Share Email Print

Proceedings Paper

Effects of sample size on burned areas accuracy estimates in the Amazon Basin
Author(s): A. Fernandez-Carrillo; M. A. Belenguer-Plomer; E. Chuvieco; M. A. Tanase
Format Member Price Non-Member Price
PDF $14.40 $18.00
cover GOOD NEWS! Your organization subscribes to the SPIE Digital Library. You may be able to download this paper for free. Check Access

Paper Abstract

Remote sensing-derived maps contain errors. The magnitude of such errors is evaluated through accuracy assessment based on sub-sampling the total ‘population’ (i.e. map). Several strategies have been proposed to define the optimal sampling design leading to a statistically robust accuracy assessment. In this work, a stratified random sampling approach as proposed by Padilla et al.1 was applied to validate two burned area (BA) products as part of the ESA’s Firecci project: a SAR-based product generated from S1 data and the MODIS MCD64. The sampling design considers sample allocation as a function of burned area proportion inside each biome. In our study the sampling size was computed as suggested by Olofsson et al.2. The objective of this study was to assess to which extent a reduction in the sampling size influences the accuracy metrics. The validation was carried out for BA detected from Sentinel-1 as well as a MODIS based-product, the MCD64, generated for the year 2017 in the Amazon region (8M km2). The reference BA dataset was generated using optical time-series acquired by the Landsat-7 ETM+, Landsat-8 OLI, and Sentinel-2 MSI sensors. The BA products were validated three times: i) over n = 44 sample units (as computed from Olofsson et al.2; ii) considering a sample size of n/2, and iii) considering a sample size of n/4, to test the sensitivity of the accuracy assessment to changes in the sample size over the tropics.

The results showed that halving the sample size while maintaining the stratified allocation method, yielded similar results when compared to the original sample size (differences in OE and CE did not exceed 5% in any of the products, while differences in DC did not exceed 2%). For a sample size of n/4, the validation results were more unstable (differences in DC reached up to 9% and confidence intervals were higher). The results provide evidence for the optimal sampling size for the accuracy assessment of different BA products over the Amazon basin.

Paper Details

Date Published: 9 October 2018
PDF: 8 pages
Proc. SPIE 10790, Earth Resources and Environmental Remote Sensing/GIS Applications IX, 107901S (9 October 2018); doi: 10.1117/12.2325686
Show Author Affiliations
A. Fernandez-Carrillo, Univ. de Alcalá (Spain)
M. A. Belenguer-Plomer, Univ. de Alcalá (Spain)
E. Chuvieco, Univ. de Alcalá (Spain)
M. A. Tanase, Univ. de Alcalá (Spain)

Published in SPIE Proceedings Vol. 10790:
Earth Resources and Environmental Remote Sensing/GIS Applications IX
Ulrich Michel; Karsten Schulz, Editor(s)

© SPIE. Terms of Use
Back to Top