Through the NDVI analysis of the Planet dry season image (2016-08-26), we delimited the four forest patches with different sizes and environmental conditions (Fig. 2). These patches are on the west facing slopes of the extinct volcano, adjacent to the rim of the collapsed crater. The north patch, the smallest, with 11.6 hectares comprising only young trees. This patch is the most exposed to wind, with rocky soils mostly, and an average slope of 17%. The central patch with 74 hectares; it has a non-burned zone (NBZ) with mature trees and a large burned zone (BZ) with young trees, the topography with an average slope of 12%, allows the accumulation of water. The central-south patch extending 9 hectares with only young individuals; it is a new zone of the natural expansion of the forest, with an average slope of 13%. The south patch, the largest (263 hectares) where a third of its extension is a NBZ, having mainly mature trees, and the rest is a BZ with mostly young trees, it has a steeper slope with an average of 18%.
On the 24 plots surveyed in 2019, we found 638 individuals taller than 2 m. Twelve of them were mature, and 626 were young; only the height of 575 trees were measured because the rest were too close to each other, making it impossible to distinguish the treetop of each one. The height range of young trees observed in the field was from 2 to 14 m: 74% less than four m, 18% between 5 and 7 m, and 8% between 8 and 14 m high. We could observe mature trees from 6 to 24 m in height. At least 68% of the counted trees had reproductive structures, in individuals from 1.5 m to 24 m.
The canopy height models cover all the patches, and the comparison between the two dates provides information about the changes in the vertical structure of the forest, allowing us to monitor the appearance, growth and loss of trees. Examples of these results are in Fig. 6.
The correlation between the number of individuals measured in the field and Forest tools results was R = 0.65, underestimating the number of young trees and overestimating the mature trees. Despite this, it shows a correlation between the field measurements and ForestTools results.
The correlation between tree heights measured in the field and from ForestTools was R = 0.92, showing that height values of the CHM are accurate to assess the forest and describe its vertical structure.
Table 3 shows the population of the cypress forest calculated using ForestTools, for the 2016 and 2019 surveys. The current densities of individuals by patches are shown in Table 4, considering the burned and non-burned zones in central and south patches (BZ and NBZ, respectively); the areas used to calculate these densities are the areas delineated for each polygon (Fig. 2). It is important to remember that the north and central-south patches have only BZ. The NBZs are covered mainly by mature trees with heights between 6 and 21 m. However, in these areas, some young trees grew after the fire due to the natural dynamic of the forest.
Table 3
The number of individuals calculated for each patch, according to the camera used and survey year.
Number of individuals |
| | Patches |
| | North | Central | Central-South | South | Total |
2016 NIR | Young | 2 459 | 33 929 | 286 | 25 899 | 62 576 |
Mature | 0 | 120 | 0 | 1684 | 1,804 |
Total | *2459 | *34,049 | *286 | *27,583 | *64,380 |
2016 RGB | Young | 5 759 | 47 348 | 565 | 39 878 | 93 550 |
Mature | 0 | 122 | 0 | 1675 | 1 797 |
Total | *5759 | *48 236 | *565 | *41 553 | *95 374 |
2019 RGB | Young | 3 986 | 38 425 | 419 | 22 558 | 65 388 |
Mature | 0 | 138 | 0 | 1814 | 1952 |
Total | 3 986 | 38 563 | 419 | 24 372 | 67 340 |
*Differences between the total number of trees reported by each type of camera in 2016. There was a difference of at least 30,000 trees between the RGB and NIR in Total results. NIR results include only live trees, while RGB include live and dead-standing trees.
Table 4
Tree densities in each patch from the 2019 survey.
Patch | Individuals per hectare |
| BZ | NBZ |
North | 344 | - |
Central | 526 | 67 |
Central-South | 40 | - |
South | 165 | 24 |
Table 4 shows that central and north patches have a higher density than the south and central patch. Also, considering the number of individuals in each patch, we note that the central patch has the largest number of individuals, for both years and type of camera.
