Amazon windthrow disturbances are likely to increase with storm frequency under global warming

Identification of windthrow events

Landsat images from January 1st 2018 to December 31st 2019 were filtered on 20% or less of cloud coverage, and only the least cloudy image at each location was selected to make an image composite covering the entire Amazon region. In total, 395 least cloudy Landsat 8 images within the Amazon boundary during 2018–2019 were selected and displayed in false color (red: shortwave infrared band, green: near-infrared band, blue: red band) on Google Earth Engine for windthrow events identification (Supplementary Fig. 6). Hollow regions on Supplementary Fig. 6 (2.8% of the total area of the Amazon region) indicated that no clear images with <20% cloud were found in these areas. Spectral mixture analysis was applied on each image to detect potential windthrows (similar methods have been used in previous windthrow studies^5,17). Each pixel was unmixed to fractions of image-derived endmembers, including green vegetation (GV), non-photosynthetic vegetation (NPV), and shade. GV and NPV fraction images were normalized without shade and then used to identify windthrows. Windthrows were identified manually as large fan-shape areas with high NPV fraction. Each potential windthrow was then visually checked using false color images and evaluated based on authors’ 15-year experience working with windthrow and remote sensing^2,3,17,19. “New” windthrows that occurred within 1 year were spectrally more visible based on their clear fan-shape^5,10 (diverging from a central area with small pixels scattered at the tail) and their relatively distinguishable reddish colors (Supplementary Fig. 7a, due to high reflectance in shortwave infrared band from woody biomass), while “old” windthrows (Supplementary Fig. 7b) occurred >1 year before the identification were displayed in bright green colors (due to reflectance in near-infrared band from the pioneer species). “Old” windthrows account for ~80% of total identified windthrows, and they were verified using historical Landsat images that can go as far as 1984 (when Landsat 5 was launched). “Old” windthrows were validated once they were found with clear shape and more distinguish color on the historical Landsat images (Supplementary Fig. 7c). 10–15% of “old” windthrows without fan-shape were eliminated from this study because it was hard to identify if they were windthrows or other types of forest disturbance. The minimum size of windthrows identified in this study was 25,000 m². This process generated the location and rough size of 1012 visible (both “old” and “new”) windthrow scars with fan-shaped patch, scattered small disturbance pixels tails, and an area of over 25,000 m² (Supplementary Fig. 8). Based on a gap-size probability distribution function that simulates the entire disturbance gradient from all sizes of windthrows¹⁹, the proportion of total tree mortality represented by large windthrows (>25,000 m²) identified in this study is 0.5–1.1%.

Among 1012 visual identified windthrows, the occurrence year of 125 windthrows were identified using Landsat 5,7,8, MODIS, and TRMM dataset (Supplementary Table 2), and 38 windthrows from these 125 windthrows had clear remote sensing evidence to validate their occurring date (Supplementary Table 3). It is difficult to get the accurate year and date of occurrence of all identified windthrows. Previous studies showed that windthrows in the northwestern Amazon took ~20 years to recover to 90% of “pre-disturbance” biomass from all damage classes while forests in the central Amazon took ~40 years to recover⁴⁰. The biomass recovery depends on the windthrow severity and time since disturbance³³. Based on the recovery time (20–40 years) and the time of windthrow identification (2018–2019), we estimated that these 1012 windthrows most likely occurred within 30 years (between 1990 and 2019), and the estimated occurrence period was validated using the range of the occurrence year (1986–2019) of 125 windthrow cases.

Windthrow density data

The windthrow density shown in Fig. 1b was generated using 1012 windthrow points in QGIS⁴⁵. We created a 2.5° by 2.5° grid map, and the windthrow density was calculated by counting the number of windthrows in each grid. These values were then converted to a density with units of counts of windthrows per 10,000 km². We chose 2.5 degrees to aggregate the data to make sure that over 50% of grids have at least 1 windthrow event while still preserving the spatial distribution of mean afternoon CAPE over the Amazon. The contour lines displayed in Fig. 1c were generated using the “Contour” function on the windthrow density map in QGIS.

Meteorological data

To derive the correlation between windthrow density and meteorological variables, we used ERA 5 global reanalysis hourly CAPE on single levels from 1979 to present at 0.25° × 0.25° resolution provided by the European Center for Medium-Range Weather Forecasts. ERA 5 CAPE was computed by considering parcels of air departing at different pressure levels below the 35 kPa level, with maximum–unstable algorithm under a pseudo-adiabatic assumption⁴⁶. Afternoon mean CAPE map was calculated as the average of hourly CAPE data from 17:00–23:00 UTC (13:00–19:00 local time in Amazon) over all the months between 1990 and 2019. We chose to average CAPE over 30 years because these windthrow events occurred in these 30 years and calculating the average can help capture the overall spatial pattern of CAPE and minimize the influence of interannual climate variability on windthrow events.

