Remote Sensing and Multi-Level Data Analyses for Hum na Sutli Landslide Impact Assessment in a Changing Climate

Highlights

What are the main findings?

• For the Hum na Sutli landslide in northern Croatia, a comprehensive development history has been established. The triggering factors were identified, and an initial landslide model was developed using both existing information and newly acquired high-resolution remote sensing data.
• By integrating multi-level datasets, a geological column and an engineering geological map were produced, enabling detailed mapping and characterization of the materials and landslide features.

What are the implications of the main finding?

• The study revealed that the materials present in the research area belong to the Golubovec fm. (clay, silt, and sand), and that a correlation with the Vrbova fm. in northern Croatia can be established based on comparable geotechnical properties and similar failure mechanisms.
• The analyzed precipitation data points towards the ongoing climate changes, and consequently, the presented landslide impact assessment takes into the consideration the changing environmental conditions. Additionally, the study provides a viable basis for developing mitigation plans.

Abstract

In Northern Croatia, the stability of slopes is increasingly compromised by a combination of anthropogenic pressures, seismic activity, and climate-driven changes in precipitation patterns. This study presents an integrated, multi-level investigation of the complex, composite Hum na Sutli landslide to characterize its failure mechanism, identify cascading triggering factors, and provide a quantitative basis for impact assessment and mitigation plan development. By reviewing the existing relevant (geo) data, information on the landslide’s historical background and triggering factors was gathered. Material properties were determined in the field and confirmed via laboratory tests. With the integration of new data and multilevel methodology, including unmanned aerial vehicle (UAV) derived light detection and ranging data (LiDAR) data and Electrical Resistivity Tomography (ERT), the characterization of a landslide type was conducted, and an initial landslide map and model were created. Analyzing precipitation data from over the last 25 years provided insights into the area’s changing precipitation trends, highlighting the importance of continuous monitoring of this site. The presented research results for the Hum na Sutli landslide provide a viable basis for mitigation plan creation. Furthermore, laboratory results establish a correlation in landslide susceptibility between two regional units: the Golubovec and Vrbova formations, based on their similar clay-silt-sand compositions and observed failure mechanisms. The research presented here highlights the benefits of multi-level data analysis, emphasizing the integration of existing data with new high-resolution remote sensing data in order to develop a rapid and reliable initial landslide model.

1. Introduction

Landslides in the Pannonian Basin of Northern Croatia represent a persistent geohazard, where slope stability is governed by the interplay of susceptible geology, anthropogenic modifications, and intensifying climatic triggers [1,2,3]. Landslide types vary, as materials and their mechanical behavior differ when in motion [4,5,6]. However, the most common landslide triggers are water (rain, snow melting, floods, etc.), earthquakes and anthropogenic activities [7,8,9]. In the research area, soil slides are frequent [6,10] and landslides commonly involve both natural (heavy rainfall, snow melting, earthquakes, etc.) and artificial (slope overloading and undercutting, water and sewer system leakage, road construction works, etc.) triggers [11]. The absence of a national landslide inventory for Croatia [12] necessitates that impact assessments and mitigation measures be developed through individual case studies [13]. Remote sensing and a multidisciplinary approach are crucial in developing adequate measures to counter the adverse effects of climate change in landslide mitigation and risk management processes [14,15,16,17].

This study addresses the Hum na Sutli municipality, an area of high landslide susceptibility but with limited detailed data [2,12,18]. The recent activation of a landslide threatening critical infrastructure presents a vital opportunity to develop an integrated, multi-level analysis framework and a landslide model usable in monitoring, mitigation and risk management processes. The site’s significance is heightened by its complex trigger history, potentially involving anthropogenic slope loading, seismic events, and intense rainfall events. The presented research results also facilitate a better understanding of landslide mechanisms in the same or similar conditions or formations in the wider area [3,10,11,13,18], as correlations between distinct landslide-prone geological formations can be interpreted. The research area is located in the Hum na Sutli municipality (Krapinsko-Zagorska County, northwest of Zagreb). This region is a hilly area known to be prone to landslides [2,18] (Figure 1).

2. Materials and Methods

Multi-level data analysis has been applied in Hum na Sutli landslide research. The existing data were reviewed, including geological maps (GMs), engineering geological maps (EGMs), landslide susceptibility maps (LSMs), available orthophotos (OPs) from multiple sources and generations, high-resolution digital elevation models (hrDEMs) from the National Geodetic Administration of Croatia (NGA), and precipitation data from the Croatian Meteorological and Hydrological Service (CMHS). New data were acquired via site mapping, with a developed geological column (GC) and engineering geological map (EGM). Additionally, material sampling from the landslide main scarp was carried out with laboratory analysis. High-resolution remote sensing data were collected in the field (2025) by unmanned aerial vehicles (UAVs), namely, high-resolution light detection and ranging data (hrLiDAR) and high-resolution OPs. On-site geophysical measurements were conducted and electrical resistivity tomography (ERT) cross-sections were developed (with further cross-sections and research planned).

