Artículos de revistas
Assessing the performance of smoothing functions to estimate land surface phenology on temperate grassland
Fecha
2016-04Registro en:
Lara, Bruno Daniel; Gandini, Marcelo Luciano; Assessing the performance of smoothing functions to estimate land surface phenology on temperate grassland; Taylor & Francis; International Journal of Remote Sensing; 37; 8; 4-2016; 1801-1813
0143-1161
1366-5901
CONICET Digital
CONICET
Autor
Lara, Bruno Daniel
Gandini, Marcelo Luciano
Resumen
NDVI (Normalized Difference Vegetation Index) time-series have beenused for permitting a land surface phenology retrieval but these timeseries are affected by clouds and aerosols,which add noise to the signalsensor. In this sense, several smoothing functions are used to removenoise introduced by undetected clouds and poor atmospheric conditions,but a comparison between methods is still necessary due todisagreements about its performance in the literature. The applicationof a smoothing function is a necessarily previous step to describe landsurface phenology in different ecosystems. The aims of this researchwere to evaluate the consistency of different smoothing functions fromTIMESAT software and their impacts on phenological attributes oftemperate grassland ? a complex mosaic of land uses with naturalvegetated and agricultural regions using NDVI-MODIS time series. Anadaptive Savitzky?Golay (SG) filter, Asymmetric Gaussian (AG) andDouble Logistic (DL) functions to fitting NDVI data were used andtheir performances were assessed using the measures root meansquare error (RMSE), Akaike Information Criterion (AIC), BayesianInformation Criterion (BIC) and bias. Besides, differences on the estimationof the start of the growing season (SOS) and the length of thegrowing season (LOS) were obtained. High and low RMSE over croplandsand grassland were observed for the three smoothing functions;in the rest of the region, the SG filter showed more reliable results.Patterns of difference on the estimation of SOS and LOS between SGfilter and the other two models were randomly distributed, wheredifferences of 20?50 days were found. This study demonstrated thatmethods from TIMESAT software are robust and spatially consistentbut must be carefully used.