Height distribution was calculated, distinguishing between mature and youth. The results are in Table 5, showing the differences between BZ and NBZ.
Table 5
Height statistics for each patch, according to the camera used and survey year. Reporting height average, maximum and standard deviation (SD). All measures in m.
| | Patches |
| | North | Central | Central-South | South |
| | | NBZ | BZ | | NBZ | BZ |
2016 NIR | Average | 2.7 | 12.58 | 2.98 | 4.67 | 7.37 | 3.16 |
Max | 15.76 | 21.96 | 14.93 | 9.36 | 23.47 | 21.15 |
SD | 0.98 | 5.11 | 0.9 | 1.64 | 4.9 | 2.19 |
2016 RGB | Average | 2.87 | 12.14 | 3.32 | 4.4 | 7.68 | 3.66 |
Max | 15.76 | 22.18 | 20.1 | 9.9 | 24 | 23.69 |
SD | 0.98 | 5.51 | 1.5 | 1.76 | 4.8 | 2.32 |
2019 RGB2019 | Average | 2.96 | 11.85 | 3.34 | 4.95 | 8.4 | 3.99 |
Max | 17.2 | 21.53 | 19.17 | 10.5 | 23 | 22.21 |
SD | 1.09 | 5.6 | 1.14 | 1.76 | 4.6 | 2.34 |
According to Table 5, the average height in the NBZs is greater than the average height in the BZs. The maximum value of trees in BZ of Central Patch is reported with the RGB camera, it includes standing dead trees and branches, while the maximum value in the same zone with the NIR camera represents young living trees (NDVI > 0.3).
In the BZ of the south patch, the maximum value reported, is from a live tree, this fact was corroborated in the field and the images. We assumed that this tree, while in the burned polygon, was not affected by fire perhaps because of the larger open spaces in this patch reported by Ramirez Serrato (2016). The average height values of all patches and zones are mostly influenced by the large number of young trees between 2 and 3 m. This fact can be observed in Fig. 7, showing the tree height histograms for each patch.
Figure 7 shows the height distribution of the forest patches for each survey year and camera used. According to the number of individuals, the central patch has the largest number, and the central-south patch has the lowest number of trees. It is possible to observe that the trees with heights between 2 and 3 m are the most representative in all the patches, including burned and non-burned zones (NBZ). However, in NBZ, the tree population is between 11 and 24 m height and has a higher frequency than in BZs. We can observe a large difference in the number of individuals reported by each type of camera, the 2016 RGB data has the highest frequencies in each category compared with the others.
Result examples obtained by applying the RUSLE model are shown in Fig. 8, where it is possible to observe the high-resolution analysis obtained using photogrammetry. The modeled erosion rates vary from 0 to 2 Kg m− 2 y− 1 (Fig. 8 − 2) for areas with trees of different ages, from 4 to 5 Kg m− 2 y− 1 (Fig. 8 − 3) for areas with grass or shrubby vegetation, and from 15 Kg to 33 Kg m− 2 y − 1 (Fig. 8 − 1) for bare soils and steep slopes areas.
The horizontal variation of a forest can be assessed by vegetation index (VI) using multispectral (high and medium resolution) imagery. In this research, we observed the recovery and expansion of the cypress forest on Guadalupe Island after the 2008 fire, by comparing the size of the polygons reported by Ramírez-Serrato (2014) and the polygons traced in this work based on 2019 Planetscope satellite images. The cypress forest showed a total increase of 134 hectares in its area, and new areas of expansion were identified like the south-central patch described in this work. This patch is a clear natural recovery zone, essential for the growth of the forest.
