Mapping forest structural heterogeneity of tropical montane forest remnants from airborne laser scanning and Landsat time series

Tutkimustuotos: ArtikkelijulkaisuArtikkeliTieteellinenvertaisarvioitu

Kuvaus

Tropical montane forests are important reservoirs of carbon and biodiversity and have a central role in the hydrological cycle. They are, however, very fragmented and degraded, leaving isolated remnants across the landscape. These montane forest remnants have considerable differences in forest structure, depending on factors such as tree species composition and degree of forest degradation. Our objectives were (1) to analyse the reliability of airborne laser scanning (ALS) in modelling forest structural heterogeneity, as described by the Gini coefficient (GC) of tree size inequality; (2) to determine whether models are improved by including tree species-sensitive spectral-temporal metrics from the Landsat time series (LTS); and (3) to evaluate differences between three forest remnants and different forest types using the resulting maps of predicted GC. The study area was situated in Taita Hills, Kenya, where indigenous montane forests have been partly replaced by single-species plantations. The data included field measurements from 85 sample plots and two ALS data sets with different pulse densities (9.6 and 3.1 pulses m(-2)). GC was modeled using beta regression. We found that GC was predicted more accurately by the ALS data set with a higher point density (a cross-validated relative root mean squared error (rRMSE(CV)) 13.9%) compared to ALS data set with lower point density (rRMSE(CV) 15.1%). Furthermore, important synergies exist between ALS and LTS metrics. When combining ALS and LTS metrics, rRMSE(CV) was improved to 12.5% and 13.0%, respectively. Therefore, if the LTS metrics are included in models, ALS data with lower pulse density are sufficient to yield similar accuracy to more expensive, higher pulse density data acquired from the lower altitude. In Ngangao and Yale, forest canopy has multiple layers of variable tree sizes, whereas elfin forests in Vuria are of more equal tree size, and the GC value ranges of the indigenous forests are 0.42-0.71, 0.20-0.74, and 0.17-0.76, respectively. The single-species plantations of cypress and pine showed lower values of GC than indigenous forests located in the same remnants in Yale, whereas Eucalyptus plantations showed GC values more similar to the indigenous forests. These results show the usefulness of GC maps for identifying and separating forest types as well as for assessing their distinctive ecologies.

Alkuperäiskielienglanti
Artikkeli105739
LehtiEcological Indicators
Vuosikerta108
Sivumäärä16
ISSN1470-160X
DOI - pysyväislinkit
TilaJulkaistu - tammikuuta 2020
OKM-julkaisutyyppiA1 Alkuperäisartikkeli tieteellisessä aikakauslehdessä, vertaisarvioitu

Tieteenalat

  • 4112 Metsätiede

Lainaa tätä

@article{59ca7c9abb564c67a7e97f633feab6f7,
title = "Mapping forest structural heterogeneity of tropical montane forest remnants from airborne laser scanning and Landsat time series",
abstract = "Tropical montane forests are important reservoirs of carbon and biodiversity and have a central role in the hydrological cycle. They are, however, very fragmented and degraded, leaving isolated remnants across the landscape. These montane forest remnants have considerable differences in forest structure, depending on factors such as tree species composition and degree of forest degradation. Our objectives were (1) to analyse the reliability of airborne laser scanning (ALS) in modelling forest structural heterogeneity, as described by the Gini coefficient (GC) of tree size inequality; (2) to determine whether models are improved by including tree species-sensitive spectral-temporal metrics from the Landsat time series (LTS); and (3) to evaluate differences between three forest remnants and different forest types using the resulting maps of predicted GC. The study area was situated in Taita Hills, Kenya, where indigenous montane forests have been partly replaced by single-species plantations. The data included field measurements from 85 sample plots and two ALS data sets with different pulse densities (9.6 and 3.1 pulses m(-2)). GC was modeled using beta regression. We found that GC was predicted more accurately by the ALS data set with a higher point density (a cross-validated relative root mean squared error (rRMSE(CV)) 13.9{\%}) compared to ALS data set with lower point density (rRMSE(CV) 15.1{\%}). Furthermore, important synergies exist between ALS and LTS metrics. When combining ALS and LTS metrics, rRMSE(CV) was improved to 12.5{\%} and 13.0{\%}, respectively. Therefore, if the LTS metrics are included in models, ALS data with lower pulse density are sufficient to yield similar accuracy to more expensive, higher pulse density data acquired from the lower altitude. In Ngangao and Yale, forest canopy has multiple layers of variable tree sizes, whereas elfin forests in Vuria are of more equal tree size, and the GC value ranges of the indigenous forests are 0.42-0.71, 0.20-0.74, and 0.17-0.76, respectively. The single-species plantations of cypress and pine showed lower values of GC than indigenous forests located in the same remnants in Yale, whereas Eucalyptus plantations showed GC values more similar to the indigenous forests. These results show the usefulness of GC maps for identifying and separating forest types as well as for assessing their distinctive ecologies.",
keywords = "4112 Forestry, Forest structure, Gini coefficient, Spectral-temporal metrics, LiDAR, Africa, EASTERN ARC MOUNTAINS, TOPOGRAPHIC NORMALIZATION, SPECIES-DIVERSITY, GINI COEFFICIENT, BOREAL FORESTS, COVER CHANGE, LIDAR, MANAGEMENT, INDEX, BIODIVERSITY",
author = "Hari Adhikari and Ruben Valbuena and Petri Pellikka and Janne Heiskanen",
year = "2020",
month = "1",
doi = "10.1016/j.ecolind.2019.105739",
language = "English",
volume = "108",
journal = "Ecological Indicators",
issn = "1470-160X",
publisher = "Elsevier Scientific Publ. Co",

}

Mapping forest structural heterogeneity of tropical montane forest remnants from airborne laser scanning and Landsat time series. / Adhikari, Hari; Valbuena, Ruben; Pellikka, Petri; Heiskanen, Janne.

julkaisussa: Ecological Indicators, Vuosikerta 108, 105739, 01.2020.

