Reconstruction of Landslide Activity Using Dendrogeomorphological Analysis in the Karavanke Mountains in NW Slovenia

: Tree ring eccentricity was used to reconstruct landslide activity in the last 138 years in the Urbas landslide located at Potoška planina in the NW part of the Karavanke Mountains, Slovenia. The research was based on the dendrochronological sampling of Norway spruce ( Picea abies (L.) Karst.) in areas of varying landslide intensity. Analysis of a sudden change in the eccentricity index of 82 curved trees concluded that there were 139 growth disturbances and 16 landslide reactivations between 1880 and 2015, with a landslide return period of 8.5 years. Using lidar data, changes in the surface of the digital terrain model (DTM) were compared with changes in the eccentricity index of trees at the same location in the period 2014–2017. On the basis of temporal changes in the eccentricity index and by using spatial interpolation, landslide activity was reconstructed for the period 1943–2015. During this period, landslide intensity increased in the central part of the landslide. Although categorization into seven categories of di ﬀ erent stem curvature was proposed, no distinction between categories with respect to their eccentricity index was found. analyzed by the growth of P. abies , Larix decidua Mill., and Pinus mugo var. mugo [26–28]; rockfalls by L. decidua , A. alba , and P. abies [29–31]; debris ﬂows by Pinus sylvestris L., Pinus cembra L., L. decidua , P. abies, and P. mugo var. mugo [28,32,33]; and especially in French landslides by Pinus uncinata Mill.


Study Area
The study area is situated on the Urbas landslide in the Karavanke Mountains in the NW part of Slovenia. The landslide-prone area is located on Potoška planina under Mt. Vajnež (2099 m), in the hinterland of the settlement of Koroška Bela. In the past, Koroška Bela was already heavily damaged by a debris flow event [70,71], and the risk still exists [72,73].
The landslide-prone area is located between 1125 and 1350 m above sea level and extends across approximately 20 ha, while the active part of the Urbas landslide is 8.53 ha [73]. The volume of the landslide is estimated to be 985,000 m 3 [74]. The Urbas landslide consists of: tectonically deformed black shaley mudstone, siltstone, sandstone and conglomerate; black massive limestone; and vast accumulations of limestone scree that is deposited over all of them. The depth of the landslide is estimated at between 5 and 20 m (14 m) [72,73]. The main body of the landslide is characterized by complex dynamics and is presumed to be a slow-motion slip. Locally, the movements indicate deep-seated rotational sliding [73].
Precipitation data were obtained from the closest weather station (Javorniški Rovt, operated by Slovenian Environmental agency) which is located approximately 3 km from the landslide area. Annual precipitation on the site is between 1600 and 1800 mm, while the maximum 24-h precipitation rate is between 180 and 210 mm [75]. The wider area of Potoška planina is characterized by a large number of springs, many of which (e.g., the Urbas spring) also originate in the active part of the Urbas landslide, causing further saturation of the landslide body [71,73].
The active landslide area is divided into five (5) zones of different landslide intensity, based on geological measurements and field observations ( Figure 1) [71,73]. The first zone is the most active part and is located in the lower part of the landslide (toe of the landslide) where mass movements are estimated to be more than 1 m per year (maximum measured horizontal movements were up to 17.9 m in the observation period) [73]. The terrain is moving at approximately 1 m per year in the second zone, greater than 0.5 m (but less than 1 m) per year in the third zone, and less than 0.5 m per year in zones four and five [73]. and different categories of tree trunk bending, and iv) analyze the relationship between changes in the DTM and tree ring eccentricity.

Study Area
The study area is situated on the Urbas landslide in the Karavanke Mountains in the NW part of Slovenia. The landslide-prone area is located on Potoška planina under Mt. Vajnež (2099 m), in the hinterland of the settlement of Koroška Bela. In the past, Koroška Bela was already heavily damaged by a debris flow event [70,71], and the risk still exists [72,73].
The landslide-prone area is located between 1125 and 1350 m above sea level and extends across approximately 20 ha, while the active part of the Urbas landslide is 8.53 ha [73]. The volume of the landslide is estimated to be 985,000 m 3 [74]. The Urbas landslide consists of: tectonically deformed black shaley mudstone, siltstone, sandstone and conglomerate; black massive limestone; and vast accumulations of limestone scree that is deposited over all of them. The depth of the landslide is estimated at between 5 and 20 m (14 m) [72,73]. The main body of the landslide is characterized by complex dynamics and is presumed to be a slow-motion slip. Locally, the movements indicate deepseated rotational sliding [73].
Precipitation data were obtained from the closest weather station (Javorniški Rovt, operated by Slovenian Environmental agency) which is located approximately 3 km from the landslide area. Annual precipitation on the site is between 1600 and 1800 mm, while the maximum 24-h precipitation rate is between 180 and 210 mm [75]. The wider area of Potoška planina is characterized by a large number of springs, many of which (e.g., the Urbas spring) also originate in the active part of the Urbas landslide, causing further saturation of the landslide body [71,73].
The active landslide area is divided into five (5) zones of different landslide intensity, based on geological measurements and field observations ( Figure 1) [71,73]. The first zone is the most active part and is located in the lower part of the landslide (toe of the landslide) where mass movements are estimated to be more than 1 m per year (maximum measured horizontal movements were up to 17.9 m in the observation period) [73]. The terrain is moving at approximately 1 m per year in the second zone, greater than 0.5 m (but less than 1 m) per year in the third zone, and less than 0.5 m per year in zones four and five [73].