2.1. Landslide Site History

The landslide under investigation has a complex history of activation. According to landowner interviews, the slope had been anthropogenically loaded with up to 3 m of fill material over several decades. Additionally, 15 years ago, sliding occurred at the location in the orchard, the landslide area was filled, and water and gas installations were replaced and moved. On 22 March 2020, Zagreb was struck by an M 5.5 earthquake [19], with near epicentral area peak ground acceleration at surface ~0.3 g [20]. The epicenter was approximately 45 km southeast of the Hum na Sutli. However, initial cracks on the slope appeared following the December 2020–January 2021 Petrinja earthquake series [21]. The foreshock took place on 28th December (M 5.0), and on 29 December 2020, an M 6.2 earthquake occurred near Petrinja [21], with near epicentral area peak ground acceleration at surface ~0.5 g [22]. The epicenter was approximately 100 km southeast of the landslide in Hum na Sutli. Reportedly, movements were constant within the landslide area from 2022 through 2024, although not with the same intensity and not across the whole landslide area. The landslide significantly accelerated in the spring and summer of 2024 following periods of heavy rainfall, leading to the situation observed on 17 January 2025 by the Croatian Geological Survey (HGI-CGS), (Figure 2).

2.2. Existing Remote Sensing Data

Available OPs from multiple sources and generations were reviewed and analyzed (Figure 3). Existing high-resolution OPs from NGA were inspected (Figure 3a) [23] and hrDEMs were acquired (based on the data collected by NGA in the 2021–2023 period; details can be found at [24]). Additionally, historical imagery from Google Earth Pro for the 2008-2025 period proved useful, as initial sliding from March 2024 was recorded (Figure 3b) [25].

2.3. New UAV Data Aquistion

A UAV survey was conducted on 6 March 2025, using a DJI Matrice 300 RTK equipped with a Zenmuse L2 LiDAR sensor. The LiDAR system was selected due to the presence of dense vegetation in the study area, which would limit the effectiveness of conventional photogrammetric methods. The flight was performed at 50 m height above ground level with a speed of 6.1 m/s. The LiDAR sensor was in nadir orientation and the side overlap was 50%. During the entire flight, the UAV utilized RTK positioning to ensure high geolocation accuracy. Additionally, four ground control points were established in the vicinity of the endangered building, and 11 further elevation control points were measured south and east of the landslide, along the road using a Kolida K3 GNSS receiver connected to the Croatian CROPOS network (±2 cm horizontal and ±4 cm vertical accuracy) to further improve georeferencing accuracy.

LiDAR data acquisition yielded a dense point cloud with an average point density exceeding 1000 points per square meter. The raw point cloud was initially processed and georeferenced in DJI Terra v4.3.0. software using options for high point cloud density and optimization for point cloud accuracy. The derived point cloud, along with UAV trajectories was then imported into TerraSolid v025 software for further processing. The point cloud was split along flight trajectories and matching of passes was performed. In the overlapping parts, points from the suboptimal flight path were discarded. Noise was removed and the point cloud was smoothed. Automatic ground classification was performed using default options with manual cleaning for smaller details. Georeferencing was then verified using four GCPs and 11 elevation control points (horizontal RMS error 0.031 m, vertical RMS error 0.042 m). The finished point cloud was used for producing a high-resolution digital terrain model (hrDTM) with a spatial resolution of 0.25 m. Simultaneously, the RGB imagery captured by the L2 sensor was used to generate a high-resolution OP mosaic in Pix4Dmapper. Full image scale was used for key points derivation using the standard calibration method. OP resolution was set as double the ground sampling distance, which resulted in a resolution of less than 3 cm/pixel. OP was then georeferenced using the same control points as for the lidar data (final mean RMS error of GCPs is 0.021 m).

The generated hrDEMs and high-resolution OPs were interpreted within a GIS environment (ESRI ArcMap) to manually delineate landslide features, including the main scarp, flanks, toe, and zones of depletion and accumulation, following established geomorphological mapping approaches [10,26,27].

2.4. Existing Geological Data Review and New Geological and Mapping Data Acquisition

According to the Engineering Geological Map at a scale of 1:500,000 (EGM500), the landslide area is located in the Upper Neogene bedded sedimentary complex. This complex has very variable porosity and permeability, is prone to erosion and sliding, and is composed of sandstone, marl, sand, and clay [28]. The Engineering Geological Map at a scale of 1:300,000 (EGM300), shows similar material characteristics in the landslide area. Materials composed of marl, sand, silt, and clay are present, and these were determined to be clastic Oligocene–Miocene (Ol-M) deposits [29].

According to the chronostratigraphic Basic Geological Map (BGM), at a scale of 1:100,000, sheet Rogatec [30], the head of the landslide is located within the Middle Miocene deposits (Badenian, M₂²), which transgressively overlie older clastic Oligocene–Miocene (Ol-M) deposits. The M₂² complex is composed of marl, clayey limestone, and sandy marl.

However, the newer data—collected as part of the project for the purpose of producing a lithostratigraphic map (LM) at a 1:50,000 scale (sheet Ptuj 3 HGI-CGS, unpublished map data, development in progress)—show somewhat different lithological characteristics in the landslide area bedrock. Specifically, the head scarp, main body, and toe of the landslide lie within the informal Golubovec formation deposit complex. The basal part of the Golubovec fm. is composed of pyroclastic breccias and andesites, concordantly overlain by alternating conglomerates, sandstone, sand, silt, sandy silt, and clayey silt containing coal. Pyroclastic material reappears in the upper part of the formation [31].