Tutkimustuotos: ArtikkelijulkaisuArtikkeliTieteellinenvertaisarvioitu

TY - JOUR

T1 - Mapping forest structural heterogeneity of tropical montane forest remnants from airborne laser scanning and Landsat time series

AU - Adhikari, Hari

AU - Valbuena, Ruben

AU - Pellikka, Petri

AU - Heiskanen, Janne

PY - 2020/1

Y1 - 2020/1

N2 - Tropical montane forests are important reservoirs of carbon and biodiversity and have a central role in the hydrological cycle. They are, however, very fragmented and degraded, leaving isolated remnants across the landscape. These montane forest remnants have considerable differences in forest structure, depending on factors such as tree species composition and degree of forest degradation. Our objectives were (1) to analyse the reliability of airborne laser scanning (ALS) in modelling forest structural heterogeneity, as described by the Gini coefficient (GC) of tree size inequality; (2) to determine whether models are improved by including tree species-sensitive spectral-temporal metrics from the Landsat time series (LTS); and (3) to evaluate differences between three forest remnants and different forest types using the resulting maps of predicted GC. The study area was situated in Taita Hills, Kenya, where indigenous montane forests have been partly replaced by single-species plantations. The data included field measurements from 85 sample plots and two ALS data sets with different pulse densities (9.6 and 3.1 pulses m(-2)). GC was modeled using beta regression. We found that GC was predicted more accurately by the ALS data set with a higher point density (a cross-validated relative root mean squared error (rRMSE(CV)) 13.9%) compared to ALS data set with lower point density (rRMSE(CV) 15.1%). Furthermore, important synergies exist between ALS and LTS metrics. When combining ALS and LTS metrics, rRMSE(CV) was improved to 12.5% and 13.0%, respectively. Therefore, if the LTS metrics are included in models, ALS data with lower pulse density are sufficient to yield similar accuracy to more expensive, higher pulse density data acquired from the lower altitude. In Ngangao and Yale, forest canopy has multiple layers of variable tree sizes, whereas elfin forests in Vuria are of more equal tree size, and the GC value ranges of the indigenous forests are 0.42-0.71, 0.20-0.74, and 0.17-0.76, respectively. The single-species plantations of cypress and pine showed lower values of GC than indigenous forests located in the same remnants in Yale, whereas Eucalyptus plantations showed GC values more similar to the indigenous forests. These results show the usefulness of GC maps for identifying and separating forest types as well as for assessing their distinctive ecologies.

AB - Tropical montane forests are important reservoirs of carbon and biodiversity and have a central role in the hydrological cycle. They are, however, very fragmented and degraded, leaving isolated remnants across the landscape. These montane forest remnants have considerable differences in forest structure, depending on factors such as tree species composition and degree of forest degradation. Our objectives were (1) to analyse the reliability of airborne laser scanning (ALS) in modelling forest structural heterogeneity, as described by the Gini coefficient (GC) of tree size inequality; (2) to determine whether models are improved by including tree species-sensitive spectral-temporal metrics from the Landsat time series (LTS); and (3) to evaluate differences between three forest remnants and different forest types using the resulting maps of predicted GC. The study area was situated in Taita Hills, Kenya, where indigenous montane forests have been partly replaced by single-species plantations. The data included field measurements from 85 sample plots and two ALS data sets with different pulse densities (9.6 and 3.1 pulses m(-2)). GC was modeled using beta regression. We found that GC was predicted more accurately by the ALS data set with a higher point density (a cross-validated relative root mean squared error (rRMSE(CV)) 13.9%) compared to ALS data set with lower point density (rRMSE(CV) 15.1%). Furthermore, important synergies exist between ALS and LTS metrics. When combining ALS and LTS metrics, rRMSE(CV) was improved to 12.5% and 13.0%, respectively. Therefore, if the LTS metrics are included in models, ALS data with lower pulse density are sufficient to yield similar accuracy to more expensive, higher pulse density data acquired from the lower altitude. In Ngangao and Yale, forest canopy has multiple layers of variable tree sizes, whereas elfin forests in Vuria are of more equal tree size, and the GC value ranges of the indigenous forests are 0.42-0.71, 0.20-0.74, and 0.17-0.76, respectively. The single-species plantations of cypress and pine showed lower values of GC than indigenous forests located in the same remnants in Yale, whereas Eucalyptus plantations showed GC values more similar to the indigenous forests. These results show the usefulness of GC maps for identifying and separating forest types as well as for assessing their distinctive ecologies.

KW - 4112 Forestry

KW - Forest structure

KW - Gini coefficient

KW - Spectral-temporal metrics

KW - LiDAR

KW - Africa

KW - EASTERN ARC MOUNTAINS

KW - TOPOGRAPHIC NORMALIZATION

KW - SPECIES-DIVERSITY

KW - GINI COEFFICIENT

KW - BOREAL FORESTS

KW - COVER CHANGE

KW - LIDAR

KW - MANAGEMENT

KW - INDEX

KW - BIODIVERSITY

U2 - 10.1016/j.ecolind.2019.105739

DO - 10.1016/j.ecolind.2019.105739

M3 - Article

VL - 108

JO - Ecological Indicators

JF - Ecological Indicators

SN - 1470-160X

M1 - 105739

ER -