, which was performed on the unfiltered measurements. For the mobile monitoring campaign, data aggregation and visualization were performed using Microsoft Excel. Spatial comparisons between cellar zones were carried out using bar plots of normalized values or each monitored variable. Normalization was applied to account for the different physical units and scales of the variables, thereby facilitating direct comparison across parameters.

3. Results and Integrated Analysis

3.1. Fixed Monitoring Results

Temperature and Relative Humidity were monitored in four different areas of the cellar (one outside and three inside) using commercial Shelly sensors over a period of approximately two and a half years (Figure 6).

In Figure 6a, the data are plotted as area fluctuations to visually emphasize the magnitude and temporal evolution of the variations recorded by each sensor. This representation facilitates comparison between zones and highlights seasonal patterns and differences in microclimatic stability.

The seasonal effect is clear in both the temperature and humidity measurements. The exterior sensor (TH4) shows greater variability, as the cellar construction goal is to maintain almost constant conditions generating a particular microclimate. Relative humidity within the cellar remains persistently high throughout the year, approaching saturation at all three internal locations. During winter months, a slight decrease in internal relative humidity is observed, despite high exterior humidity levels. This behavior is attributed to reduced temperature gradients between the interior and exterior, which limit condensation processes and reduce moisture accumulation within the cellar air.

Figure 7 shows a zoomed view of the data corresponding to December 2025 and January 2026. This period was selected because the airflow sensors were installed at the end of November, and these two months provide the first complete and continuous interval in which the air inlets and outlets could be monitored reliably.

TH1 consistently recorded higher temperature and relative humidity values than the other internal sensors. The elevated humidity measured by TH1 is attributed to the greater isolation of the kitchen, where air recirculation is limited and ventilation relies primarily on a single chimney, resulting in reduced air exchange compared to the rest of the cellar. Sensors TH2 and TH3, located in more centrally connected areas of the cellar, exhibit very similar thermal and hygrometric behavior, indicating relatively homogeneous conditions in those zones. In contrast, TH4, positioned outside the cellar, exhibits larger variations in both temperature and relative humidity, reflecting direct exposure to ambient atmospheric conditions.

Overall, the cellar demonstrates a strong capacity to maintain stable microclimatic conditions; however, small but persistent spatial differences are observed between zones. The kitchen was selected for internal humidity (IH) analysis because it consistently showed the highest humidity levels, but only IH was measured in that space. In contrast, the safe zone and the collapse zone were selected because both locations contain two water-potential sensors (WP) and two internal-humidity sensors (IH), which allows a direct comparison between the material moisture measured by the IH sensors and the water-uptake tendency of the soil–wall interface captured by the WP sensors. Figure 8 shows the IH and WP measurements for three representative zones: a safe zone (1), a collapse zone (2), and a high humidity zone (3): the kitchen.

With respect to the IH, the collapse zone exhibits the highest values, exceeding 50%, whereas the dangerous zone (kitchen) shows intermediate levels of approximately 35%. In contrast, the safest zone presents substantially lower values, slightly above 10%. Regarding WP, values close to 97 kPa are recorded in the safe zone, whereas significantly lower values, between 20 and 50 kPa, characterize the collapse zone. Since water in unsaturated porous materials moves toward regions of lower water potential, this combination of high superficial moisture content and low water potential at greater depth indicates the presence of a downward-directed hydraulic gradient in the collapse zone. Consequently, the near-surface layer in this area is not only highly humid but also acts as a potential source of moisture for deeper regions of the wall, promoting internal moisture redistribution. Conversely, the safe zone remains near the maximum water potential (≈97 kPa), suggesting a limited tendency for water inflow. Most monitored variables display stable temporal behavior, with the notable exception of WP2, which shows a decreasing trend over time, indicative of an increasing capacity to attract water in the collapse zone.

The ventilation measurements obtained at the five air-exchange points are summarized in Figure 9, where airflow velocity is represented with color coding to distinguish inlets (blue) from outlets (red). Initial expectations, based on preliminary assessments and local stakeholder knowledge, suggested that during winter the entrance door (AF0) would alternate between inlet and outlet behavior, while the chimneys would primarily function as outlets. However, the measured data reveal a more complex ventilation pattern. AF0 consistently functions as an outlet, while AF1 alternates between inlet and outlet behavior depending on external conditions. AF2 behaves as a low-pressure inlet. This is because the hatch connects with the dining room area where the door is most of the time closed. AF3 acts as the primary inlet supplying fresh air to the cellar. Finally, AF4, located at the far end of the cellar, operates predominantly as an outlet. With respect to the flow magnitude, AF2, which is placed on a collapse zone, shows the lowest values. It also shows almost constant values.

