• Bonan, G. B., , R. S. DeFries, , M. T. Coe, , and D. S. Ojima. 2004. Land use and climate. Land Change Sciences, G. Gutman, et al., Eds., Kluwer Academic, 301–314.

    • Search Google Scholar
    • Export Citation
  • Brown, J. F., , T. R. Loveland, , D. O. Ohlen, , O. Donald, , and Z. Zhu. 1999. The global land-cover characteristics database: The user’s perspective. Photogramm. Eng. Remote Sens. 65:10691074.

    • Search Google Scholar
    • Export Citation
  • Chen, J. M., and J. Cihlar. 1996. Retrieving leaf area index of boreal conifer forests using Landsat TM images. Remote Sens. Environ. 55:153162.

    • Search Google Scholar
    • Export Citation
  • Cihlar, J. 2000. Land cover mapping of large areas from satellites: Status and research priorities. Int. J. Remote Sens. 21:10931114.

    • Search Google Scholar
    • Export Citation
  • Congalton, R. G. 1991. A review of assessing the accuracy of classification of remotely sensed data. Remote Sens. Environ. 37:3546.

  • DeFries, R., , M. Hansen, , M. Steininger, , R. Dubayah, , R. Sohlberg, , and J. Townshend. 1997. Subpixel forest cover in Central Africa from multisensor, multitemporal data. Remote Sens. Environ. 54:209222.

    • Search Google Scholar
    • Export Citation
  • Dickinson, R. E. 1995. Land-atmosphere interaction. Rev. Geophys. 33:917922.

  • Dickinson, R. E., , A. Henderson-Sellers, , P. J. Kennedy, , and M. F. Wilson. 1986. Biosphere-atmosphere transfer scheme for the NCAR community climate model. NCAR Tech. Note NCAR/TN-275+STR, 69 pp.

    • Search Google Scholar
    • Export Citation
  • Di Gregorio, A., and L. J. M. Jansen. 2000. Land Cover Classification System (LCCS). F.A.O., 179 pp.

  • Feddema, J. J., , K. W. Oleson, , G. B. Bonan, , L. O. Mearns, , L. E. Buja, , G. A. Meehl, , and W. M. Washington. 2005. The importance of land-cover change in simulating future climates. Science 310:16741678.

    • Search Google Scholar
    • Export Citation
  • Foley, J. A., , M. H. Costa, , C. Delire, , N. Ramankutty, , and P. Snyder. 2003. Green surprise? How terrestrial ecosystems could affect earth’s climate. Front. Ecol. Environ. 1:3844.

    • Search Google Scholar
    • Export Citation
  • Foody, G. M. 2002. Status of land cover classification accuracy assessment. Remote Sens. Environ. 80:185201.

  • Friedl, M. A. Coauthors 2002. Global land cover mapping from MODIS: Algorithms and early results. Remote Sens. Environ. 83:287302.

  • Ge, J., , J. Qi, , B. Lofgren, , N. Moore, , N. Torbick, , and J. M. Olson. 2007. Impacts of land use/cover classification accuracy on regional climate simulations. J. Geophys. Res. 112:D05107. doi:10.1029/2006JD007404.

    • Search Google Scholar
    • Export Citation
  • Ge, J., , J. Qi, , and B. Lofgren. 2008. Use of vegetation properties from EOS observations for land-climate modelling in East Africa. J. Geophys. Res. 113:D15101. doi:10.1029/2007JD009628.

    • Search Google Scholar
    • Export Citation
  • Giri, C., , Z. Zhu, , and B. Read. 2005. A comparative analysis of the Global Land Cover 2000 and MODIS land cover data sets. Remote Sens. Environ. 94:123132.

    • Search Google Scholar
    • Export Citation
  • Jansen, L. J. M., and A. Di Gregorio. 2003. Land-use data collection using the “land cover classification system”: Results from a case study in Kenya. Land Use Policy 20:131148.

    • Search Google Scholar
    • Export Citation
  • Lark, R. M. 1995. Components of accuracy of maps with special reference to discriminant analysis on remote sensing data. Int. J. Remote Sens. 16:14611480.

    • Search Google Scholar
    • Export Citation
  • Latifovic, R., and I. Olthof. 2004. Accuracy assessment using sub-pixel fractional error matrices of global land cover products derived from satellite data. Remote Sens. Environ. 90:153165.

    • Search Google Scholar
    • Export Citation
  • Lee, T. J. 1992. The impact of vegetation on the atmospheric boundary layer and convective storms. Atmosphere Science Paper No. 509, Colorado State University, 155 pp.

    • Search Google Scholar
    • Export Citation
  • Loveland, T. R., , Z. Zhu, , D. O. Ohlen, , J. F. Brown, , B. C. Reed, , and L. Yang. 1999. An analysis of the IGBP global land-cover characterization process. Photogramm. Eng. Remote Sens. 65:10211032.

    • Search Google Scholar
    • Export Citation
  • Loveland, T. R., , B. C. Reed, , J. F. Brown, , D. O. Ohlen, , Z. Zhu, , L. Yang, , and J. W. Merchant. 2000. Development of a global land cover characteristics database and IGBP DISCover from 1 km AVHRR data. Int. J. Remote Sens. 21:13031330.

    • Search Google Scholar
    • Export Citation
  • Mathews, E. 1983. Global vegetation and land use: New high resolution data bases for climate studies. J. Climate Appl. Meteor. 22:474487.

    • Search Google Scholar
    • Export Citation
  • Mayaux, P., , E. Bartholome, , S. Fritz, , and A. Belward. 2004. A new land-cover map of Africa for the year 2000. J. Biogeogr. 31:861877.

  • Myneni, R. B. Coauthors 2002. Global products of vegetation leaf area and fraction absorbed PAR from year one of MODIS data. Remote Sens. Environ. 83:214231.

    • Search Google Scholar
    • Export Citation
  • Olson, J. S. 1994a. Global ecosystem framework-definitions. Internal Rep., USGS EROS Data Center, Sioux Falls, SD, 37 pp.

  • Olson, J. S. 1994b. Global ecosystem framework-translation strategy. Internal Rep., USGS EROS Data Center, Sioux Falls, SD, 39 pp.

  • Olson, J. S., , J. A. Watts, , and L. J. Allison. 1983. Carbon in live vegetation of major world ecosystems. Oak Ridge National Laboratory, 51 pp.

