Vegetation and Fluvial Geomorphology Dynamics after an Urban Fire

: The goal of this research was to characterize the impact of invasive riparian vegetation on burn severity patterns and ﬂuvial topographic change in an urban Mediterranean riverine system (Med-sys) after ﬁre in San Diego, California. We assessed standard post-ﬁre metrics under urban conditions with non-native vegetation and utilized ﬁeld observations to quantify vegetation and ﬂuvial geomorphic processes. Field observations noted both high vegetation loss in the riparian area and rapidly resprouting invasive grass species such as Arundo donax (Giant Reed) after ﬁre. Satellite-based metrics that represent vegetation biomass underestimated the initial green canopy loss, as did volumetric data derived from three-dimensional terrestrial laser scanning data. Field measurements were limited to a small sample size but demonstrated that the absolute maximum topographic changes were highest in stands of Arundo donax (0.18 to 0.67 m). This work is the ﬁrst quantiﬁcation of geomorphic alterations promoted by non-native vegetation after ﬁre and highlights potential grass–ﬁre feedbacks that can contribute to geomorphic disruption. Our results support the need for ground-truthing or higher resolution when using standard satellite-based indices to assess post-ﬁre conditions in urban open spaces, especially when productive invasive vegetation are present, and they also emphasize restoring urban waterways to native vegetation conditions. invasive vegetation-driven erosional pathways after ﬁre urban areas. This research provides new information on spatial variability in burn severity, vegetation regrowth, and ﬂuvial geomorphic hazards that should be considered by resource managers and engineers to inform management decisions in urban riverine environments prone to Arundo donax -initiated riparian ﬁres.


Introduction
Across the world, wildfires are increasing in frequency and magnitude under a changing climate and increased human interaction, which in turn impacts natural resources, infrastructure, and millions of people [1]. Continuous and extreme landscape conversion due to the expansion of the human population and establishments in southern California (United States) has fragmented chaparral ecosystems and proliferated the Wildland-Urban Interface (WUI). This has increased the potential for ignition and damages to human communities and surrounding ecosystems [2,3]. The magnitude of fire effects on Mediterranean riverine systems is often related to fire frequency, severity, and timing, which are driven by climate conditions, vegetation type, fuel loads, and landscape conditions [4]. The variable hydrologic, geomorphic, and ecosystem responses induced by fire introduce a high degree of uncertainty in modeling and predictions for management.
In Mediterranean systems, it is well documented that wildfire initiates sedimentation and flooding due to the loss of vegetation [5,6], reduced infiltration [7], soil cohesion [8], and soil water repellency [9]. These processes expedite surface runoff and flooding potential [6,10] as well as suspended sediment discharge and sedimentation [11], which often induce significant geomorphic responses such as aggradation, incision, bank widening, channel narrowing, and braiding [12,13]. At excessive levels, sediment and turbidity become pollutants, in comparison to undisturbed watersheds that are adapted to naturally occurring levels [14]. Soil and sediment are composed of a variety of components including

Precipitation
Precipitation data (15-min intensity; I15) were collected from the Lake Murray site 3892 (approximately 2.9 km from the study site), which is a part of the San Diego County Flood Control District ALERT Flood Warning System, from May 1, 2018 to May 1, 2019. The maximum 15-min intensity for each storm during the study period was associated with NOAA ATLAS 14-point precipitation frequency estimates. Six significant storm events occurred during the 2018-2019 wet season (October-March) following the Del Cerro Fire (Table 1).

