Global Tree Cover and Biomass Carbon on Agricultural Land: The contribution of agroforestry to global and national carbon budgets

Agroforestry systems and tree cover on agricultural land make an important contribution to climate change mitigation, but are not systematically accounted for in either global carbon budgets or national carbon accounting. This paper assesses the role of trees on agricultural land and their significance for carbon sequestration at a global level, along with recent change trends. Remote sensing data show that in 2010, 43% of all agricultural land globally had at least 10% tree cover and that this has increased by 2% over the previous ten years. Combining geographically and bioclimatically stratified Intergovernmental Panel on Climate Change (IPCC) Tier 1 default estimates of carbon storage with this tree cover analysis, we estimated 45.3 PgC on agricultural land globally, with trees contributing >75%. Between 2000 and 2010 tree cover increased by 3.7%, resulting in an increase of >2 PgC (or 4.6%) of biomass carbon. On average, globally, biomass carbon increased from 20.4 to 21.4 tC ha−1. Regional and country-level variation in stocks and trends were mapped and tabulated globally, and for all countries. Brazil, Indonesia, China and India had the largest increases in biomass carbon stored on agricultural land, while Argentina, Myanmar, and Sierra Leone had the largest decreases.


Supporting Information -Methods:
To quantify estimates of biomass carbon on agricultural land, IPCC Tier 1 default estimates for carbon stored in a variety of land cover types across bioclimatic strata were combined with tree cover estimates based on 250 m resolution MODIS satellite imagery, to provide a global Tier-1 spatial mapping and tabulation, by region and countries, of biomass carbon on agricultural land for the period 2000-2010.
The spatial modeling procedure was developed and implemented in ArcGIS 10.2 (ESRI Inc.) using both ArcAML and Python programming language. Datasets were re-projected to a sinusoidal projection (World Sinusoidal) in order to calculate zonal statistics and carry out areal computations, as it represents area extent accurately across latitudes (i.e., equal-area projection). The cell size for analyses in sinusoidal projection is 1 km 2 ., These datasets are presented in geographic coordinates in the figures, for mapping and presentation purposes.

Assessment of Global Tree Cover on Agricultural Land:
The global geospatial analysis to identify tree cover on agricultural land combined a global assessment of tree cover, based upon a MODIS 250 m resolution satellite remote sensing dataset 31 , with the Global Land Cover 2000 (GLC 2000) land-use classification . Tree cover on agricultural land was identified and results mapped and tabulated; globally, by global region, and by countries. A detailed description of this analysis is available online in a working paper report (Zomer et al, 2014): http://www.worldagroforestry.org/sites/default/files/WP89_text_only.pdf

Geodatasets
The geodatasets used in the analysis are listed below. o

Tree-cover data
The MOD44B MODIS/Terra Vegetation Continuous Fields Dataset (VCF) (Hansen 2003) was developed by the University of Maryland and provides global estimates of vegetation cover in terms of woody vegetation, herbaceous vegetation and bare-ground percentages. The updated MOD44B MODIS VCF -Collection 5 dataset (DiMiceli et al 2011) used in the current analysis improves upon the earlier versions and provides data at the resolution of 250 m. A limited amount of validation performed using field data from two sites in Maryland and three sites in Brazil, South America show that the Collection 5VCF product is substantially more accurate then previous versions, with accuracy significantly improved within agricultural areas and forest clearings. (User Guide for VCF Collection 5 -Version 1). This data (and a User Guide for VCF Collection 5 -Version 1) is available online at: http://www.landcover.org/data/vcf/

Land-cover categories
Three agricultural land-use types from the Global Land Cover Class scheme used for the Global Land Cover 2000 database were selected as relevant for the specific objectives of this work: • Cultivated and Managed Areas (agriculture -intensive), • Cropland/Other Natural Vegetation (non-trees: mosaic agriculture/degraded vegetation) • Cropland/Tree Cover Mosaic (agriculture/degraded forest).
Although at first the Cropland/Tree Cover Mosaic type seems to identify agroforestry systems, the mix of forest and agriculture does not occur at discrete intervals but is a gradient where the two components of landscape-level agroforestry mix within the landscape. The mix of tree cover over agriculture land is depicted along a continuous gradient by the MODIS VCF tree-cover dataset, within the relevant GLC2000 land-cover type. Tree cover shows the percentage of the 1 km 2 grid cell occupied by trees, therefore, at this resolution of 1000 m 2 , the tree-cover percentage can be expressed as hectares (ha) of tree cover per km 2 . At 100% tree cover, the whole grid cell is occupied, that is, 100 ha/km 2 .
The Global Land Cover 2000 database is available online here: http://www.gvm.jrc.it/glc2000

Administrative boundaries
The GADM database of Global Administrative Areas was used to define both regional and country boundaries. The GADM database is available online here: http://www.gadm.org

Aridity-Wetness Index
A global model of aridity (Zomer et al. 2007) was used to stratify ecological conditions based on climatic and agro-ecological characteristics. Aridity is expressed as a function of precipitation, temperature and potential evapotranspiration (PET). Based upon an attempt to classify climatic zones by moisture regime, the Aridity-Wetness Index (AWI) quantifies precipitation deficit over atmospheric water demand as: • Aridity-Wetness Index (AWI) = MAP / MAE] where: o MAP = mean annual precipitation o MAE = mean annual evapotranspiration The AWI dataset is available online here: http://csi.cgiar.org/aridity/