Field Work
Increment cores were taken from P. abies trees with a 400 mm Pressler borer with a core diameter of 5 mm. Cores were taken at the height of visible maximum bending of the tree or at breast height if the tree was not tilted [50,51,76,77]. Increment cores were taken from three different directions: on the lower part of the trunk, where compression wood was expected; on the upper part of the trunk; and perpendicular to the first sample (control sample) [14,52] (Figure 2). Two hundred and fifty-two increment cores were taken from 97 tilted P. abies trees that showed no visible signs of damage related to non-geomorphological disturbances (e.g., wind, snow, bark beetles). Dominant and codominant trees were sampled, thus eliminating any disturbance in growth resulting from competition [52]. Sampling was done in zones two, three, and five. Zone 1 was excluded from sampling due to a lack of trees, and Zone 4 was too similar to Zone 5 and was therefore also excluded from sampling ( Figure 1).

Figure 1.
Location of the Urbas landslide, locations of disturbed and reference trees, and landslide intensity zones. Intensity zones are based on geological measurements and field observations [71,74] (Datasets: Geological Survey of Slovenia, 2017).

Field Work
Increment cores were taken from P. abies trees with a 400 mm Pressler borer with a core diameter of 5 mm. Cores were taken at the height of visible maximum bending of the tree or at breast height if the tree was not tilted [50,51,76,77]. Increment cores were taken from three different directions: on the lower part of the trunk, where compression wood was expected; on the upper part of the trunk; and perpendicular to the first sample (control sample) [14,52] (Figure 2). Two hundred and fifty-two increment cores were taken from 97 tilted P. abies trees that showed no visible signs of damage related to non-geomorphological disturbances (e.g., wind, snow, bark beetles). Dominant and codominant trees were sampled, thus eliminating any disturbance in growth resulting from competition [52]. Sampling was done in zones two, three, and five. Zone 1 was excluded from sampling due to a lack of trees, and Zone 4 was too similar to Zone 5 and was therefore also excluded from sampling ( Figure  1). In order to eliminate the influence of other non-geomorphological factors (e.g., climate, insect outbreaks) which also influence the width of the tree rings of landslide disturbed trees, 21 undisturbed P. abies trees were sampled in an area outside of the landslide body area with similar climate conditions [1,14,53], distance between reference trees and disturbed trees was between 50 and 200 m (Figure 1). Cores were sampled perpendicular to the slope (2 cores per tree).
Locations of both landslide disturbed and reference trees were collected using a Trimble T1 GNSS receiver (Trimble, Sunnyvale, CA, USA) [78] with accuracy up to 1 m and a mobile application collector for ArcGIS (ESRI, Redlands, CA, USA) [79]. Additional attributes collected for each tree included diameter at breast height (DBH), tree height, possible tree damage, number of bends, category of tree bending, height of increment cores, health of the tree, social standing of the tree, and photographs of the trunk. In order to eliminate the influence of other non-geomorphological factors (e.g., climate, insect outbreaks) which also influence the width of the tree rings of landslide disturbed trees, 21 undisturbed P. abies trees were sampled in an area outside of the landslide body area with similar climate conditions [1,14,53], distance between reference trees and disturbed trees was between 50 and 200 m (Figure 1). Cores were sampled perpendicular to the slope (2 cores per tree).

Laboratory Work
Locations of both landslide disturbed and reference trees were collected using a Trimble T1 GNSS receiver (Trimble, Sunnyvale, CA, USA) [78] with accuracy up to 1 m and a mobile application collector for ArcGIS (ESRI, Redlands, CA, USA) [79]. Additional attributes collected for each tree included diameter at breast height (DBH), tree height, possible tree damage, number of bends, category of tree bending, height of increment cores, health of the tree, social standing of the tree, and photographs of the trunk.

