Accuracy assessment of the National Forest Inventory map of Mexico: sampling designs and the fuzzy characterization of landscapes

Abstract

There is no record so far in the literature of a comprehensive method to assess the accuracy of regional scale Land Cover/ Land Use (LCLU) maps in the sub-tropical belt. The elevated biodiversity and the presence of highly fragmented classes hamper the use of sampling designs commonly employed in previous assessments of mainly temperate zones. A sampling design for assessing the accuracy of the Mexican National Forest Inventory (NFI) map at community level is presented. A pilot study was conducted on the Cuitzeo Lake watershed region covering 400 000 ha of the 2000 Landsat-derived map. Various sampling designs were tested in order to find a trade-off between operational costs, a good spatial distribution of the sample and the inclusion of all scarcely distributed classes (‘rare classes’). A two-stage sampling design where the selection of Primary Sampling Units (PSU) was done under separate schemes for commonly and scarcely distributed classes, showed best characteristics. A total of 2 023 punctual secondary sampling units were verified against their NFI map label. Issues regarding the assessment strategy and trends of class confusions are devised.

Keywords

• Muestreo doble
• clase escasa
• sistema de clasificación
• Landsat
• interpretación estereoscópica
• fotografía aérea
• caracterización difusa