Strong temporal variation in treefall and branchfall rates in a tropical 1 forest is related to extreme rainfall: results from five years of monthly 2 drone data for a 50-ha plot 3

. A mechanistic understanding of how tropical tree mortality responds to climate variation is urgently needed to predict how tropical forest carbon pools will respond to anthropogenic global change, which is altering the frequency and intensity of 18 storms, droughts, and other climate extremes in tropical forests. We used five years of approximately monthly drone-acquired RGB imagery for 50 ha of mature tropical forest on Barro Colorado Island, Panama, to quantify spatial structure, temporal 20 variation, and climate correlates of canopy disturbances, i.e., sudden and major drops in canopy height due to treefalls, branchfalls, 21 or collapse of standing dead trees. Canopy disturbance rates varied strongly over time and were higher in the wet season, even 22 though wind speeds were lower in the wet season. The strongest correlate of monthly variation in canopy disturbance rates was 23 the frequency of extreme rainfall events. The size distribution of canopy disturbances was best fit by a Weibull function, and was 24 close to a power function for sizes above 25 m 2 . Treefalls accounted for 74 % of the total area and 52 % of the total number of 25 canopy disturbances in treefalls and branchfalls combined. We hypothesize that extreme high rainfall is a good predictor because 26 it is an indicator of storms having high wind speeds, as well as saturated soils that increase uprooting risk. These results demonstrate 27 the utility of repeat drone-acquired data for quantifying forest canopy disturbance rates at fine temporal and spatial resolutions over large areas, thereby enabling robust tests of how temporal variation in disturbance relates to climate drivers. Further insights could be gained by integrating these canopy observations with high-frequency measurements of windspeed and soil moisture in mechanistic models to better evaluate proximate drivers, and with focal tree observations to quantify the links to tree mortality and woody turnover. et al.,


21
or collapse of standing dead trees. Canopy disturbance rates varied strongly over time and were higher in the wet season, even 22 though wind speeds were lower in the wet season. The strongest correlate of monthly variation in canopy disturbance rates was 23 the frequency of extreme rainfall events. The size distribution of canopy disturbances was best fit by a Weibull function, and was 24 close to a power function for sizes above 25 m 2 . Treefalls accounted for 74 % of the total area and 52 % of the total number of 25 canopy disturbances in treefalls and branchfalls combined. We hypothesize that extreme high rainfall is a good predictor because 26 it is an indicator of storms having high wind speeds, as well as saturated soils that increase uprooting risk. These results demonstrate 27 the utility of repeat drone-acquired data for quantifying forest canopy disturbance rates at fine temporal and spatial resolutions 28 over large areas, thereby enabling robust tests of how temporal variation in disturbance relates to climate drivers. Further insights 29 could be gained by integrating these canopy observations with high-frequency measurements of windspeed and soil moisture in 30 mechanistic models to better evaluate proximate drivers, and with focal tree observations to quantify the links to tree mortality and 31 woody turnover. 32 33 1 Introduction

34
Moist tropical forests account for 40% of the global biomass carbon stocks (Xu et al., 2021), and uncertainty regarding 35 the future of these stocks is a major contributor to uncertainty in the future global carbon cycle (Cavaleri et al., 2015). Tropical 36 forest carbon stocks depend critically on tree mortality rates, and recent studies suggest tropical tree mortality rates may be 37 increasing due to anthropogenic global change (Brienen et al., 2015;McDowell et al., 2018). Tropical tree mortality can be caused 38 by a diversity of drivers including windthrow (Fontes et al., 2018), droughts (McDowell et al., 2018Silva et al., 2018), fires (Silva 39 et al., 2018), lightning strikes (Yanoviak et al., 2017), and biotic agents (Fontes et al., 2018). The frequency of extreme rainfall 40 and drought events is expected to increase in tropical regions, potentially increasing associated tree mortality (IPCC, 2014;Deb et 41 al., 2018;Aubry-Kientz et al., 2019). An improved understanding of the processes of forest disturbance is critical to constrain 42 estimates of current and future carbon cycling in tropical forests under climate change (Leitold et al., 2018;Johnson et al., 2016; 43 Muller-Landau et al., 2021).

