2.1 Study area

The Blanco River, located in southern Chile (Figure 1a), Los Lagos Region, originates from the flanks of the Chaitén volcano (72° 39′ 7′′W, 42° 50′ 1′′S) and Michinmahuida volcano and flows into the Pacific Ocean. The basin covers an area of 77 km² (Figure 1b), with elevations ranging from 0 to 1545 m above sea level. The area receives an annual precipitation of 2500 to 5000 mm, predominantly during the winter months, causing the Blanco River to exhibit a rain-dominated regime (Dirección Meteorológica de Chile, https://climatologia.meteochile.gob.cl/; 2023). The forest is defined as an evergreen type with high species diversity, especially Nothofagus spp. Trees with diameters larger than 1 m are common in the native old-growth forest that grows along the slopes. The understory is dominated by Gunnera tinctoria, an endemic species known for its rapid growth and large leaves. The Chaitén Volcano erupted in 2008, causing a significant impact on the Blanco River basin. The explosive eruption wiped out 4 km² of forest around the caldera, while the tephra fall caused widespread damage along the slopes in the whole basin. Additionally, volcaniclastic sediments conveyed by Pyroclastic Density Currents (PDCs) deposited on the valley bottom, burying the floodplain (Alfano et al., 2011; Major et al., 2016; Ulloa et al., 2016; Ulloa, Iroumé, Mao, et al., 2015). Secondary disruptions were predominantly generated by excessive rains after the initial period of explosive activity (Pierson et al., 2013), leading to lahar-flood events and the deposition of fine sedimentary material with the consequent avulsion of the main channel into the downstream village of Chaitén. The forest was affected by the eruption in various ways. In areas where the forest was not entirely wiped away, trees remained standing although severely damaged (e.g. scorched, defoliated, abraded and buried) or dead; thus, stand characteristics were measured (Swanson et al., 2013). In the old-growth forest along the slopes, the density ranges from 250 to 550 stems ha⁻¹, and the basal area ranges from 48 to 124 m² ha⁻¹. In the river corridor, younger high-density stands (2100 to 2600 stems ha⁻¹) are also present, with smaller dimensions and lower basal area (63 to 83 m² ha⁻¹) (Swanson et al., 2013). Ten years after the eruption, Iroumé et al. (2020) observed that riparian vegetation had not yet recovered enough to stabilise channel banks and terraces, despite signs of recolonisation in active sedimentary areas elsewhere on the valley floor.

In this work, a study reach of the Blanco River was investigated. The reach is in the downstream part of the main channel (Figure 1b), characterised by an average width of 134.5 m and a slope of 0.5%–1%. The reach spans 29.5 ha and is 2.2-km long (Figure 1c). After the eruption, the river experienced massive aggradation that caused the riverbed and floodplain to rise 6 m. In the following years, the channel started to incise and wander into the deep and unconsolidated layer of fine sediments (Major et al., 2013). As a result, the floodplain is now exposed to large collapses because of the channel's lateral shifts even during ordinary floods. Depending on the annual variations of water discharge, sediment and wood input, the planform changes continuously between wandering and braided configurations (Iroumé et al., 2020), making the fluvial morphology highly variable (Figure 1d). The study reach presents large quantities of wood, and it partially falls into a specific area referred to as the ‘logjam zone’ by Umazano et al. (2014), who attributed this zone with a key role in the avulsion of the channel after the eruption in 2008.

Water discharge data for the Blanco River is limited, with only a short 4-month record available from January to April 2015. However, daily flow discharges (Q) for the period 2015–2023 were estimated by Iroumé et al. (2024), based on the statistical relationship between the 4-month discharge values of the Blanco River and the neighbouring Palena River, located south of Chaitén. The Q series for the period analysed in this study (December 2018 to January 2023) is presented in Figure 2, highlighting the three unmanned aerial vehicle (UAV) surveys used to assess morphological changes and WJ mobility (Section 3.2). The estimated Q during these three surveys were 8.7, 9.4, and 6.8 m³ s⁻¹, respectively. Although these values have been adjusted for an average coefficient of overestimation of 8%, they may still be overvalued by up to 59% (Iroumé et al., 2020). Considering this, it is likely that the water discharges in all 3 days were closer to the Blanco River's base-flow condition of 4 m³ s⁻¹. Based on the hydrological analysis by Iroumé et al. (2020), which established a high-flow threshold of 15 m³ s⁻¹, the Blanco River experienced three high-flow events between December 2018 and January 2023 (Figure 2). These high-flow events occurred during the second time interval (2019–2023), with two remarkable Q peaks of 22.2 and 20.8 m³ s⁻¹ on 16 May 2021 and 12 June 2021, respectively. Despite no high flows >15 m³ s⁻¹ occurring during the first interval (2018–2019), 14.77 m³ s⁻¹ were estimated for 29 November 2019.