Laboratory Work
Laboratory work consisted of preparing fresh increment cores for digitalization using standard dendrochronological procedures [80]. Polished increment cores were scanned using the ATRICS © tool (Tom Levanič and Slovenian Forest Institute, Ljubljana, Slovenia) [81]. The widths of individual tree rings were measured in the CooRecorder program, which is a subprogram of CDendro TM (Cybis Electronik & Data AB, Saltsjöbaden, Sweden), with 0.01 mm accuracy. The synchronization of chronologies was done in the PAST4 TM program [82]. First, we constructed a reference chronology by on-screen visually and statistically comparing individual increment curves of undisturbed trees to each other with respect to the statistical significance of the t value after Baillie-Pilcher [83] and Gleichläufigkeit (GLK%) [84]. Reference chronology was created with sufficient sample depth (>5) [52] between 2017 and 1880. To cross-date the increment curves of disturbed trees, we visually and statistically compared them to the reference chronology. Comparison between the reference chronology and disturbed trees was done for individual trees, for zone chronologies, and for chronology of all disturbed trees. Crossdating identified missing and/or false rings in the tree ring series. Additionally, comparison with the reference chronology of the undisturbed site also helped to distinguish non-geomorphic from geomorphic signals in the tree ring series (e.g., [1,19,85]). This distinction is possible because variations in growth trends between disturbed and reference trees should be cleared by cross-dating, and any remaining variation in growth trends should be the result of landslide activity. In addition, quality of cross-dating was evaluated using the COFECHA program [86]. Any problematic samples were re-checked in the CooRecorder and PAST4 TM program and removed from further analysis if deemed unusable (damaged). In that way, from 97 disturbed trees, 15 trees were excluded.

Spatial and Temporal Reconstruction of Landslide Activity
Landslide active years in the tree ring series were identified using the eccentricity index Ei. Ei is a computational tool that can be applied to dendrochronology, usually to analyze certain phenomena (e.g., wind stress, avalanche, landslide), but always in relation to tree ring eccentricity [14]. In this article, Ei was calculated using the formula described in [52,87]: where R B is the width of a tree ring where eccentricity is expected, R C is the width of a tree ring perpendicular to the R B side. Ei values are without units, an exception being Ei defined by [14]. The Ei index is further categorized into three groups: 0 (Ei < 0.25), 1 (0.25 < Ei <0.5), and 2 (Ei > 0.5) [88]. Using a combination of the Ei indexes for the whole tree ring series, signals can be additionally classified into four intensity classes [18], here, the goal is to identify abrupt changes in the intensity of a signal (for further explanation, see [52]). Abrupt changes in signal intensity can be interpreted as a tree reacting to tilting [18,52].
The next step consisted of identifying landslide reactivation years, where a threshold of GD ≥ 3 and It ≥ "3%" were used. GD represents the number of trees with an abrupt eccentricity signal in a given year. Threshold values were selected based on the recommendations of [36], which state that GD ≥ 3 and It ≥ "3%" is best for the signal-to-noise ratio when using a sample size of approximately 100 trees. The It index was calculated using the formula by [49,62]: where R T is the number of trees showing responses in their tree ring record in year T, and A T is the number of sampled trees in year T. Additionally, spatial interpolation was applied to the Ei values of landslide disturbed trees for the period from 2014 to 2016, where the interpolated area of extent was compared to changes in the DTM for the period between 2014 and 2017. Interpolation was done using the kriging tool in ArcMap 10.7 (ESRI, Redlands, CA, USA) [89]. Because the focus was on landslide events, Ei values from the first rings with eccentricity signalizing a landslide event were used.

Landslide Magnitude and Eccentricity
The Urbas landslide was previously delineated into five zones of different landslide magnitudes based on field observations and in-situ measurements [71,73,90]. We tested if the geological zonation of landslide magnitudes is also reflected by tree ring eccentricity. It was expected that trees growing in the landslide zone where the greatest movements were observed would be most influenced by the landslide. For this purpose, mean Ei values from the first rings with eccentricity signalizing a landslide event were calculated and then compared between different zones (2, 3, and 5). A comparison was done for a shorter period between 2014 and 2017 and a longer period between 1880 and 2017.

Landslide Intensity and Tree Trunk Bending
Based on field observations of tree crowns and trunk curvature, stem curvature was categorized following the suggestions of [14,91] (Figure 3). Homogeneity of variance and the Tukey's HSD (honestly significant difference) test were used to determine if there is a statistical difference between categories and their eccentricity. Mean Ei values were computed for individual trees, and negative Ei values were also included because we hypothesized that if a tree was not curved in the downslope direction, ring widths would have higher values in that direction.

Landslide Magnitude and Eccentricity
The Urbas landslide was previously delineated into five zones of different landslide magnitudes based on field observations and in-situ measurements [71,73,90]. We tested if the geological zonation of landslide magnitudes is also reflected by tree ring eccentricity. It was expected that trees growing in the landslide zone where the greatest movements were observed would be most influenced by the landslide. For this purpose, mean Ei values from the first rings with eccentricity signalizing a landslide event were calculated and then compared between different zones (2, 3, and 5). A comparison was done for a shorter period between 2014 and 2017 and a longer period between 1880 and 2017.

Landslide Intensity and Tree Trunk Bending
Based on field observations of tree crowns and trunk curvature, stem curvature was categorized following the suggestions of [14,91] (Figure 3). Homogeneity of variance and the Tukey's HSD (honestly significant difference) test were used to determine if there is a statistical difference between categories and their eccentricity. Mean Ei values were computed for individual trees, and negative Ei values were also included because we hypothesized that if a tree was not curved in the downslope direction, ring widths would have higher values in that direction.