Burn Severity and Canopy Metrics
To address the first goal of this study, standard methods were used to characterize the burn severity and canopy of the study area. Collection 1 level 1 data from Landsat 8 Operational Land Imager (OLI) and Thermal Infrared Sensor (TIRS) image data were collected and processed prior to calculating burn severity and vegetation metrics. The Landsat digital number (DN) was converted into a top of atmosphere (TOA) reflectance or the amount of light reflected to the satellite [36], which corrects for atmospheric conditions and the position of the sun, mitigates the effects of light scattering in the atmosphere, and reduces haze wavelength distortion [36]. Atmospheric correction was not required, as the images selected for this study did not contain clouds or any other disturbances. Five Landsat images were acquired and have a spatial resolution of 30 m and temporal resolution or frequency of 32 days (Table 2). The Normalized Burn Ratio (NBR) is an effective measure of burn severity in a variety of landscapes ranging from forest to chaparral [36,37]. This index can be related to the severity of a wildfire on the ecosystem by quantifying the transition from vegetated terrain to dry, ashy soil that is interspersed with blackened vegetation [38]. The NBR is calculated by using the relative difference in reflectance between the Near Infrared (NIR) and Short-Wave Infrared (SWIR) (Equation (1)). Equation (1) is based on the physical properties of vegetation, where green plant growth reflects NIR well, while dry, burned soil reflects highly in the SWIR [38]. NBR is the ratio of the difference in percent reflectance between the two spectra and ranges between -1 and 1. The differenced Normalized Burn Ratio (dNBR) immediately following fire (0-year) was calculated by differencing Landsat Images 1 and 2 (Equation (2); Table 2).
Burn severity is approximated by dNBR through established thresholds that relate the change in reflectance from pre-fire to post-fire conditions, to the surveyed ecological and socio-economic impact of the fire [38][39][40]. We utilized standard burn severity levels established by [38,39] to represent enhanced regrowth (−500 to −101), unburned (−100 to 99), low severity (100 to 269), moderate severity (270 to 659), and high severity (660 to 1300), where the dNBR range was scaled by 10 3 . High dNBR values indicate high severity burn damage, and negative to low values indicate low burn severity to increasing vegetation productivity.
Red and NIR spectrums from Landsat 8 OLI/TIRS were used to estimate plant biomass through the relationship of chlorophyll light absorption for photosynthesis in the red (Red) wavelength and high reflectance in the NIR [41]. The Normalized Difference Vegetation Index (NDVI) is a proxy for the presence of "greenness" or green canopy cover (Equation (3)). NDVI was estimated for each vegetation class for the following five time-points (Table 2): May 16, 2018 (pre-fire), June 21, 2018 (immediately post-fire), July 23, 2018 (one-month post-fire), November 12, 2018 (five-months post-fire), and June 19, 2019 (one-year post-fire). The lower NDVI represents bare soil to sparse vegetation (0.025-0.09) and higher NDVI represents green vegetation (0.25-0.5). The Differenced NDVI (dNDVI) was calculated between time-points to demonstrate spatial and ecological shifts in vegetation distribution and estimate vegetation health through "greenness" (Equation (4)) [42,43]. In forested environments, dNDVI has a weaker relationship with field-based measurements of burn severity than dNBR [38]. However, in [44], it was an effective measure of burn severity in riparian environments and thus included in this study.
To monitor both the immediate and longer-term (1-year) post-fire vegetation trends, the dNDVI for each fire was calculated using pixel-to-pixel analysis between a preceding and subsequent time-point using five Landsat imagery dates (Table 2). Based on dNDVI and in-field observations of vegetation loss observed in [45], we categorized levels of canopy loss (approximated by dNDVI) as unburned (<0.005); low (0.005 to 0.049); moderate (0.05 to 0.199); and high (>0.2).

Topographic Surveys
To quantify changes in the fluvial morphology across the upland and riparian zones (objective two), cross-sectional topographic profiles were collected using standard surveying techniques with an auto-level scope and stadia rod. A 62-m transect perpendicular to Alvarado Creek was established on June 29, 2018, within the burned area ( Figure 2F). Precipitation events are the main drivers of geomorphic change; thus, we assumed that the landscape immediately following fire, before any precipitation, was representative of pre-fire conditions. The transect ran south to north from a prominent floodplain bench on the upland hillslope, across the riparian zone to another smaller floodplain bench, abutting the fences of a residential area ( Figure 2F; white dashed line represents the transect extent). Elevation was recorded every 0.5 m, and all elevations were normalized to a benchmark established by a monument located on the upland bench near the cross-section transect based on the methods described by [46,47]. Four cross-sectional surveys were conducted successively on the same transect before and after major storm events during the period of June 30, 2018 to March 15, 2019 (Table 3). Field data collection began immediately post-fire on June 29, 2018. The vegetation survey methods followed the Growth Habit Codes and Definitions for the identification of plant growth forms [48]. Vegetation was identified as "present" or "absent" systematically along the transect at 0.5 m intervals. Specifically, Arundo donax presence or absence was also recorded. In the absence of plants, the substrate was categorized by grain size clastic classes, as described in [49]. These data, in conjunction with volumetric measurements of canopy derived from TLS data, were used to quantify and track the vegetative recovery as well as relate vegetation cover to measurements of topographic change described in Section 2.4.3.

Topographic Analysis by Vegetation Cover
Cross-sectional topographic change following storm events after the fire was related to patterns of ground cover by position on the transect. All the channel surveys were normalized to the height of the local benchmarked monument as the datum (0,0). Coupling the topographic relief to vegetation growth form type is especially important for documenting geomorphic change and channel routing within the riparian region and the extent of change in invasive and infested urban areas. The event-based topographic change in height (∆H Event ) between each successional cross-section was calculated by differencing the elevation (meters) at each survey point by the following: The maximum seasonal change in topography (∆H Maximum ) was calculated by differencing the maximum elevation (H maximum elevation ) by the minimum elevation (H minimum elevation ) at each survey point on the transect for the study period between June 2018 to March 2019: All ∆H Maximum and ∆H Event topographic height data were separated by vegetation growth forms or substrate types. A threshold at the 75th percentile of all topographic change measured within a cover class was used to distinguish measurements of significant elevation change for both event-based and maximum seasonal change per substrate or vegetation growth type. The 75th percentile was also established to account for noise in the topographic data such as the misalignment of the multiple cross-sections due to instrument error, sample size (only one cross-section was observed), human error, or environmental factors (i.e., wind, vegetation growth, etc.). The cross-sectional area for the transect extent ( Figure 2F; white dashed line) was calculated and compared across all storm events through the end of the wet season (March 2018). Cross-sectional areas were calculated by linearly interpolating all cross-sectional data to a 0.1 m resolution and using a trapezoidal numerical integration.

Point Cloud Processing and Vegetation Volume
Three-dimensional imaging of the area of interest through terrestrial laser scanning (TLS) was used to document fine-scale vegetation density after the fire and the first storm season. Light Distancing and Ranging (LiDAR) is the emission of laser pulses that interact with a surface or object. The scattering or reflection of the laser signal back to the instrument sensor provides a distance and point location (x,y,z) or point cloud. Terrestrial Laser Scanning utilizes LiDAR to acquire high-resolution, ground-based elevation data points, which can be used to estimate topographic and volumetric changes along a surface or complex terrain through repeated scanning over time [50]. A Trimble GX 3D Scanner was used on June 29, 2018, July 23, 2018, September 28, 2018, and January 19, 2019 and a Riegl VZ-400 was used on May 24, 2019 to scan the same 100 square meter area that encompasses the immediate upland area of Alvarado Creek and the southern portion of the riparian area ( Figure 2). Point cloud datasets from the scans were processed in Trimble RealWorks Survey and Cloud Compare to identify regions of volumetric density change in vegetation over time.
All scans, in the form of point clouds, were cleaned using the Statistical Outlier Removal tool in Cloud Compare, which uses the average distance of each point to its neighbors. Outlier points were defined as the points over the average distance between k = 10 nearest neighboring points multiplied by one standard deviation (nSigma) [51,52]. The vegetation was separated from the bare ground using the Cloth Simulation Filter (CSF) in Cloud Compare to extract ground from "not ground" points that represent vegetation points under the topmost boundary of the canopy. The CSF uses an algorithm to create a mesh on the underside of a point cloud surface, which is compared to the deformation on the topside of the scan. We defined the terrain relief parameters as a 0.1 m cloth resolution, with 1000 max iterations, and a 0.1 m height threshold for vegetation classification [53].
Each scan was classified into riparian or upland and vegetation or ground for which vegetation volume density [m 3 m −2 ] was calculated. This relative metric was estimated between each scan for upland and riparian land classes. Using the vegetation and ground surface point clouds for each scan generated from the CSF, the average distance from the ground surface and to the top of the vegetation point cloud in the z-direction was calculated along a 0.01 m resolution grid. The gridded vegetation height surface created was used to estimate volume by Reimann sums. Each volume was divided by the confining area to give a relative measurement of vegetation volume per square meter. The confining area was defined as the number of 0.01 m grid pixels populated by either ground or vegetation. Holes in the point cloud data due to shadows were omitted from the volume and confining area calculations. All reported means are accompanied by standard deviations, which are included in the figures as error bars.

Statistical Analysis
The numeric values of both vegetation indices, dNBR and dNDVI (burn severity and green canopy loss), were compared between time-point conditions by each land class using average values derived across 30 m pixels. Burn severity and canopy loss within all vegetation classes (invasive, riparian, and upland) were compared using unpaired two-tailed t-tests. Our field observations were also analyzed using a two-tailed student's t-test. To address objective three of this study, the following hypotheses were tested: (1) the difference between the means of topographic elevation change in the upland versus the riparian area; (2) the difference between the means of topographic elevation change by vegetation cover classes; and (3) the difference between the means of volumetric density in the upland and riparian area. The null hypotheses for all t-tests were that the two-population means were equal and rejected if the resulting p-values were less than 0.05. All significant results are defined by p-values less than 0.05.

Satellite-Based Burn Severity and Canopy
We estimated burn severity and vegetation conditions with standard post-fire satellite-based metrics in the riparian and upland zones. The average immediate burn severity (dNBR) in the riparian area was higher (0.26 ± 0.14) than in the upland (0.17 ± 0.12) and invasive areas (0.22 ± 0.16) ( Figure 3A). The immediate burn severity of the invasive vegetation was not statistically different from the upland or riparian areas ( Table 4). The smallest immediate loss in the green canopy (dNDVI) occurred in the areas with invasive vegetation (0.17 ± 0.09), and the greatest loss in canopy occurred in the upland (0.27 ± 0.12) ( Figure 3B; Table 4). The immediate loss in green canopy in the riparian (0.22 ± 0.11) was not statistically different from the invasive vegetation class (Table 4).
To track the vegetation canopy over time, NDVI was estimated for five dates after the fire ( Figure 3C). The lowest NDVI immediately post-fire (0-year; June 29, 2018) occurred in the upland region (0.18 ± 0.07). Invasive, riparian, and upland classes have distinct differences in canopy for June, July, and November 2018. The NDVI ranged from 0.18 ± 0.07 to 0.44 ± 0.13, where the invasive area had generally larger values than the riparian and upland areas. By June 2019, the NDVI values were statistically similar between all classes ( Table 5). The upland class was also the only region with NDVI that reflected bare ground levels at any time-point measured following fire (0.18 ± 0.07). In contrast, the riparian region had the highest immediate burn severity indicated by dNBR ( Figure 3A), but the average NDVI values immediately post-fire (0-year; June 21, 2018) did not reduce below the green vegetation threshold of 0.25 ( Figure 3C).
The highest immediate (0-year) dNDVI or loss of green canopy occurred in the upland area ( Figure 2B); however, the spatial distribution of canopy loss was highly heterogeneous. The majority of the pixels that detected green canopy loss were located in the upland hillslope directly adjacent to the Alvarado Creek riparian zone ( Figure 3D1). There was an abundance of regrowth detected within the riparian and invasive areas along Alvarado Creek from June 2018 to July 2018, before any precipitation occurred (mean dNDVI: −0.04 ± 0.01), in contrast to the minimal change that occurred in the upland region (−0.03 ± −0.01), ( Figure 3B,D2,3). From mid-November 2018 to June 2019, five storms occurred and ranged between less than 1-and 1-year return intervals for 15-min intensities ( Table 1). Over this time, intensive regrowth was detected in the upland region (−0.17 ± −0.01), which contrasts the relatively low regrowth in the riparian area and minimal growth in the invasive pixels (−0.05 ± −0.01; Figure 3D4).