The existing engineering, geological, and geological maps reviewed here are small-scale maps in the range of 1:50,000–1:500,000 [28,29,30]. For the landslide location, additional field activities were performed, and new geological data were acquired in order to improve the available geo-data resolution. The geological column was developed at the landslide main scarp (10 m high), with sampling and engineering geological mapping of the landslide and its vicinity. Additionally, a ~9.5 ha area was reviewed, and a landslide can be tracked and identified on the NGA OP from 2011 (Figure 4). That landslide presents the event described by the owner (Section 2.1).

2.5. Geophysical Measurements—ERT in Landslide Research

The 2025 field activities included mapping (geological, engineering geological, and by UAVs), sampling, and on-site geophysical measurements (Figure 5). As the wider Hum na Sutli landslide area is prone to slope movements, ERT was used to gain insight into material properties (resistivity) and, consequently, into the sediments in which these movements occur.

We employed electrical resistivity tomography (ERT) using a POLARES 2.0 system to delineate subsurface structures and the potential sliding surface [32,33,34]. Two cross-sections (ERT1-ERT1′ and ERT2-ERT2′, Figure 4) were acquired in 2025 using a 48-electrode array with 5 m spacing and a Wenner-Schlumberger configuration to optimize resolution for both horizontal and vertical features [35,36,37,38]. Data were inverted using Res2DInv software (v. 4.9.3). The software includes two distinct routines for generating 2D resistivity models, employing either the L1-norm (blocky or robust) or the L2-norm (smoothness-constrained least-squares) inversion methods. For the dataset analyzed in this study, the L2-norm–based inversion was applied, as it provides optimal results when subsurface resistivity varies gradually [39,40]. This approach minimizes the sum of squares of both the model resistivity spatial variations and the data misfit. The inversion process was terminated once the RMS error dropped below 10%.

2.6. Laboratory Analysis

Material characteristics were determined in the field during geological and engineering geological mapping and ERT measurements. Additionally, four samples (HKL-1, HKL-2, HKL-2/1, and HKL-3) were taken from the landslide’s main scarp to the laboratory in order to determine the sliding plane material ratios. Particle size distribution analysis was conducted on these samples to determine the ratios of gravel, sand, silt, and clay present on the sliding surface area. Granulometry was determined through a combination of dry sieving (for the coarse fraction) and hydrometer analysis (for the fine fraction, <0.063 mm) following ASTM standards [41,42,43,44]. Sediments were classified based on the relative proportions of clay, silt, sand, and gravel particles, using the ISO classification system in practical use by HGI-CGS [45].

2.7. Precipitation Data

Precipitation data for the 2000–2024 period (the last 25 years) were acquired from CMHS [46] for the nearest rain gauge station (Desinić), approximately 7.5 km south of the Hum na Sutli landslide location. The intention was to analyze precipitation trends, seasonal changes, and heavy rainfall events as landslide triggering events, to find patterns indicating ongoing climate changes, as well as to correlate the results acquired with relevant recent findings for Croatia [3]. Detailed historical precipitation data review and future trends expectations for Croatia are given in [47].

3. Results

The combination of performed field and UAV mapping enabled determination of the landslide area (~5400 m²), main scarp height (~10 m), length of the landslide (~260 m with ~50 m slide and ~210 m flow), width of the landslide (~20 to ~25 m) and landslide height difference (~70 m in total). Conducted mapping, ERT data interpretation, and laboratory analysis results revealed that the materials in the landslide area should be determined as Golubovec fm. deposits composed of sands, silts, and clays (mixtures). Geomorphological features indicate that the landslide is composite type, and the upper part is a rotational slide, while the middle and lower parts are a flow, more precisely, a mudflow. Additionally, sliding plane depth is assumed to be in the range of 5 to 10 m based on terrain and landslide geomorphological features, ERT data, sediment composition, and the conducted main scarp investigation.

3.1. Interpreted High Resolution Remote Sensing Data from Field 2025

UAV-derived datasets—including the hrOP (with a ground sampling distance < 3 cm) and the 0.25 m resolution hrDTM—were used to geomorphologically map the landslide body and its structural components in detail (Figure 6). The combination of high spatial resolution and accurate elevation data enabled us to identify distinct morphological features indicative of different movement mechanisms [5,6,9]. Specifically, the head scarp area exhibited signs of a rotational landslide, characterized by arcuate cracks and stepped topography. The lower portion of the slope displayed geomorphic signatures consistent with a mudflow, such as flow lobes and a smoother surface profile [5,6,9]. At the landslide area, different features and areas are distinguishable: main scarp, flanks, landslide body, and toe area (Figure 6).