Changes in the DTM in Relation to Tree Ring Eccentricity
In this article, the relationship between landslide activity and changes in terrain were analyzed by comparing positive and negative changes in the DTM to mean Ei values of trees in the same geolocation for the period between 2014 and 2017. The DTM was obtained by lidar scanning. Data from 2014 was acquired from the Surveying and Mapping Authority of Slovenia by airborne laser scanner RIEGL LMS-Q780, and 2017 data from the Geological Survey of Slovenia by terrestrial laser

Changes in the DTM in Relation to Tree Ring Eccentricity
In this article, the relationship between landslide activity and changes in terrain were analyzed by comparing positive and negative changes in the DTM to mean Ei values of trees in the same geolocation for the period between 2014 and 2017. The DTM was obtained by lidar scanning. Data from 2014 was acquired from the Surveying and Mapping Authority of Slovenia by airborne laser scanner RIEGL LMS-Q780, and 2017 data from the Geological Survey of Slovenia by terrestrial laser scanner RIEGL VZ-400. The precision of measurement between data sets differ; 2014 data had a precision of ±0.02 ∆X, ±0.02 ∆Y, ±0.025 ∆Z, and 2017 data had a precision of ±0.01 m. As an indicator of the spatio-temporal landslide activity, cumulative surface displacement between 2014 and 2017 was calculated [69]. For every tree that was disturbed by a landslide reactivation event in the period between 2014 and 2017, Ei was calculated using tree ring widths from the first year after the reactivation event. Later on, negative elevation change (material loss) and positive elevation change (accumulation of material) were compared to Ei values in the same geolocation of the tree.
Further testing of the relationship between DTM change and Ei included the possibility that elevation change could have an impact on tree roots and consequently influence tree growth. Damage to root systems due to landslide activity has been well documented (e.g., [1,37]) although in our case we tested if changes in the DTM in the proximity of a tree could have an impact on Ei. In order to test if changes in the DTM influence Ei, mean differences in the DTM between 2014 and 2017 were extracted within buffer zones of 2, 5, 7.5, and 10 m around the tree location. A radius of 2 m as a buffer zone was based on the documented average radius of the root system of P. abies [92], while the other buffer zone diameters were based on the fact that roots can extend laterally nearly as far as a tree's height [93]. Therefore, approximations of 5, 7.5, and 10 m were also used as buffer diameters.

Results
On the active area of the Urbas landslide, 82 trees with an average age of 116.4 years were included in the eccentricity analysis. The oldest tree was around 175 years old and the youngest was 16 years old. The average DBH was 39.7 cm and the average height was 21.7 m. The average age of the reference trees was 110.6 years. The oldest reference tree was around 160 years old and the youngest around 84 years old. The reference chronology was based on 38 increment cores from 20 trees and covers the period from 1880 to 2017. There was a statistically significant similarity between the chronology of disturbed trees and the reference chronology (GLK index = 71.3%, Coefficient t BP = 5.63, Years of overlap = 144).

Temporal Reconstruction of Landslide Activity-Landslide Reactivation Years and Intensity of Events
In the last 136 years (1880 to 2015), 139 GD were recognized. Landslide activity was expressed by the It index and GD. The It index exceeded the threshold value (>"3%") for the first time in 1891, but a very low number of trees recorded that event. The It index exceeded the threshold value (of "3%") in the following years: 1891, 1897, 1899,1900,1906,1908,1917,1918,1923,1943,1944,1956,1971,1977,1988,1991,1992,1994,1995,2007,2008,2009,2012, and 2016 ( Figure 4). Based on both the It index and number of GD, 16 yearly anomalies could be recognized. Their return period is 8.5 years. In the period between 1880 and 2015, the first landslide event is detected between 1942 and 1943, while the following nine potential landslide reactivation periods were in 1955, 1970, 1976, 1987, 1990 to 1994, 2006 to 2008, 2010, and 2015. scanner RIEGL VZ-400. The precision of measurement between data sets differ; 2014 data had a precision of ± 0.02 ∆X, ± 0.02 ∆Y, ± 0.025 ∆Z, and 2017 data had a precision of ± 0.01 m. As an indicator of the spatio-temporal landslide activity, cumulative surface displacement between 2014 and 2017 was calculated [69]. For every tree that was disturbed by a landslide reactivation event in the period between 2014 and 2017, Ei was calculated using tree ring widths from the first year after the reactivation event. Later on, negative elevation change (material loss) and positive elevation change (accumulation of material) were compared to Ei values in the same geolocation of the tree.
Further testing of the relationship between DTM change and Ei included the possibility that elevation change could have an impact on tree roots and consequently influence tree growth. Damage to root systems due to landslide activity has been well documented (e.g., [1,37]) although in our case we tested if changes in the DTM in the proximity of a tree could have an impact on Ei. In order to test if changes in the DTM influence Ei, mean differences in the DTM between 2014 and 2017 were extracted within buffer zones of 2, 5, 7.5, and 10 m around the tree location. A radius of 2 m as a buffer zone was based on the documented average radius of the root system of P. abies [92], while the other buffer zone diameters were based on the fact that roots can extend laterally nearly as far as a tree's height [93]. Therefore, approximations of 5, 7.5, and 10 m were also used as buffer diameters.

Results
On the active area of the Urbas landslide, 82 trees with an average age of 116.4 years were included in the eccentricity analysis. The oldest tree was around 175 years old and the youngest was 16 years old. The average DBH was 39.7 cm and the average height was 21.7 m. The average age of the reference trees was 110.6 years. The oldest reference tree was around 160 years old and the youngest around 84 years old. The reference chronology was based on 38 increment cores from 20 trees and covers the period from 1880 to 2017. There was a statistically significant similarity between the chronology of disturbed trees and the reference chronology (GLK index = 71.3%, Coefficient tBP = 5.63, Years of overlap = 144).

Temporal Reconstruction of Landslide Activity -Landslide Reactivation Years and Intensity of Events
In the last 136 years (1880 to 2015), 139 GD were recognized. Landslide activity was expressed by the It index and GD. The It index exceeded the threshold value (> "3%") for the first time in 1891, but a very low number of trees recorded that event. The It index exceeded the threshold value (of "3%") in the following years: 1891,1897,1899,1900,1906,1908,1917,1918,1923,1943,1944,1956,1971,1977,1988,1991,1992,1994,1995,2007,2008,2009,2012, and 2016 ( Figure 4). Based on both the It index and number of GD, 16 yearly anomalies could be recognized. Their return period is 8.5 years. In the period between 1880 and 2015, the first landslide event is detected between 1942 and 1943, while the following nine potential landslide reactivation periods were in 1955, 1970, 1976, 1987, 1990 to 1994, 2006 to 2008, 2010, and 2015.  Based on Ei values, landslide events differ by the degree of the signal. Most landslide events had a very weak intensity (83.9%), followed by weak (7.5%), strong (5.4%), and very strong intensity (3.2%). The strongest events with respect to It and GD thresholds occurred in 1943, 1944, 1956, 1971, 1988, 1992, 1995, 2007, 2008, and 2016. The intensity of landslide events decreases when observed chronologically in the retrograde sense. This is the result of decreasing sample depth the further back in time we observe events. The intensity of landslide signals relative to the It index in a particular year is shown in Figure 5. Stronger signals occurred in 1899, 1906, 1910, 1943, 1971, 1988, 1991, 1993, 1995, 1997, 2002, and 2016. Based on Ei values, landslide events differ by the degree of the signal. Most landslide events had a very weak intensity (83.9%), followed by weak (7.5%), strong (5.4%), and very strong intensity (3.2%). The strongest events with respect to It and GD thresholds occurred in 1943, 1944, 1956, 1971, 1988, 1992, 1995, 2007, 2008, and 2016. The intensity of landslide events decreases when observed chronologically in the retrograde sense. This is the result of decreasing sample depth the further back in time we observe events. The intensity of landslide signals relative to the It index in a particular year is shown in Figure 5. Stronger signals occurred in 1899,1906,1910,1943,1971,1988,1991,1993,1995,1997,2002, and 2016.

Spatial Reconstruction of Landslide Activity
The extent of landslide activity is somewhat spatially limited, landslide reactivation events occurred locally, and no massive reactivation events were discovered. In the last 73 years, the extent and intensity of the landslide increased, especially from 1990 onwards ( Figure 6). The greatest landslide activity can be observed in the third zone (in the central part), and these results seem to support the geological observation of the landslide, but not in its entirety. Since the first, second, and third zones with observed and measured higher movement velocities are located on the central landslide longitudinal axis, correlation with Ei values is possible.

Spatial Reconstruction of Landslide Activity
The extent of landslide activity is somewhat spatially limited, landslide reactivation events occurred locally, and no massive reactivation events were discovered. In the last 73 years, the extent and intensity of the landslide increased, especially from 1990 onwards ( Figure 6). The greatest landslide activity can be observed in the third zone (in the central part), and these results seem to support the geological observation of the landslide, but not in its entirety. Since the first, second, and third zones with observed and measured higher movement velocities are located on the central landslide longitudinal axis, correlation with Ei values is possible.

Categories of Tree Tilting and Eccentricity Index
A comparison between Ei values was done for different categories of tree tilting for the period between 1880 and 2017 ( Figure 7). The 'multiple bends' category of tree tilting has the highest median (median = 0.080), followed by 'straight downwards' (median = 0.076), 'straight upwards' (median = 0.043), 'bent downwards' (median = 0.011), 'bent upwards' (median = 0.006), and 'undefined' (median = −0.08). Homogeneity of variance was tested for all categories where we rejected the null hypothesis (p = 0.77). The Tukey HSD test showed that the mean values are statistically insignificant (p = 0.9).

Categories of Tree Tilting and Eccentricity Index
A comparison between Ei values was done for different categories of tree tilting for the period between 1880 and 2017 ( Figure 7). The 'multiple bends' category of tree tilting has the highest median (median = 0.080), followed by 'straight downwards' (median = 0.076), 'straight upwards' (median = 0.043), 'bent downwards' (median = 0.011), 'bent upwards' (median = 0.006), and 'undefined' (median = -0.08). Homogeneity of variance was tested for all categories where we rejected the null hypothesis (p = 0.77). The Tukey HSD test showed that the mean values are statistically insignificant (p = 0.9).