44
Despite the importance of tree mortality to forest structure and carbon turnover rates, the mechanisms underlying tree 45 mortality remain unclear (McDowell et al., 2018). A key problem is that remeasurement intervals of permanent plots average 5 or 46 more years, making it difficult to link mortality variation with particular climatic events (Phillips et al., 2010;Davies et al., 2021; 47 Arellano et al., 2019). The high rates of decomposition in tropical forests further obscure evidence of underlying mechanisms and 48 risk factors . The few studies that have quantified temporal variation of tree mortality at monthly and bi-49 monthly scales using ground-based data have all found higher tree mortality in times of higher rainfall (Brokaw, 1982;Fontes et 50 al., 2018;Aleixo et al., 2019). This is consistent with the understanding that many trees die in treefalls, which are proximately 51 caused by trunk breakage or uprooting, and are associated with storms (Marra et al., 2014;Araujo et al., 2017;Fontes et al., 2018; 52 Negrón-Juárez et al., 2017Esquivel-Muelbert et al., 2020). The collection of additional high temporal resolution mortality 53 data over large areas, together with high temporal resolution climatological data, can aid in linking mortality to particular climatic 54 events and thereby elucidating mortality mechanisms McMahon et al., 2019).

55
Drone-acquired imagery and digital aerial photogrammetry software now provide excellent tools for monitoring of forest 56 canopies (Araujo et al., 2020) and repeat drone flights can quantify canopy dynamics over large areas at high temporal resolution.

57
Photogrammetric analysis of simple RGB imagery enables reconstruction of the appearance and three-dimensional structure of the 58 top of the canopy at high spatial resolution (Dandois and Ellis, 2013;Araujo et al., 2020;Zahawi et al., 2015). Comparison of 59 photogrammetry products from successive drone flights allows easy detection and quantification of canopy disturbances due to 60 treefalls and branchfalls of canopy trees. Canopy trees constitute a high proportion of stems, aboveground carbon stocks and woody 61 productivity (Araujo et al., 2020), and thus information on their mortality rates is disproportionately useful to understanding forest 62 dynamics and carbon cycling. Treefalls do not necessarily result in tree mortality (trees may survive and resprout), but almost all 63 treefalls and branchfalls result in a large flux of carbon (wood) from biomass to necromass within a short time period after the 64 event, which translates to reduced woody residence time. Periods of higher canopy disturbance rates thus represent periods of 65 higher biomass turnover, and likely correlate with higher tree mortality rates. Further, even when trees do not die from a canopy 66 disturbance event, suffering crown loss or damage increases the risk of subsequent mortality .

67
Monitoring canopy disturbances with drones also provides the opportunity to precisely quantify the size distributions of 68 these canopy disturbances, and to distinguish branchfalls from treefalls. Here we define a canopy disturbance as a substantial 69 decrease in canopy height in a contiguous patch of canopy occurring over one measurement interval, such as typically results from 70 a treefall or branchfall. Marvin and Asner (2016) and Dalagnol et al. (2021) referred to these as "dynamic canopy gaps." By 71 definition, canopy disturbances reduce canopy height and thereby change light regimes for understory and neighboring trees, and 72 the magnitude of the change depends on the disturbance size in area and depth (Hubbell et al., 1999). In general, larger canopy 73 disturbances cause larger canopy gaps as traditionally measured on the ground. Previous studies have analyzed the size distributions 74 of static gaps -areas with canopy height below a threshold -for insights into forest structure, habitat niches, and disturbance 75 regimes (e.g., Manrubia and Solé, 1997;Dalling, 2013, 2014;Fisher et al., 2008). Tree species respond differently to 76 canopy gaps of different sizes, with small gaps favoring a different set of species than large gaps (Brokaw, 1985;Denslow, 1980Denslow, , 77 1987Dalling et al., 2004). Branchfalls, like treefalls, are important in generating canopy gaps and contributing to woody turnover, 78 but also often go unmeasured (Marvin and Asner, 2016;Leitold et al., 2018). Quantifying tree mortality and other damage 79 contributes to a better understanding on change of forest structure, necromass estimates and nutrient cycling.