During winter, temperature and relative humidity inside the cellar are closer to external conditions than in summer (see Figure 6), resulting in weak buoyancy-driven ventilation. Under these circumstances, airflow is largely stagnant and ventilation is mainly driven by external wind forces. Higher airflow rates are therefore expected during summer, when the thermal and hygrometric contrast between the interior and exterior is greater, enhancing natural ventilation through buoyancy effect. This remains a hypothesis and would require further investigation after recording airflow data during summer months.

Figure 10 shows the PPV values measured by the accelerometer installed outside the cellar (Acc1) with those recorded by the accelerometer located inside the cellar beneath a vehicle traffic path (Acc2). The data indicate that high vibration events outside, primarily associated with vehicle passages and other surface movements, are not propagated inside the cellar, where the accelerometer records negligible vibrations. The observed attenuation of vibrations is likely influenced by the clay-rich materials surrounding the cellar, which are known to exhibit plastic deformation and significant energy dissipation when moisture content varies within the plastic range [32,33]. In addition, the sandstone of the cellar is characterized by high moisture content, which can further reduce vibration propagation by increasing damping and lowering wave velocities. Moreover, the geometry of the cellar system, including its vaulted shapes, wall thicknesses, and spatial configuration, may contribute to the redistribution and attenuation of dynamic stresses. Complex geometries can act as natural vibration filters, reducing the direct transmission of surface waves into the underground voids. Finally, the presence of pre-existing fissures, joints, and microcracks within the sandstone and surrounding materials may also absorb vibrational energy, further limiting the amplitude of vibrations reaching the cellar interior.

The combined effect of high internal humidity and limited ventilation appears to be the main driver of internal landslides, whereas vibrations seem to play a lesser role. Nevertheless, a more detailed investigation is required to fully understand the mechanisms and quantify the associated risks. This emphasizes the need for careful assessment of environmental conditions when planning preservation strategies. Additionally, a dedicated geological study of the surrounding materials is essential to characterize the mechanical properties, degree of consolidation, and moisture sensitivity of the substrate. Such information is critical for evaluating the slope stability, identifying zones prone to weakening, and understanding how water infiltration and seasonal humidity cycles interact with the excavated cellars. The combination of environmental and structural monitoring with geological characterization provides the basis for a comprehensive risk assessment and designing effective conservation measures.

3.2. Mobile Monitoring Results

Figure 11 shows the temperature, humidity, gases and particle concentrations for each of the four zones in which the monitoring took place. Zones are defined in Figure 3 and correspond to Galleries 1, 2 and 3, for Zones 1–3, and the Kitchen for Zone 4. Values shown are normalized to the max-min values in order to have an easier comparison across variables. The normalized values show that variables differ between zones, confirming that the cellar does not behave as a homogeneous environment. However, pronounced contrasts observed in some parameters, such as temperature, should be interpreted with caution. In this case, temperature is highly stable and exhibits very low absolute variability; differences in less than one degree Celsius between zones can appear amplified after normalization, despite representing minor physical variations.

Temperature is relatively uniform across Zones 1–3 but significantly lower in Zone 4, consistent with its different microclimatic behavior. On top of that, Zone 4 shows the highest humidity levels, which is consistent with the TH values reported by TH1, relative humidity of 100% most of the time. It can also be appreciated that from Zone 1 to Zone 3 humidity increases progressively which is consistent with a lower ventilation as Zone 1 is the closest and most direct zone from entrance. Similarly, from Zone 1 to Zone 3 VOCs and PM_2.5 increases. However, for the VOCs Zone 4 exhibits the highest normalized values while for the PM_2.5 the lowest ones. This combination of high VOCs and low PM could be related to the elevated humidity in this area (kitchen), which may favor the development of microorganisms that emit VOCs while simultaneously reducing airborne PM_2.5. This remains a hypothesis and would require further investigation.

With respect to the monitored gases, SO₂ concentrations remain relatively uniform across all inspected zones, suggesting the absence of strong localized sources within the cellar. The remaining gases’ values seem to decrease from Zone 1 to Zone 4. This phenomenon may be attributed to the location of the different zones, with Zone 1 being the closest to the cellar entrance and Zone 4 in a more isolated area where human activity gases, produced outside the cellar, do not reach so easily.

Large error bars for several variables indicate high temporal variability. This is also determined by the low number of data points for each zone as it took a few minutes to cover each area. Nevertheless, it should be noted that the high variability observed over such small spatial scales strongly reinforces the value of mobile monitoring systems. Fixed sensors, which provide measurements at discrete and highly localized points, may fail to capture relevant environmental information occurring in the surrounding area. As a result, point-based measurements can be unrepresentative of the overall conditions within a cellar characterized by strong spatial heterogeneity.