    • Search Google Scholar
    • Export Citation
  • Pielke Sr., R. A., , R. Avissar, , M. Raupach, , A. J. Dolman, , X. Zeng, , and A. S. Denning. 1998. Interactions between the atmosphere and terrestrial ecosystems: Influence on weather and climate. Global Change Biol. 4:461475.

    • Search Google Scholar
    • Export Citation
  • Pielke Sr., R. A., , G. Marland, , R. A. Betts, , T. N. Chase, , J. L. Eastman, , J. O. Niles, , D. D. S. Niyogi, , and S. W. Running. 2002. The influence of land-use change and landscape dynamics on the climate system: Relevance to climate-change policy beyond the radiative effect of greenhouse gases. Philos. Trans. Roy. Soc. London A360:17051719.

    • Search Google Scholar
    • Export Citation
  • Running, S. W., , T. R. Loveland, , L. L. Pierce, , R. R. Nemani, , and E. R. Hunt Jr.. 1995. A remote sensing based vegetation classification logic for global land cover analysis. Remote Sens. Environ. 51:3948.

    • Search Google Scholar
    • Export Citation
  • Scheffe, H. 1959. The Analysis of Variance. John Wiley & Sons, 477 pp.

  • Sellers, P. J., , Y. Mintz, , Y. C. Sud, , and A. Dalcher. 1986. A simple biosphere model (SiB) for use within general circulation models. J. Atmos. Sci. 43:505531.

    • Search Google Scholar
    • Export Citation
  • Sellers, P. J. Coauthors 1996a. A revised land surface parameterization (SiB2) for atmospheric GCMs. Part I: Model formulation. J. Climate 9:676705.

    • Search Google Scholar
    • Export Citation
  • Sellers, P. J. Coauthors 1996b. A revised land surface parameterization (SiB2) for atmospheric GCMs. Part II: The generation of global fields of terrestrial biophysical parameters from satellite data. J. Climate 9:706737.

    • Search Google Scholar
    • Export Citation
  • Sellers, P. J. Coauthors 1997. Modelling the exchanges of energy, water, and carbon between continents and the atmosphere. Science 275:502509.

    • Search Google Scholar
    • Export Citation
  • Stapleton, J. H. 1995. Linear Statistical Models. John Wiley & Sons, 444 pp.

  • Torbick, N. M., , D. Lusch, , J. Qi, , N. Moore, , J. Olson, , and J. Ge. 2006. Developing land use land cover parameterization for climate-land modelling in East Africa. Int. J. Remote Sens. 27:42274244.

    • Search Google Scholar
    • Export Citation
  • Walko, R. L. Coauthors 2000. Coupled atmosphere–biophysics–hydrology models for environmental modelling. J. Appl. Meteor. 39:931944.

    • Search Google Scholar
    • Export Citation
  • Wang, Y. Coauthors 2004. Evaluation of the MODIS LAI algorithm at a coniferous forest site in Finland. Remote Sens. Environ. 91:114127.

    • Search Google Scholar
    • Export Citation
  • Zhuang, X., , B. A. Engel, , X. Xiong, , and C. J. Johannsen. 1995. Analysis of classification results of remotely sensed data and evaluation of classification algorithms. Photogramm. Eng. Remote Sens. 61:427433.

    • Search Google Scholar
    • Export Citation
  • View in gallery

    A simulated 4 × 4 pixel area. (a) Two classes and (b) three classes. Numbers shown are LAI values.

  • View in gallery

    Study area presented by GLC2000.

  • View in gallery

    The Q maps for four land covers at the quadrate size of 30 × 30 km: (a) GLC2000, (b) MODIS land cover, (c) OGE, and (d) LEAF.

  • View in gallery

    Mean Q for all land covers at three scales.

  • View in gallery

    Multiple comparison (Tukey’s method) results for mean Q values for a quadrate size of 30 × 30 km. MOD means MODIS land cover. The mean Q values of GLC and LEAF are significantly different (GLC LEAF). Confidence intervals were built on significance level of 0.05.

  • View in gallery

    Here Q was applied to a single class, croplands (>50%), in GLC2000. (a) A Q map at 30 × 30 km quadrate size, (b) croplands in the hot spot in (a), pointed by two lines, and (c) Africover corresponding to (b).

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 38 38 4
PDF Downloads 10 10 0

Biophysical Evaluation of Land-Cover Products for Land–Climate Modeling

View More View Less
  • 1 Department of Geography, Oklahoma State University, Stillwater, Oklahoma
  • | 2 Applied Geosolutions LLC, Durham, New Hampshire
  • | 3 Department of Geography, and Center for Global Change and Earth Observations, Michigan State University, East Lansing, Michigan
© Get Permissions
Full access

Abstract

The need for accurate characterization of the land surface as boundary conditions in climate models has been recognized widely in the climate modeling community. A large number of land-cover datasets are currently used in climate models either to better represent surface conditions or to study the impacts of surface changes. Deciding upon land-cover datasets can be challenging because the datasets are made with different sensors, ranging methodologies, and varying classification objectives. A new statistical measure Q was developed to evaluate land-cover datasets in land–climate interaction research. This measure calculates biophysical precision of land-cover datasets using 1-km monthly Moderate Resolution Imaging Spectroradiometer (MODIS) leaf area index (LAI) product. This method aggregates within-class biophysical consistency, calculated as LAI variation, across a study domain and over multiple years into a single statistic. A smaller mean Q value for a land-cover product indicates more precise biophysical characterization within the classes. As an illustration, four land-cover products were assessed in the East Africa region: Global Land Cover 2000 (GLC2000), MODIS land cover, Olson Global Ecosystems (OGE), and Land Ecosystem–Atmosphere Feedback (LEAF) model. The evaluation was conducted at three different spatial scales corresponding to 30 × 30, 50 × 50, and 100 × 100 km quadrates. The Q measure found that GLC2000 ranked higher compared to the other three land-cover products for every quadrate size. For the 30 × 30 km quadrate size GLC2000 was significantly better than LEAF, which is currently used in the Regional Atmospheric Modeling System. The statistic ranks MODIS land cover above OGE, which is above LEAF. As quadrate size increases, differences between Q decrease indicating greater uncertainty at coarser resolution. The utility of the measure is that it can be applied to any continuous parameter over any scale (space or time) to evaluate the biophysical precision of any land-cover dataset.