Landslide Intensity by DTM Changes
Based on changes in the DTM in the period between 2014 and 2017, we can detect the highest positive (3.32 m) and negative change (2.19 m) in materials in the SE area of the landslide, where the toe of the landslide is located. This area corresponds to the first zone. Within the zones with sampled trees, the highest DTM changes (−1.68 m, +1.06 m) were recorded in the third zone, where the average negative change was 0.135 m (SD ± 0.14 m) and average positive change was 0.1878 m (SD ± 0.168 m). Ei values were compared to changes in the DTM, which was clipped by the extent of spatial interpolation (Figure 8).
In the analysis of elevation change of landslide activity, no trend was discovered when comparing Ei values with values of material change (accumulation/loss) based on DTM change (Table 1), with respect to different buffer zones.
The statistical tests (Levene; p = 0.38 and Tukey's HSD; p = 0.97) showed no correlation between variances of Ei values between zones for the 1943 to 2017 period. The third zone had the highest median values (median = 0.584, n = 23), followed by the second zone (median = 0.555, n = 13) and the third zone (median = 0.550, n = 1) (Figure 9a). Levene's test (p = 0.18) and Tukey's HSD test (p = 0.58) were used to prove that there are no statistical differences between the second, third, and fifth zones based on Ei values for the period between 2014 and 2016. The highest median Ei was in the third zone (median = 0.535, n = 4), followed by the second zone (median = 0.4425, n = 1) and the fifth zone (median = 0.276, n = 1) (Figure 9b). Because Ei values of only reactivation years were used, sample sizes differ and were too little for any relevant statistical analysis. positive (3.32 m) and negative change (2.19 m) in materials in the SE area of the landslide, where the toe of the landslide is located. This area corresponds to the first zone. Within the zones with sampled trees, the highest DTM changes (-1.68 m, +1.06 m) were recorded in the third zone, where the average negative change was 0.135 m (SD ± 0.14 m) and average positive change was 0.1878 m (SD ± 0.168 m). Ei values were compared to changes in the DTM, which was clipped by the extent of spatial interpolation (Figure 8). In the analysis of elevation change of landslide activity, no trend was discovered when comparing Ei values with values of material change (accumulation/loss) based on DTM change (Table  1), with respect to different buffer zones.   The statistical tests (Levene; p = 0.38 and Tukey's HSD; p = 0.97) showed no correlation between variances of Ei values between zones for the 1943 to 2017 period. The third zone had the highest median values (median = 0.584, n = 23), followed by the second zone (median = 0.555, n = 13) and the third zone (median = 0.550, n = 1) (Figure 9, a). Levene's test (p = 0.18) and Tukey's HSD test (p = 0.58) were used to prove that there are no statistical differences between the second, third, and fifth zones based on Ei values for the period between 2014 and 2016. The highest median Ei was in the third zone (median = 0.535, n = 4), followed by the second zone (median = 0.4425, n = 1) and the fifth zone (median = 0.276, n = 1) (Figure 9b). Because Ei values of only reactivation years were used, sample sizes differ and were too little for any relevant statistical analysis.

Discussion
In this paper, bent trees on an active landslide area were sampled in order to spatially and temporally reconstruct landslide activity. The main focus was to correlate tree ring eccentricity with landslide magnitude (change in the DTM) in different zones of the same landslide body. Firstly, we should comment on constraints regarding geological (magnitude) zonation of the Urbas landslide,

Discussion
In this paper, bent trees on an active landslide area were sampled in order to spatially and temporally reconstruct landslide activity. The main focus was to correlate tree ring eccentricity with landslide magnitude (change in the DTM) in different zones of the same landslide body. Firstly, we should comment on constraints regarding geological (magnitude) zonation of the Urbas landslide, which was based on field observations and in-situ measurements in the last few years [71,73,90]. Movements in each zone are only rough estimations, as the geological observations were not spatially distributed over the entire landslide surface and covered a relatively short time span.

