Accuracy assessment methodology for the Mexican national forest inventory: a pilot study in the Cuitzeo lake watershed

Stéphane Couturier 1, Jean-François Mas 2, Erna López 2, Gabriela Cuevas 2, Álvaro Vega 1 and Valdemar Tapia 1
1 Geography Institute, Universidad Nacional Autónoma de México (UNAM), 
Ciudad Universitaria, Coyoacán 04510, Mexico City, Mexico
Tel.: + 00 52 56 22 3443; Fax: + 00 52 56 16 2145
andres@igiris.igeograf.unam.mx
2 Geography Institute, UNAM, Unidad Académica Morelia,
Aquiles Serdan 382, Col. Centro 58000 Morelia, Michoacán, Mexico
Tel.: + 00 52 443 317 9423; Fax: + 00 52 443 317 9425
jfmas@igiris.igeograf.unam.mx

Abstract
A methodology for assessing the accuracy of the Mexican National Forest Inventory (NFI) map is presented. This methodology emerges as the most adequate strategy found after various trials along the successive steps of the assessment design. A main challenge was to integrate the high diversity of classes encompassed in the classification scheme within a cost-controlled statistically sound assessment. A pilot study focused on the Cuitzeo Lake watershed region covering 400,000 ha of the 2000 Landsat-derived NFI. The availability of detailed quasi- synchronous reference data and the high variability of mapped classes allowed a careful thematic analysis on the selected region, relevant for national extrapolation. The assessment strategy incorporated an original two stage sampling design. The selection of Primary Sampling Units (PSU) was done under separate schemes for commonly and scarcely distributed classes. A compromise was statistically found for maximizing PSU spatial distribution while including all classes. The verification protocol included stereoscopic photo- interpretation and a digital restitution towards  the geometry of the Landsat data. A scale adjustment operator, based on the epsilon  probabilistic band approach, was applied to the PSU verification maplets in order to reduce the inclusion of errors due to scale. A total of 2023 punctual secondary sampling units were then compared with their NFI map label, according to conventional Boolean and linguistic fuzzy criteria. Issues regarding the assessment strategy and trends of class confusions are devised. Conclusions are drawn in terms of separability of classes on remote-sensing supports, classification system, geographic stratification and scale. This methodology is to be applied to a larger territory including a wider set of classes in the classification system.

Keywords: double sampling, rare class, scale, fuzzy, classification system

In: Caetano, M. and Painho, M. (eds). Proceedings of the 7th International Symposium on Spatial Accuracy Assessment in Natural Resources and Environmental Sciences, 5 – 7 July 2006, Lisboa, Instituto Geográfico Português

AttachmentSize
Couturier2006accuracy.pdf1.09 MB