* Corresponding author address: Jianjun Ge, Department of Geography, Oklahoma State University, 225 Scott Hall, Stillwater, OK 74078-4073. jianjun.ge@okstate.edu

Abstract

The need for accurate characterization of the land surface as boundary conditions in climate models has been recognized widely in the climate modeling community. A large number of land-cover datasets are currently used in climate models either to better represent surface conditions or to study the impacts of surface changes. Deciding upon land-cover datasets can be challenging because the datasets are made with different sensors, ranging methodologies, and varying classification objectives. A new statistical measure Q was developed to evaluate land-cover datasets in land–climate interaction research. This measure calculates biophysical precision of land-cover datasets using 1-km monthly Moderate Resolution Imaging Spectroradiometer (MODIS) leaf area index (LAI) product. This method aggregates within-class biophysical consistency, calculated as LAI variation, across a study domain and over multiple years into a single statistic. A smaller mean Q value for a land-cover product indicates more precise biophysical characterization within the classes. As an illustration, four land-cover products were assessed in the East Africa region: Global Land Cover 2000 (GLC2000), MODIS land cover, Olson Global Ecosystems (OGE), and Land Ecosystem–Atmosphere Feedback (LEAF) model. The evaluation was conducted at three different spatial scales corresponding to 30 × 30, 50 × 50, and 100 × 100 km quadrates. The Q measure found that GLC2000 ranked higher compared to the other three land-cover products for every quadrate size. For the 30 × 30 km quadrate size GLC2000 was significantly better than LEAF, which is currently used in the Regional Atmospheric Modeling System. The statistic ranks MODIS land cover above OGE, which is above LEAF. As quadrate size increases, differences between Q decrease indicating greater uncertainty at coarser resolution. The utility of the measure is that it can be applied to any continuous parameter over any scale (space or time) to evaluate the biophysical precision of any land-cover dataset.

* Corresponding author address: Jianjun Ge, Department of Geography, Oklahoma State University, 225 Scott Hall, Stillwater, OK 74078-4073. jianjun.ge@okstate.edu

Introduction

By changing the fluxes of mass and energy between ecosystems and the atmosphere, human modification of the land surface impacts regional and global climate processes (Pielke et al. 2002; Foley et al. 2003). Using land–climate modeling techniques, impacts of land-use and land-cover changes on the Earth system can be studied and monitored. Most regional and global atmospheric models developed 20 years ago either ignored or oversimplified the interactions of the atmosphere with underlying soil and vegetation surfaces. As numerical modeling of atmospheric processes has progressed and research on human impacts on climate change has become more urgent over the past decade, the need for accurate characterization of the land surface as boundary conditions in climate modeling is becoming widely recognized (e.g., Dickinson 1995; Sellers et al. 1997; Pielke et al. 1998; Bonan et al. 2004; Feddema et al. 2005; Ge et al. 2007).

Land-cover products have been used in the soil–vegetation–atmosphere transfer (SVAT) schemes in climate models for a long time. Current SVAT schemes, such as the Biosphere–Atmosphere Transfer Scheme (BATS), simple biosphere model (SiB and SiB2), and Land Ecosystem–Atmosphere Feedback (LEAF) model, represent the land surface through a discrete number of predetermined land-cover classes, each with a suite of biophysical parameters including leaf area index (LAI), vegetation fractional cover (VFC), albedo, surface roughness, and root depth, among others (Dickinson et al. 1986; Sellers et al. 1986; Lee 1992; Sellers et al. 1996a; Sellers et al. 1996b; Walko et al. 2000). Among these parameters, LAI quantifies the amount of foliage area per unit ground surface area, and it controls many biological and physical processes such as photosynthesis, transpiration, and rainfall interception (Chen and Cihlar 1996). Until the last decade, land-cover products used in most climate models were initially compiled from maps, ground surveys, and various national sources, which have inherent limitations (Mathews 1983; Olson et al. 1983; Cihlar 2000). In the mid-1990s, global-scale land-cover products generated from remote sensing images became available. Land-cover datasets with the BATS, SiB, and SiB2 schemes can be derived from the global land-cover characteristics database, which is generated from 1-km Advanced Very High Resolution Radiometer (AVHRR) data (Loveland et al. 2000). Land cover used in LEAF is cross mapped from Olson Global Ecosystems (OGE) dataset (Olson 1994a; Olson 1994b), which can also be derived from the global land-cover characteristics database. Recently, new Moderate Resolution Imaging Spectroradiometer (MODIS) land cover (MOD12) and the Global Land Cover 2000 (GLC2000) have been developed with enhanced spectral, spatial, radiometric, and geometric quality (Friedl et al. 2002; Latifovic and Olthof 2004). These two products, along with others, are more suitable for monitoring land surface properties at regional to global scales and have great potential to be employed in climate modeling systems in the future (Giri et al. 2005).

With many land-cover products from different sources becoming available for a given region of the Earth, a challenge arises as to which product is optimal for a land–climate modeling study. Traditional classification accuracy assessment is primarily based on ground-based surveys or interpretation of high spatial resolution aerial photos and satellite images. By comparing the classified land-cover dataset with the ground-truthing data, error metrics can be developed to report the commission and omission errors. Measures of accuracy, such as the kappa coefficient of agreement, are frequently calculated to express classification accuracy (Congalton 1991; Foody 2002). However, current land-cover classification assessment methods are not triggered by the ultimate users’ practical needs. One classification product will hardly be optimal from the viewpoint of different users (Lark 1995; Brown et al. 1999). Errors in land-cover classifications thus should be viewed in the context of the application of the land-cover information. For different applications, such as biodiversity assessments, natural resource management, or more sophisticated land–climate modeling, it is very likely that misclassification errors will have different consequences. For land–climate modeling, biophysical attributes are crucial and are actually used in model calculations. However, no studies have addressed the biophysical characteristics in land-cover classification assessment from the perspective of land–climate modeling.

At the same time, a great number of satellite-based biophysical products (e.g., LAI and albedo) have been developed with increasing accuracy. Some recent studies have attempted to use satellite-based biophysical products directly in climate modeling to improve the representation of surface processes (e.g., Ge et al. 2008). Because of the limitation of computing resources, land-cover datasets are still used as a means for averaging biophysical attributes and calculated fluxes. The use of land-cover datasets along with satellite-based biophysical products will likely continue as long as land–climate models have spatial resolutions coarser than biophysical products (e.g., 1 km). Potential conflicts between observed biophysical information and existing land-cover datasets developed from various sources are becoming even more apparent. Evaluating land-cover products in terms of biophysical characteristics is necessary, and evaluation methods are greatly needed.