Methodology for Landslide Reconstruction
In this paper, in order to detect landslide years, we used a fixed threshold value of It ("3%") and GD (3). Consideration could be made about using different threshold values and the interpretation of lower threshold values (both It and GD). Firstly, if we were to apply higher threshold values, a great deal of signal would be lost, resulting in misinterpretation of landslide activity, especially if we account for the high proportion of low-intensity years with It index above 0 (Figures 4 and 5). Secondly, as indicated in Figure 4, the abundance of years after 1940 with values of It and GD between 1 and 3 indicates either (a) constant movement of the landslide body with occasional higher or increased activity or (b) poor statistical representation, where threshold values of around 1 and 2 could be interpreted as no/missing landslide activity, and threshold values above 3 as landslide activity.
In this paper, the analysis of landslide movement focused on abrupt changes in eccentricity, although reaction wood, traumatic resin ducts or growth suppression analysis could also be applied [38,94,95]. However, dating landslide activity by analyzing reaction wood or growth reduction are considered to be less accurate than using eccentricity [52,53,95]. Therefore, the same authors advise against using reaction wood analysis due to the documented limitations. Compared to the macroscopic analysis of reaction wood, eccentricity signals provide more data on smaller events and can be considered as a more sensitive recorder of landslide activity [52]. Because reaction wood analysis is mainly based on macroscopic analysis, the approach is subjective and problems can therefore occur because of it's non-measurable features [16]. Thus, reaction wood analysis should be applied only to confirm events, not to date them [96].
Uncertainty in application of eccentricity methods to analyzing landslide dynamics is eccentricity inertia. It occurs due to the fact that the growth of the tree rings is affected for a couple of years after landslide activity has ended (e.g., [69]). Multiple landslide reactivation events that result in formation of eccentricity inertia can lead to a problem of an overlapping signal (e.g., [18,19,52]), which occurs due to heterogeneous and long-lasting landslide movements where eccentric growth cannot fully follow the landslide activity. The new landslide induced growth anomaly can overlap the previous one which was not completely recorded in eccentric growth. Thus, changes in Ei are not recognized and cannot be related to a specific sliding event because the tree does not record any further tilting events if it is already under pronounced tilting pressure. This overlapping signal in the time series can lead to misinterpretation of dating landslide reactivation years [19]. Consequently, the application of eccentricity analysis for the reconstruction of landslide activity that shows overlapping signals is not entirely reliable [38,52,96] and can be considered as minimal (e.g., [1]). Our results might also indicate uncertainty regarding classifying landslide years because in the years that were recognized as GD, eccentricity on average lasted 6.5 years. Events of weak intensity have the longest duration (15.7 years on average). The duration of signals could reflect landslide dynamics (e.g., slow-motion slip or creeping), where low intensity and high-frequency events cause signal overlapping [97]. This also indicates that in the reconstruction of Urbas landslide activity, based on the eccentricity signal the prevailing type of geomorphic process was creeping and not landsliding. Inducing noise in the eccentricity signal can be due to root system morphology of P. abies, which seems to be more sensitive to record soil creeping [97]. By selection of higher values of It index, noise can be filtered out, although often on account of not dating true events [97].
Overlapping signal and eccentricity inertia indicate that a great deal of noise was included in the dendrogeomorphologic records. In order to avoid misinterpretation of events on account of an overlapping signal, some authors [18,52] advise the application of the method used in this article for extracting landslide years only for landslides with rather low frequency and magnitude, and the use of greater sample depth [19]. Further insights into the problem of overlapping signal and inertia could also be made possible by a new signal extraction method where moderate to high changes in eccentricity value would also be acknowledged [53].
Assuming an association between creeping movements and precipitation (Figure 10), the influence of precipitation on creeping can be observed in years 1970 (summer and autumn precipitation), 1976 (spring and winter precipitation), 2010 (winter and summer precipitation), and 2015 (winter precipitation). Although the relationship between increased precipitation records and onset of growth disturbances can be observed in some years, no positive correlation can be observed for all the years where It index indicated landslide activity ( Figure 10).

Landslide Activity Within Zones
Based on eccentricity analysis, the greatest landslide intensity was observed in the central part -Zone 3. Despite the fact that zones of different landslide intensity have been distinguished based on geological observations and measurements in the last decade, trees showed no statistically significant distinction between these zones with respect to Ei. The results of spatial interpolation indicate that trees in the third zone experience similar or even greater physical stress than trees in the second zone. This is probably related to small differences between landslide activity in the second and third zone or continuous sliding intensity transitions between zones. The sampling of trees was not done in Zone 1 since it is not forested due to the erosion processes, but it would otherwise be especially favorable given that the highest landslide magnitude was measured in this area in previous studies