The estimate of the number of trees using photogrammetry and ForestTools correlated reasonably with the results obtained in the field (R = 0.65). However, it underestimated the number of young trees and overestimated the number of mature trees. Gallardo et al. (2020) used the same methodology (ForestTools algorithm) to count trees in a crop, and reported a higher correlation between field and ForestTools data. Ninety-five percent of the trees measured in the field were detected by the ForestTools algorithm, proving the efficiency of photogrammetry to identify and count trees. The difference in correlation results is because Gallardo (2020) analyzed a crop where the trees are regularly distributed and are the same age; unlike this study, which was performed in a forest affected by several factors and trees at different growth stages. Field observations allowed us to verify the high density of young trees in some areas of the cypress forest, where up to 30 individuals were registered in a square meter, making it difficult to measure the height of individuals in the field. The comparison of this work with Gallardo’s (2020) results allowed us to determine the reach of photogrammetry to model specific ecosystems and stimulated us to seek ways of improving the technique.
On the other side, this study’s correlation value of height obtained using photogrammetry was R = 0.92. Therefore, the CHM model derived from photogrammetry proved to be a good tool to assess the recovery and evolution of the canopy structure and horizontal distribution of the cypress forest. It was able to generate a plausible map of tree crown geometry for the different forest patches combined with tree height histograms.
The difference between both surveys exposes how photogrammetry results can change depending on the study area, and camera type. In the present work, the correlation was not too strong, but it was positive. Thus, current results are a reliable 3D description of the Guadalupe Island cypress forest, the first and only of this type with high resolution.
Compared with vegetation studies using satellite images, photogrammetry using UAV has some advantages. The results of canopy cover and the crown areas obtained in this work are compared with those of Rodríguez-Malagón (2007), who estimated the number of individuals using high-resolution satellite multispectral images, suggesting an average canopy coverage area of 98.86 m2 and 132.38 m2 for mature trees. By photogrammetry, results showed coverage values of 2 to 104 m2 in trees from 13 to 24 m in height, and an average of 25 m2 of cup coverage. This demonstrates variability and precision in photogrammetric products. Figure 9 shows the crown polygons generated by ForestTools using the photogrammetry-derived CHM, which delineates the canopy projection on the ground, with a more accurate approximation to the canopy cover area than the use of VI from multi-spectral images, where the vertical component is not included in the results. The CHM obtained by UAV photogrammetry considers the height of any type of vegetation, allowing the discrimination of the evergreen herbaceous vegetation as Calystegia macrostegia ssp. macrostegia. The full estimate of the number of individuals by Rodríguez-Malagón (2007) is not comparable with present results, because the first were obtained before the fire and only mature trees were considered. The resolution of satellite images has lately been improved. Multispectral images like Sentinel or even satellite photogrammetry are tools that nowadays enhance the temporal and spatial resolution; however, their high cost may hinder research.
Point clouds generated by photogrammetry, do not describe the vertical structure of the forest, because they register only one return from the surface, the highest. In contrast, analysis with laser scanners provides multiple returns from one pulse. Yépez-Rincón et al. (2021) used point clouds from a terrestrial laser scanner to measure tree species in specific areas of Guadalupe Island. In that study, they analyzed the cypress and pine trees and reconstructed the crown geometry to differentiate trees. Their results showed a high correlation between field measurements and LASER data: diameter at breast height R2 = 0.94, crow diameter R2 = 0.97 and tree heights R2 = 0.97.
LiDAR aerial surveys and terrestrial laser scans are remote sensing techniques with capacities more suitable for a precise 3D reconstruction of the vertical structure of the forest, enabling the registration of multiple returns per pulse and the precise location of reflectors, although resources to acquire them are not always available. Terrestrial laser scans are limited to spatial coverage due to operational limitations. Satellite photogrammetry from high-resolution stereo pairs or tri-stereos is an option for large areas, although the spatial resolution of reconstruction may not be appropriate for the construction of the CHM.
Comparing present results from NIR and RGB images of 2016, we found a large difference between the number of individuals counted. The differences correspond to standing dead trees located in BZ in each patch. Results showed the advantage of using an NIR camera to apply an NDVI filter on the resulting point cloud. With this filter, the live and dead trees within a zone may be detected.
