A land cover map of Latin America and the Caribbean in the framework of the SERENA project

Abstract

Land cover maps at different resolutions and mapping extents contribute to modeling and support decision making processes. Because land cover affects and is affected by climate change, it is listed among the 13 terrestrial essential climate variables. This paper describes the generation of a land cover map for Latin America and the Caribbean (LAC) for the year 2008. It was developed in the framework of the project Latin American Network for Monitoring and Studying of Natural Resources (SERENA), which has been developed within the GOFC-GOLD Latin American network of remote sensing and forest fires (RedLaTIF). The SERENA land cover map for LAC integrates: 1) the local expertise of SERENA network members to generate the training and validation data, 2) a methodology for land cover mapping based on decision trees using MODIS time series, and 3) class membership estimates to account for pixel heterogeneity issues. The discrete SERENA land cover product, derived from class memberships, yields an overall accuracy of 84% and includes an additional layer representing the estimated per-pixel confidence. The study demonstrates in detail the use of class memberships to better estimate the area of scarce classes with a scattered spatial distribution. The land cover map is already available as a printed wall map and will be released in digital format in the near future. The SERENA land cover map was produced with a legend and classification strategy similar to that used by the North American Land Change Monitoring System (NALCMS) to generate a land cover map of the North American continent, that will allow to combine both maps to generate consistent data across America facilitating continental monitoring and modeling.

Introduction

During the past decades Latin America and the Caribbean (LAC) have undergone unprecedented land cover and land use changes. Cropland expansion and forest conversion, accelerated by economic globalization and climate change, are the dominant land-use trends in the region (Grau & Aide, 2008). LAC have lost more forests since 1990 than any other major world region (FAO, 2005), and the rapid conversion of forests to agriculture has been especially evident in areas such as South America's Amazon region and the lowland forests of Central America. Latin America is responsible for 4.3% of global greenhouse gas emissions (Magrin et al., 2007), and of these, 48.3% result from deforestation and land use changes (UNEP, 2000). In this context land use and land cover (LULC) maps are vital for monitoring, understanding, and predicting the effects of complex human–nature interactions. Information on LULC changes needs to be as accurate and timely as possible if it is to be incorporated into management and policy decisions. To meet these requirements it is necessary to develop cost-effective ways for automated processing of satellite images and production of LULC maps with high temporal resolution (DeFries and Belward, 2000, Skole et al., 1997).

Traditionally, data from medium to high spatial resolution sensors (10–60 m), e.g. Landsat, Système Pour l'Observation de la Terre (SPOT) and Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER), have been used for mapping LULC at local to regional scales (Rogan & Chen, 2004). This level of spatial resolution is generally adequate to detect fine-scale (circa 1:100,000) land use patterns. However, potential data costs, the small image extent, haze and cloudiness, and sporadic acquisitions can make data from medium spatial resolution sensors impractical for macro-regional, continental, and global mapping (Asner, 2001, Hansen et al., 2008). Therefore, data with lower spatial resolution (250–1000 m) but high temporal frequency of image acquisition are in many cases selected as an alternative to medium resolution data for large-area land cover mapping, because they are affordable, require less effort for data manipulation and processing, and their near-to-daily coverage increases the probability of cloud-free composites (Arino et al., 2008, Bartholomé and Belward, 2005, Friedl et al., 2002, Hansen et al., 2002, Loveland et al., 2000). Also many classes can only be mapped with high accuracy using time series that characterize vegetation phenology rather than single-date spectral response (Coppin et al., 2003, Friedl et al., 2002, Friedl et al., 2010, Hansen et al., 2000).

Various global land cover maps have been produced from low resolution satellites, e.g. 1.1 km Advanced Very High Resolution Radiometer (AVHRR; IGBP DISCover, Loveland et al., 2000; UMD GLCC, Hansen et al., 2000), 1 km SPOT-Vegetation (GLC2000, Bartholomé & Belward, 2005), 500 m and 1000 m Moderate Resolution Imaging Spectroradiometer (MODIS; MOD12Q1, Friedl et al., 2002; MCD12Q1, Friedl et al., 2010), and 300 m Medium Resolution Imaging Spectrometer (MERIS; Globcover, Bicheron et al., 2008). Specifically for the LAC region, a vegetation map for South America (SA) for the year 2000 with twenty-two classes was developed using 1 km SPOT-Vegetation as part of a project to map Global Land Cover for the year 2000 (GLC 2000, Eva et al., 2004), and more recently a land cover map of SA with nine classes was derived from the 300 m MERIS for the years 2008 and 2010 (Hojas Gascon et al., 2012). A study by Clark et al. (2012) mapped annual land-cover data with 8 classes from 2001 to 2010 using 250 m MODIS data, and tracks change patterns of three generalized classes including woody, mix woody/plantation, and agriculture/herbaceous vegetation by linear regression at the municipality level. However, a discrete land cover map of coarse spatial resolution has several disadvantages (DeFries et al., 1995a, DeFries et al., 1999, Fernandes et al., 2004). Discrete classes cannot represent spatially complex areas (Hansen et al., 2002), because spatial complexity increases with spatial resolution (Moody & Woodcock, 1994). As a result, homogeneous landscapes e.g. large stands of evergreen broadleaf forest in the Amazon tend to be well classified, but accuracy is poor for mixed pixels that represent small-patch landscapes and transitional zones with various spectral and temporal signals of trees, shrubs, and herbaceous vegetation (e.g. shrub-grass steppes in southern Patagonia; Herold et al., 2008).

The Global Observation of Forest and Land Cover Dynamics (GOFC-GOLD) program is a coordinated international effort, working to provide ongoing space-based and in situ observations of forests and other vegetation cover for sustainable management of terrestrial resources and to obtain an accurate and reliable quantitative understanding of the terrestrial carbon budget (http://www.fao.org/gtos/gofc-gold/overview.html). The creation of regional networks has been encouraged, which provide a mechanism for sharing resources and expertise. Within this framework, the Latin American Remote Sensing and Forest Fires Network (Red Latinoamericana de Teledetección e Incendios Forestales — RedLaTIF) was created in 2002. This paper presents the production and analysis of a land cover map for Latin America and the Caribbean for the year 2008 that was part of the Latin American Network for Monitoring and Studying of Natural Resources (Red Latinoamericana de Seguimiento y Estudio de los Recursos Naturales — SERENA, web page http://www.proyectoserena.com.ar/). The SERENA project, developed within the RedLaTIF network, aims to monitor, study, and disseminate information associated with biomass burning (wildfires) and changes in land use and cover for LAC utilizing satellite data. The network brings together 18 institutions from 10 countries with 53 researchers. The network envisages the development of common methodologies for developing homogeneous products that build upon and are validated with local data.

The SERENA land cover map for LAC incorporates: 1) the local expertise of SERENA network members to generate training and validation data, 2) a land cover mapping approach based on decision trees using MODIS time series, and 3) estimates of class memberships to partly overcome pixel heterogeneity issues of coarse resolution land cover maps. The suite of SERENA land cover products also includes a discrete map with an additional layer that represents the estimated confidence. The land cover map was recently finished and is available as a wall map. The digital version will be released in the near future and is hosted on the SERENA web page. The SERENA land cover mapwas produced using a legend and classification strategy similar to the North American Land Change Monitoring System (NALCMS) that generated a land cover map for the North American continent (Colditz et al., 2012, Latifovic et al., 2012). The similarity between both maps will facilitate legend harmonization and map combination to form a consistent land cover map from Ellesmere Island, Canada to Tierra del Fuego, Argentina that will be of high value for monitoring and modeling across the American continent.