To project future windthrow density in the Amazon for the end of the 21st century, we analyzed meteorological output from 10 ESMs that participated in CMIP6 (https://www.wcrp-climate.org/wgcm-cmip/wgcm-cmip6). The models used in this research were listed in Supplementary Table 1. We extracted daily surface temperature (tas), specific humidity (huss), surface pressure (ps), temperature (ta) from these models to calculate daily nondilute, near-surface-based, adiabatic CAPE. CMIP6 CAPE was calculated by considering the buoyancy of a near-surface parcel lifted adiabatically to a series of discrete pressure levels (100 kPa to 10 kPa in increments of 10 kPa). CMIP6 CAPE is calculated as follows:

Where ({{{{{rm{d}}}}}}p) = 10 kPa, H is the Heaviside unit step function, and ({b}_{i}=frac{1}{{rho }_{i}}-frac{1}{{rho }_{e,i}}), with ({rho }_{i}) being the parcel density at pressure level i and ({rho }_{e,i}) being the environmental density at pressure level i.

The future projections in our analysis were based on SSP585, a high-emission scenario with high radiative forcing by the end of the century. We calculated mean daily CAPE over 1990–2015 as current CMIP6 CAPE and mean daily CAPE over 2070–2099 as future CMIP6 CAPE. Since different approaches were used to calculate ERA 5 CAPE and CMIP6 CAPE⁴⁷, the absolute CAPE values of the two datasets are not comparable. Therefore, for each ESM model, we scaled future CMIP6 CAPE by multiplying, grid-wise, the delta CAPE generated from an individual model in CMIP6 with the ERA 5 current mean afternoon CAPE (Fig. 1c) as follows:

The delta CAPE indicated the projected increase in CAPE from 1990–2015 to the end of the 21st century. In this way, a scaled CMIP6 future CAPE map was generated for each model, and an ensemble-mean scaled CMIP6 CAPE map over 10 ESM models can be found in Supplementary Fig. 5b. The scaled CMIP6 future CAPE values were within plausible range compared to the ERA 5 current mean afternoon CAPE values, and both current and future CAPE maps were used to produce the increase in area with high CAPE values (>1023 J kg⁻¹) in Table 1. However, it is worth noting that the scaling with relative changes in delta CAPE (%) is more sensitive to CMIP historical baseline conditions than absolute changes of CAPE (J kg⁻¹), which will likely introduce a larger scaled spread (min/max CAPE changes).

The increase in area with storm-favorable environments was calculated as follows:

Where area_current is the area of CAPE > 1023 J kg⁻¹ for current ERA 5 CAPE, and area_future is the area of CAPE > 1023 J kg⁻¹ for the scaled CMIP6 future CAPE.

A model of windthrow density

We developed a model based on the relationship between satellite-derived windthrow density and mean afternoon CAPE from the ERA 5 reanalysis over 1990–2019. The non-parametric model provides a look-up table of windthrow density as a function of CAPE within the range of observations. Counts of observed windthrow events and Amazon’s area were separately binned by CAPE using the same bins, producing two histograms of CAPE. The ratio of the former to the latter gives the density of windthrow events (windthrow events per area) as a function of CAPE. To avoid noise at the tails of the histograms, the six CAPE bins were chosen such that each bin would have about the same number of windthrow events (either 168 or 169). The total number of windthrow events is given by the sum over bins of the product of windthrow density and area. The minimum and maximum of current ERA 5 mean afternoon CAPE was 42 and 1549. The minimum CAPE value of the first bin was extended to 0 and the maximum CAPE value of the last bin was extended to infinity under the assumption that the windthrow density is similar for neighboring values. Based on the windthrow density and CAPE relationship used in the model, it is the increase in the area with high CAPE that then leads to an increase in the number of windthrow events.

It is worth noting that the future windthrow density produced by models may be underestimated because the windthrow observations within regions with high CAPE were incomplete due to high cloud coverage. Moreover, the non-parametric model makes the conservative assumption that the windthrow density does not increase at higher, as-yet unobserved values of mean afternoon CAPE.

Future projections of windthrow density

We combined the non-parametric relationship (Fig. 2a) with the future CAPE map generated from the ten CMIP6 ESMs (adjusted by ERA 5 mean CAPE values) to estimate the changes in windthrow density at the end of the 21st century. We estimated uncertainties for windthrow density projections by combining information about model-to-model differences. The analysis yielded a set of 10 estimates. The overall windthrow density increase and uncertainty were estimated using the mean increase and one standard deviation from the ensemble of the 10 models.

Source: Ecology - nature.com