The difference between the number counted in 2016 and 2019 using RGB cameras (at least 30,000 individuals) may be due to the removal of dry material (dead-standing trees and their fallen branches), either by natural factors or human activity. Natural factors involve the removal of dry matter by wind and humidity; for example, in the northern patch, where there is no human activity, we recorded the fall of dry trees of up to 17 m. On the other hand, human activity is evident in the south patch, where management practices are applied to prevent fires by removing fuel material (knocking down, cutting, and removing dead trees). The difference between dry and live vegetation in 2016 is in Table 1, where the number of individuals counted with NIR image data (live trees) is similar to the count in 2019.
Although the loss of dead trees or large branches may explain the significant difference between 2016 and 2019, the loss of green vegetation must also be considered, especially for mature individuals. Due to their large size and weight, in addition to their shallow roots, mature individuals can be knocked down by strong winds, which are common in winter.
The number and density of trees vary between patches; the central patch accommodated the highest number of individuals and density; the north patch showed the second largest density despite its extent and substrate (mainly rock); the south patch density was at least three times lower than the central patch despite having a greater extension, and the central-south patch displayed the lowest density and number of individuals. The differences in distribution and growth of trees may be due to the environmental conditions at each patch, determined by topography and exposure to ocean moisture. The north patch is the most exposed to humidity, and thus has enough water for the trees; however, soils and exposure to winds are limiting conditions for tree growth. The central patch receives and retains the highest humidity due to its topography. Although it is exposed to winds, this patch can sustain a continuous distribution of trees. The central-south and south patches have the largest space to accommodate new trees; however, topography makes them the most exposed to soil erosion (Ramos-Franco, 2007), generating many runoff areas due to their proximity to the inactive crater rim. This also allows the growth of the tallest individuals due to the open spaces and was the reason why a great number of trees survived the fire in 2008 (Ramírez, 2014), even with their low density.
The height distribution in each patch, in burned or non-burned zones, shows the forest strata and evidences its recovery. The highest stratum is represented by mature trees, with heights between 6 and 23 m (from field observations). The most representative height frequency in this stratum is between 13 and 17 m. Table 5 shows that tree height is more variable in the areas not affected by fire (NBZs), independently of the camera used and year of survey. The NBZs contain mostly mature trees. There are two groups of young trees, those that grew immediately after the fire (2008–2009 cohort), and those that grew in the following years. The first group had heights between 4 and 12 m in 2019; their growth is determined by the abiotic conditions (topography, humidity, or wind), and the competition with other trees. The second group is represented mainly by trees between 2 and 3 m. These are trees of different generations, probably due to the seed bank that did not germinate immediately after the fire as well as the seed production of the new trees. In this work, we did not estimate the age of each stratum, but by the presence of reproductive structures in trees of 2 m height, we can assume they are a new generation. The existence of the second group of young trees is evidence of the continuous natural regeneration of the forest.
According to RUSLE analysis results, the most susceptible areas to water-borne erosion are those with the steepest slopes or bare surfaces, while the areas covered with trees, shrubs, or grass are more resistant to soil loss. The vertical projection and distribution of vegetation is an important feature included in the analysis that determines the effect of water on soil erosion. Canopy cover reduces the kinetic energy of falling raindrops, and the surface cover (shrubs and grass) reduces the transport capacity of water, allowing the accumulation of humidity and preventing the formation of gullies (Renard et al., 1997). Integrating vegetation height into the RUSLE model to assess water-borne erosion is an advantage over models that only consider the horizontal distribution of plants, because it considers the heterogeneity of vegetation structure and its different effects on the water dynamics and soil retention.
Smith (2007) and Puente (2011) developed studies in areas with similar environmental conditions to Guadalupe Island; their methodologies were used to incorporate the CHM in the C factor of the RUSLE model. These studies are based on multispectral images and vegetation index to evaluate the C factor; the resolution and variability of their results depend on satellite image resolution (30–60 m) and the estimated parameters for each vegetal species identified in the area, omitting the variability of the vegetation structure. Compared with those methodologies, photogrammetric analysis (using unmanned aerial vehicles) presents an advantage over multispectral analysis, because it incorporates surface variability in the CHM. Besides, the resolution of products can be adjusted depending on the camera and altitude of flights.