Burned area detection based on Landsat time series in savannas of southern Burkina Faso

Forskningsoutput: TidskriftsbidragArtikelVetenskapligPeer review

Sammanfattning

West African savannas are subject to regular fires, which have impacts on vegetation structure, biodiversity and carbon balance. An efficient and accurate mapping of burned area associated with seasonal fires can greatly benefit decision making in land management. Since coarse resolution burned area products cannot meet the accuracy needed for fire management and climate modelling at local scales, the medium resolution Landsat data is a promising alternative for local scale studies. In this study, we developed an algorithm for continuous monitoring of annual burned areas using Landsat time series. The algorithm is based on burned pixel detection using harmonic model fitting with Landsat time series and breakpoint identification in the time series data. This approach was tested in a savanna area in southern Burkina Faso using 281 images acquired between October 2000 and April 2016. An overall accuracy of 79.2% was obtained with balanced omission and commission errors. This represents a significant improvement in comparison with MODIS burned area product (67.6%), which had more omission errors than commission errors, indicating underestimation of the total burned area. By observing the spatial distribution of burned areas, we found that the Landsat based method misclassified cropland and cloud shadows as burned areas due to the similar spectral response, and MODIS burned area product omitted small and fragmented burned areas. The proposed algorithm is flexible and robust against decreased data availability caused by clouds and Landsat 7 missing lines, therefore having a high potential for being applied in other landscapes in future studies.
Originalspråkengelska
TidskriftInternational Journal of Applied Earth Observation and Geoinformation
Volym64
Sidor (från-till)210-220
Antal sidor11
ISSN0303-2434
DOI
StatusPublicerad - feb 2018
MoE-publikationstypA1 Tidskriftsartikel-refererad

Vetenskapsgrenar

  • 1172 Miljövetenskap
  • 114 Fysik
  • 4112 Skogsvetenskap

Citera det här

@article{4c6226ee541a419c8b381f403ffd2369,
title = "Burned area detection based on Landsat time series in savannas of southern Burkina Faso",
abstract = "West African savannas are subject to regular fires, which have impacts on vegetation structure, biodiversity and carbon balance. An efficient and accurate mapping of burned area associated with seasonal fires can greatly benefit decision making in land management. Since coarse resolution burned area products cannot meet the accuracy needed for fire management and climate modelling at local scales, the medium resolution Landsat data is a promising alternative for local scale studies. In this study, we developed an algorithm for continuous monitoring of annual burned areas using Landsat time series. The algorithm is based on burned pixel detection using harmonic model fitting with Landsat time series and breakpoint identification in the time series data. This approach was tested in a savanna area in southern Burkina Faso using 281 images acquired between October 2000 and April 2016. An overall accuracy of 79.2{\%} was obtained with balanced omission and commission errors. This represents a significant improvement in comparison with MODIS burned area product (67.6{\%}), which had more omission errors than commission errors, indicating underestimation of the total burned area. By observing the spatial distribution of burned areas, we found that the Landsat based method misclassified cropland and cloud shadows as burned areas due to the similar spectral response, and MODIS burned area product omitted small and fragmented burned areas. The proposed algorithm is flexible and robust against decreased data availability caused by clouds and Landsat 7 missing lines, therefore having a high potential for being applied in other landscapes in future studies.",
keywords = "1172 Environmental sciences, 114 Physical sciences, 4112 Forestry, Burned area, Landsat time series, Harmonic model, Breakpoint identification, MODIS, WEST-AFRICA, SPECTRAL INDEXES, MODIS IMAGERY, CLOUD SHADOW, FIRE HISTORY, COVER CHANGE, FOREST, ALGORITHM, PRODUCTS, VEGETATION",
author = "Jinxiu Liu and Janne Heiskanen and Maeda, {Eduardo Eiji} and Pellikka, {Petri K. E.}",
year = "2018",
month = "2",
doi = "10.1016/j.jag.2017.09.011",
language = "English",
volume = "64",
pages = "210--220",
journal = "International Journal of Applied Earth Observation and Geoinformation",
issn = "1569-8432",
publisher = "Elsevier Scientific Publ. Co",

}

TY - JOUR

T1 - Burned area detection based on Landsat time series in savannas of southern Burkina Faso

AU - Liu, Jinxiu

AU - Heiskanen, Janne

AU - Maeda, Eduardo Eiji

AU - Pellikka, Petri K. E.