The importance of biophysical attributes for land-cover classification is not new. Researchers have proposed to use biophysical products with observable, unambiguous, and continuous structural variables (e.g., LAI, canopy height, etc.) to make an optimized classification for a specific need (Running et al. 1995; DeFries et al. 1997; Cihlar 2000). The purposes, however, were to address the subjective nature of a traditional classification paradigm, which assigns each image pixel to one land-cover type based on a predefined scheme. The traditional paradigm tends to cause confusion, and the ambiguity usually continues to exist in the conventional accuracy assessment procedure, and as a result there is no single acknowledged standard for accuracy assessment (Loveland et al. 1999; Foody 2002). In the development of the MODIS land-cover product (MOD12), the importance of biophysical characteristics to land-cover types has been considered to facilitate regional and global climate modeling studies (Friedl et al. 2002). However, the classification still relies on ground samples and supervised classification algorithms. This is partly because of the complex nature of land-cover classification and relatively limited number of biophysical products available.

Both accuracy and precision are relevant to the quality of classification. This paper focuses on biophysical precision of land-cover products. The objective is to develop a new statistical method with one straightforward statistic to evaluate the biophysical consistency from the perspective of land–climate modeling. LAI is used in this new method as it is a crucial land surface parameter in land–climate modeling. A description of the land scheme in climate models and the development of the statistical measure are given in section 2. Data used are discussed in section 3. The results are presented in section 4, and the major conclusions are reviewed in the final section.

Method

One common approach to represent subgrid heterogeneity within grid cells of climate models is to treat different land-cover types as “patches” or “tiles” (Walko et al. 2000). Through a set of specified biophysical variables, each patch has its own vegetation, soil, canopy air, etc. and responds to and influences the overlying atmosphere in its own way. Fluxes in a grid cell are integrated over all patches weighted by the corresponding patch fractional areas. The benefit of this approach is the ability to represent subgrid heterogeneity and save computational resources. Complete resolution of surface features is not feasible in most climate modeling studies over large areas. Biophysical variables assigned for land-cover types usually include LAI, VFC, albedo, surface roughness, and root depth, among others.

The primary purpose of land-cover classification is to divide the land surface into distinct types based on their biophysical characteristics to simplify the complexity of the landscape. More precise biophysical characteristics within land-cover classes indicate a classification that is in harmony with landscape features and thus a more realistic representation of the land surface. Biophysical precision is critical for climate modeling because it is biophysical variables that are used for calculating fluxes not arbitrary labels of land-cover types. Furthermore, biophysical precision is a function of spatial scales and varies with time. This has not been fully addressed in the traditional classification assessment methods. Climate modeling studies vary substantially in spatial and temporal scales. Global climate models (GCMs) usually have grid cells larger than 100 km, while regional climate models (RCMs) have grid cells less than 100 km. The simulation time period can be months, years, and decades.

To address the specific needs in land–climate modeling studies, a new statistical measure Q is developed in this paper. The application in this paper utilizes the recently available 1-km MODIS LAI product. Biophysical precision is calculated as within-class LAI variation. Patch-level LAI variation is then aggregated across the domain and over time. Spatial scale is also addressed by adjusting the size of grid cells. As a result, smaller Q indicates a classification that is more suitable for land–climate modeling studies. Ideally, land-cover classification for land–climate modeling needs to be evaluated based on a suite of biophysical variables including LAI, VFC, albedo, surface roughness, and root depth. This paper focuses on the application of LAI.

Development of Q

For an area of interest (one model grid cell) with k classes, LAI variation V within one class i can be calculated using the statistical error sum of squares (Stapleton 1995):
i1087-3562-13-6-1-e1
where pi is the total number of pixels for class i, n refers to any particular pixel, and LAIi is the mean LAI value. Also, beyond the absolute LAI value, the seasonal dynamics of LAI defines important phenological characteristics. An appropriate land-cover classification needs to consider not only spatial consistency of biophysical characteristics but also seasonal vegetation dynamics. Therefore, seasonal changes of LAI were taken into account in the assessment. By adding this temporal LAI information, Equation (1) can be rewritten as
i1087-3562-13-6-1-e2
where t represents time. In this paper, two years of monthly LAI data (February 2000 to December 2001) were used with t ranging from 1 to 23, and T therefore equals the maximum time periods used. (For some areas not all 23 time periods can be used because of quality control for LAI pixels, as discussed below.) Summing Wi for all classes gives an indication of within-class LAI variation. However, for a given model grid cell, different land-cover products may classify an unequal number of classes to it, because classification schemes are usually different. Therefore, summation of Wi must be normalized by (Nk) to yield the final statistical measure Q, where N is the total number of pixels for a grid cell and is same for different land-cover datasets, and k is total number of classes. This is expressed as
i1087-3562-13-6-1-e3
It is assumed here that N is larger than one. In most climate modeling studies, the spatial resolution is typically lower than 10 km because of the limitation on computing resources. By combining Equations (1) and (2), Equation (3) can be written as
i1087-3562-13-6-1-e4
In summary, Q aggregates spatial and temporal LAI variation within each land-cover type at given spatial scales as one straightforward measure. The LAI value is usually between 0 and 7. The theoretical minimum Q is 0 when each land-cover type has uniform LAI values. The maximum Q can be calculated by looking at a grid cell with only one vegetation type (k = 1) without considering temporal dynamics (T = 1). When LAI values are either 0 or 7 and the number of 0s and 7s is same (LAI = 3.5), Q reaches its theoretical maximum: 3.52 = 12.25. Between 0 and 12.25 a smaller Q value indicates more consistent biophysical characteristics and the more appropriate classification for climate modeling.

It needs to be pointed out that some land-cover types have natural variation in biophysical attributes, for example, savannas. Savannas are grasslands that have occasional trees and shrub patches. However, the tree canopy covers the surface with a limited range, for example, between 10% and 30% (Friedl et al. 2002). The variation of biophysical attributes for savanna is thus expected to be within a range. Therefore, Q is still useful for evaluating savanna biophysical precision for land–climate applications.

Illustration

As an illustration, Figure 1 represents a simulated area with 4 × 4 pixels. The LAI value at one particular time is also shown for each pixel. For the purpose of clarity, this rectangular area will be referred as quadrate to differentiate from pixel in an image and grid cell in a climate model. One quadrate is composed of smaller image pixels and corresponds to one grid cell. In this illustration two quadrates in Figure 1 represent two different classifications. Quadrate a has two classes and its Q value is 0.02. In comparison quadrate b has three classes and its Q value is 0.47. These two Q values suggest Figure 1a is more precise in terms of its intraclass biophysical consistency.