2.2 Remote sensing and data preparation

Three UAV surveys were carried out during the summer season (December–February) in three different periods, spaced approximately 1 and 3 years apart: 01/12/2018, 04/12/2019, and 10/01/2023, hereinafter referred to simply as 2018, 2019, and 2023. UAV surveys were carried out using a DJI Phantom 4 Pro (stock camera with 20 MP and FOV 8.8 mm/24 mm) at an altitude of 150 m above ground. The flight plans generated photosets of 105–236 images with 70% front and side overlaps. Between 80 and 160 ground control points (GCPs) were used, depending on the conditions of the channel and the accessibility of the field site. The GCPs were then georeferenced using a GNSS device (Trimble® R6 dGPS; horizontal and vertical accuracy of 8 and 15 mm, respectively). Photo datasets were then processed using the typical SfM–SMV workflow in Agisoft Metashape® to generate three high-resolution orthomosaics (2-cm pixel) and three point clouds.

Vegetation and wood were filtered out from the point clouds using a combination of manual identification and ready-to-use plugins in CloudCompare (https://www.danielgm.net/cc/) to derive DTMs. First, the point clouds were handled in smaller chunks to reduce computation time, increase manoeuvrability, and separate the macro-units, such as the floodplain, which could potentially interfere with the filtering process. This is because the steep flanks of the floodplain are often recognised as a surface to be removed. The Cloth Simulator Filter (CSF; Zhang et al., 2016) was then used to remove vegetation, applying progressively lower mesh resolutions (from 5 to 0.5 m) with variable local thresholds but never below 0.4 m to preserve steep features like bar edges. This process was manually supervised and adjusted. Finally, wood was removed by cropping the point clouds using the polygons of the wood jams, as described in Section 3.4. Moreover, to generate robust and comparable multi-temporal DTMs, residual inaccuracies and alignment issues were reduced through a co-registration procedure. To co-register the point clouds, stable unchanged areas were first identified, including natural and anthropogenic features such as roads, stable meadows and infrastructures, where no changes occurred between the multi-temporal SfM surveys. The Iterative Closest Point algorithm (ICP, Besl & McKay, 1992) was run over subsets of the point clouds cropped in the stable areas. Once the misalignments were minimised among the subsets, the same co-registration matrix was extended to the point clouds. A co-registration procedure was performed to align the more recent point cloud over the older one, according to a pairwise chain (i.e. 2023 to 2019 and 2019 to 2018). Finally, interpolation of point clouds generated three DTMs resampled at 25 cm spatial resolution in ArcGIS Pro 3.1.0.

2.3 Characterisation of river morphology and dynamics

2.4 WJ analysis

The WJs were manually identified in 3 years using the orthomosaics previously described at a resolution of 2 cm. A single operator carried out the digitalisation in ArcGIS Pro 3.1.0, following a consistent methodological approach to reduce the unavoidable subjectivity. Based on the definition of WJs adopted (sensu Gurnell 2013), we considered any accumulation of wood to be a jam if it included at least one piece of large wood (0.1 m in diameter; 1 m in length) clustered with other logs or woody debris, forming a detectable jam at the pixel resolution of the orthomosaics. Following this approach, the minimum size of a detected WJ was 0.9 m². For each year, all the WJs found anywhere in the study reach were digitalised as polygons, and their basic characteristics were recorded.

2.4.1 WJ mobility

To study mobility between consecutive surveys (i.e. 2018–2019 and 2019–2023), polygons were updated in subsequent images to identify wood jams that were missing, newly formed or persistent, where these terms were defined as follows (Figure 3):

• Missing: a WJ that was mapped in the previous survey but not found in the subsequent one. In this context, we assumed that the accumulation might have been dispersed, transported downstream, or altered beyond the operator's ability to recognise it.
• Newly formed: a WJ that was not present in the previous survey but found in the subsequent one. We assumed that this accumulation formed after the deposition of upstream material, either recruited from outside the domain or resulting from the partial separation of a previous WJ.
• Persistent: A WJ that was mapped in the previous survey and found again in the subsequent one. This category includes two types of persistent WJ: type I WJs found in the same exact shape and position (i.e. unmodified stable polygon); type II WJs found in a different shape and/or position (i.e. modified and updated polygon), indicating a translation and/or the addition/removal of elements. The operator classified a WJ as persistent only if at least one element composing the jam was recognised from the previous survey. However, given the size of the study reach and the high WJ density, a search radius of 25 m, equivalent to the average size of the geomorphic units, was established to limit the spatial distance over which a persistent WJ could be identified.

