Limits...
Predictive Models of Primary Tropical Forest Structure from Geomorphometric Variables Based on SRTM in the Tapajós Region, Brazilian Amazon.

Bispo Pda C, Dos Santos JR, Valeriano Mde M, Graça PM, Balzter H, França H, Bispo Pda C - PLoS ONE (2016)

Bottom Line: Validation of the models with nine independent sample plots revealed a Root Mean Square Error (RMSE) of 3.73 m2/ha (20%) for BA, 1.70 m (12%) for H, and 1.78% (21%) for CO.The coefficient of determination between observed and predicted values were r2 = 0.32 for CO, r2 = 0.26 for H and r2 = 0.52 for BA.The models obtained were able to adequately estimate BA and CO.

View Article: PubMed Central - PubMed

Affiliation: Ciência e Tecnologia Ambiental, Universidade Federal do ABC (UFABC), Santo André, São Paulo, Brazil.

ABSTRACT
Surveying primary tropical forest over large regions is challenging. Indirect methods of relating terrain information or other external spatial datasets to forest biophysical parameters can provide forest structural maps at large scales but the inherent uncertainties need to be evaluated fully. The goal of the present study was to evaluate relief characteristics, measured through geomorphometric variables, as predictors of forest structural characteristics such as average tree basal area (BA) and height (H) and average percentage canopy openness (CO). Our hypothesis is that geomorphometric variables are good predictors of the structure of primary tropical forest, even in areas, with low altitude variation. The study was performed at the Tapajós National Forest, located in the Western State of Pará, Brazil. Forty-three plots were sampled. Predictive models for BA, H and CO were parameterized based on geomorphometric variables using multiple linear regression. Validation of the models with nine independent sample plots revealed a Root Mean Square Error (RMSE) of 3.73 m2/ha (20%) for BA, 1.70 m (12%) for H, and 1.78% (21%) for CO. The coefficient of determination between observed and predicted values were r2 = 0.32 for CO, r2 = 0.26 for H and r2 = 0.52 for BA. The models obtained were able to adequately estimate BA and CO. In summary, it can be concluded that relief variables are good predictors of vegetation structure and enable the creation of forest structure maps in primary tropical rainforest with an acceptable uncertainty.

Show MeSH

Related in: MedlinePlus

Illustration of (a) a hemispherical photograph of forest cover and (b) a binary image resulting from GLA analysis.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4835096&req=5

pone.0152009.g002: Illustration of (a) a hemispherical photograph of forest cover and (b) a binary image resulting from GLA analysis.

Mentions: DBH was measured using a measuring tape. CO was estimated from hemispherical images obtained using a digital camera with 10.2 Megapixel resolution (Nikon D60) coupled to a wide-angle lens (fisheye) (Soligor 0.25x52 mm). The images were captured 1.8 m from the ground, with a 90 degree angle relative to the ground, under the canopy and avoiding direct sunlight, i.e., under conditions of uniform sky before sunrise (from 6:30 to 8:30 am) or after sunset (3.30 to 5:00 pm). These images were taken along central axis transects (25 x 100 m), every 20 m (following the same method used by Galvão et al. [41]), totaling in 5 images per transect. Images were analysed using Gap Light Analyser software (GLA) [42], and the fraction of canopy openness (Gap fraction) was calculated. Gap fraction is defined as the percentage of canopy gaps, i.e., the probability of sunlight passing through the forest canopy without meeting leaves or another plant part [43]. The GLA package uses an image classifier based on pixel thresholds to calculate the fraction of canopy openness. Therefore, a threshold needs to be defined for each image, which can vary with sun exposure. The images of the canopy and sky were adequately converted and separated into black and white pixels. The resulting images in binary format were used for determining the fraction of canopy openness (Fig 2).


Predictive Models of Primary Tropical Forest Structure from Geomorphometric Variables Based on SRTM in the Tapajós Region, Brazilian Amazon.

Bispo Pda C, Dos Santos JR, Valeriano Mde M, Graça PM, Balzter H, França H, Bispo Pda C - PLoS ONE (2016)

Illustration of (a) a hemispherical photograph of forest cover and (b) a binary image resulting from GLA analysis.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4835096&req=5

pone.0152009.g002: Illustration of (a) a hemispherical photograph of forest cover and (b) a binary image resulting from GLA analysis.
Mentions: DBH was measured using a measuring tape. CO was estimated from hemispherical images obtained using a digital camera with 10.2 Megapixel resolution (Nikon D60) coupled to a wide-angle lens (fisheye) (Soligor 0.25x52 mm). The images were captured 1.8 m from the ground, with a 90 degree angle relative to the ground, under the canopy and avoiding direct sunlight, i.e., under conditions of uniform sky before sunrise (from 6:30 to 8:30 am) or after sunset (3.30 to 5:00 pm). These images were taken along central axis transects (25 x 100 m), every 20 m (following the same method used by Galvão et al. [41]), totaling in 5 images per transect. Images were analysed using Gap Light Analyser software (GLA) [42], and the fraction of canopy openness (Gap fraction) was calculated. Gap fraction is defined as the percentage of canopy gaps, i.e., the probability of sunlight passing through the forest canopy without meeting leaves or another plant part [43]. The GLA package uses an image classifier based on pixel thresholds to calculate the fraction of canopy openness. Therefore, a threshold needs to be defined for each image, which can vary with sun exposure. The images of the canopy and sky were adequately converted and separated into black and white pixels. The resulting images in binary format were used for determining the fraction of canopy openness (Fig 2).

Bottom Line: Validation of the models with nine independent sample plots revealed a Root Mean Square Error (RMSE) of 3.73 m2/ha (20%) for BA, 1.70 m (12%) for H, and 1.78% (21%) for CO.The coefficient of determination between observed and predicted values were r2 = 0.32 for CO, r2 = 0.26 for H and r2 = 0.52 for BA.The models obtained were able to adequately estimate BA and CO.

View Article: PubMed Central - PubMed

Affiliation: Ciência e Tecnologia Ambiental, Universidade Federal do ABC (UFABC), Santo André, São Paulo, Brazil.

ABSTRACT
Surveying primary tropical forest over large regions is challenging. Indirect methods of relating terrain information or other external spatial datasets to forest biophysical parameters can provide forest structural maps at large scales but the inherent uncertainties need to be evaluated fully. The goal of the present study was to evaluate relief characteristics, measured through geomorphometric variables, as predictors of forest structural characteristics such as average tree basal area (BA) and height (H) and average percentage canopy openness (CO). Our hypothesis is that geomorphometric variables are good predictors of the structure of primary tropical forest, even in areas, with low altitude variation. The study was performed at the Tapajós National Forest, located in the Western State of Pará, Brazil. Forty-three plots were sampled. Predictive models for BA, H and CO were parameterized based on geomorphometric variables using multiple linear regression. Validation of the models with nine independent sample plots revealed a Root Mean Square Error (RMSE) of 3.73 m2/ha (20%) for BA, 1.70 m (12%) for H, and 1.78% (21%) for CO. The coefficient of determination between observed and predicted values were r2 = 0.32 for CO, r2 = 0.26 for H and r2 = 0.52 for BA. The models obtained were able to adequately estimate BA and CO. In summary, it can be concluded that relief variables are good predictors of vegetation structure and enable the creation of forest structure maps in primary tropical rainforest with an acceptable uncertainty.

Show MeSH
Related in: MedlinePlus