®®®® SIIA Público

Título del libro: 38th Asian Conference On Remote Sensing - Space Applications: Touching Human Lives, Acrs 2017
Título del capítulo: Detection of forest disturbances by time series analysis of NDVI from MODIS sensor for Michoacan State, Mexico (2000 - 2014)

Autores UNAM:
ALEXANDER QUEVEDO CHACON; YAN GAO;
Autores externos:

Idioma:
Inglés
Año de publicación:
2017
Palabras clave:

Deforestation; Harmonic functions; Radiometers; Remote sensing; Satellite imagery; Space applications; Space optics; Bfast spatial; Data preprocessing; Forest cover change; Forest disturbances; High spatial resolution; Land-cover change; LANDSAT; Savitzky-Golay filter; Time series analysis


Resumen:

We present forest disturbance detection using time-series MODIS NDVI (2000-2014) data for Michoacán state, Mexico. First, we carried out a data preprocessing by applying a quality layer (QA), a spline interpretation and a Savitzky-Golay filter. Then we applied the method Bfast monitor (Verbesselt et al. 2012) for the detection of forest disturbance. Bfast monitor decomposes the time-series data into harmonic function, trend and remainder (noise). It works by first constructing a linear function using data from a presumably stable period. It applies a moving sum to detect a breakpoint where a change occurred, and calculates a change magnitude by the difference between the observed and the predicted NDVI values at the breakpoint. In our case, we manually defined the data from 2000 - 2007 as reference period, and data of 2007 - 2014 as change detection period, and forest disturbances were detected spatially and temporally by combining the maps of break points and change magnitude. We define a forest gain as a positive change magnitude larger the threshold 0.05 and a forest loss as a negative change magnitude smaller than -0.05. We compared the results with the changes derived from high spatial resolution land-cover maps (10 m) from 2007 and 2014. NDVI time-series detected less changes than the reference data, ranging from 46% - 48%. On one land, MODIS does not detect changes smaller than 6.25 ha, on the other hand, MODIS time-series detects more than land cover change, such as deforestation, but degradation as well, which is forest that remains as forest. However, the separation of both is still an unsolved problem, with no proved relation between magnitude and change categories. Copyright © 2017 ISRS, All Rights Reserved.


Entidades citadas de la UNAM: