Suitability of the global forest cover change map to assess climatic megadisturbance impacts on remote tropical forests | Scientific Reports -…
Temporal analysis of megadisturbance impacts in the AHNP
Before 2016, very little impact was detected by the GFCC product (Table 1). Hernndez Rodrguez and Cruz Flores15 likewise reported very small spectral variations between 2000 and 2010 over the National Park, suggesting little impacts on the vegetation cover. In the 20002015 time period, seven tropical storms were registered whose track passed through the earternmost region of Cuba (Fig.6,27), including Tropical Storm Isaac (2012) whose trajectory crossed the Park. Maximum wind speed registered for Tropical Storm Isaac (2012) was about 50 knots, significantly lower than the 112 knots of Hurricane Matthew.
Track of the tropical storms registered in the 20002017 time period, easternmost region of the Republic of Cuba27. The Alejandro de Humboldt National Park is highlighted in red colour. The map was created using ArcGIS 10.7 software (https://support.esri.com/en/products/desktop/arcgis-desktop/arcmap/10-7-1).
Two tropical storms in this region reached hurricane category in the 20002015 time period: Hurricane Ike (2008, 111 knots) passed about 60km North of the National Park area, and Hurricane Sandy (2012, 100 knots) passed about 90km West of the National Park area (Fig.6). Wind speeds of these meteors in the National Park area were typical of tropical storms. The megadisturbance impacted area estimated for 2008 and 2012 were only 21.6ha and 9.8ha, respectively. A thorough visual inspection of high resolution imagery in the GoogleEarth archive confirmed this low visible impact in 2008 and 2012.
By contrast, Hurricane Matthew passed about 30km East of the National Park area (Fig.7,28), with close to maximum hurricane force wind speeds affecting the Easternmost part of the National Park.
Hurricane Matthews track and distribution of wind fields during its passage over Eastern Cuba on 5th of October 201616,28. Most of the Park area is within Hurricane force windfield. The map was created using ArcGIS 10.7 software (https://support.esri.com/en/products/desktop/arcgis-desktop/arcmap/10-7-1).
The significative decrease in spectral indices in the AHNP in 20162017 with respect to 20002015 (Table 4) is in contrast with the general increase in NDVI found by Cruz Flores et al.29 accross the national protected areas between 20112015 and 20162018. The latter study was based on NDVI maps at 300m spatial resolution, and was not meant to capture local trends. In our approach, the GFCC and NDVI products at 30m spatial resolution provide the means of quantifying megadisturbance at the scale of National Protected Areas.
In 2016, the estimation of the area impacted using NDVI and GFCC is similar (12.6% and 11.8% of the AHNP total area, respectively). NDVI tends to capture temporary vegetation impacts whereas GFCC, based on the wetness, brightness and greenness indices, should be sensitive to longer lasting impacts11. This similarity suggests that megadisturbance occurred in 2016 was mainly associated with defoliation caused by the Hurricane Matthew impact16, remaining after three month (early OctoberDecember 2016). A depression in NDVI three month after Hurricane Mara was likewise reported in Puerto Rico by Hu and Smith8.
By contrast, in 2017, no major extreme event occurred, and yet, megadisturbance was detected in an additional 1276 hectares (1.8% of the National Park area). Perturbance, possibly related to prolonged drought, was registered in Cuban national protected areas, including AHPN, before 201613,30 and in 201631. The degradation detected in the "forest loss" product in 2017 could relate to a long lasting effect of the megadisturbance combining Hurricane Matthew and prolonged droughts in previous years. This interesting finding could corroborate de Beurs et al.'s hypothesis12 that studies based on the change detection of appropriate remote sensing spectral indices at medium resolution (1030m) may detect persistent defoliation and degradation following the combination of several extreme events (in this case a hurricane event and previous prolonged droughts).
Hurricane Matthew passed a few kilometers East of the Alejandro de Humboldt National Park in the Atlantic ocean (Fig.7,28). The large area impacted in the easternmost part of the AHNP (especially in the Toa watershed and along the coast) seems largely related to the high wind speeds and the strength of the vortices in the vicinity of the hurricane trajectory.
Additionally, according to the exposition map in Fig.8, forests on slopes with exposition near to the South-east were most vulnerable to impacts, presumably because the general direction of Hurricane Matthew was from South-east to North-west. For example, slopes with predominant exposition to the South in the Jaguani watershed were particularly impacted (Fig.8). By contrast, few forests with exposition to the North were impacted.
Predominant slope expositions in megadisturbance impacted forests (mapped in Fig.3) of the Alejandro de Humboldt National Park. The map was created using ArcGIS 10.7 software (https://support.esri.com/en/products/desktop/arcgis-desktop/arcmap/10-7-1).
Temporary impact on lowland rainforest areas is illustrated in Fig.9 using Sentinel-2 colour composites and NDVI images before and after the Hurricane Matthew event. The NDVI map indicates redensification of vegetation 15months later in these areas (Fig.9).
Sentinel-2 colour composites and NDVI images of the Jaguan watershed (a tributary of the Toa river), Alejandro de Humboldt National Park, before and after the Hurricane Matthew event in October 2016.
Prolonged impact on the coastal swamp forest is illustrated in Fig.10 using Sentinel-2 colour composites and NDVI images before the Hurricane Matthew event and in December 2018 (15 Months after the event). On the right hand side of the images, persistent defoliation was observed in December 2018 in patches that showed dense vegetation in September 2016, just before the event. The local increase in sea level and the scattering of saline water during the hurricane event may both have caused high mortality of trees in parts of the coastal swamp forest.
Sentinel-2 colour composites and NDVI images of a coastal area (between Taco Bay to the Southeast, and the Jaragu Point to the Northwest), Alejandro de Humboldt National Park, before and after the Hurricane Matthew event in October 2016.
NDVI distribution has been statistically documented at the national level in the Republic of Cuba12,32 or national park level15,29. Additionally, remotely sensed spectral indices have been assessed in Cuba for forest loss versus forest persistence in forest management areas33. Based on a similar use of spectral indices, our study provides the first methodology for degradation assessment in national protected areas in the Caribbean. Our methodology and cartographic dataset could enrich the impact assessment framework of local forest management companies in charge of national protected areas in Cuba (e.g. in Baracoa for AHNP34).
In our study, the area impacted by megadisturbance was estimated applying good practices of area change estimation22,23 to the GFCC product. Accordingly, the "forest loss" layer only slightly underestimated (by 9%) the megadisturbed area (11,110 hectares). This result contrasts with results of studies on the forest loss layer over areas of anthropic deforestation, degradation and selective logging sites where much more underestimation was registered35,36,37. Megadisturbance events (e.g. hurricanes, prolonged droughts) may affect forests on a much larger extent and more homogeneously than anthropic intervention, which makes the GFCC forest loss layer more accurate at estimating impacted areas in the case of megadisturbance.
Recent degradation studies in the neotropical forests do not make use of the GFCC product to estimate degradation because most disturbance is due to shifting cultivation38. By contrast, in the case of little anthropic disturbance, our study suggests that the GFCC product can be useful for the assessment of megadisturbance impacts. In a long-term Typhoon study in Taiwan, Lin et al.39 document thatit took two years for litterfall to return to pre-Typhoon levels after a major event in 1994, and annual peak leaf area index only returned to pre-event levels after ten years. This recovery timescale corroborates that the yearly forest loss GFCC product could successfully capture the spatial distribution of megadisturbance impacts in subtropical settings. According to de Beurs et al.12, recovery from megadisturbance appeared much slower using the disturbance index (DI), than using NDVI. The GFCC "forest loss" product is partly based on the greenness, wetness and brightness indices of the Landsat sensor bands, which are used to compute the DI index.
Limitations of our study include difficulties in the accuracy assessment process: Verification sites, visible on the high resolution imagery of the Google Earth archive, are not necessarily identifiable in the Landsat imagery used to generate the GFCC "forest loss" product. As a consequence, in mountainous settings, geometric errors due to cumulated uncertainties in the georeferenciation of the Landsat imagery and of the high resolution imagery could generate errors in the accuracy assessment process. This difficulty is hard to overcome with the visual assessment of sites in some homogenous forested environmentson the Landsat imagery because of the (too coarse) 30m spatial resolution. Annual forest loss maps at 10m resolution (near to crown scale) derived from Sentinel-2, for example, should be more adapted to the application of forest degradation estimation in the future.