The mobility rates for the two time intervals (2018–2019 and 2019–2023) were calculated as the ratio of the number of missing and newly formed WJs relative to the number of persistent WJs.

To investigate the drivers of mobility at the reach scale, we first carried out an exploratory analysis to detect patterns among the mobility groups in relation to the morphodynamic mechanisms and to WJs' intrinsic characteristics, namely size and shape compactness. In particular, size is intended as the planimetric area covered by the WJ, while shape compactness was calculated using the Convex Hull Ratio (CHR). The CHR is the ratio between the area of the polygon and the minimum convex polygon that encloses the object's geometry. It falls from 0 (low compactness) to 1 (high compactness), and it was chosen over other parameters of complexity and compactness because it does not exhibit size bias. For instance, parameters like the Perimeter-Area Ratio are affected by polygon size, meaning that even if the shape remains unchanged, increasing the polygon size will alter the parameter value (McGarigal et al., 2002). Then, a logistic regression model was set up to evaluate the significance of the relationship and to evaluate the explanatory ability of these factors. Hence, morphodynamic mechanisms, size and shape compactness were set as independent variables, while the binary dependent variable was represented by two categories: moved (i.e. missing and newly formed) and persistent WJs. The logistic models were developed solely for exploratory purposes, aiming to investigate potential controls, rather than to evaluate predictive capacity or generalise to other case studies. The significance of the predictors (p-value < 0.05), their relative importance and the overall significance of the model were tested. The statistical analyses were performed in RStudio (Posit team, 2023).

Finally, to better understand mobility at a smaller scale, we chose accumulations that remained persistent throughout the study period (both 2018–2019 and 2019–2023) but changed shape and position, resulting in type II. Of these elements, the differences in size, distance and the direction of displacement were investigated. In particular, distance and direction were calculated using the vector of displacement of the centroids. The distance was represented by a segment, whereas the direction was classified with respect to the orientation of the flowing channel in three groups: parallel-upstream, parallel-downstream and perpendicular-transversal. The relationship among size difference, distance and direction was analysed to extrapolate conclusions on the dynamics of this subgroup of WJs.

3.1 Morphological characterisation

The Blanco River transformed from 2018 to 2023, changing its morphology because of channel adjustments (Table 1). The area covered by banks decreased over the year, especially between 2018 and 2019, when the percentage lowered from 5.3% to 4.1%. More evident is the change of the floodplain, which experienced a consistent decrease from 2019 to 2023 (−6.0%), suggesting the collapse of large portions of pyroclastic deposits. On the contrary, the proportion of areas covered by bars increased from a total of 44.4% in 2018 to 47.6% in 2019 and finally to 58.2% in 2023. Another significant variation can be observed between the two types of bars. While in 2018 lateral bars covered an area slightly higher than mid-channel bars, in 2019 and 2023, this difference accentuated: 48.1% against 10.1%.

In terms of relative elevation, channel adjustments mainly involved the loss of floodplain areas with a consequent increase of low bars: from 34.6% in 2018 to 43.1% in 2023. High bars increased slightly, reaching up to 21.5% in 2023. The spatial arrangement of geomorphic units and relative elevation classes can be seen in the source maps provided in the Supporting Information (Figures S1 and S2).

3.2 Geomorphic changes and morphodynamics

The DoD maps, presented in Figure 4a, point out different processes in the two periods, while Figure 4b shows the corresponding morphodynamic mechanisms. Between 2018 and 2019 (Figure 4a1), the Blanco River was affected by erosional processes, mainly localised in the downstream part of the reach, whereas deposition was restricted to small spots with a few meters of elevation difference. The shallower erosions in the lower sections involved a combination of mechanisms, including the formation of chutes and channels across the bars (i.e. chute cut-off and lobe dissection), as well as bar-edge trimming (Figure 4b1). On the contrary, the small depositions were associated with mid-channel and lateral bar developments. Furthermore, channel filling was also an extensive mechanism, even though it was only detected by the spatial variation of geomorphic units and not by the DoD. Concerning deeper erosion (elevation difference < −3 m), a large area is visible along the left bank, which is caused by the erosion and collapse of the floodplain as pointed out again in the map of morphodynamics (Figure 4b1). As a result, the net volume of sediment displaced is −9.59 × 10⁴ m³ ± 3.07 × 10⁴ m³, emphasising the magnitude of geomorphic changes undergone by the fluvial system and the amount of sediment transported outside the reach.