3.2. Developed Geological Column and Engineering Geological Map: Integrating Multi-Level Data

After performing field mapping of the landslide area in 2025, a detailed geological column (GC) and engineering geological map (EGM) were developed (Figure 7). Field investigation of the landslide main scarp revealed a ten-meter-thick clastic deposit sequence (Figure 7a). At the base of the examined profile, a four-meter-thick sequence of poorly sorted sandy silt/clay and silty/clayish very fine sand is exposed. This basal unit occasionally contains thin lenticular gravelly sand interbeds. The deposits are massive, predominantly non-carbonate, and bioturbated, displaying a slight upward coarsening trend. Their color varies from ochre and yellowish to gray. Based on their lithological characteristics, the lower four meters of the profile correspond to the clastic portion of the Golubovec fm. complex. Within the first 1–2 m of these deposits, a sliding surface is visible, inclined at 43° northwards (azimuth 10°). The upper boundary of the Golubovec fm. deposits on the head scarp area are erosional and irregular. The overlying unit consists of very poorly sorted silty/clayish sand, with numerous embedded fragments of carbonized wood at its base. This sediment is non-carbonate, gray to yellowish, and lacks visible internal structure. The thickness of this interval ranges from 3 to 3.5 m, most likely representing colluvial deposits of the Pliocene to Quaternary age. The uppermost part of the profile comprises 2.5 to 3 m of unconsolidated, stratified anthropogenic deposits. At least three generations (layers) of backfilled material are distinguishable. These anthropogenic deposits consist of various debris, including bricks, stone blocks, glass, and other rubble. The profile terminates with a recent soil layer, up to 10 cm thick, covered by low grass vegetation (Figure 7a).

On the developed EGM of the Hum na Sutli landslide area, we mapped the investigated landslide in detail, along with other landslides in the research area (Figure 7b). In the developed landslide inventory, active, dormant, and relict landslides can be differentiated. Landslide distinction was based on landslide features visibility during mapping [10,26,48,49]. The 2011 landslide can be clearly identified on hrDEM from 2023, while on hrDEM from 2025, objects (house, courtyard, infrastructure, and the road) are damaged or endangered by the active 2025 landslide. One EG unit is defined for the area: Lower Miocene Golubovec fm. mixtures composed of clay, silt, and sand (CMS). This unit is prone to landslides. On a relatively small area, numerous landslides, or at least some landslide features, can be mapped. Additionally, on the developed EGM, both measured and planned ERT cross-section positions are marked, along with the position of the assumed fault.

3.3. Geophysical Measurements—Interpreting ERT Cross-Sections

The results of the ERT measurements provide a general overview of the landslide area and its surroundings (Figure 8). Data inversion resulted in RMS errors of 7.8% and 3.2% for ERT1-ERT1′ and ERT2-ERT2′, respectively. Resistivity distribution shows relatively low values, generally from 5 to 240 Ωm.

Two domains can be distinguished in the ERT1-ERT1′ cross-section: an upper domain (Layer 1) with a variable thickness of 5–15 m and characterized by resistivity values from 15 to >135 Ωm. This zone contains local anomalies with both low (15–20 Ωm) and higher (generally 20–70 Ωm, and up to 240 Ωm at 180 m distance) resistivity values. The other is a bottom domain (Layer 2) located at a depth higher than 15 m with resistivity values of 40–70 Ωm (locally up to 200 Ωm). In the bottom layer, three distinct zones characterized by small-scale resistivity anomalies are identifiable at distances of 20–40 m, 70–100 m, and 130–160 m, with resistivity values ranging from 5 to 15 Ωm. Additionally, this layer exhibits local high-resistivity anomalies, generally ranging from 80 to 140 Ω·m, with values reaching up to 200 Ω·m at approximately 185 m.

ERT2-ERT2′ shows a similar resistivity distribution, with inversion values ranging from 5 to 110 Ωm and two distinct identifiable layers. The upper layer (Layer 1) exhibits discontinuous geometry, with a thickness of at least 10 m in the southeastern part (20–50 Ωm), thinning to 2–5 m in the central section (resistivity >75 Ωm), and thickening again to approximately 25 m in the western part, where it is characterized by low resistivity values (5–20 Ωm). The lower layer (Layer 2), approximately 40 m thick, contains two zones with elevated resistivity values, generally ranging from 60 to 75 Ωm and reaching up to 110 Ωm locally. These two zones are separated by a sharp lateral decrease in resistivity, with values between 10 and 30 Ωm.

The interpreted ERT1-ERT1′ indicates multiple sliding planes with multiple generations of sliding (Figure 8a). A distinct, sub-vertical low-resistivity anomaly observed in the ERT2-ERT2′ profile (Figure 8b) is interpreted as a potential weakened zone, possibly corresponding to a fault. This interpretation is supported by the alignment of the anomaly with geomorphological lineaments identified in the hrDTM (Figure 7b) and field observations (steep relief section, formed gully, and a valley in a forming with a stream).

3.4. Laboratory Analysis Results

Four samples were taken from the landslide’s main scarp area (Figure 7a) for laboratory analysis; this was in order to both clarify and verify the material characteristics determined in the field and confirm the EG unit classification. Particle size distribution analysis (granulometry) was conducted on these samples to determine the ratios of gravel, sand, silt, and clay present on the sliding surface area. Actual material ratios (according to the laboratory test results) are shown in

Table 1. Laboratory analysis results: particle size distribution with classification for the analyzed samples.