Coupled Topographic and Vegetation Growth Form Surveys
Four field surveyed cross-sections and vegetation growth form surveys were conducted after the Del Cerro Fire. Cross-section 1 (CS1) was completed on June 30, 2018 (28 days following the fire) and served as the baseline (pre-storm) geomorphic condition ( Table 3). The thalweg of Alvarado Creek is visible at approximately 20 m horizontally from the benchmark at the datum (0,0) ( Figure 4). Other prominent geomorphic features moving outward in both directions from the main channel of Alvarado Creek were a series of over-banks and secondary channels. The maximum seasonal topographic change (∆H Maximum ) is estimated across the transect (Figure 4, in light gray). The 75th percentile change in topographic height or greater is equivalent to 0.403 m (Figure 4, dark gray region). The approximate cross-sectional area measured immediately following fire was the smallest recorded over the entirety of the study period (CS1; 40.5 m 2 ). The area increased after Storm 1 (CS2; 46.2 m 2 ) and expanded even more after Storm 2 to the largest recorded area during CS3 (51.0 m 2 ). The last cross-section (CS4) following Storms 3-9 resulted in a reduction of cross-sectional area similar to the area observed during CS2 (47.6 m 2 ).  Vegetation and stream channel cross-sectional surveys were conducted simultaneously and revealed six major categories of ground cover present along the transect (Table 6): (1) Upland hillslope, (2) Invasive I (mixture of Arundo donax and Washingtonia spp.), (3) Main channel (defined by wetted perimeter), (4) Forbes and cobbles, (5) Secondary channels, and (6) Invasive II (predominantly Arundo donax). There was no statistically significant change in the presence or absence of vegetation along the transect between surveys except in the upland area. We identified vegetation cover types such as trees (Washingtonia spp.), graminoids (Arundo donax), and forbes (i.e., Erigeron sp., Brassica nigra and Foeniculum sp.). Bare substrates across the transect were identified as sand, cobble, or mixed sand and cobble. Further, all sand substrate was associated with Arundo donax.  Vegetation and stream channel cross-sectional surveys were conducted simultaneously and revealed six major categories of ground cover present along the transect (Table 6): (1) Upland hillslope, (2) Invasive I (mixture of Arundo donax and Washingtonia spp.), (3) Main channel (defined by wetted perimeter), (4) Forbes and cobbles, (5) Secondary channels, and (6) Invasive II (predominantly Arundo donax). There was no statistically significant change in the presence or absence of vegetation along the transect between surveys except in the upland area. We identified vegetation cover types such as trees (Washingtonia spp.), graminoids (Arundo donax), and forbes (i.e., Erigeron sp., Brassica nigra and Foeniculum sp.). Bare substrates across the transect were identified as sand, cobble, or mixed sand and cobble. Further, all sand substrate was associated with Arundo donax. The highest mean ∆H Maximum occurred in the secondary channels (no vegetation present; 0.39 ± 0.12 m) and the Invasive II over-banks (0.42 ± 0.09 m), which were not statistically different ( Figure 5A; primary axis). The Upland hill slope, Invasive I, and Alvarado Creek cover classes had the lowest mean ∆H Maximum of all the covers throughout the whole wet season. The Gravel and Forbes cover class had a statistically higher mean ∆H Maximum than the aforementioned three cover classes (0.34 ± 0.04 m). Although the secondary channels and Invasive II cover average ∆H Maximum were not statistically different, the Invasive II cover class had 31% higher seasonal maximum topographic changes that were above the 75th percentile threshold than the secondary channels (32%; Figure 5A; secondary axis). All other cover classes experienced some topographic change equal to or above the 75th percentile, except for the Upland hillslope.
Geosciences 2020, 10, x FOR PEER REVIEW 13 of 24 axis). All other cover classes experienced some topographic change equal to or above the 75th percentile, except for the Upland hillslope. The highest event-based topographic change (|∆HEvent|) occurred after the first storm ( Figure  5B). Positive mean elevation change represents a gain in elevation, while negative is a loss in elevation from the preceding measurement. The highest absolute mean |∆HEvent| across all storms occurred in the secondary channels after Storm 1 (0.22 ± 0.18 m; June 2018 to October 2018; Table 1). The cover classes that also experienced the largest |∆HEvent| after Storm 1 were Alvarado Creek (0.20 ± 0.07 m The highest event-based topographic change (|∆H Event |) occurred after the first storm ( Figure 5B). Positive mean elevation change represents a gain in elevation, while negative is a loss in elevation from the preceding measurement. The highest absolute mean |∆H Event | across all storms occurred in the secondary channels after Storm 1 (0.22 ± 0.18 m; June 2018 to October 2018; Table 1). The cover classes that also experienced the largest |∆H Event | after Storm 1 were Alvarado Creek (0.20 ± 0.07 m loss) and the Upland hillslope (0.18 ± 0.05 m loss). The |∆H Event | for Invasive II after Storm 1 (0.19 ± 0.10 m) and after Storm 2 (October to 2018 to December 2018; 0.21 ± 0.14 m) were not statistically different, and both means were greater than the 75th percentile ( Table 7). The cover classes that experienced the greatest |∆H Event | after Storm 2 were both Forbes and cobbles (0.20 ± 0.04 m) and Invasive I (0.15 ± 0.08 m). The lowest |∆H Event | means overall were recorded after Storms 3-6 (December 2018 to March 2019). The Upland hillslope, Invasive I, and Alvarado Creek were not significantly different from each other ( Table 7), but they all experienced a mean ∆H Event of −0.02 ± 0.05 m (elevation loss). In contrast, a positive mean ∆H Event (elevation gain) occurred in three cover classes: Forbes and cobbles, Secondary channels, and Invasive II (0.10 ± 0.11 m). Table 7. p-values and t-statistics for all hypotheses associated with topographic analysis data by cover class. The shaded p-value denotes rejected null hypotheses (p < 0.05).

p-Value t-Statistic
Hillslope

Volumetric Vegetation Density
Volumetric densities were derived from TLS for the following four dates: with the TLS scanner in the field.
The vegetation densities in the riparian areas did not substantially increase immediately following fire. The July 2018 scan revealed a loss of vegetation density in the upland area from 0.38 m 3 m −2 (June 2018) to 0.24 m 3 m −2 . There was also a loss of vegetation density from 2.56 m 3 m −2 (June 2018) to 1.66 m 3 m −2 (July 2018) in the riparian area. Although the riparian measurement was omitted, the January 2019 upland scan continued to show vegetation density loss in the upland area (0.15 m 3 m −2 ). From January 2019 to May 2019, after Storms 7-9, the upland vegetation more than tripled to 0.51 m 3 m −2 , and the May 2019 vegetation density in the riparian area was over three times greater than that in July 2018 and twice the vegetation density observed in June 2018.

Satellite-Based Metrics after an Urban Fire
As anticipated, rapid vegetation regeneration has significant implications on the accuracy of remote sensing techniques. Contrary to previous literature [24][25][26], traditional satellite-derived metrics did not capture the substantial increase of green canopy in the invasive riparian zone after fire. As hypothesized, the immediate burn severity (dNBR) in the riparian area was statistically higher than the upland area. However, the invasive class was more variable and neither significantly higher nor lower than either the riparian or upland areas. The upland areas experienced the highest canopy loss and expressed the only indications of bare ground. This is expected given the typical sparse distribution of chaparral near urban areas [54][55][56] in comparison to the denser canopy cover of riparian areas [16].
We also observed an unexpected low average immediate canopy loss within the invasive class. There was also an overall disagreement in trends between the two vegetation indices at 0 years due to the rapid regeneration of non-native vegetation within the invasive and riparian regions as well as the theory by which these indices were developed. Burn severity (dNBR) is a measure of the fire impact on the combination of a reduction of mesophyll leaf structure (NIR) and biomass water content (SWIR), while green canopy loss (dNDVI) is a measure of fire impact on the reduction of mesophyll leaf structure (NIR) and greenness as chlorophyll (red) [36]. After the fire, the biomass and water content were low, but the greenness was detectable due to the presence of small and new Arundo donax shoots. As a result, the immediate post-fire measurement of dNDVI underestimated the green canopy loss due to the rapidly resprouting invasive grass species [57]. The deterioration of the relation between dNDVI and dNBR has not been previously reported. Thus, we advocate that the accuracy of remotely sensed immediate canopy loss (dNDVI) in urban Med-sys is reliant on the availability of observations less than a week following the fire. If satellite images are unavailable, metrics derived from satellite data to guide management in post-fire areas should be avoided or should be supplemented with immediate and detailed field surveys to provide a better assessment [58].
In urban riparian areas, the combination of the highly flammable chaparral [33,37,54,59] near extra fuel loads presented by both the woody debris and abundant grasses of riparian-originated fires [59,60] encourage intense and sustained fires that can severely impact vegetation health [36,61,62]. This suggests that the riparian-upland interface is vulnerable during and after urban fires, which is supported by our observations (Section 4.4), where the largest immediate (0-year) dNDVI or loss of green canopy was in the upland area ( Figure 3B) adjacent to the riparian zone ( Figure 3D1). In general, there was a bias in dNDVI toward new and green regrowth (May 4, 2018 to June 21, 2018), where the canopy loss signal is dampened in the riparian corridor dominated by invasive vegetation. Meanwhile, the relative change in dNDVI by vegetation class from June 21, 2018 to June 19, 2019 revealed a successive order of regrowth in the riparian vegetation before the wet season, followed by regrowth in the upland area after the wet season ( Figure 3D1-4). The location in the watershed, time of year, and species type directly impact the water availability within Med-sys for vegetation regrowth patterns in the post-fire environment [62,63]. While metrics based on widely available and cost-effective satellite products such as Landsat are commonly used in post-fire assessment, we advocate caution when applied to urban fires. Higher temporal and spatial resolution monitoring strategies are needed to accurately assess the initial impact of fire as well as the lasting ecological impact in urban fluvial systems vulnerable to invasive vegetation infestation.

The Role of Vegetation in Upland and Riparian Post-Fire Topographic Change
Immediately after the Del Cerro Fire, burn severity was highest in the riparian area and lowest in the upland areas, which could have implications for geomorphic impacts. Similar to previous works, we observed that vegetation regrowth patterns with respect to burn severity can significantly impact the hydro-geomorphic dynamics in Med-sys [14,16,60,64]. In particular, the impact of Med-sys geomorphology on the spatial distribution of burn severity patterns [60,62] creates a feedback dynamic between geomorphology, vegetation, and fire that is promoted by invasive vegetation infestation. Although only speculated by previous studies [65], we presented the first attempt to quantify post-fire topographic alterations and volumetric vegetation in the urban med-sys with invasive cover in relation to commonly used fire management metrics such as burn severity.
The natural mitigation of fire-driven erosion and the resultant geomorphic change is the reestablishment of native vegetation cover [8]. In Alvarado Creek, very little vegetation recovery occurred on the upland hillslope for most of the observation period (June 2018 to the end March 2019), and the magnitude of seasonal maximum geomorphic change in the upland was 23% lower than in the riparian area. The upland hillslopes were generally steep, 17 • to 32 • , and they had the potential for erosional hazards after fire [6]. The largest geomorphic response for the upland hillslope was measured following the first post-fire storm event, which was the lowest I15 storm and progressively decreased thereafter. This would be expected, as the geomorphic response to a lower threshold of storm intensity is typically heightened directly following fire [66]; however, the peak I15 during the first storm was much lower (~1.3 mm/h) relative to typical values of saturated hydraulic conductivity for even the finest clast sizes in moderate to low-intensity fires [67]. While the rain gauge utilized during this study was nearby (2.9 km), this distance can introduce uncertainty in the actual rainfall intensity over the study site and consequently the true erosional pressure from the first storm.
Field observations suggested that fine sediments were removed from the hillslope ( Figure 7A,B) and rainfall intensity surpassed the saturated hydraulic conductivity of the finest layer of the hillslope. The immediate release of sediment, coupled with an armored hillslope due to deep chaparral root systems [68] and large clast grain structure [69], can result in a smaller sediment discharge, despite intense storms. The removal of the topsoil and ash deposits after the first storm event was documented in the field (Figure 7). The top layer ( Figure 7A) was removed after the first storm, leaving only the charred layer that was beneath the initial horizon ( Figure 7B) and leaving a coarser grain size distribution. The removal of the smaller grain size clasts without significant rilling likely occurred by erosional mechanisms such as rain splash, overland sheet flow, or thin debris flows [70,71]. These processes can wick away smaller clasts and result in a coarsening of grain sizes in sediments. In our study area, the higher frequency of larger grain sizes may have contributed to the geomorphic stability of the landscape after the first storm event by increasing the resilience to hydraulic shear stress that could initiate particle transport during future precipitation events [72,73]. This phenomenon in conjunction with the low burn severity on the upland hillslopes and intact native chaparral root systems may have fortified the structural integrity of the hillslope from hydraulically initiated erosion events [69].
Field observations suggested that fine sediments were removed from the hillslope (Figure 7A,B) and rainfall intensity surpassed the saturated hydraulic conductivity of the finest layer of the hillslope. The immediate release of sediment, coupled with an armored hillslope due to deep chaparral root systems [68] and large clast grain structure [69], can result in a smaller sediment discharge, despite intense storms. The removal of the topsoil and ash deposits after the first storm event was documented in the field (Figure 7). The top layer ( Figure 7A) was removed after the first storm, leaving only the charred layer that was beneath the initial horizon ( Figure 7B) and leaving a coarser grain size distribution. The removal of the smaller grain size clasts without significant rilling likely occurred by erosional mechanisms such as rain splash, overland sheet flow, or thin debris flows [70,71]. These processes can wick away smaller clasts and result in a coarsening of grain sizes in sediments. In our study area, the higher frequency of larger grain sizes may have contributed to the geomorphic stability of the landscape after the first storm event by increasing the resilience to hydraulic shear stress that could initiate particle transport during future precipitation events [72,73]. This phenomenon in conjunction with the low burn severity on the upland hillslopes and intact native chaparral root systems may have fortified the structural integrity of the hillslope from hydraulically initiated erosion events [69]. We expected the largest topographic change in both the upland and riparian area to occur after the first post-fire storm event. On average, the magnitudes of topographic change between June 2018 to October 2018 and October 2018 to December 2018 for the riparian area were not significantly different, despite the larger average I15 and total rainfall from October 2018 to December 2018. The max I15 of the first storm event in October (1.3 mm h −1 ) was over 50% less intense than the I15 observed in Storms 2 and 3 (4.1 mm h −1 and 6.3 mm h −1 ), yet the overall magnitude of topographic change was not significantly different between CS1 and CS2.
Changes in both the riparian and upland areas due to the multiple storms observed from CS1 to CS2 demonstrated a potential "first-flush" (first storm of the season) response. The "first-flush" created the most pronounced erosional and geomorphic changes of the season following the fire despite precipitation intensity. In addition, the overall low intensity of all storms observed through the 2018-2019 wet season suggests that the geomorphic dynamics observed in both the riparian area were driven by pre-existing geomorphology and vegetation regrowth patterns rather than the magnitude of precipitation events [64]. However, the first storm event did not instigate the largest topographic change within the riparian area. In contrast to the upland hillslope, a significant regeneration of vegetation occurred in the riparian area, which was predominantly non-native vegetation species. The elevated topographic variation in the Invasive II cover class over the wet season demonstrated the influence of non-native vegetation on the stability of the fluvial topographic structure in urban riparian environments. This study is limited to topographic changes from one cross-section; however, our approach allowed for more frequent observations of changes with respect to precipitation and highlighted emerging challenges in characterizing and managing urban waterways. We also provided new observations as well as a quantification of topographic changes in post-fire riparian areas with non-native vegetation, which can be used to guide future monitoring and assessment.