Evaluation design

For the evaluation of this paper, nonoverlapping equal-area quadrates are tiled completely for the study area. The Q values were calculated quadrate by quadrate for each land-cover product. The mean Q value was then utilized as the final evaluator to rank the land-cover products. By doing so, the effect of spatial scale of the climate model can be studied by adjusting the quadrate size. As the quadrate size becomes smaller, the Q value may decrease. In the case when the size of quadrates turns out to be one pixel, there would be a single LAI value and thus no variation at all. For this study, the assessment was conducted at three different spatial scales: 30 × 30, 50 × 50, and 100 × 100 km. The consideration is based on the balance between computation time and spatial resolution of typical climate modeling exercises. Simulations with spatial resolutions higher than 20 ∼ 30 km can be time consuming, particularly when they are run at the regional to continental scales with time spans longer than several months. Also, regional climate models usually have grid cells smaller than 100 km. From our experiences (Ge et al. 2007), 50 km is the optimal resolution for the East African region. Therefore, three spatial scales (30, 50, and 100 km) are selected to illustrate the potential of the Q measure at different scales. The Q values for all four land-cover products at these three quadrate sizes were calculated and analyzed.

Data description

Study area

The study area was in East Africa and covers a domain of 1200 km × 900 km. This area was chosen because it is a primary area of our land–climate interaction studies by a multi-institution research team (additional information is available online at www.clip.msu.edu). Countries include parts of Kenya, Uganda, Tanzania, and the Democratic Republic of the Congo and all of Rwanda and Burundi (Figure 2). The geographical position of the center point is 0.43°S, 32.33°E.

Land-cover products

Here Q was applied to the following four land-cover products.

GLC2000 for Africa

This product was developed by the Joint Research Centre’s Global Vegetation Unit based primarily on Systeme Pour l’Observation de la Terre (SPOT) Vegetation daily 1-km data, which were acquired from 1 November 1999 to 31 December 2000 (Mayaux et al. 2004). Other data sources such as radar and Defense Meteorological Satellite Program (DMSP) were also used. GLC2000 uses the land-cover classification system developed by the Food and Agriculture Organization (FAO) of the United Nations and the United Nations Environment Programme (UNEP) and contains 27 land-cover classes (Di Gregorio and Jansen 2000). The GLC2000 for Africa was obtained online at http://bioval.jrc.ec.europa.eu/products/glc2000/data_access.php. All land-cover products were preprocessed into Lambert azimuthal equal-area projection.

MODIS land cover

This product (MOD12Q1) was developed by the MODIS Land-Cover group at Boston University. It was prepared using MODIS Terra daily data acquired from 15 October 2000 to 15 October 2001, excluding June 2001, which is missing because of instrument down time. This product was based on the International Geosphere-Biosphere Programme (IGBP) global vegetation classification scheme and has 17 classes (Friedl et al. 2002). MOD12Q1 for Africa was downloaded from http://duckwater.bu.edu/lc/mod12q1.html.

OGE

This is the Global Land Cover Characterization database version 2.0 with OGE legend (Olson 1994a; Olson 1994b). It was developed by the U.S. Geological Survey (USGS), the University of Nebraska—Lincoln (UNL), and the European Commission’s Joint Research Centre (JRC), based on 1-km AVHRR data spanning April 1992 through March 1993. It is archived at the Earth Resources Observation and Science (EROS) Data Center (https://wist.echo.nasa.gov/~wist/api/imswelcome/). OGE has 96 classes globally but only 25 classes in the study area

LEAF

This is the default land cover used in the land component of the Regional Atmospheric Modeling System (RAMS) model, termed LEAF. LEAF reaggregates the OGE dataset into 31 classes to use BATS’ land surface parameters (Walko et al. 2000).

MODIS LAI product

This Q application used 1-km monthly MODIS LAI data from February 2000 to December 2001. The newly reprocessed MOD15_BU LAI C4.1 data were obtained online at ftp://primavera.bu.edu/pub/datasets/MODIS/. All data were reprojected to Lambert azimuthal equal-area projection with 1-km resolution. For LAI pixels with normal values (0–7), quality assessment (QA) flags were used to detect cloud contamination in each pixel (Myneni et al. 2002). Only high quality LAI pixels having QA values from one to four were selected for Q. These pixels were produced by the main algorithm without saturation under clear-sky conditions. For LAI pixels with fill values (200, pixels outside projection; 253, barren, desert, or very sparsely vegetated; 254, water; 255, noncomputed pixels or missing pixels), only 253 and 254 were used by replacing them with zero. It needs to be noted that there was persistent cloud over part of Congo forest in our study area (Figure 2), especially in the wet season. This highlights an advantage of Q since it can be implemented with only 1 month out of 23 months.

Results

Values of Q

The Q values were calculated quadrate by quadrate for all four land-cover products at three quadrate sizes (30 × 30, 50 × 50, and 100 × 100 km). There are 1200 quadrates for the quadrate size of 30 × 30 km, 432 for the quadrate size of 50 × 50 km, and 108 for the quadrate size of 100 × 100 km. Only Q values at quadrate size of 30 × 30 km are graphically presented here (Figure 3). In Figure 3, the cell size of these four maps equals 30 × 30 km, which corresponds to the size of the quadrate used to calculate Q. It is noticeable that Q has a similar spatial pattern for all land-cover products. It has larger values for areas with complex landscapes such as mountains or patchy agricultural landscapes and has smaller values for areas with less heterogeneity such as dense, large-area forests. Large water areas (e.g., Lake Victoria) have a Q of zero.

The mean Q value was calculated and used as the indicator of evaluation. The mean Q values for each land-cover product at three different quadrate sizes are plotted in Figure 4. Mean Q value for GLC2000 was found to be smaller while LEAF was found to be greater than the other two land-cover products at every quadrate size. The rankings for mean Q values for all four land-cover products were QGLC < QMODIS < QOGE < QLEAF. This suggests that GLC2000 has the most precise land-cover classification and LEAF has the least in biophysical attributes. MODIS land cover was found to rank more suitably than OGE. As expected, for a given land-cover product, Q values increase as the size of quadrates increases. For example, the mean Q value for GLC2000 increases from 0.863 for 30 × 30 km to 1.008 for 100 × 100 km.