Sample	Depth (m)	Gravel, G (%)	Sand, S (%)	Silt, M (%)	Clay, C (%)	Classification
HKL-3	5.5	9	41	23	27	S with CM
HKL-2/1	7.0	6	33	37	24	MSC (mixture)
HKL-2	7.2	2	24	42	32	MCS (mixture)
HKL-1	9.5	0	19	39	42	CM with S

. The in situ field material classification is in alignment with laboratory results. It is worth pointing out the following (Table 1):

• As the depth increases, so does the ratio of small particles. The material from “sandy”, becomes mixtures, and “clayish” and “silty” afterward, as the ratio of small particles (M + C) increases with depth (50% in HKL-3 at 5.5 m of depth, 61% in HKL-2/1 at 7.0 m of depth, 74% in HKL-2 at 7.2 m of depth and 81% in HKL-1 at 9.5 m of depth).
• Consequently, gravel-size particles decrease as the depth increases.
• In the field, “clayish” and “silty” material is clearly visible at the sliding plane (Figure 5d).

3.5. Analyzing Precipitation Data from Desinić Rain Gauge Station

As precipitation was assumed to be the main force behind the propagation and further development of the landslide, we performed analyses on precipitation data from the Desinić rain gauge station. Precipitation data for the last 25 years (2000-2024 period, [46]) were reviewed and interpreted in the context of ongoing climate change [3,47]. An overview of these data is provided in

Table 2. Precipitation data from the Desinić rain gauge station for the 2000–2024 period.

Year	Precipitation (mm/Year) 1	Precipitation Minimum (Monthly Values)	Precipitation Maximum (Monthly Values)
2000	841	0 mm in August	132 mm in November
2001	997	5 mm in February	167 mm in June
2002	918	13 mm in January	148 mm in April
2003	647 2	16 mm in June	140 mm in October
2004	1156	48 mm in November	194 mm in October
2005	1159	21 mm in October	164 mm in July
2006	898	23 mm in December	166 mm in August
2007	1107	7 mm in April	177 mm in September
2008	1096	9 mm in January	186 mm in June
2009	1012	28 mm in September	136 mm in January
2010	1161	44 mm in October	249 mm in September
2011	792	0 mm in November	163 mm in July
2012	1010	2 mm in March	170 mm in September
2013	1182	17 mm in December	230 mm in November
2014	1372 3	14 mm in March	239 mm in September
2015	1014	0 mm in December	233 mm in October
2016	1129	1 mm in December	188 mm in June
2017	993	26 mm in January	245 mm in September
2018	997	18 mm in December	164 mm in July
2019	1264	29 mm in January	182 mm in May
2020	968	18 mm in January	159 mm in July
2021	1080	8 mm in June	242 mm in May
2022	1027	13 mm in March	337 mm in September 4
2023	1368 3	21 mm in February	181 mm in May
2024	1304	20 mm in February	211 mm in September

¹ The average (yearly) precipitation for the Desinić rain gauge station for the 2000-2024 period is 1060 mm. ² As can be seen, 2003 was a “dry” year; ³ 2014 and 2023 were “wet” years; and ⁴ September 2022 was an extremely “wet” month.

. The annual precipitation totals (Table 2) exhibit considerable inter-annual variability, with notably high precipitation recorded in 2014 and 2023. The minimum and maximum precipitation values (months) vary greatly over the year(s), with a somewhat “drier period” (2000-2003) followed by “wetter period” (2004-2024, except 2006 and 2011), and extreme values for 2014 and 2023 (“wet years”, Table 2). The seasonal distribution of precipitation is further explored in Figure 9.

According to data from the Desinić rain gauge station, the average seasonal precipitation values for the 2000–2024 period are as follows: 247 mm in winter (December–February), 295 mm in spring (March–May), 323 mm in summer (period June–August), and 194 mm in autumn (September–November). The highest precipitation is generally seen during summer (with heavy rainfall events), followed by spring, and the lowest values are in autumn. More than 400 mm cumulative precipitation was recorded during the summers of 2010, 2012, 2014, 2017, and 2022 and the springs of 2005, 2008, 2014, and 2023 (Figure 9). The data shows high inter-annual and seasonal variability and higher average seasonal precipitation values for summer and spring than for winter and autumn.

4. Discussion

This research explores landslide development in the Hum na Sutli area over the last 15 years. It can be stated that the landslide occurred in a landslide-prone area with a history of such events. Triggering factors are discussed (heavy rainfall events, earthquakes, and anthropogenic influence), and an initial landslide model is developed with spatial coverage, sliding materials, and movement type defined. Additionally, a geological column and a new EG map are developed. However, as the landslide is active, further research and monitoring are planned, including the development of ERT cross-sections within the landslide body and the installation of movement sensors. With acquired and planned data, we hope to quantify the movement rate, provide a more precise impact assessment, and develop an efficient and economically reasonable mitigation plan.

4.1. Mapping Benefits and Geomorphological Insights from High-Resolution Remote Sensing Data

The combined use of RTK positioning and GNSS-based ground control points ensured high geolocation accuracy, minimized spatial errors, and enabled dataset alignment, something which is essential for reliable comparison between (planned) survey epochs [50,51].