The Impact of Invasive Vegetation on Post-Fire Fluvial Topographic Change in the Urban Riparian Corridor
We expected that the largest topographic change along the transect would coincide with invasive vegetation (including Invasive I). Arundo donax dominated invasive cover (Invasive II) had higher geomorphic variability than a mixed cover of Arundo donax and Washingtonia spp. (Invasive I). Both the greatest event-based topographic elevation loss and gain were recorded in Invasive II. Less is known about the hydrologic and geomorphic processes associated with palm species such as Washingtonia spp. However, the root depth structure and tensile strength of Washingtonia spp. more closely resemble Salix spp. (willows), which is a native Med-sys riparian species that promotes riverine bank stability and topographic recalcitrance [16,31,74]. Although originally introduced to California's flood plains for erosion control [30], the low tensile strength and shallow root systems of Arundo donax can invoke undercutting, bank collapse, and lateral migration [26,27]. These processes are exacerbated in flashy semi-arid hydrologic systems after fire and can transform riverine hydrology [20,24], and they were observed in Alvarado Creek (Figure 8). The secondary channels parallel to Alvarado Creek were flanked by Arundo donax stands (Invasive II), where the average seasonal maximum topographic change was the highest across all cover classes and statistically indistinguishable ( Figure 5B). This suggested that the geomorphic impacts of Arundo donax observed in the riparian area extended beyond the immediate stand of vegetation and into the surrounding upland features [31]. Between June 2018 and December 2018, topographic elevation decreased for all cover classes, and the cross-sectional area of Alvarado Creek continually increased from CS1 to CS3. The locations of greatest topographic elevation loss in the riparian environment generally occurred in conduit geomorphic features (such as all secondary channels and the main channel) between CS1 and CS2. This was in addition to the uniform topographic variation of the Forbes and cobbles floodplain area, which was validated by field observations. However, between CS2 and CS3, the greatest topographic elevation loss transitioned to Arundo donax-dominated stands and the outer forb and cobbledominated floodplain. Whereas, between December 2018 and March 2019 (CS4), the Forbes and cobbles, Secondary channels, and Invasive II cover classes gained topographic elevation. Across the transect, all three cover classes made up the majority of the secondary over-banks and outer Between June 2018 and December 2018, topographic elevation decreased for all cover classes, and the cross-sectional area of Alvarado Creek continually increased from CS1 to CS3. The locations of greatest topographic elevation loss in the riparian environment generally occurred in conduit geomorphic features (such as all secondary channels and the main channel) between CS1 and CS2.
This was in addition to the uniform topographic variation of the Forbes and cobbles floodplain area, which was validated by field observations. However, between CS2 and CS3, the greatest topographic elevation loss transitioned to Arundo donax-dominated stands and the outer forb and cobble-dominated floodplain. Whereas, between December 2018 and March 2019 (CS4), the Forbes and cobbles, Secondary channels, and Invasive II cover classes gained topographic elevation. Across the transect, all three cover classes made up the majority of the secondary over-banks and outer floodplain area. This topographic gain in elevation resulted in a reduction in the overall cross-sectional area even after continuous incision from CS1 to CS3. The elevated roughness imposed by Arundo donax infestations decreased hydraulic velocities and flooding, which contributed to sediment deposition and lateral bank movement [75,76]. Consistent with [77], the aggradation of the floodplain, continued channel incision, and obliteration of step-pool morphology observed during this study may also contribute to the sustained topographic change in the riparian area surrounding Alvarado Creek (Figure 9). The elevation disparity between the main channel of Alvarado Creek and the surrounding floodplain may promote the desiccation of the riparian area due to a lowered water table, which reduces the available water supply in the floodplain [78,79]. This is exacerbated by the extreme demand on water resources by Arundo donax stands [80,81] further contributing to the aridification of the riparian environment and promoting the grass-fire feedback [24].

Volumetric Vegetation Density after Fire
Vegetation densities approximated by TLS in the riparian areas did not substantially increase immediately following fire. This was contrary to the observed rapid regrowth of Arundo donax around Alvarado Creek ( Figure 3E1-3) as well as the NDVI ( Figure 3C). The volumetric vegetation density derived from the TLS decreased by 35% in the riparian area and 37% in the upland area from June 2018 to July 2018. Instead of capturing the rapid regrowth of Arundo donax shoots, which is proportionally a very small component of the total riparian volume, these data captured secondary burn severity impacts. In addition to tree death, secondary impacts include leaf fall and branch decomposition of fire-damaged trees within the riparian region [82].
A similar relationship was observed in the upland area, where the volumetric density of vegetation decreased from June 2018 to January 2019. These results contradicted the satellite-based NDVI ( Figure 3C) as well as photographic evidence of scattered chaparral resprouting as early as July 2018. In this study, seasonal leaf fall and senescence played a minor role in the lower vegetation density. The structural loss of the chaparral through the continued decomposition of vegetation material due to char and fire damage contributed to a net coverage loss by volume throughout the first six months following the fire, despite the vegetative volume added by resprouting chaparral [82][83][84]. The structural information of vegetation provided by TLS data can link vegetation analysis by satellite remote sensing to the geomorphology of Med-sys following fire. However, there are also