Significance test

Initial results based on mean Q values have shown these land covers have different classification performances for climate modeling. However, the differences in terms of mean Q values need to be further tested for significance. If the Q differences are not significant, the advantage of one land-cover product over another one may be negligible.

A statistical method to test the null hypothesis that mean Q values of four land-cover products are all equal was required. For this purpose, one-way analysis of variance (ANOVA) was selected (Scheffe 1959). The advantage of using ANOVA rather than multiple t tests is that it gives one p value for a large number of groups. It would have required six pairs for t tests to evaluate four land-cover products in this study. Like other statistical tests, ANOVA assumes that within each sample the values are independent and normally distributed. This may not be met for Q values since they may be spatially correlated, which is common for most spatially distributed data. In this study we assume there is no spatial correlation for Q values. A significance level of 0.05 was chosen to conduct the test for each quadrate size. Sample sizes (number of quadrates) for four land covers were equal on all three scales: 1200 for 30 × 30 km, 432 for 50 × 50 km, and 108 for 100 × 100 km. The resulting p values for each test are 0.0053, 0.0912, and 0.2370, respectively. This analysis suggests that mean Q values at the quadrate size of 30 × 30 km are significantly different (0.0053 ≪ 0.05), while mean Q values at the other two scales are not. This result seems to contradict the plot in Figure 4 by qualitative inspection, where the difference for the quadrate size of 30 × 30 km is small—this is because the sample size (1200) at this scale is much larger than the other two (432 and 108).

To identify specific differences between pairs of groups, a multiple comparison procedure (MCP) was then used following the ANOVA test. ANOVA only indicates whether there is a difference among the four land-cover products but not where the difference occurs. MCP indicates which land cover is different if a significant difference has been found by ANOVA. Tukey’s method is one such MCP and was used in this study (Zhuang et al. 1995; Stapleton 1995).

Figure 5 presents part of the result from MCP for a quadrate size of 30 × 30 km. In this figure, the positions of dots represent the differences between sample means for Q. Parentheses and dashed lines indicate the extent of the confidence intervals. If a confidence interval does not contain zero, the difference for that pair was significant. Figure 5 illustrates that the only significant difference among all paired land-cover products is between GLC and LEAF. Other paired differences are not significant for the quadrate size of 30 × 30 km. ANOVA has already shown that there is no significant difference for the quadrate size of 50 × 50 and 100 × 100 km; therefore, it is not necessary to conduct MCP at these two scales.

Single class investigation

For illustrative purposes, Q was also applied to a single class in GLC2000 to investigate the LAI variation within that class. The class selected for this study is croplands (>50%), which is the largest land-cover type and occupies about 19% of the whole study area (Figure 2). It is defined as regions of intensive cultivation and/or sown pasture.

In Figure 6a, Q values of this class calculated at the quadrate size of 30 × 30 km are presented. Map cells in white represent those quadrates not containing any croplands. The mean Q value is 1.287 for this class, which is higher than that for all classes of GLC2000. This is because agricultural fields in this region are usually small and mixed with savannas and fallows, which preclude a reliable mapping. The spatial pattern is very similar to the map in Figure 3a. High Q values (yellow and red) tend to occur in areas with complex landscapes.

A hot spot with high Q was identified in Figure 6a, which is a rectangular area pointed by two dashed lines. Geographically, it is in the Mount Elgon area that straddles the border between Kenya and Uganda. The Q map consists of four nearby quadrates and has size of 60 × 60 km. The mean Q value for this area is 3.265 and is much higher than the total mean value (1.287). The spatial distribution of croplands (>50%) in this hot spot area is also presented (Figure 6b). It occupies 1069 one-km pixels (about 29.7%) originally in GLC2000. To further investigate the cause of high Q, Africover at 100-m resolution was examined at locations indicated as croplands (Figure 6c). Africover, developed by the United Nations FAO, was produced from a visual interpretation of thematic mapper (TM) data from 2000 to 2001 (Jansen and Di Gregorio 2003). It is considered as ground truth here because of much higher resolution and has been found to adequately capture human land uses (Torbick et al. 2006). By comparing with Africover, GLC croplands in this hot spot area are actually composed of up to five diversified land-cover types: forest, open shrubland, closed shrubland, savanna, and crop. Integrating such complex landscape to one single type produced high within-class LAI variation.

In Figure 6a, there are other high Q areas because of high surface heterogeneity. For example, at the coast of Lake Victoria in Uganda an area with 30 × 30 km has a high Q of 6.281. By looking at Africover, croplands in GLC2000 include marshes and shrubs (not shown). Therefore, the precision of classification decreased significantly. This was captured by Q as well.

Discussion

There are several comments to be made on the proper use of the proposed approach. First, the time period of LAI data should be fairly long. Based on quality control flags only high quality LAI pixels were utilized. Thus, not every LAI pixel is valid for Q calculation. If, for example, only a couple of months of LAI are used, there may not be enough LAI pixels to calculate Q in some quadrates, such as where persistent cloud cover could exist for months, especially for tropical areas. As a result, full evaluation could be affected. Second, the usefulness of Q depends on the quality of input LAI data. Two measures were taken to address this issue. One was using monthly LAI data, composited from multitemporal data by selecting high quality pixels over a month period. In this example, the accuracy of 8-day LAI is about 0.5 LAI (Wang et al. 2004). The other was further filtering the LAI pixels according to the quality assessment flags. Only pixels produced by the main algorithm under clear-sky conditions were selected. However, the overall LAI quality at this region is still not well known. Some publications have already shown that the LAI product in itself has limitations (Wang et al. 2004).

In addition, LAI is only one of the biophysical parameters that are important to land–climate modeling. Other parameters such as albedo, VFC, and surface roughness have been shown to be important too. As more products are being developed, the Q approach could be applied to these variables to know more complete biophysical characteristics of land-cover classifications. Simple adjustment, for example, averaging Qs of all parameters, could be made to achieve this goal.

It is difficult to assess land-cover classification using only one measure. Thus, the Q statistic has its limitations. As presented previously, Q only quantifies precision based on the attribute of interest. Accuracy or “correctness” of classification is not addressed. To have a more complete evaluation of land-cover products, the Q statistic may be used together with other methods such as those based on traditional ground sampling. In a statistical context, precision and accuracy are always associated. The accuracy of land-cover classification may affect the value of Q. A land-cover type with low accuracy (or high bias) may have high precision, which means small Q value. Further research is needed to investigate their relationship. Comparing Q and accuracy using simulated ensemble land-cover maps might be a way to address this issue. In addition, Q does not consider the exact structure of biophysical attributes within land-cover types. Similar Q values might have totally different spatial distributions of LAI. Although this may not be an issue for land–climate modeling as biophysical attributes are typically averaged over each land-cover type, this may be important for other applications. Some other statistics such as spatial correlation could be explored to constrain Q.