Acquired high-resolution UAV-based LiDAR and photogrammetric data significantly improved topographic data accuracy and landslide mapping [52]. The application of UAV-LiDAR provided a critical, high-resolution dataset that was essential for mapping and characterizing the landslide’s type and complex morphology. The resulting 0.25 m DTM, with a point density exceeding 1000 pts/m², enabled a detailed differentiation of movement mechanisms, such as rotational sliding near the head scarp and mudflow activity in lower zones. This level of detail, which was not discernible in the available 1 × 1 m national DEM, was fundamental in landslide type determination as a composite.

UAV-based approaches offer cost-effective, rapid, and repeatable data acquisition, especially in cases of difficult or vegetation-covered terrain [51,53]. These advantages support accurate mapping, more responsive hazard assessment, and slope stability management [16].

4.2. Geological Column and Engineering Geological Map—Crucial Basic Data Needed for Research

Geological maps of various scales—both basic and engineering—serve as a fundamental starting point for assessing landslide risk in specific areas. Although chronostratigraphic in nature, the BGM of the investigated area (at a scale of 1:100,000) [30], enables the identification of lithological complexes where landslides are likely to occur, provided that triggering mechanisms are activated. However, the principal limitation of this map lies in its scale, which lacks the resolution needed to reliably correlate landslide occurrences with the geological basement at the micro-location level, such as at the landslide site in the Hum area. Specifically, the 1:100,000 map indicates that the landslide head scarp is located within the Miocene M₂² unit, composed of marls and limestones [30], which are generally considered to be stable. In contrast, field observations—supported by a developed geological column and landslide inventory for the research area (Figure 7)—suggest a different geological scenario. To establish the actual lithological conditions, a geological column was recorded at the landslide head scarp. The exposed deposits were found to belong to the clastic complex of the Golubovec fm. (Figure 7), overlain by younger colluvial and anthropogenic materials.

The LSM of the investigated area (at a scale of 1:25,000) [2], enables the identification of areas or zones where landslides are likely to occur and provides information about slope stability in general. However, the developed detailed EGM gives insight into the locality with spatial coverage of the landslide and materials present in the area (clay, silt, and sand).

The Golubovec fm. is characterized by large lithological, as well as structural and textural, heterogeneities. Layer thickness varies between 5 and 50 cm, often bounded by erosional surfaces. Numerous channel forms and normally graded sedimentary bodies of coarse clastics alternate with laminated or massive bioturbated fine-grained sediments, which are often deformed by slumping processes. The total thickness of the Golubovec fm. reaches up to 350 m. The depositional environments of these sediments ranged from marine–brackish settings—particularly in coastal areas influenced by tides—to prodelta and delta environments in the upper parts of the formation [31]. Similar lithological characteristics have been documented within the informal Vrbova fm., represented by an interbedded heterolithic succession of sands, silts, clays, gravels, and, more rarely, coal layers of Pliocene age [54]. In several studies from the broader Zagreb area, particularly within the Vukomeričke gorice region, these sediments have been well documented and described in terms of their lithological properties [55,56], whose detailed mineralogical composition was subsequently analyzed by Kurečić et al. [57]. A developed landslide model within deposits of the Vrbova fm. [3,10], shows a correlation with our model from the Hum na Sutli area. Golubovec and Vrbova formations, based on their similar clay-silt-sand compositions and observed failure mechanisms are landslide susceptible and comparable formations.

4.3. ERT Cross-Sections: Subsurface Insights

ERT cross-sections effectively illustrate landslide sliding surface morphology and burial depth. The primary objective was to identify the boundary between permeable and impermeable layers, which often corresponds to the slip surface depths [58]. Due to the contrasting electrical properties of the subsurface materials, the resistivity profiles can delineate the morphological features of the sliding surface(s) with high reliability. ERT cross-section results were complemented by those obtained from geological and laboratory analysis.

Resistivity distributions allow us to classify the subsurface into two principal domains: (i) an impermeable clay-rich domain (comprising clay and silt with a minor amount of sand) characterized by low resistivity values ranging from 5 to 50 Ω·m; and (ii) a permeable sandy domain (dominated by sand with varying proportions of silt and clay) exhibiting higher resistivity values exceeding 90 Ω·m.

Based on the lithological characteristics of the study area, it is inferred that slip surfaces typically develop at the interface between layers with markedly contrasting resistivity, specifically, where a low-resistivity- overlies or underlies a high-resistivity layer. These resistivity contrasts reflect the vertical heterogeneity commonly observed in such depositional environments.

Slip zones are typically associated with these resistivity contrasts, where an interchange of relatively high-resistivity layers and low-resistivity units is present [59]. The presence of water within these low-resistivity layers can reduce shear strength, creating slippery conditions that facilitate downslope movement, ultimately leading to landslides and ground subsidence [13].

A persistent low-resistivity anomaly is observed in both cross-sections and may represent a mechanically unstable zone, particularly under conditions of increased moisture content due to rainfall infiltration or rising groundwater. Specifically, ERT2-ERT2′ reveals a low-resistivity anomaly at an approximate depth of 25 m and distance of 90 m, with values between 10 and 30 Ω·m; ERT1-ERT1′ shows a similar anomaly at around a depth of 25 m and distance of 90 m and at depth of 40 m and distance of 150 m, with resistivity values ranging from 20 to 30 Ω·m. These anomalies are interpreted as fault or highly permeable zones, promoting water accumulation and enhanced weathering, both of which contribute to slope instability.