In the 2023–2019 DoD (Figure 4a2), deposition processes are shown in the middle part of the reach, with wide and shallow accretion of the lateral bar on the right (elevation difference 0–1 m) and thicker aggradation along the curve on the left (elevation 1–3 m). The former was associated mainly with channel filling and overbar deposition (Figure 4b2), and the latter with mid-channel and lateral bar developments, thus to new bars. However, deep elevation differences (< −3 m) because of floodplain erosions also occurred, with a major area on the downstream right bank. As a result, the net volume of sediment displaced is still negative and equal to −4.75 × 10⁴ m³ ± 3.28 × 10⁴ m³.

Overall, in both time intervals analysed, 50% of the study reach underwent geomorphic change, with lobe dissection, channel filling, and floodplain erosion being the most represented in terms of surface, regardless of the large share of unsolved changes (full areas in Table S2). A total net displacement of −1.43 × 10⁵ m³ ± 6.35 × 10⁴ m³ was derived for the whole 4 years study period (December 2018 to January 2023).

The budget segregation analysis highlighted the importance of floodplain erosions also in terms of sediment volume, with 2.78 × 10⁴ m³ mobilised by this process in 2018–2019 and 6.76 × 10⁴ m³ in 2019–2023 (Figure 5). Specifically considering the 2019–2018 DoD (Figure 5a), lobe dissection, channel incision, and bar edge trimming generated a large amount of sediment volumes (> 1 × 10⁴ m³ of erosion each). As for the maps, also the budget segregation underlined the minor role of channel filling, mid-channel and lateral bars development, and overbar deposition, which show a combination of depositions and erosions. The same four processes were instead characterised by larger deposition volumes according to the 2023–2019 budget (Figure 5b). In particular, channel filling, mid-channel bar development, and overbar mechanisms showed more than 1 × 10⁴ m³ of sediment deposition each. The complete characterisation of morphodynamic mechanisms, areas and budget segregation results, can be seen in Table S2.

3.3 Multi-temporal spatial distribution of wood jams

The number of WJs in the study reach increased from 483 (16.4 per ha) in 2018 to 562 (19.1 per ha) in 2019, and remained quite stable in 2023 with 560 (19.0 per ha) accumulations (Table 2). In terms of sizes, the averages were stable throughout the whole period. The largest WJ was mapped in 2019, corresponding to an area of 2566.7 m². Finally, the overall WJ coverage (percent of the area of the study reach) followed the same trend observed: lowest coverage in 2018 (9.9%) and highest in 2019 (11.7%).

The displacement of WJs on geomorphic units and according to the different classes of elevation was then analysed. In absolute numbers, most of the WJs were found on lateral bars and at low elevations, as these two classes were the most represented (Table 1). However, the normalised abundance of WJs per each class (Figure 6) gave different results. Overall, banks are characterised by the largest number of elements in the 3 years (total > 100 WJ ha⁻¹), even though the number decreased from 2018 to 2023 (Figure 6a). On the contrary, floodplain and the flowing channel host the lowest number of WJs (total < 25 WJ ha⁻¹) in 3 years. Considering the bars, the highest number of WJs per surface unit was found in 2019. Geomorphic units at high elevations are visibly associated with the largest number of jams (total > 100 WJ ha⁻¹), followed by geomorphic units at low elevations (Figure 6b). At channel and floodplain levels, the abundance was again significantly lower (total < 25 WJ ha⁻¹).

3.4 WJ mobility

Between 2018 and 2019, the number of newly formed WJs was higher than the number of missing or persistent elements (Table 3). The result was a normalised number of 10.6 newly formed WJ ha⁻¹ against 7.9 and 8.6 WJ ha⁻¹. However, considering the size, it is evident that the larger WJs were persistent, with an average size and maximum size of 88.8 and 1676.1 m², respectively. On the contrary, newly formed and missing accumulations were characterised by relatively small sizes. Consistently, the area represented by the three classes showed persistent WJs covering 7.6% of the study reach against the 3.4% of the other two classes combined. Regarding the 2019–2023 mobility, the pattern is similar. Persistent WJs were lower in number, but they are the largest accumulations, and they cover a higher percentage of the area of the study reach. However, in this time window, mobility is mainly represented by missing elements (360 WJ; 12.2 WJ ha⁻¹). In the end, a conservative mobility rate (without considering those persistent but changed size and position) over 1 year was 68% (2018–2019), and 78% (2019–2023) over 3 years.