Global Tree Cover on Agricultural Land
To facilitate the global analysis, the VCF Tree Cover -Collection 5 dataset (250 m resolution) grid cells were aggregated to 1 km 2 resolution. All the geodatasets were masked to exclude areas which are either non-agricultural land-use types or urban areas.
Tree-canopy cover on agricultural land has been tabulated for all years available in the VCF-C5 dataset, that is, from 2000 to 2010. Variation in the estimates from year to year appears to be high and not consistent with the expected year-to-year change. There is a fair amount of 'noise' in the year-to-year estimates, which can be expected from having a significant variability associated with the quality of the remote-sensing dataset and seasonal and other confounding factors affecting the classification algorithm used in the VCF-C5 processing. In order to reduce the effect of this variability in estimates of change during the period, we have averaged the first three years of the dataset (2000)(2001)(2002) and the last three years (2008)(2009)(2010) and use these averaged results as the beginning and end points for the change analysis. They are further referred to within the text as 2000 and 2010, respectively, to simply presentation of the results. . The extent of agricultural land and various associated tree-canopy cover values have been analyzed, compared, mapped and tabulated globally, and by global regions, countries, and aridity-wetness index zones. Within each stratum, or within specific aggregation of strata, zonal statistical values (mean, sum, total area, percentiles, areal distribution etc) were summarized to describe those factors of interest for this study.
Cumulative agricultural area is presented at decreasing tree-canopy cover to infer at global and subcontinent scales the total area engaged above any specific tree-canopy cover values. In a second stage, the same cumulative distribution of total agricultural land in function of tree-canopy cover has been disaggregated for five different aridity classes (AWI < 0.45 or arid, 0.45 < AWI < 0.6 or semi-arid, 0.6 < AWI < 0.8 or subhumid, 0.8 < AWI < 1.0 or humid, AWI >1.0 or very humid) to show how climate regimes might differentiate specific patterns of interdependence between tree-canopy cover and bioclimatic conditions for different geographical areas. , based on FAO ecofloristic zones, and which continent that zone is found. In each of those "carbon zones" a carbon value has been calculated for each GLC_2000 landuse class in that zone. These values are available in tables, and apply across the whole of each carbon zone.
The authors state that this "… spatial database is likely the best available, globallyconsistent map depicting vegetation carbon stocks, circa 2000". It is based on widely accepted IPCC methods for estimating carbon stocks at the national level. However, the methods employed were not directly linked to ground-based measures of carbon stocks and have not been validated with field data. It is noted that croplands received the same carbon stock value regardless of the type of crop that might be growing.
To construct the Global Biomass Carbon Map, Ruesch and Gibbs (2008) used the IPCC GPG Tier-1 method for estimating vegetation carbon stocks using the globally consistent default values provided for aboveground biomass (IPCC 2006). Belowground biomass (root) carbon stocks were added using the IPCC root to shoot ratios for each vegetation type, and then total living vegetation biomass was converted to carbon stocks using the carbon fraction for each vegetation type ( which varies between forests, shrublands and grasslands). All estimates and conversions were specific to each continent, ecoregion and vegetation type (stratified by age of forest). Thus, a total of 124 carbon zones or regions, each with a unique carbon stock value for each of the GLC_2000 Landcover Classes found in that zone, were delineated, based on the IPCC Tier-1 methods and default values.

Deriving the Global Tier 1 Estimates of Biomass Carbon on Agricultural Land:
The In order to account for the added contribution of tree cover on agricultural land, we use the default Tier 1 biomass carbon value for agricultural land (5tC/ha) as the baseline value, i.e. at 0% treecover the biomass carbon is 5tC/ha (in all carbon zones).
We use the biomass carbon value of the GLC_2000 Mixed Forest class (or similar class in case this class is not present) in that same carbon zone as a surrogate biomass carbon value where there is full tree cover on agricultural land (i.e. tree cover percentage = 100). We then assume a linear increase in biomass carbon from 0 to 100 percent tree cover where, within a specific grid cell in a specific carbon zone: • Biomass carbon is equal to the default tier 1 value for agricultural land (5 tC/ha) when there are no trees on that land, o (i.e. tree cover = 0%) • There is an incremental linear increase of tC/ha proportionally as tree cover increases up to the maximum value for Mixed Forest in that specific carbon zone, o (i.e. biomass carbon values on agricultural land with 100% tree cover are equal to the related Mixed Forest class.) Results were tabulated and mapped globally, by region, and by country. Spatial datasets resulting from this global analysis of carbon biomass on agricultural land are available online at: http://www.worldagroforestry.org/global-tree-cover/index.html Figure S1: Average biomass carbon per hectare (tC ha -1 ) shown by the cumulative area of agricultural land (km 2 ) with that amount of average biomass carbon, and as a percent (%) of total agricultural land globally. About 79% (17.5 million km 2 ) had <25 t C ha -1 , and 53% (5.7 million km 2 ) had <10 t C ha -1 .