4.4. Hum na Sutli Landslide Area Material Properties Findings

Hum na Sutli landslide area materials can be generally described as sand with silt and clay or clay and silt with sand (depending on the site-specific material ratios present). The materials exposed in the landslide head scarp and body can generally be classified as an unconsolidated clastic mixture with low to negligible carbonate content. These sediments range from clayey silt to very fine silty sand, with occasional lenticular beds of sandy gravel. The exact ratios present are hard to determine by field mapping, but laboratory tests can give a definite answer.

At this micro location, the most significant factor contributing to landslide occurrence is the laboratory-confirmed increase in clay content with depth (Table 1). At the base of the head scarp, the clay content reaches >40% (sample HKL-1), gradually decreasing to <30% in the colluvial unit (sample HKL-3). The documented sliding surface (Figure 5d) is situated within the lowermost interval, composed of material containing >80% clay and silt. Such elevated clay and silt content indicates rapid saturation and reduced permeability, inhibiting water percolation through the sediment.

Field observations confirm the same lithological trends, with a general coarsening-upward sequence throughout the entire profile (Figure 7a, Table 1). The uppermost part of the head scarp is composed of a highly porous and permeable material (Figure 2a), classified as an unbound anthropogenic unit (slope fill-up material). These characteristics enable relatively rapid seepage through the upper ~6 m of deposit, which is followed by a gradual decline in percolation rate into deeper zones characterized by progressively higher clay/silt contents.

As “sandy” materials are more permeable than “clayish” materials, sliding commonly occurs upon (often irregular) inter-layer contact, as in Golubovec fm. (and in Vrbova fm. [3]). With the findings presented here (for the Hum na Sutli landslide), and in [3] (examining the Gajevo landslide), two distinct formations of Northern Croatia can be correlated (informal Golubovec fm. and Vrbova fm.). Within these formations, material (lithological) characteristics are similar, and landslides are common; however, relief energy indexes (REIs) are different [18]. The somewhat higher REIs of Golubovec fm. generally indicate the higher altitudes where these deposits can be found. However, the material properties are interpreted as the key preconditioning factor for landslide development at this site, with failure “set-up” by earthquake(s) damaging effects and ultimately triggered by prolonged periods of elevated precipitation in the form of heavy rainfall events.

4.5. Importance of Heavy Rainfall Events

As heavy rainfall events are the most common landslide trigger in Northern Croatia [3,60], daily heavy rainfall events were singled out from January 2000 to December 2024 from the available daily data from the Desinić rain gauge station (

Table 3. Precipitation value ≥ 40 mm per heavy rainfall event (H.R.E.) for the Desinić rain gauge station for the 2000–2024 period.

Year	March	April	May	June	July	August	September	October	November	December	No. of H.R.E. 1
2005		44	44			46					3
2007							52				1
2008					57	55					2
2009	48										1
2010				52			79				2
2011					53			42		47	3
2012					46		64	70			3
2013									42, 48		2
2014						45	50	49			3
2015			82	63			42	43			4
2016				60					69		2
2017								61		42	2
2018			43			42					2
2019						40				45	2
2020					42		44		60		3
2021					43		44				2
2022							60, 63				2
2023	41	41									2
2024	59				67	40	81	47			5

¹ The table presents all recorded events with precipitation ≥ 40 mm per heavy rainfall event. Years and months with no such events are omitted for brevity. Values ≥ 60 mm are shown in bold; values ≥ 79 mm are shown in bold red.

). Regional landslide-triggering precipitation thresholds from prior literature or statistical analysis of the local rainfall record are nonexistent for the area of interest. Consequently, a threshold value for heavy rainfall events, precipitation of ≥40 mm cumulative rainfall per day was set based on analyzed data, engineering judgment, and similarity with landslide research presented in [3]. The data showed that, from 2000 to 2009 (10-year period), only seven heavy rainfall events were recorded (with relatively low values for daily precipitation of up to 57 mm of rain, as seen in Table 3). From 2010 to 2024 (a 15-year period), there were 39 heavy rainfall events. Generally, yearly precipitation values increased and from 2010 to 2024, with gradually more months with a higher amount of (cumulative) rainfall and more critical heavy rainfall events (Table 3). The described context is very similar to the Gajevo landslide situation [3], and the same conclusion can be drawn: “These events are spread throughout the year without a visible pattern, therefore continuous precipitation monitoring is needed” [3]. Precipitation trends for 5-year periods showed that extreme events are increasing, with 0 events in the 2000–2004 period, 7 events in 2005–2009, 13 events in 2010–2014, 12 events in 2015–2019, and 14 events in 2020–2024. Additionally, in the period 2000-2009, there were no events with precipitation values ≥60 mm of heavy rainfall, while in the period 2010–2024, 13 such events occurred, with three extreme events (with ≥79 mm of rainfall in 2010, 2015, and 2024), Table 3. If this data trend continues, it can be interpreted as a potential future risk for slope stability and points toward ongoing climate changes [47].