Figure 7 reports the distribution of size and CHR values for the three mobility groups. Persistent WJs showed larger sizes with respect to those newly formed or missing (Figure 7a). On the contrary, overall lower CHR values were observed in persistent WJs, while newly formed ones showed higher compactness (Figure 7b). Ultimately, in both time intervals, the same pattern can be observed: newly formed and missing jams have smaller but more compact shape, and persistent ones having a larger but less compact shape.

The interaction between wood jam mobility classes and morphodynamics is presented in Figure 8, which presents the normalised number of accumulations for each type of mechanism. In the first period, 2018–2019 (Figure 8a), overbar deposition and bank erosion were the two mechanisms associated with the largest number of WJs, 58.1 and 52.7 WJ ha⁻¹, respectively, especially to persistent elements. Newly formed WJs were primarily related to mid-channel bar development, channel filling, lateral bar development, pool scour and floodplain erosion. Bar edge trimming, lobe dissection and chute cut-off instead were associated with missing WJs. Persistency was found primarily in correspondence to bank erosion, overbar deposition and areas of no change. Between 2019 and 2023 (Figure 8b), the number of wood jams was more evenly distributed, with all mechanisms associated with less than 40 WJ ha⁻¹. The representativeness of mobility classes within each mechanism followed a similar pattern to the previous time interval. Persistent elements were found relatively abundant within bank erosion, and no change classes but overbar deposition was associated with newly formed WJs rather than persistent accumulations. Also, newly formed elements were found highly represented within channel filling, mid-channel bar development, pool scour and floodplain erosion. Missing WJs were mainly related to lobe dissection, chute cut-off, bar edge-trimming mechanisms and areas of no change.

Figure 9 presents the size distributions of all WJs based on morphodynamic mechanisms, also reflecting the non-normalised version of the results previously presented. Overall, no remarkable differences among the mechanisms were visible, although bank erosion and lateral bar development showed higher medians in 2018–2019 (Figure 9a) and 2019–2023 (Figure 9b), respectively. Instead, within the mechanisms, the only clear pattern was the presence of clusters of persistent and very large accumulations (up to 2566.7 m², Table 3) in areas with no changes. Considering the numerosity for each mechanism, it is visible that the largest number of WJ was found in areas of no change, as 50% of the study reach did not undergo geomorphic changes.

The logistic regression analysis of mobility 2018–2019 and 2019–2023 reported a good overall fit for both models, with the chi-square tests significant with p < 0.05. However, pseudo-R² was low (0.25 and 0.24), indicating a modest level of explained variance. Several predictors showed consistent and significant effects on mobility across the models. Size had a significant negative effect (p < 0.001), suggesting that larger size reduces the likelihood of mobility. Conversely, CHR exhibited a significant positive influence (p < 0.01), with higher values (higher compactness) increasing the probability of mobility. Among the mechanisms, bar edge trimming (p < 0.05), channel filling (p < 0.001), lobe dissection (p < 0.01) and mid-channel bar development (p < 0.01) all showed significant effects on mobility, each positively affecting mobility. Some predictors, such as floodplain erosion and lateral bar development, were significant in one model but showed consistently positive effects when significant, while others, such as bank erosion and chute cut-off, were not significant in either model, indicating their less critical role. Variable importance using z-score highlighted the primary role of WJ size in both periods, followed by lobe dissection and bar edge-trimming dynamics. Compactness was evaluated as fairly important even though not in the top positions (Figure S3).

Persistent WJs included not only those accumulations that remained completely stable but also those that changed position and/or shape while retaining common elements. The latter accounted for 61% during the 2018–2019 period and 82% during the 2019–2023 period. Figure 10 shows the distance and direction of movement associated with the difference in size. Considering the mobility 2018–2019 (Figure 10a), the chart demonstrates that the distance of movement is dependent on the difference in size primarily when the movement is upstream (R² = 0.65), while transversal movement is poorly caused by the change in WJ size (R² = 0.23). Consistent outcomes were found in the analysis 2023–2019 (Figure 10b), in which movements parallel to the channel, that is, upstream and downstream, are showing stronger relationships, R² equal to 0.65 and 0.63, respectively. Therefore, while upstream and downstream mobility of a WJ is mostly related to the reworking of the WJ itself, perpendicular–transversal mobility is mostly a result of its translation and only a minor part of its reworking.

4.1 Channel morphological changes 10 to 15 years after the eruption

The Blanco River was deeply affected by the Chaitén volcanic eruption, which suddenly released huge amounts of sediments and wood. Cascading processes have been occurring in the years after the eruption as often reported after such kind of large infrequent disturbances (Mazzorana et al., 2019). An increase of landslide activity (Paredes et al., 2023), wood recruitment (Iroumé et al., 2024) and river adjustments (Iroumé et al., 2020; Ulloa, Iroumé, Picco, et al., 2015) was reported to be a significant cascading process in the catchment.