Volumetric Vegetation Density after Fire
Vegetation densities approximated by TLS in the riparian areas did not substantially increase immediately following fire. This was contrary to the observed rapid regrowth of Arundo donax around Alvarado Creek ( Figure 3E1-E3) as well as the NDVI ( Figure 3C). The volumetric vegetation density derived from the TLS decreased by 35% in the riparian area and 37% in the upland area from June 2018 to July 2018. Instead of capturing the rapid regrowth of Arundo donax shoots, which is proportionally a very small component of the total riparian volume, these data captured secondary burn severity impacts. In addition to tree death, secondary impacts include leaf fall and branch decomposition of fire-damaged trees within the riparian region [82].
A similar relationship was observed in the upland area, where the volumetric density of vegetation decreased from June 2018 to January 2019. These results contradicted the satellite-based NDVI ( Figure 3C) as well as photographic evidence of scattered chaparral resprouting as early as July 2018. In this study, seasonal leaf fall and senescence played a minor role in the lower vegetation density.
The structural loss of the chaparral through the continued decomposition of vegetation material due to char and fire damage contributed to a net coverage loss by volume throughout the first six months following the fire, despite the vegetative volume added by resprouting chaparral [82][83][84]. The structural information of vegetation provided by TLS data can link vegetation analysis by satellite remote sensing to the geomorphology of Med-sys following fire. However, there are also potential errors in this type of calculation such as the interpolation of the mesh in areas of low point density (e.g., LiDAR shadow) and the bias toward surficial canopy cover. Superficial canopy cover may not provide information about vegetation densities across multi-level canopies such as those present in the riparian area. Future calculations of volumetric densities in the riparian environment should incorporate vertically stratified canopy classes from the basal ground cover to the mid-level canopy of immature trees and shrubs to the upper canopy of mature tree cover [84]. For example, as Arundo donax matures after fire, it begins to dominate the mid to upper canopy, reaching approximately 6 m high at full maturity [30,77].

Conclusions
Our results (1) support the need for ground-truthing or higher resolution when using standard satellite-based indices to assess the burn severity of a managed open-space area, especially when highly productive invasive species such as Arundo donax are present and (2) emphasizes the need to restore urban waterways to native vegetation conditions. While the containment-type management strategies for Arundo donax as presented in the Regional Implementation for San Diego County may mitigate the spatial spread of Arundo donax, the extreme fire and geomorphic hazards associated with this invasive vegetation often found in Med-sys WUIs may not be thoroughly addressed [84]. Prioritization of Arundo donax eradication, although challenging, would more effectively mitigate the flooding, fire, and geomorphic hazards that we have observed as being propagated by its presence. The Del Cerro Fire provided a unique opportunity to quantify and evaluate the impact of invasive vegetation regrowth on the recovery and stability of an urban channel. There were limitations inherent in our field methodologies; however, through a combination of satellite imagery, precipitation analysis, topographic surveys, and three-dimensional terrestrial laser scanning (TLS), we quantified coupled the vegetation and geomorphic dynamics present during the recovery of an urban Med-sys throughout the first year after fire. Our results supported previous speculations [17,25,26] and suggested that the pre-fire vegetation structure, burn severity, and regrowth patterns strongly influenced the geomorphic responses observed after fire. We applied TLS to document the morphological dynamics of the land surface terrain and the quantification of volumetric change of vegetation biomass and growth forms. In contrast to the rapid increase of healthy vegetation captured by satellite, the TLS data highlighted a decrease of vegetation cover due to secondary burn severity impacts such as leaf fall, branch decomposition, and tree death in both the upland and riparian areas. Such results provided evidence of the dynamic structural changes that occur in vegetation in both the upland and riparian regions of Med-sys independent of regrowth patterns and suggested that remotely sensed data may be biased by the upper canopy condition. The presence of Arundo donax in the lower canopy significantly increased channel instability in the riparian region and encouraged the deposition of sediments in the floodplain (accretion), which can promote riparian desiccation and fire risk.
The presence of Arundo donax had a significant impact on the magnitude of topographic response to storm events after fire and encouraged a geomorphological regime that may promote riparian desiccation and fire recurrence. Management efforts should focus on the early detection and eradication of exotic plants that can contribute to the destruction of natural med-sys step-pool structures, elevated sedimentation, and increased fire risk. This work builds upon our current knowledge of wildfire and recovery processes by providing an integrated procedure to understand post-fire mechanisms and anthropogenic feedbacks that may occur in urban Mediterranean riverine systems [84]. As this is a novel case study approach, we recommend that future investigations expand monitoring to represent variable landscape conditions and geomorphic processes within the burned area to further elucidate invasive vegetation-driven erosional pathways after fire in urban areas. This research provides new information on spatial variability in burn severity, vegetation regrowth, and fluvial geomorphic hazards that should be considered by resource managers and engineers to inform management decisions in urban riverine environments prone to Arundo donax-initiated riparian fires.