Conclusions

Land surface changes have been identified as an important driving force of regional and global climate systems. An increasing number of land-cover datasets, including historically reconstructed and future projected, have been and will be used in the modeling of land–climate interactions. The quality of land-cover datasets becomes an important issue. In this study, a statistical measure Q was developed to evaluate the biophysical precision of different land-cover products: GLC2000, MODIS land cover, OGE, and LEAF. Monthly MODIS LAI products were used in the evaluation, as it is crucial to land–climate modeling. Evaluations were conducted at three spatial scales (quadrate sizes): 30 × 30, 50 × 50, and 100 × 100 km. This evaluation found that in terms of Q, GLC2000 ranks the best and LEAF ranks the lowest at every scale. MODIS land cover ranks better than OGE. As quadrate size increases, Q differences between land-cover products tend to decrease. For quadrate size of 30 × 30 km, GLC2000 is significantly better than LEAF. This suggests that the LEAF dataset needs to be updated by GLC2000 for the land–climate modeling in East Africa to better represent the surface condition.

Acknowledgments

We thank two anonymous reviewers, who have helped greatly to improve the quality of this paper. This research has been funded as part of the National Science Foundation Biocomplexity of Coupled Human and Natural Systems Program, Award No. BCS-0308420.

REFERENCES

  • Bonan, G. B., , R. S. DeFries, , M. T. Coe, , and D. S. Ojima. 2004. Land use and climate. Land Change Sciences, G. Gutman, et al., Eds., Kluwer Academic, 301–314.

    • Search Google Scholar
    • Export Citation
  • Brown, J. F., , T. R. Loveland, , D. O. Ohlen, , O. Donald, , and Z. Zhu. 1999. The global land-cover characteristics database: The user’s perspective. Photogramm. Eng. Remote Sens. 65:10691074.

    • Search Google Scholar
    • Export Citation
  • Chen, J. M., and J. Cihlar. 1996. Retrieving leaf area index of boreal conifer forests using Landsat TM images. Remote Sens. Environ. 55:153162.

    • Search Google Scholar
    • Export Citation
  • Cihlar, J. 2000. Land cover mapping of large areas from satellites: Status and research priorities. Int. J. Remote Sens. 21:10931114.

    • Search Google Scholar
    • Export Citation
  • Congalton, R. G. 1991. A review of assessing the accuracy of classification of remotely sensed data. Remote Sens. Environ. 37:3546.

  • DeFries, R., , M. Hansen, , M. Steininger, , R. Dubayah, , R. Sohlberg, , and J. Townshend. 1997. Subpixel forest cover in Central Africa from multisensor, multitemporal data. Remote Sens. Environ. 54:209222.

    • Search Google Scholar
    • Export Citation
  • Dickinson, R. E. 1995. Land-atmosphere interaction. Rev. Geophys. 33:917922.

  • Dickinson, R. E., , A. Henderson-Sellers, , P. J. Kennedy, , and M. F. Wilson. 1986. Biosphere-atmosphere transfer scheme for the NCAR community climate model. NCAR Tech. Note NCAR/TN-275+STR, 69 pp.

    • Search Google Scholar
    • Export Citation
  • Di Gregorio, A., and L. J. M. Jansen. 2000. Land Cover Classification System (LCCS). F.A.O., 179 pp.

  • Feddema, J. J., , K. W. Oleson, , G. B. Bonan, , L. O. Mearns, , L. E. Buja, , G. A. Meehl, , and W. M. Washington. 2005. The importance of land-cover change in simulating future climates. Science 310:16741678.

    • Search Google Scholar
    • Export Citation
  • Foley, J. A., , M. H. Costa, , C. Delire, , N. Ramankutty, , and P. Snyder. 2003. Green surprise? How terrestrial ecosystems could affect earth’s climate. Front. Ecol. Environ. 1:3844.

    • Search Google Scholar
    • Export Citation
  • Foody, G. M. 2002. Status of land cover classification accuracy assessment. Remote Sens. Environ. 80:185201.

  • Friedl, M. A. Coauthors 2002. Global land cover mapping from MODIS: Algorithms and early results. Remote Sens. Environ. 83:287302.

  • Ge, J., , J. Qi, , B. Lofgren, , N. Moore, , N. Torbick, , and J. M. Olson. 2007. Impacts of land use/cover classification accuracy on regional climate simulations. J. Geophys. Res. 112:D05107. doi:10.1029/2006JD007404.

    • Search Google Scholar
    • Export Citation
  • Ge, J., , J. Qi, , and B. Lofgren. 2008. Use of vegetation properties from EOS observations for land-climate modelling in East Africa. J. Geophys. Res. 113:D15101. doi:10.1029/2007JD009628.

    • Search Google Scholar
    • Export Citation
  • Giri, C., , Z. Zhu, , and B. Read. 2005. A comparative analysis of the Global Land Cover 2000 and MODIS land cover data sets. Remote Sens. Environ. 94:123132.

    • Search Google Scholar
    • Export Citation
  • Jansen, L. J. M., and A. Di Gregorio. 2003. Land-use data collection using the “land cover classification system”: Results from a case study in Kenya. Land Use Policy 20:131148.

    • Search Google Scholar
    • Export Citation
  • Lark, R. M. 1995. Components of accuracy of maps with special reference to discriminant analysis on remote sensing data. Int. J. Remote Sens. 16:14611480.

    • Search Google Scholar
    • Export Citation
  • Latifovic, R., and I. Olthof. 2004. Accuracy assessment using sub-pixel fractional error matrices of global land cover products derived from satellite data. Remote Sens. Environ. 90:153165.

    • Search Google Scholar
    • Export Citation
  • Lee, T. J. 1992. The impact of vegetation on the atmospheric boundary layer and convective storms. Atmosphere Science Paper No. 509, Colorado State University, 155 pp.

    • Search Google Scholar
    • Export Citation
  • Loveland, T. R., , Z. Zhu, , D. O. Ohlen, , J. F. Brown, , B. C. Reed, , and L. Yang. 1999. An analysis of the IGBP global land-cover characterization process. Photogramm. Eng. Remote Sens. 65:10211032.