The present multi-temporal analysis of river morphology demonstrated that channel adjustments are still occurring around 15 years after the eruption, adding another piece to the understanding of the post-disturbance evolution of the Blanco River. The outcomes pointed out the significant loss of floodplain area, highlighting the role of lateral shifts in reshaping the channel morphology and, overall, the dynamism of the river system even without extreme flood events, as observed during the first time interval analysed (2018–2019). In total, the percentage of study reach classified as bars increased, reaching a maximum in 2023. A potential explanation for this is the slightly lower water discharge during the 2023 survey, which may have reduced the areas at water level, as indicated in Table 1, thereby increasing the visibility of depositional units. However, it is worth noting that the areas at the water level first decreased between 2018 and 2019, despite water discharges in the study reach being slightly higher in 2019 (simulated Q of 9.4 m³ s⁻¹). In the end, considering that water discharges were not that different among the periods (Q of 8.7, 9.4, and 6.8 m³ s⁻¹; Section 3.1) and also potentially all close around the base flow (Iroumé et al., 2020), we can assume that temporal fluctuations in water levels cannot justify the increase in bar units observed in this study. Instead, this increase might be attributed to planform and vertical changes (i.e. incision) and to the collapse and transformation of floodplain areas. The large variation in the difference between areas covered by mid-channel and lateral bars further supports high fluvial activity. The remarkable reduction in the surface of mid-channel bars mainly in favour of high lateral ones (19.9% against 24.5% in 2018; 10.1% against 48.1% in 2023, respectively) suggests a progressive change in the planform configuration, with the channel perhaps seeking a more wandering configuration like the one previous to the eruption (Ulloa, Iroumé, Mao, et al., 2015). Transitions from multi- to single-thread configuration are typically reported for highly disturbed rivers (Comiti, 2012) and are associated with channel incision and narrowing (Surian & Rinaldi, 2003). In the Blanco River, although channel incision was reported as a major mechanism by the DoD and supported by past evidence (Major et al., 2016), apparently widening is occurring rather than narrowing as indicated by the large floodplain erosions. However, a possible explanation can be provided by integrating our findings with the knowledge of the upstream Blanco River documented by Iroumé et al. (2020). During floods, the active channel fills completely, causing the steep and unconsolidated floodplain to fail. The material deposited at the toe forms extensive high lateral bars, narrowing the main watercourse during ordinary flow conditions. Therefore, although the Blanco River appears to widen locally because of the collapse of large portions of the floodplain, the progressive formation and subsequent abandonment of active sedimentary units force the transition to a single-thread configuration. The absence of extreme flood events, however, associated with extensive and deep floodplain collapses during the period 2018–2019, supports this theory. Our perception is that this process will become irreversible once the Blanco River is no longer able to reach and collapse the floodplain even during ordinary flood events. At that point, the floodplain will have definitively become a terrace, lateral bars will form the new floodplain, and the Blanco River will revert to a wandering water course.

It is clear that the lower Blanco River is still highly dynamic and far from a condition of equilibrium, if it even exists. The analyses of geomorphic change detection, morphodynamics, and budget segregation confirmed the dynamism of the Blanco River and quantified its activity in terms of morphological changes between 2018 and 2023. In total, the net sediment output during these 4 years was 1.43 × 10⁵ m³, 67% coming from floodplain erosions, resulting in an average annual output of 3.58 × 10⁴ m³. This corresponds to a sediment yield of 1.21 × 10⁵ m³ km⁻² year⁻¹. Although these data cannot directly represent the sediment yield of the entire catchment because of possible undetected sediment transport and non-representative study reach dynamics, they still provide a valuable order of magnitude. This is particularly relevant in a catchment where quantitative sediment and hydrological data are still little investigated (Basso-Báez et al., 2020). Post-eruption river basins have significantly higher annual sediment yields than non-volcanic basins affected by different large disturbances (Korup, 2012). Volcanic-affected streams show remarkable sediment yields especially the first 5–10 years after the eruption (up to 10⁷ m³ km⁻² year⁻¹), but they can persist at high rates even in the next decades (10²–10⁵ m³ km⁻² year⁻¹) owing to bed and bank erosion and from mass wasting from the terraces (Pierson & Major, 2014). The same mechanisms were observed in the Blanco River, which proved an acute response in terms of morphological changes and sediment displacement even after 10–15 years from the eruption.

