User Guide Outline

User Guide Outline

• 1 Dataset overview
• 1.1 Background and purpose
• 1.2 Crop mask from previous maps of Africa
• 1.3 Summary of approach to creating data
• 1.3.1 Inputs
• 1.3.2 Algorithm
• 1.3.3 Outputs
• 2 Characteristics per Dataset
• 3 FAQ: Data Knowledge
• 3.1 Are there any other considerations to make when working with the data?
• 3.2 What information should be known regarding Quality Assurance of the data?
• 4 Applicable Data Tools
• 5 Contact Information
• 6 Citations

1 Dataset overview

This Global Food Security-support Analysis Data (GFSAD) at nominal 250m result from a collaborative effort by the National Aeronautics and Space Administration (NASA) and the United States Geological Survey (USGS), to provide high resolution global cropland data and their water use that contributes towards global food security in the twenty-first century.

1.1 Background and purpose

The GFSAD30 is a NASA funded project to provide high resolution global cropland data and their water use that contributes towards global food security in the twenty-first century. The GFSAD30 products are derived through multi-sensor remote sensing data (e.g., Landsat, MODIS, AVHRR), secondary data, and field-plot data and aims at documenting cropland dynamics from 1990 to 2017.

Monitoring global croplands is imperative for ensuring sustainable water and food security to the people of the world in the Twenty-first Century. The currently available cropland products suffer from major limitations such as:

• Absence of precise spatial location of the cropped areas; Coarse resolution nature of the map products with significant uncertainties in areas, - locations, and detail;
• Uncertainties in differentiating irrigated areas from rainfed areas;
• Absence of crop types and cropping intensities; and
• Absence of a dedicated webportal for the dissemination of cropland products.

1.2 Crop mask from previous maps of Africa

Ten existing state-of-art studies of African and Global cropland studies (Table 1) were combined to a baseline 250m cropland mask of Africa (Table 1). First, global 1-km cropland extent map (Teluguntla et al., 2015) was used as a starting point because it provides the distribution of the most consistent global cropland extent map based of 4 well known studies (1. Thenkabail et al., 2009b, Biradar et al., 2009, Thenkabail et al., 2011, 2. Pittman et al., 2010, 3. Yu et al., 2013, and 4. Friedl et al 2010). However, it is clear that each of these four studies have their own inconsistencies (Teluguntla et al., 2015). Thereby, we expanded our investigation of the cropland extent, specifically focused on Africa, by incorporating 10 studies (Table 1) to ensure most comprehensive assessment. This included products from 250m (MODIS), 300m (MERIS) and even higher 30m (Landsat). Vector datasets such as Africover are converted into a 250 m resolution raster using the “maximum area” criteria, i.e., the feature with the largest area in the cell yields the attribute assigned to that cell. Some of the products (Globcover, Africover and the MODIS-JRC dataset) have used the Land Cover Classification System (LCCS) while others have not. Therefore, those products that have not adopted the LCCS differ in how they define agriculture (i.e., land cover description, land use intensity) since the aims of these products differ. In order to avoid ignoring any significant cropland areas as mapped by different products, all the significant cropland classes were aggregated, plus the classes that include mosaic vegetation like shrubs and grassland where even a small cropland fractions exist. For those products which provide different levels of agricultural land use intensity (0–100%), a visual analysis of the products in comparison with recent sub-meter to 5 meter very high-resolution imagery (VHRI) was then carried out to delineate croplands from non-croplands.

Table 1. Datasets used in creating 250m cropland mask of Africa in terms of their reference, data source, resolution, and time interval.
Name / Institution / Data Source / Resolution / Time interval / Classes / Reference
Globcover / ESA / MERIS / 300m / 2005, 2009 / LULC / (Arino et al., 2007)
Africover / FAO / Landsat 7 / 30m / 1995-2002 / LULC / (Kalensky, 1998)
LULC 2000 / USGS / AVHRR / 2000m / 2000 / LULC / (Soulard et al., 2014)
GLC 2000 / JRC / SPOT / 1/112° / 2000 / LULC / (Bartholome et al., 2002)
MCD12Q1 / NASA / MODIS / 500m / 2004 - now / LULC / (Leroux, Jolivot, Bégué, Seen, & Zoungrana, 2014)
MODIS-JRC / JRC/MARS / MODIS, Landsat / 250m / 2009 / LULC / (Vancutsem, Marinho, Kayitakire, See, & Fritz, 2012)
GCEV1 / USGS / MODIS, Landsat / 1000m / 2010 / Cropland / (Thenkabail & Teluguntla, 2015)
Global30 / NGCC / Landsat 7 / 30m / 2010 / LULC / (J. Chen et al., 2015)
FROMGC / CESS / Landsat 7 / 30m / Circa 2010 / LULC / (Gong et al., 2013)
GRIPC / BU / MODIS / 500m / Circa 2005 / Cropland / (Salmon, Friedl, Frolking, Wisser, & Douglas, 2015)

1.3 Summary of approach to creating data

1.3.1 Inputs

The MODIS 250m 16-day composite NDVI product (Didan, 2015) were found to have sufficient spatial and temporal resolutions to detect cropland over large area (Wardlow, Egbert, & Kastens, 2007). The year 2014 was chosen as a reference year because most of the ground samples and very high spatial resolution imagery (VHRI) collected in the same year and the annual rainfall in 2014 is also at normal level.