Landslide Activity Within Zones
Based on eccentricity analysis, the greatest landslide intensity was observed in the central part-Zone 3. Despite the fact that zones of different landslide intensity have been distinguished based on geological observations and measurements in the last decade, trees showed no statistically significant distinction between these zones with respect to Ei. The results of spatial interpolation indicate that trees in the third zone experience similar or even greater physical stress than trees in the second zone. This is probably related to small differences between landslide activity in the second and third zone or continuous sliding intensity transitions between zones. The sampling of trees was not done in Zone 1 since it is not forested due to the erosion processes, but it would otherwise be especially favorable given that the highest landslide magnitude was measured in this area in previous studies [71,73].
Trees also showed no statistically significant differences in Ei between these zones in the shorter observation period (2014-2017). The mismatch between Ei and the DTM could be attributed to the missing eccentricity data because in order to determine the landslide year, two subsequent ring widths measurements are necessary to calculate the Ei value. Therefore, we weren't able to determine landslide reactivation events for years 2016 and 2017 with this methodology. Another limitation was the inaccuracy of the GNSS receiver under the tree canopy, as the collected location of a single tree could be off by more than a meter. Another explanation for the similarity in Ei between magnitude zones could be attributed to a combination of the deep-seated character of the landslide (around 15 m) [71][72][73][74] and the plate-root system of P. abies. The plate-root system is typically shallow and was observed to record less GD compared to the heart-root system, which is typically deeper and expected to suffer more damage because it reaches the slip surface after landslide initiation [85]. This could also be the reason why no relationship between the DTM change in buffer zones and Ei index was found (Table 1).
It is also possible that a tree will not record a landslide signal at all. In dendrogeomorphology, there is considerable discussion regarding the selection of appropriate trees for sampling in terms of location, age, DBH, tree species, and damage (e.g., [36,77]). [19,53] recognized the relationship between tree age and tree species and its ability to record a landslide signal. It was documented that P. abies's ability to record a landslide signal usually drops between the 9th and 10th decade of age [95]. In the case of the Urbas landslide, it is possible that trees did not record a landslide signal due to their age. Furthermore, it is also possible that there was sampling error because almost a quarter (20) of the trees did not record a GD, although they were located in zones where landslide magnitude was presumably the highest (Figure 11), or that geological zonation is even more complicated. Hence, it can be concluded that in our case, a portion of the landslide signal was probably lost. observed to record less GD compared to the heart-root system, which is typically deeper and expected to suffer more damage because it reaches the slip surface after landslide initiation [85]. This could also be the reason why no relationship between the DTM change in buffer zones and Ei index was found (Table 1). It is also possible that a tree will not record a landslide signal at all. In dendrogeomorphology, there is considerable discussion regarding the selection of appropriate trees for sampling in terms of location, age, DBH, tree species, and damage (e.g., [36,77]). [19,53] recognized the relationship between tree age and tree species and its ability to record a landslide signal. It was documented that P. abies's ability to record a landslide signal usually drops between the 9th and 10th decade of age [95]. In the case of the Urbas landslide, it is possible that trees did not record a landslide signal due to their age. Furthermore, it is also possible that there was sampling error because almost a quarter (20) of the trees did not record a GD, although they were located in zones where landslide magnitude was presumably the highest (Figure 11), or that geological zonation is even more complicated. Hence, it can be concluded that in our case, a portion of the landslide signal was probably lost. Figure 11. Location of trees that did not record GD. Although trees were located in sites where high elevation change can be observed, they did not record GD. Moreover, their Ei values were all close to 0.
Although no correlation between the DTM and Ei was found, further studies on this aspect could be done on a larger sample size which includes different tree species using a more precise GNSS receiver and multitemporal DTMs. In the case of this study, the time span of the two consecutive lidar measurements (2014 and 2017) was also short, and the data acquisition accuracy of both scanning missions was not the same. Therefore, we were not able to directly compare DTM changes in detail. Figure 11. Location of trees that did not record GD. Although trees were located in sites where high elevation change can be observed, they did not record GD. Moreover, their Ei values were all close to 0.
Although no correlation between the DTM and Ei was found, further studies on this aspect could be done on a larger sample size which includes different tree species using a more precise GNSS receiver and multitemporal DTMs. In the case of this study, the time span of the two consecutive lidar measurements (2014 and 2017) was also short, and the data acquisition accuracy of both scanning missions was not the same. Therefore, we were not able to directly compare DTM changes in detail.

Tree Trunk Categorization in Relation to the DTM and Ei
Our results indicate that there are no statistical differences between Ei and different categories of trunk shape (Figure 7). The possible relation between categories of stem inclination and Ei could be further investigated by analyzing cores from multiple directions and comparing Ei in their respective directions. Categories of tree tilting could be linked to DTM change, where the relationship between elevation change and the direction of tree tilting could be analyzed. Furthermore, the degree of tree inclination and variation between tree species could also be included [85]. Finally, the direction and degree of tree inclination could be linked to specific landslide geomorphology (scarps, ridges, and fissures), possibly indicating landslide dynamics [91]. In our case, the link between landslide dynamics and tree trunk bending was already observed in the field because in the upper parts of the landslide (Zone 2 and upper part of Zone 3), a greater number of bends was observed (category 'pistol-butted','s-shaped'), and trees categorized as 'bent downwards' were predominately found in the lower part of the third zone.

Conclusions
In conclusion, based on our dendrogeomorphical study, relatively weak landslide activity of the Urbas landslide can be observed. Landslide intensity has nonetheless been increasing since 1940, especially in the central part of the landslide. Although geological and remote sensing measurements done in the past years indicate various intensities in different parts of the landslide, presented eccentricity analysis cannot confirm these results, or it is not a suitable tool for landslide zonation. This is most probably related to the inability of a tree to record minor landslide reactivations or the deep-seated nature of the landslide which could move the entire tree and root system without tilting. Despite the fact that no correlation was found between elevation change in the digital terrain model (DTM) and Ei of trees in the same location, further investigation should yield better insights, especially given the limited capacity of the case-specific study. In future research, the coupling of direct (DTM) and indirect (dendrogeomorphology) measurements could provide better insights by enabling the analysis of landslides of greater magnitude variability; by including other growth disturbances, greater sample depth, and species variability; and by providing comparable spatial resolution data from longer time intervals.