4.2 WJs: abundance, distribution and mobility

Between 2009 and 2012, Ulloa, Iroumé, Picco, et al. (2015) reported variations of WJ abundance (number of WJ) in the range of 0–60% along the main Blanco channel, although their assessment was carried out with satellite data, hence not directly comparable with ours. Subsequently, the variation in WJ abundance over 1 year, from 2015 to 2016, ranged between 11% and 91%, with greater changes observed in upstream reaches (Tonon et al., 2017). In the present work, the variation in WJ abundance and size was greater in the single year from 2018 to 2019 (+16.3%) than in the 3-year period from 2019 to 2023 (−0.4%). However, the methodological approach applied in this study allowed for the derivation of the mobility rate, which shed more light on true WJ dynamics. As a result, mobility rates of 68% and 78% were estimated for the 2018–2019 and 2019–2023 periods, respectively. In the same downstream segment of the channel, Iroumé et al. (2024) calculated that the percentage of WJs that disappeared or reworked between 2017 and 2022 was higher than 88%.

The difference in mobility between the two time intervals can be explained by the number of high-flow events that occurred during the second time interval versus the first. Between 2019 and 2023, three floods (Q > 15 m³ s⁻¹) were reported, with two exceeding 20 m³ s⁻¹. These are higher than the 99.9% of daily Q reported by the flow duration curve of Iroumè et al. (2020), calculated from 2005 to 2018. On the other side, the absence of extreme floods during the first period is associated with a 68% mobility rate, indicating that WJs in the Blanco River are highly mobile even during ordinary flow conditions.

Considering WJ abundance within geomorphic units, the largest number of WJ was found on lateral bars at low elevations, as these were the most common morphological units. After normalising the data, banks stood out as the geomorphic unit with the highest density of WJs, exceeding 100 WJ ha⁻¹ across the 3 years studied, followed by bars and secondary channels. This high density on banks could be attributed to their typical role as transition zones between the floodplain, where dead forest provides logs for recruitment through erosion, and the channel, where water flow transports and deposits the wood (Gurnell et al., 2002). If we consider the normalised data, our results follow the state of the art about preferential sites of wood presence for single thread to wandering rivers, where wood is primarily stored along the outer channel margins, on concave banks, point bars, and along the edges of islands (when present) and secondary channels (Abbe & Montgomery, 2003; Gurnell, 2013).

As mentioned previously, in the two time windows analysed, more WJs moved than persisted. WJ persistency, or vice versa mobility, depends on numerous variables (Wohl, 2013). Our results showed that persistent WJs were larger and less compacted. Larger WJs are more likely to remain stable for longer periods and have a greater chance of recruiting additional pieces and developing vegetation, which in turn further stabilises the accumulations, creating a positive feedback mechanism. In the Blanco River, this mechanism is evident because of the fast colonisation of vegetation on larger accumulations, as visible in the example provided in Figure 11, where vegetation quickly developed on a persistent WJ with a size greater than 1000 m². Additionally, shape complexity was identified as a key factor by Abbe and Brooks (2011), as complex key pieces enhance stability by anchoring the entire accumulation. Similarly, we demonstrated that the less compact the overall shape of the WJ, the longer it persists.

Besides areas with no change, which favoured the persistence of wood jams, the primary morphodynamic mechanisms associated with persistent jams were overbar deposition and bank erosion, in particular during the first time interval. In the Blanco River, thicker overbar sediment sheets were frequently observed in areas with large WJs. This suggests that wood accumulations have a significant ability to interact with sediment fluxes, slowing water velocity, promoting sediment deposition and stabilising of fluvial landforms. Gurnell et al. (2001) highlighted the pivotal role of wood in the development of vegetated islands. Ravazzolo et al. (2015) reported a positive relationship between wood volumes and deposition processes in the bar units of the Tagliamento River (Italy). Curran (2010) discussed a positive feedback mechanism in the San Antonio River (USA), where increased sediment deposition around the wood facilitates the development of a mid-channel bar, which in turn becomes a preferential deposition site for more wood, further stabilising the geomorphic units. In contrast, the linkage with bank erosions is more unclear, as such a process is typically reported to favour recruitment and mobility rather than stability (Comiti et al., 2016). However, if we shift the perspective and consider that in-channel wood can also redirect flow and induce localised bank erosion (Massé & Buffin-Bélanger, 2016; Wohl, 2013), it can be suggested that it is the persistence of large wood jams that drives bank erosion. Nevertheless, the connection between persistence and bank erosion was most evident between 2018 and 2019, so further sampling is needed to support this hypothesis.