1.3.2 Algorithm

ACM2016 was implemented as following steps:

First, MODIS 250 NDVI imagery composite of every 16-days was stratified by: (1) Mask out the non-crop area using 250m baseline crop mask of Africa, (2). Subset into 8 consolidated FAO agroecological zones (AEZs) and (3) clustered into unique clusters within individual AEZs using K-means method. Second, ground samples from reference samples repository were split into traing part and validation part. The former was used to generate crop signature to identify unique clusters and later was used to validate the identified results, which is called baseline Cropland Layer (RCL). Third, given the accuracy of RCL is satisfactory, the way to carry out classification in individual AEZs was generalized to a replicable decisio-tree, all the trees are ensemble to a complete automated algorithm which don’t need recalibration for entire Africa.

Methodology schematics for developing reference cropland product

Instead of relying on in situ training data, the proposed method privileges the use of established baseline reference cropland layer, since it consists in the best possible land cover information available for each Agro-ecological Zone. The construction of decistion-tree algorithm is a procedure that recursively partitions a dataset into smaller subdivisions on the basis of a set of tests defined at each branch or node in the tree. The tree is composed of a root node (formed from training data), a set of internal nodes (splits), and a set of terminal nodes (leaves). A zonal tree rules are constructed by recursively partitioning the spectral distribution of the training dataset using WEKA (Sharma, Ghosh, & Joshi, 2013) and then expert-timed till the derived cropland product for the year 2014 (ACP2014) accurately matches with RCL2014. In zones where land cover features were misclassified and classification output was considered unsatisfactory, we added training data, redeveloped the decision tree models, and reapplied the models.

We used a decision tree approach to hierarchically classify crop types. The decision tree for each AEZ consisted of three steps: a) seperated using irrigation/rainfed masks, b) fallow cropland identification, c) decision-tree for primary classes in the individual AEZ based on . Fallow croplands were filtered out seperately for irrigation and rainfed: for irrigation area, area whose NDVI value lower than 0.2 in six months of one calendar year being excluded from the cropland area as fallow-land; for rainfed area, pixels whose NDVI falls below a threshold during the peak growing seasons of the crop will be coded as cropland fallows.

Figure 2. Example of ACMA algorithm established for AEZ 3

1.3.3 Outputs

We applied ACM2016 algorithm for the years 2003 through 2014 using MODIS 250 m every 16-day time-series data on Google Earth Engine.The net cropland areas (NCAs) Africa increase by about 10 Mha from 2003 to 2014, going from 250 Mha to 270 Mha and, the gross cropland areas (GCAs) Africa increase by about 10 Mha from 2003 to 2014, going from 310 Mha to 325 Mha. To summerize, the cropland fallows of Africa decrease by about 10 Mha from 2003 to 2014, going from 43 Mha to 30 Mha. From 2003 to 2014 there is 10 Mha in cropland area increase and another 10 Mha of cropland fallow decrease. Roughly an increase of 1 Mha of croplands per year. This can only increase further with rapid increase in population and increasing food and nutritional demands of the populations.

2 Characteristics per Dataset

All 250m GFSAD data for continental Africa are stored in Geotiff, were produced in Geographic projection (WGS84) at a spatial resolution of 0.0022458 degrees (equivalent to 250 m at the equator) as Table 2.

Table 2. Characteristics of Dataset
Characteristics
Collection / global food security support analysis data @ 250m for Africa
Short name / GFSAD250 Africa
Temporal granularity / Yearly
Temporal extent / 2003-2014
Spatial extent / -25.3694999999999986,-34.8302631203000033 63.5053596242299960,37.3494000000000028
Coordinate reference system and datum / +proj=longlat +datum=WGS84 +no_defs
File format / GeoTIFF
Rows/Columns / 38368 x 34699
Number of Science Dataset (SDS) Layers / 4
Pixel size / 0.00224579 degree
Table 3. Descriptions of SDS Layer
Band # / Band Name / Class # / Class Label / Description
1 / Cropland extent
0 / Non-Croplands
1 / Croplands
2 / Watering method
0 / Non-Croplands
1 / Rainfed
2 / Irrigated
3 / Crop intensity
0 / Non-Croplands
1 / Single Crop
2 / Double Crop
3 / Triple Crop
4 / Continuous Crop
4 / Crop dominance / SC: single Crop, DC: double Crop, season 1: Oct - Mar, season 2: May - Sep
1 / Irrigated, SC, season 2 / wheat, barley
2 / Irrigated, SC, season 1 / maize, rice, millet
3 / Irrigated, DC, / rice/chili-vegetable, rice-rice
4 / Irrigated, Continuous / sugarcane, plantation
5 / Rainfed, SC, season 2 / millet, barley, maize, beans, cassava, yam
6 / Rainfed, SC, season 1 / maize, sorghum, tef, wheat, barley, cassava, yam
7 / Rainfed, DC, / rice-rice, maize-maize, rice-beans/potato/chickpea/pulses
8 / Rainfed, Continuous / sugarcane, plantation
9 / Fallow-lands