80
Here, we use five years of ~monthly drone-acquired RGB imagery for a 50 ha area of mature tropical forest on Barro 81 Colorado Island, Panama to investigate canopy dynamics at high temporal resolution. We aim to (1) quantify temporal variation 82 in canopy disturbance rates and its relationship to climate variation; (2) characterize the size structure of canopy disturbances; and

83
(3) evaluate the role of branchfalls in canopy dynamics. We expect that disturbance rates will be higher in the wet season than the 84 dry season, we hypothesize disturbance rates will increase with the frequency of extreme rainfall and wind events, and we compare 85 the correlations of various rainfall and wind statistics with temporal variation in disturbance rates. To characterize the size structure 86 of canopy disturbances, we quantify the size (area) distribution and evaluate whether it is best fit by power, Weibull, or exponential 87 functions. Finally, we quantify the proportion of canopy disturbance due to branchfalls (rather than treefalls), and test whether 88 branchfalls and treefalls exhibit similar patterns of temporal variation. Our results provide new insights into the patterns and 89 drivers of canopy disturbance and tree mortality in this tropical forest, and illustrate the utility of drones for quantifying canopy 90 dynamics over large areas at high temporal resolution.  (Holdridge, 1947). Annual precipitation averages approximately 2600 mm, with a pronounced dry 98 season between January and April (a mean of about 3.5 months with < 100 mm mo -1 ).  (Windsor, 1990). The 50 ha forest dynamics plot (1000 m x 500 m) was 102 established on BCI in 1981 and is located in an old-growth forest (Leigh, 1999), with the exception of a small area of 1.92 ha of 103 old secondary forest (~100 years old) in the north central part of the plot (Harms et al., 2001

117
We used approximately monthly orthomosaics and canopy surface models produced from drone-acquired imagery to 118 analyze temporal variation in canopy disturbance rates in the 50 ha forest dynamics plot between 2 October 2014 and 28 November 119 2019. RGB imagery was collected using a variety of drones and cameras over the years, with a horizontal spatial resolution of 3-120 7 cm. Imagery for each sampling date was processed using the photogrammetry software Agisoft Metashape to obtain orthomosaics 121 and surface elevation models, which were then aligned vertically and horizontally.

122
We defined a canopy disturbance as a substantial decrease in canopy height in a contiguous patch of canopy occurring 123 over one measurement interval, such as typically results from a treefall or branchfall. We identified canopy disturbances through 124 a combination of analysis of the canopy surface model changes and visual interpretation of the orthomosaics (Fig. 1). We first 125 differenced surface elevation models for successive dates to obtain a raster of the canopy height changes for the associated interval 126 ( Fig. 1, Text S1). We then pre-delineated major canopy disturbances by filtering for areas in which canopy height decreased more 127 than 10 m in contiguous areas of at least 25 m 2 , and that had an area-to-perimeter ratio greater than 0.6. We note that 25 m 2 is the 128 minimum gap area used in previous studies of this site by Brokaw (1982) and Hubbell et al. (1999). The area-to-perimeter condition 129 removes artifacts associated with slight shifts in the measured positions of individual trees from one image set to another, whether 130 due to wind or alignment errors (note that this criterion involves a combination of shape and size). Finally, we systematically 131 examined 1-ha square subplots for each pair of successive dates and edited the pre-delineated polygons, removed false positives, 132 and added visible new canopy disturbances that were not previously delineated (whether because they were too small in area or in 133 canopy height drop). We also classified disturbances as being due to treefalls (a whole previously live tree fell, creating a clearly

141
We calculated the total number and area of canopy disturbances within the BCI 50 ha plot during the 5 years of the study.

142
In calculating the number and total area of disturbances, we included all disturbed areas that were inside the plot boundaries (if a 143 disturbance was on the boundary, only the area inside the plot was included). Our analyses of temporal variation employed the 144 same definitions for numbers and areas of canopy disturbances within the 50 ha plot. For analyses of the size structure of 145 disturbances, we included the complete areas of disturbances whose centroids were located within the plot (i.e., we excluded 146 disturbances centered outside the plot, and included area outside the plot for disturbances centered inside the plot to avoid artifacts 147 related to reducing disturbance size by trimming at the plot boundaries).

150
We calculated canopy disturbances rates for each measurement interval as the % of area disturbed per month (i.e., per 30-151 day period). Specifically, we summed the total area disturbed during the measurement interval, and divided by the total area of 152 the plot and the length of the time interval. We excluded one excessively long interval (237 days -image acquisition gap) from all 153 analyses of temporal variation; the remaining intervals ranged from 14 to 91 days, with a median of 31.5 days (Table S1). We also 154 calculated an incidence canopy disturbance rate as the number of canopy disturbances per hectare per month. We calculated the 155 mean, minimum, maximum, and the 25 th , 50 th , and 75 th percentiles of interval length, number and area of canopy disturbances, and 156 the respective monthly rates.

157
We compared canopy disturbance rates between wet and dry seasons and between early wet and late wet seasons. We 158 defined the dry season as January 1 to April 30 (rainfall < 100 mm mo -1 , Fig. S3), the early wet season as 1 May to 31 August, and 159 the late wet season as 1 September to 31 December. Intervals that straddled more than one season were classified to the season in 160 which they had more days. We tested for differences in canopy disturbance rates between seasons using two-tailed Student's t-test 161 on the log-transformed canopy disturbance rates for each measurement interval, after first confirming that these rates met 162 assumptions for normality (Shapiro-Wilk test) and homogeneity of variance (Levene test).

163
We evaluated the relationship of temporal variation in canopy disturbance rates with temporal variation in the frequencies 164 of climate extremes using parametric correlations. We calculated the .
188 where and are fitted parameters, is canopy disturbance area in m 2 , is the natural exponential basis, and are normalization 189 constants such that the truncated distribution integrates to 1. Recognizing that our methods are likely to miss smaller disturbances, 190 we fit these distributions to truncated datasets, excluding disturbances below 2, 5, 10 or 25 m 2 . Note that 25 m 2 is the minimum 191 area for defining a canopy disturbance in our automated pre-delineation algorithm, and we are confident we captured all 192 disturbances above this area. We are progressively less confident of our ability to capture smaller disturbances. We also truncated 193 the fitted distributions above at the maximum possible disturbance area we could have observed using our methods (50 ha, or 194 500,000 m 2 ). We fit each type of distribution (exponential, power, Weibull) to each dataset ( We selected the model that minimized Akaike's Information Criterion (AIC) (Burnham and Anderson, 2002). We also evaluated 198 goodness of fit using the Kolmogorov-Smirnov statistic, the maximum difference in the cumulative probability distributions 199 between the observed data and the fitted distribution (Carvalho, 2015).

201
We classified each canopy disturbance as being a branchfall, treefall, or standing dead tree, except for those disturbances 202 occurring in the exceptionally long time interval. In 35 cases we could not distinguish the type of disturbance, and these cases were 203 omitted from analyses that required disturbance classification. We evaluated the relative contributions of branchfalls vs. treefalls, 204 and we did not include standing dead trees in the analysis because our methods possibly missed standing dead trees. We separately 205 calculated treefall and branchfall disturbance rates for each interval, and relative contributions to their summed number and area.

206
We calculated the Pearson correlations of branchfall disturbance rates with treefall disturbance rates, for both area-and number-207 based rates. 208 209

210
We identified 1048 canopy disturbances with a combined area of 56,134.37 m 2 (5.61 ha) that affected the area within the 211 BCI 50 ha plot between 2 October 2014 and 28 November 2019 (Fig. 2). During the 5 years of the study, 11.2 % of the area of the 212 BCI 50-ha plot was affected by canopy disturbances (Fig. 2), and 0.6 % was disturbed more than once (Fig. S4). There was strong temporal variation in canopy disturbance 220 rates, with similar temporal variation in the total area disturbed (Fig. 3) and in the number of disturbances (Fig. S5). The mean rate 221 of canopy disturbance creation was 905.1 m 2 mo -1 (range of 75 m 2 mo -1 to 8040.9 m 2 mo -1 ) and the median 499 m 2 mo -1 (other 222 statistics in Table S1).

224
The single highest disturbance rate was observed between 1 June and 13 July 2016, when 11,257 m 2 of disturbances were created 225 in just 42 days (a rate of 268 m 2 day -1 ). A full 2.3 % of the total area of the plot was converted to new canopy disturbances during

241
The best correlate of temporal variation in canopy disturbance rates was the frequency of 15-min rainfall events above 242 the 98.2 th percentile, which explained 22 % of the variation (Fig. 5a). This relationship was mainly driven by events occurred 243 during wet seasons (Fig. 5a). This threshold outperformed all other tested rainfall thresholds (all percentiles from 90.0 to 99.9, by 244 0.1 % of the different frequency time scales - Fig. 5b). The 98.2 th percentile corresponds to a rainfall rate of 24.3 mm hour -1 (Fig.   245   5c). There were a total of 141 15-min rainfall events exceeding this threshold, which occurred on 98 different days (Table S2). The 246 measurement interval with the highest disturbance rate (June 1 to July 13 2016) included eleven such high 15-min rainfall events 247 on six days (Table S2). The frequency of high maximum wind speed events was not significantly related with canopy disturbance 248 rates. Indeed, Pearson correlations were negative for most wind speed variables (Fig. S7).  (Fig. 6a).

262
The size distribution of observed canopy disturbances was close to a power function for areas above 25 m 2 , and was 263 relatively flat over the range of 5 to 25 m 2 (Fig. 6b). The fitted exponent of the power function was -2.16 for canopy disturbances 264 above 25 m 2 , but the Weibull distribution provided a better fit than the power function (Table 1). When distributions were fit to 265 data including smaller size classes (> 2 m 2 , > 5 m 2 or > 10 m 2 ), the distribution is further from a power function; the Weibull 266 remains the best fit, the exponential becomes the second-best fit, and the power function the worst fit of the three (Fig. S8, Fig. S9, 267 Table 1). Canopy disturbances with larger areas tended to have larger mean decreases in canopy height (Fig. 6c, Fig. 6d).

288
although the ratios of their rates varied among measurement periods (Fig. 7, Fig. S10)

296
The use of high frequency drone imagery enabled us to quantify temporal variation in canopy disturbance rates and to 297 quantify the sizes of canopy disturbances at high temporal and spatial resolutions. We found that canopy disturbance rates of the 298 BCI 50 ha plot varied strongly over time, and were higher in the wet season. The frequency of extreme rainfall events was the best 299 correlate of monthly variation in canopy disturbance rate during the 5-year study period. In contrast, maximum wind speed was 300 not significantly correlated. The size distribution of canopy disturbances was close to a power function for larger canopy 301 disturbances, but best fit by a Weibull function overall. Branchfalls accounted for 26 % of the total area of disturbances from 302 treefalls and branchfalls combined, and branchfall rates varied largely in parallel with treefall rates over time. These findings 303 contributed to improve the understanding of the size distribution, temporal variation and meteorological drivers of canopy 304 disturbances in tropical forests.

306
Canopy disturbance rates varied strongly over time in this moist tropical forest, and were higher in the wet season. A 307 single time interval (June 1 to July 13 2016) accounted for 20 % of the total disturbed area of the BCI 50-ha plot. The frequency 308 of extreme rainfall events was a strong correlate of the variation in canopy disturbance rates among measurement intervals, whereas 309 the frequency of high maximum wind speeds was not related. At our site, wind speeds are higher during the dry season, when 310 canopy disturbance rates are lower (Fig. 4a, Fig. S1), and it is possible that wind speed is systematically underestimated in periods 311 of high rainfall. We also note that wind speed and rainfall measurements were from a site 1.7 km from the boundary of the plot.

312
Given the highly local nature of convective storms in the tropics, these measurements are imperfect proxies for conditions in the 313 focal plot. Treefall and branchfall disturbance rates varied largely in parallel, but not entirely. Differences in temporal patterns 314 could in part reflect different sensitivity to particular abiotic drivers (e.g. wind regime, soil saturation).

315
These results are consistent with previous findings on seasonal variation and the role of rainfall in gap formation in tropical 316 forests. A previous 4-year study on BCI found seasonal peaks in August and September, in the middle of the wet season, with 317 monthly treefall rates significantly correlated with rainfall (r = 0.47, p < 0.02) (Brokaw, 1982). Monthly tree mortality was also 318 strongly and positively correlated with rainfall (r = 0.85) in a 1-year study of a 10-ha site in the Central Amazon (Fontes et al., 319 2018). Similarly, a study monitoring canopy trees monthly over five decades in the Central Amazon found that trees died more 320 often during wet months, even in drought years (Aleixo et al., 2019). A regional study of the Central Amazon based on 12 years 321 of satellite data found that major windthrows (visible on LANDSAT) occurred more frequently between September and February, 322 months characterized by heavy rainfall, than the rest of the year (Negrón-Juárez et al., 2017). Studies have highlighted the 323 importance of mesoscale convective systems, such as squall lines, for windthrows (Garstang et al., 1998;Negrón-Juárez et al., 324 2010, 2017Araujo et al., 2017). In Panama, the period of June to August has the higher number of mesoscale convective systems 325 (Jaramillo et al., 2017), and these were the months when we observed the highest canopy disturbance rates. The threshold rainfall 326 rate of 24.3 mm hour -1 , which defined the extreme rainfall rate that was the best predictor of canopy disturbance formation in our 327 study, is four times higher than the mean rate for mesoscale convective systems in the Panama region (Jaramillo et al., 2017), 328 highlighting the importance of extreme events. Analysis of spatial variation in forest damage from Hurricane María in Puerto Rico 329 found that total rainfall was the most important meteorological risk factor and maximum sustained one-minute wind speeds the 330 second-most-important; these two variables were moderately correlated (r = 0.43) (Hall et al., 2020 analyses, which focus on short-term changes that indicate loss of major canopy elements. In contrast, the decay of dead trees and 336 senescing branches generally involves more subtle changes in the canopy over a longer period of time, and is possibly mostly 337 missed by our methods. Treefalls account for a majority of canopy tree mortality in most tropical forests, but standing tree mortality 338 also plays a major role, especially in drought periods. Overall, treefalls (in which trees were uprooted or their trunks snapped) 339 accounted for 51.2 % of all mortality of trees > 10 cm DBH in a large-scale study of tree mortality in 189 Amazonian plots 340 (Esquivel-Muelbert et al., 2020) and 65 % in a study that monitored tree mortality in 10 ha of forest in the Central Amazon bi-341 monthly over one year (Fontes et al., 2018). Treefalls can involve a single canopy tree, or multiple canopy trees. Multi-tree treefalls 342 can result from coordinated disturbances over a large area (e.g., large footprint wind disturbance) and/or from domino effects in 343 which the failure of one canopy tree directly stresses one or more neighboring trees and causes them to fall as well (e.g., when 344 additional trees are knocked down by the first tree, or pulled down because of connections via lianas). It has been hypothesized 345 that canopy disturbances may also be contagious over longer time intervals, with increased risk of treefall near canopy gaps, but 346 evidence for this in tropical forests is mixed (Jansen et al., 2008). Given that our measurement intervals are relatively short (~one 347 month), almost all of our mapped canopy disturbances are likely to reflect single catastrophic events.