Missing and newly formed WJs, thus mobility classes, were primarily associated with morphodynamic mechanisms of erosion and deposition, respectively. Missing elements were found in correspondence to lobe dissections, chute cut-offs, bar edge trimmings, and channel incisions, indicating that the removal of morphological units caused the entrainment and removal of woody material. On the contrary, newly formed jams were mostly found in depositional areas, as the formation of new units or the accretion of existing ones can directly favour wood deposition as well. A significant exception was floodplain erosion areas that, although showing outstanding loss of sediments, were linked to newly formed jams rather than missing ones. However, the reasons are two: the initial scarcity of WJs on the floodplain and, more importantly, the expanded space within the active channel that became available for new WJ formation.

Combining the dynamics of jams with the variation of fluvial landforms helped link the predominant processes to wood mobility. However, it is important to point out that the largest part of WJs, both mobile and persistent, was found in areas with no changes, as 50% of the reach did not undergo geomorphic changes, so not in every case is possible to infer a relation between morphodynamics and mobility. Overall, the logistic regression models were found statistically significant when all three components, namely size, CHR and morphodynamic mechanisms, were involved, although the low level of variance was explained. Exploratory analysis assessed the role and importance of these three factors in relation to the mobility (or persistency) of WJs in the Blanco River, but further investigations are surely required to involve other variables and to expand awareness about cause–effect relationships.

In this study, conservative mobility values of 68% (2018–2019) and 78% (2019–2023) were calculated based on the number of WJs classified as either missing or newly formed, basically reflecting movements at the reach scale. However, we acknowledge that mobility rates may also include persistent jams that changed shape and/or position, which represent a significant portion (61% and 81%) of all persistent WJs during the respective time intervals. In active river systems, WJs are often mobile, frequently exchanging pieces or being transported and re-forming in the same location with new pieces but a similar structure (Kramer & Wohl, 2017).

4.3 Opportunities for improvement

The future of wood identification in river systems is trending towards more efficient, automated procedures, thanks to technological advances. Traditional methods such as visual inspections, field measurements and local monitoring (Ravazzolo et al., 2015; Tonon et al., 2017) are labour-intensive, expensive, and potentially not representative of the whole wood dynamics especially where data collection was confined to sub-basin sizes (Hassan et al., 2005). Remote sensing data sources have enabled high-resolution wood detection at larger scales, significantly improving coverage in complex environments where traditional methods are not feasible. In the present study, the use of UAV guaranteed a considerable spatial coverage while minimising the costs of field surveys, which can be an important factor in remote areas like the Blanco River catchment. Our approach still relied on the manual delineation of WJ' polygons, which is time-consuming although precise. To overcome these limitations, researchers are increasingly turning to machine learning techniques for more automated wood detection. For instance, Support Vector Machine (SVM) and XGBoost machine learning algorithms were successfully employed to classify wood in different riverine environments (Liang et al., 2022; Sendrowski & Wohl, 2021). Furthermore, even more advanced computer vision approaches are emerging, like those involving deep learning neural network architectures, using semantic segmentation of wood pieces at the pixel level. In the end, the initial training phase still relies on human intervention to create the annotated datasets that guide the model's learning process. However, once trained, the model can automatically predict the class of each pixel in new, unseen images, opening up numerous new opportunities for studying the dynamics of wood in rivers worldwide. Potential improvements could also target the detection of WJs beneath canopies, a challenge particularly prevalent in densely vegetated floodplain areas. In the Blanco River, although visibility is higher because of the presence of dead trees along the banks, some WJs may still have been overlooked because of the rapid regrowth of shrubs and herbaceous species (e.g. Gunnera tinctoria). Additionally, quantifying the volumes of WJs would offer valuable insights into their mobility, especially when wood accumulates in large jams with a significant vertical dimension (Sanhueza et al., 2019). However, this aspect was not addressed in the present study to prevent additional sources of uncertainty. The methodology used in this study to track WJ mobility has a fair degree of subjectivity, as the classification of persistent, moved, and newly formed WJs relies on the operator's perception. The difficulty lays in recognising those WJs that retained part of the individual elements and classifying them as persistent. Only the accumulations clearly maintaining key pieces in a relatively short spatial range were classified as persistent, while the rest were classified otherwise. Therefore, we assume that the results might underestimate the number of persistent WJs or, vice versa, overestimate the mobility rates. Future improvements could involve combining field monitoring with remote sensing approaches. For example, tracking WJ' key pieces and using them as samples to validate mobility analyses conducted remotely over larger datasets and more extensive areas.

Finally, although daily Q was estimated from the relationship with a neighbouring catchment, the absence of direct and continuous hydrological data in the Blanco River limits a more in-depth analysis of WJ dynamics, especially in understanding the crucial role of floods and planning proper mitigation measures for wood hazards (Fei & Wang, 2024).