PY - 2018/2

Y1 - 2018/2

N2 - West African savannas are subject to regular fires, which have impacts on vegetation structure, biodiversity and carbon balance. An efficient and accurate mapping of burned area associated with seasonal fires can greatly benefit decision making in land management. Since coarse resolution burned area products cannot meet the accuracy needed for fire management and climate modelling at local scales, the medium resolution Landsat data is a promising alternative for local scale studies. In this study, we developed an algorithm for continuous monitoring of annual burned areas using Landsat time series. The algorithm is based on burned pixel detection using harmonic model fitting with Landsat time series and breakpoint identification in the time series data. This approach was tested in a savanna area in southern Burkina Faso using 281 images acquired between October 2000 and April 2016. An overall accuracy of 79.2% was obtained with balanced omission and commission errors. This represents a significant improvement in comparison with MODIS burned area product (67.6%), which had more omission errors than commission errors, indicating underestimation of the total burned area. By observing the spatial distribution of burned areas, we found that the Landsat based method misclassified cropland and cloud shadows as burned areas due to the similar spectral response, and MODIS burned area product omitted small and fragmented burned areas. The proposed algorithm is flexible and robust against decreased data availability caused by clouds and Landsat 7 missing lines, therefore having a high potential for being applied in other landscapes in future studies.

AB - West African savannas are subject to regular fires, which have impacts on vegetation structure, biodiversity and carbon balance. An efficient and accurate mapping of burned area associated with seasonal fires can greatly benefit decision making in land management. Since coarse resolution burned area products cannot meet the accuracy needed for fire management and climate modelling at local scales, the medium resolution Landsat data is a promising alternative for local scale studies. In this study, we developed an algorithm for continuous monitoring of annual burned areas using Landsat time series. The algorithm is based on burned pixel detection using harmonic model fitting with Landsat time series and breakpoint identification in the time series data. This approach was tested in a savanna area in southern Burkina Faso using 281 images acquired between October 2000 and April 2016. An overall accuracy of 79.2% was obtained with balanced omission and commission errors. This represents a significant improvement in comparison with MODIS burned area product (67.6%), which had more omission errors than commission errors, indicating underestimation of the total burned area. By observing the spatial distribution of burned areas, we found that the Landsat based method misclassified cropland and cloud shadows as burned areas due to the similar spectral response, and MODIS burned area product omitted small and fragmented burned areas. The proposed algorithm is flexible and robust against decreased data availability caused by clouds and Landsat 7 missing lines, therefore having a high potential for being applied in other landscapes in future studies.

KW - 1172 Environmental sciences

KW - 114 Physical sciences

KW - 4112 Forestry

KW - Burned area

KW - Landsat time series

KW - Harmonic model

KW - Breakpoint identification

KW - MODIS

KW - WEST-AFRICA

KW - SPECTRAL INDEXES

KW - MODIS IMAGERY

KW - CLOUD SHADOW

KW - FIRE HISTORY

KW - COVER CHANGE

KW - FOREST

KW - ALGORITHM

KW - PRODUCTS

KW - VEGETATION

UR - http://www.sciencedirect.com/science/article/pii/S0303243417302003

U2 - 10.1016/j.jag.2017.09.011

DO - 10.1016/j.jag.2017.09.011

M3 - Article

VL - 64

SP - 210

EP - 220

JO - International Journal of Applied Earth Observation and Geoinformation

JF - International Journal of Applied Earth Observation and Geoinformation

SN - 1569-8432

ER -