To obtain a truly comprehensive understanding of cellar behavior using only fixed monitoring, an extremely dense sensor network would be required to adequately cover the full extent of the space. In practice, this would imply the deployment of hundreds of sensors, which is technically complex, economically unfeasible, and difficult to maintain in heritage underground environments. Mobile monitoring therefore offers a practical and efficient alternative, allowing successful access to partially collapsed and unsafe zones and proving existing differences across zones. Moreover, it is important to highlight that unsafe zones are not only those affected by collapses or structural instability, but also areas with high concentrations of gases—such as those that can occur during wine production due to the inhalation of carbon dioxide (CO₂) generated by alcoholic fermentation. Mobile monitoring therefore offers a practical and efficient alternative, allowing successful access to partially collapsed and gas-hazardous zones and revealing existing differences across areas. This underscores the critical importance of mobile inspection for achieving a realistic and safe characterization of these environments.

4. Conclusions and Future Work

4.1. Key Findings

This study presents preliminary results from a long-term preventive monitoring strategy applied to an underground wine cellar in Baltanás. The deployment of a sensor network presented significant operational constraints related to power availability, connectivity, and environmental instability. Limited physical accessibility, high humidity, and confined geometries complicated sensor installation and maintenance, while intermittent collapses and unstable surfaces posed additional risks to both equipment reliability and personnel safety. From these challenges, several lessons emerged. Repeated site visits proved essential, as the complex geometry and environmental variability of subterranean spaces require iterative sensor installation, adjustment, and calibration rather than one-time deployment. Communication planning also proved critical: WiFi and LoRaWAN performance must be carefully adapted to cellar geometry, often necessitating hybrid communication strategies or relay nodes to mitigate signal attenuation caused by thick earthen walls and depth. Close cross-disciplinary coordination among engineers, heritage specialists, and local authorities was fundamental in addressing technical constraints, ensuring site safety, and accelerating problem-solving in this complex subterranean heritage context.

Continuous temperature and relative humidity measurements collected over approximately 2.5 years provide a robust baseline for understanding the cellar’s microclimatic behavior. The results confirm a strong thermal inertia and persistently high relative humidity, with limited seasonal modulation, reflecting the stable microclimatic conditions traditionally sought in underground wine cellars, but which may also contribute to moisture-related deterioration when combined with insufficient ventilation and high wall moisture.

Although the cellar generally maintains stable environmental conditions, small but persistent spatial differences were identified between zones. In particular, areas associated with past or ongoing collapses exhibit higher internal humidity and lower water potential values. The combined use of internal humidity and water potential measurements proved effective for distinguishing zones with differing moisture dynamics, suggesting that these parameters are especially valuable for identifying areas at elevated risk of degradation or instability. Internal wall humidity monitoring therefore appears to be a useful and relatively simple indicator for the early identification of potentially problematic zones.

Ventilation measurements revealed a more complex airflow behavior than initially anticipated based on traditional assumptions and stakeholder knowledge. Air exchange patterns varied across openings, with some chimneys acting as primary inlets and others as outlets, and with limited airflow observed in collapse-prone areas. While winter conditions are characterized by weak buoyancy-driven ventilation and generally stagnant airflow, these results remain seasonally limited, and a full characterization will require detailed analysis of summer conditions, when stronger thermal contrasts are expected to enhance natural ventilation.

Vibration monitoring indicates that surface-level vibration events are strongly attenuated before reaching the cellar interior. This suggests that, under the monitored conditions, vibrations are unlikely to be a primary driver of structural damage when compared to the combined effects of high humidity, sustained wall moisture, and limited ventilation. Nevertheless, this conclusion remains tentative and would benefit from longer monitoring periods and complementary geological and geotechnical characterization.

Finally, the integration of mobile monitoring using a quadruped robot demonstrated clear added value. Mobile measurements revealed pronounced spatial variability in temperature, humidity, and gas concentrations over short distances, confirming that the cellar does not behave as a homogeneous environment. These findings underline the limitations of fixed-point monitoring in complex subterranean geometries and show that mobile inspection can efficiently capture localized conditions that would otherwise require an impractically dense sensor network.

4.2. Future Work

As a preliminary study, the work presented here represents the first phase of a broader, ongoing investigation of the Baltanás underground cellar network. Future efforts will expand both the temporal and spatial scope of monitoring, with particular emphasis on capturing full seasonal cycles, including summer periods when ventilation dynamics are expected to differ substantially from winter conditions.

Building on the existing long-term dataset, future work will focus on developing machine-learning-based anomaly detection and predictive models to identify early signs of abnormal microclimatic behavior, such as emerging moisture accumulation, condensation events, or increasing water attraction within walls. These tools aim to support early warning systems and proactive intervention strategies.

In parallel, sensor data will be integrated into digital twin models of the cellar, enabling simulation of ventilation modifications, environmental interventions, and usage scenarios. Complementary computational fluid dynamics (CFD) studies will further investigate airflow patterns and support the design of targeted ventilation improvements to mitigate persistent humidity.

Finally, future phases will incorporate geological and geotechnical characterization of the surrounding materials to better understand moisture–mechanical interactions and slope stability. Together, these developments will contribute to a scalable and transferable preventive conservation framework applicable to other underground wine cellars and subterranean heritage sites.