4.6. Hum na Sutli Landslide Impact Assessment in the Context of a Changing Environment

Even today, a common regulatory framework for appropriately dealing with landslide geohazard is needed [14]. Identifying endangered areas and developing geohazard maps are valuable initial assessment tools in early planning stages of land development projects [61,62,63,64]. “How to” case studies are valuable assets in, for example, optimizing areas for (intensive) natural hazard management [16], landslide monitoring as a mandatory step in landslide risk assessment [65], and feasible operational forecasting for weather-induced landslides; however, these remain complex, difficult, and uncertain tasks [66].

In the context of a dynamic and changing environment, landslides can—and do—commonly develop on slopes with conditional stability [67,68,69]. The Hum na Sutli wider area is prone to landslides [2], and in the recently active landslide area, older landslide(s) can be tracked. The activation of the landslide visible on the OP from 2011 can be associated with the heavy rainfall event from September 2010, with 79 mm of rainfall (Table 3). The Zagreb earthquake occurred in March 2020 [19] and, in the January-June period in 2020, there were no heavy rainfall events, nor deformations or cracks on the slope area. A heavy rainfall event occurred in November 2020 (with 60 mm of rainfall), and the Petrinja earthquake(s) occurred in December 2020 and January 2021 [21], when initial deformations and cracks were formed on the slope area. In March 2024, a heavy rainfall event occurred (with 59 mm of rainfall), forming a visible main landslide scarp. The wet period of July–October 2024 caused the landslide to propagate and further develop. The analysis reveals a multi-stage process that leads to failure. The slope appears to have been preconditioned by long-term anthropogenic loading from slope-fill activities, which likely reduced its baseline factor of safety. The subsequent occurrence of regional seismic events and an increasing frequency of heavy rainfall events appear to have acted as preparatory and triggering factors. This ultimately culminated in the 2024 landslide.

For the Hum na Sutli landslide impact assessment, it can be said that

• Changing environmental conditions played a crucial role in landslide activation(s);
• In the last 15 years, site stability has only worsened;
• Precipitation, in the form of heavy rainfall events, is the main generator of movement;
• The building (house and bakery) is endangered by the landslide;
• Indirectly, the wider area is problematic from the stability viewpoint (road, existing infrastructure network, and other objects in the area);
• Faced with this (stability) problem, a proactive approach must be adopted; for this reason, movement sensor installation and further monitoring are planned for the location.

It is important to underline that, in this case, a relatively minor problem (the slide identified on OP from 2011)—following no mitigating actions taken and the adverse effects of multiple heavy rainfall events—led to major problems (the 2025 slide and situation), posing safety concerns and damaging infrastructure and objects.

At the beginning of this research, the available existing datasets (maps) were insufficient for landslide model development. However, with detailed field and remote-sensing investigations undertaken, landslide impact can be assessed: the house and the bakery are endangered, while the wider area is at long-term risk.

5. Conclusions

Through a review and analysis of the existing and gathered data, Hum na Sutli’s landslide development history was established. It can be concluded that the landslide occurred in a landslide-prone area with a history of such events. Specifically, on the landslide location from 2025, a landslide existed in 2011, and a reactivation of it, with a larger area of impact, appears to be the result of a complex sequence of factors. The slope was likely preconditioned by long-term anthropogenic loading. The timing of initial crack formation, which was reported following the 2020 Petrinja earthquake series, suggests that regional seismicity may have also played a contributing role in the destabilization process. Slope failure was ultimately triggered by heavy rainfall events in 2024.

We developed an initial landslide model using the acquired and analyzed high-resolution remote sensing data (OPs, DEMs, and ERT data), as presented here. The model was further enhanced with field data (mapping and sampling), as well as geological column and EG map development. Additionally, we performed laboratory tests to determine the material properties on the sliding plane (~80% clay and silt, with ~20% sand). In this way, we defined the present EG unit (Lower Miocene Golubovec fm. mixtures composed of clay, silt, and sand) as being prone to landslides. A key scientific outcome of this work is the correlation in landslide susceptibility between the informal Golubovec and Vrbova formations of Northern Croatia. Based on their similar geotechnical properties and failure mechanisms, this finding suggests that landslide hazard models and mitigation strategies developed for one formation may be transferable to the other. This transfer significantly enhances regional landslide risk management capabilities.

Analyzing precipitation trends over the last 25 years shows high inter-annual and seasonal variability and higher average seasonal precipitation values for summer and spring than for winter and autumn. The yearly precipitation values generally increased, with more months with a higher amount of (cumulative) rainfall and more critical heavy rainfall events in the last 15 years. In the period 2010–2024, 13 events occurred with precipitation values ≥60 mm of heavy rainfall, with three extreme events with ≥79 mm of rainfall. This dataset trend can be interpreted as an effect of ongoing climate change. It is unfavorable from the perspective of slope stability, as well as landslide activation and reactivation processes.

The research presented here highlights the benefits of multi-level data analysis, emphasizing the integration of existing data with new high-resolution remote sensing data and providing a viable basis for mitigation plan development. However, as some unanswered questions remain (for example, regarding the actual movement rate), and as there are possibilities for further model enhancements, we recommend continuation of monitoring and research of the Hum na Sutli landslide.