    • Search Google Scholar
    • Export Citation
  • Loveland, T. R., , B. C. Reed, , J. F. Brown, , D. O. Ohlen, , Z. Zhu, , L. Yang, , and J. W. Merchant. 2000. Development of a global land cover characteristics database and IGBP DISCover from 1 km AVHRR data. Int. J. Remote Sens. 21:13031330.

    • Search Google Scholar
    • Export Citation
  • Mathews, E. 1983. Global vegetation and land use: New high resolution data bases for climate studies. J. Climate Appl. Meteor. 22:474487.

    • Search Google Scholar
    • Export Citation
  • Mayaux, P., , E. Bartholome, , S. Fritz, , and A. Belward. 2004. A new land-cover map of Africa for the year 2000. J. Biogeogr. 31:861877.

  • Myneni, R. B. Coauthors 2002. Global products of vegetation leaf area and fraction absorbed PAR from year one of MODIS data. Remote Sens. Environ. 83:214231.

    • Search Google Scholar
    • Export Citation
  • Olson, J. S. 1994a. Global ecosystem framework-definitions. Internal Rep., USGS EROS Data Center, Sioux Falls, SD, 37 pp.

  • Olson, J. S. 1994b. Global ecosystem framework-translation strategy. Internal Rep., USGS EROS Data Center, Sioux Falls, SD, 39 pp.

  • Olson, J. S., , J. A. Watts, , and L. J. Allison. 1983. Carbon in live vegetation of major world ecosystems. Oak Ridge National Laboratory, 51 pp.

    • Search Google Scholar
    • Export Citation
  • Pielke Sr., R. A., , R. Avissar, , M. Raupach, , A. J. Dolman, , X. Zeng, , and A. S. Denning. 1998. Interactions between the atmosphere and terrestrial ecosystems: Influence on weather and climate. Global Change Biol. 4:461475.

    • Search Google Scholar
    • Export Citation
  • Pielke Sr., R. A., , G. Marland, , R. A. Betts, , T. N. Chase, , J. L. Eastman, , J. O. Niles, , D. D. S. Niyogi, , and S. W. Running. 2002. The influence of land-use change and landscape dynamics on the climate system: Relevance to climate-change policy beyond the radiative effect of greenhouse gases. Philos. Trans. Roy. Soc. London A360:17051719.

    • Search Google Scholar
    • Export Citation
  • Running, S. W., , T. R. Loveland, , L. L. Pierce, , R. R. Nemani, , and E. R. Hunt Jr.. 1995. A remote sensing based vegetation classification logic for global land cover analysis. Remote Sens. Environ. 51:3948.

    • Search Google Scholar
    • Export Citation
  • Scheffe, H. 1959. The Analysis of Variance. John Wiley & Sons, 477 pp.

  • Sellers, P. J., , Y. Mintz, , Y. C. Sud, , and A. Dalcher. 1986. A simple biosphere model (SiB) for use within general circulation models. J. Atmos. Sci. 43:505531.

    • Search Google Scholar
    • Export Citation
  • Sellers, P. J. Coauthors 1996a. A revised land surface parameterization (SiB2) for atmospheric GCMs. Part I: Model formulation. J. Climate 9:676705.

    • Search Google Scholar
    • Export Citation
  • Sellers, P. J. Coauthors 1996b. A revised land surface parameterization (SiB2) for atmospheric GCMs. Part II: The generation of global fields of terrestrial biophysical parameters from satellite data. J. Climate 9:706737.

    • Search Google Scholar
    • Export Citation
  • Sellers, P. J. Coauthors 1997. Modelling the exchanges of energy, water, and carbon between continents and the atmosphere. Science 275:502509.

    • Search Google Scholar
    • Export Citation
  • Stapleton, J. H. 1995. Linear Statistical Models. John Wiley & Sons, 444 pp.

  • Torbick, N. M., , D. Lusch, , J. Qi, , N. Moore, , J. Olson, , and J. Ge. 2006. Developing land use land cover parameterization for climate-land modelling in East Africa. Int. J. Remote Sens. 27:42274244.

    • Search Google Scholar
    • Export Citation
  • Walko, R. L. Coauthors 2000. Coupled atmosphere–biophysics–hydrology models for environmental modelling. J. Appl. Meteor. 39:931944.

    • Search Google Scholar
    • Export Citation
  • Wang, Y. Coauthors 2004. Evaluation of the MODIS LAI algorithm at a coniferous forest site in Finland. Remote Sens. Environ. 91:114127.

    • Search Google Scholar
    • Export Citation
  • Zhuang, X., , B. A. Engel, , X. Xiong, , and C. J. Johannsen. 1995. Analysis of classification results of remotely sensed data and evaluation of classification algorithms. Photogramm. Eng. Remote Sens. 61:427433.

    • Search Google Scholar
    • Export Citation

Figure 1.
Figure 1.

A simulated 4 × 4 pixel area. (a) Two classes and (b) three classes. Numbers shown are LAI values.

Citation: Earth Interactions 13, 6; 10.1175/2009EI276.1

Figure 2.
Figure 2.

Study area presented by GLC2000.

Citation: Earth Interactions 13, 6; 10.1175/2009EI276.1

Figure 3.
Figure 3.

The Q maps for four land covers at the quadrate size of 30 × 30 km: (a) GLC2000, (b) MODIS land cover, (c) OGE, and (d) LEAF.

Citation: Earth Interactions 13, 6; 10.1175/2009EI276.1

Figure 4.
Figure 4.

Mean Q for all land covers at three scales.

Citation: Earth Interactions 13, 6; 10.1175/2009EI276.1

Figure 5.
Figure 5.

Multiple comparison (Tukey’s method) results for mean Q values for a quadrate size of 30 × 30 km. MOD means MODIS land cover. The mean Q values of GLC and LEAF are significantly different (GLC LEAF). Confidence intervals were built on significance level of 0.05.

Citation: Earth Interactions 13, 6; 10.1175/2009EI276.1

Figure 6.
Figure 6.

Here Q was applied to a single class, croplands (>50%), in GLC2000. (a) A Q map at 30 × 30 km quadrate size, (b) croplands in the hot spot in (a), pointed by two lines, and (c) Africover corresponding to (b).

Citation: Earth Interactions 13, 6; 10.1175/2009EI276.1

Save