348
Our study is one of several that have documented size distributions of canopy disturbances (dynamic gaps) or of static 349 canopy gaps above some size that are approximately power functions, both on BCI (Solé and Manrubia, 1995;Lobo and Dalling, 350 2014) and in other tropical forests (Marvin and Asner, 2016;Asner et al., 2013;Kellner and Asner, 2009;Silva et al., 2019;Fisher 351 et al., 2008). Static canopy gaps are areas in which the forest canopy is below a threshold height, e.g., 10 m, at a given time. A 352 power function distribution of disturbance event sizes (here canopy disturbances) and of the sizes of disturbed areas (canopy gaps) 353 can emerge from self-organization of dynamic systems such as forests in which individual tree growth and death depend on the 354 sizes of neighbors (Solé and Manrubia, 1995

359
However, we consider it unlikely that this is a sufficient explanation for the shortfall in small disturbances, and suggest that it is 360 more likely explained largely by the low frequency of small trees and branches in the canopy of this mature tropical forest, and 361 thus a dearth of small treefall and branchfall events.

362
Although rarely quantified, branchfall is an important ecological process, with major contributions to woody turnover and 363 necromass production. We found that branchfalls were almost as common as treefalls in number, although they contributed a 364 substantially smaller total area of disturbance. Similarly, a ground survey of 78 canopy turnover events in a Brazilian Amazon 365 forest found that 44 % were branchfalls, and accounted for 15 % of the total affected area (Leitold et al., 2018). In contrast, a 366 landscape level analysis of LiDAR data concluded that branchfalls were seven times more frequent than treefalls and accounted 367 for five times more area (Marvin and Asner, 2016). However, Marvin and Asner (2016)  are disproportionately important for determining stand-level woody residence time (Araujo et al., 2020). Advances in drone 379 hardware and photogrammetric software now make it relatively inexpensive and straightforward to quantify forest canopy structure 380 and dynamics at high spatial and temporal resolution through digital aerial photogrammetry and repeat drone imagery acquisitions.

381
Here we applied these methods to 50 ha of old-growth tropical forest for 5 years, and analyzed the resulting products to quantify 382 major drops in canopy height such as those created by branchfalls and treefalls, and thus calculate the canopy disturbance rate. We 383 found that canopy disturbance rates are highly temporally variable, and are well-predicted by extreme rainstorms. Spatial 384 resolutions of 3-7 cm in the orthomosaics, as used here, are now easily attained, and proved sufficient to capture canopy dynamics 385 and visually classify disturbances as treefalls, branchfalls, or decomposition of standing dead trees.

386
Future research building on these approaches and expanding them to additional sites has much to contribute to our 387 understanding of tropical forest dynamics. The relationship of standing dead tree mortality to temporal climate variation could be 388 investigated from these same data by conducting additional analyses of the orthomosaics to quantify temporal changes in leafing 389 status of standing dead trees, prior to these trees decomposing. A better understanding of the relationship of storm conditions to 390 treefall and branchfall rates could be obtained by combining such drone-acquired data with mechanistic models of wind damage 391 risk (Jackson et al., 2019), collecting higher frequency three-dimensional wind data, and/or measuring canopy dynamics at even 392 higher temporal resolution. The use of drones with high accuracy GPS systems, either post-processed kinematic (PPK) or real-393 time kinematic (RTK) systems, would also be advantageous, and could enable elimination of the alignment step of the processing 394 as well as automation of the identification of canopy disturbances based on elevation model differences alone. Finally, we 395 recommend carrying out flights under cloudy conditions when possible, as these diffuse lighting conditions improve visibility 396 deeper in the canopy and reduce complications associated with shadows. The expansion of these methods to additional and larger 397 areas, potentially in part through citizen science initiatives, has great potential to improve our understanding of tropical forest tree 398 mortality, and the future of tropical forests under changing climate regimes.

400
Code and data availability. Analysis codes, input data and output results are available at https://github.com/Raquel-