3 FAQ: Data Knowledge

3.1 Are there any other considerations to make when working with the data?

Products can be used as gridded data for research or application. When compared with areal data from other sources, full pixel areas (FPAs) are not actual areas. The actual areas are equivalent to sub-pixel areas (SPAs) and are calculated by multiplying SPAs with cropland area fractions (CAFs). This is because, a MODIS pixel even when cropped may have different proportion of crop within the pixel.

3.2 What information should be known regarding Quality Assurance of the data?

We will add a ‘QC’ band to indicate the uncertaintity. TBD.

4 Applicable Data Tools

Project Website, Visualization Tools:

5 Contact Information

If you have any questions, feel free to write to

More details about our project and product can be found at: croplands.org

6 Citations

Xiong J., Thenkabail, P.S, Gumma, M.K., Teluguntla P., Poehnelt J., Congaltond R., Yadav K. Automated Cropland Mapping of Continental Africa using Google Earth Engine Cloud Computing (manuscript)

Arino, O., Gross, D., Ranera, F., Leroy, M., Bicheron, P., Brockman, C., … others. (2007). GlobCover: ESA service for global land cover from meris. In 2007 ieee international geoscience and remote sensing symposium (pp. 2412–2415). IEEE.

Bartholome, E., Belward, A., Achard, F., Bartalev, S., Carmona-Moreno, C., Eva, H., … Stibig, H. (2002). GLC 2000: Global land cover mapping for the year 2000. Project Status, November.

Chen, J., Chen, J., Liao, A., Cao, X., Chen, L., Chen, X., … Mills, J. (2015). Global land cover mapping at 30m resolution: A POK-based operational approach. ISPRS Journal of Photogrammetry and Remote Sensing, 103, 7–27.

Didan, K. (2015). MOD13A2 modis/terra vegetation indices 16-day l3 global 1km sin grid v006. NASA EOSDIS Land Processes DAAC. Org/10.5067/MODIS/MOD13A2, 6.

Gong, P., Wang, J., Yu, L., Zhao, Y., Zhao, Y., Liang, L., … Chen, J. (2013). Finer resolution observation and monitoring of global land cover: first mapping results with Landsat TM and ETM+ data. International Journal of Remote Sensing, 34(7), 2607–2654.

Kalensky, Z. (1998). AFRICOVER land cover database and map of africa. Canadian Journal of Remote Sensing, 24(3), 292–297.

Leroux, L., Jolivot, A., Bégué, A., Seen, D. L., & Zoungrana, B. (2014). How reliable is the modis land cover product for crop mapping sub-saharan agricultural landscapes? Remote Sensing, 6(9), 8541–8564.

Salmon, J. M., Friedl, M. A., Frolking, S., Wisser, D., & Douglas, E. M. (2015). Global rain-fed, irrigated, and paddy croplands: A new high resolution map derived from remote sensing, crop inventories and climate data. International Journal of Applied Earth Observation and Geoinformation, 38, 321–334.

Sharma, R., Ghosh, A., & Joshi, P. K. (2013). Decision tree approach for classification of remotely sensed satellite data using open source support. Journal of Earth System Science, (5), 1237–1247.

Soulard, C. E., Acevedo, W., Auch, R. F., Sohl, T. L., Drummond, M. A., Sleeter, B. M., … others. (2014). Land cover trends dataset, 1973–2000. US Geological Survey.

Thenkabail, P. S., & Teluguntla, P. (2015). Global food security analysis data nominal 1km (GFSAD) derived from Remote Sensing in Support of Food Security in the Twenty-first Century: Current Achievements and Future Possibilities. In.

Vancutsem, C., Marinho, E., Kayitakire, F., See, L., & Fritz, S. (2012). Harmonizing and combining existing land cover/land use datasets for cropland area monitoring at the african continental scale. Remote Sensing, 5(1), 19–41.

Wardlow, B. D., Egbert, S. L., & Kastens, J. H. (2007). Analysis of time-series MODIS 250 m vegetation index data for crop classification in the US Central Great Plains. Remote Sensing of Environment.