INVESTIGATORS: Name and Title

Dr. Jonathan A. Foley
Center for Sustainability and the Global Environment
University of Wisconsin-Madison, U.S.A.

Title of Investigation: Global land use data reconstruction. Contacts (For Data Production Information):

2.4 Requested Form of Acknowledgment

Please cite the following publication when this data is used:

Ramankutty, N., and J.A. Foley (1999). Estimating historical changes in global land cover: croplands from 1700 to 1992. Global Biogeochemical Cycles 13(4), 997-1027.

3. INTRODUCTION

3.1 Objective/Purpose

The data set was developed to understand the consequences of historical changes in land use and land cover for ecosystem goods and services. In particular, this data set can be used to study how global changes in cultivated area has influenced climate, biogeochemical cycles, biodiversity, etc. This data set can be used directly within spatially-explicit climate and biogeochemical models.

3.2 Summary of Parameters

This is a gridded data set describing the fraction of each grid cell in the globe that is occupied by cultivated land from 1700 to 1992

3.3 Discussion

There are two sources of global land cover/land use data. The most recent estimates are derived from satellite measurements, and are available in a spatially-explicit fashion for roughly the last 30 years. The other estimate is based on ground-based sources such as census statistics, land surveys, estimates by historical geographers, etc. These land inventory data are only available at the scale of political units, but have the advantage of being historical.

Ramankutty and Foley (1998) derived a spatially-explicit data set of croplands in 1992 by synthesizing remotely-sensed land cover data with contemporary land inventory data. Furthermore, Ramankutty and Foley (1999) extended this data set into the past (back to 1700) using historical land inventory data.

The data set should only be used for continental-to-global scale analysis and modeling. The data set captures the broad patterns of cropland change over history, but not necessarily the fine details at local to regional scales please check the data quality before using it at fine spatial scales. The quality of historical data for the Russian Federation is poor. The quality of data prior to 1850 is poor -- only continental-scale historical data were used for that period.

4. THEORY OF ALGORITHM/MEASUREMENTS

In the Seasonal Land Cover Regions (SLCR) classification scheme of the Loveland et al. (2000) data set, there are many classes that have some degree of crop cover. For example, of the 205 SLCR classes for North America, 64 include some form of croplands. In this study, we regrouped the SLCR legend into six categories (or labels) according to their degree of crop cover: (0) other vegetation, (1) other vegetation with crops, (2) other vegetation/crop mosaic, (3) crop/other vegetation mosaic, (4) crops with other vegetation, and (5) crops. Here other vegetation denotes all types of land cover other than croplands, including natural vegetation and other types of land use such as pastures and shifting cultivation. These groups arose naturally from the SLCR legend for all continents except the Australia-Pacific region; for the latter we had to use some subjective judgment.

Reclassifying the SLCR legend in this way gave us a 1 km resolution global map with six categories indicating the relative density of crop cover. However, this map is still qualitative. For instance, crop/other vegetation mosaic indicates that there are roughly equal amounts of crops and other vegetation in that pixel, but no information is available about the actual fractional extent of crop cover. To determine this information, we used statistical inventory data on global land cover. Land cover inventory data are available at the national and subnational level (or political unit level) from various international and national organizations, such as the Food and Agricultural Organization (FAO) and the United States Department of Agriculture (USDA). In this analysis, we use cropland inventory data at the national level for various countries available from the FAO, supplemented with data at the state/province/region level for the United States of America, Canada, Mexico, Brazil, Argentina, India, China, and Australia. A more complete collection of agricultural inventory data, including subnational data for all the large countries of the world, would be ideal. However, to our knowledge, no such data set has been compiled at this time.

To combine the inventory data sets with the 1 km land cover maps, we calibrated the amount of fractional crop cover implied by the six labels to the fractional cover implied by the inventory data for each political unit. In this procedure, we assumed that each of the six labels could have any fractional crop cover value ranging from 0.05 to 1.0, except for the label crops which was assigned a value of 1.0 and the label other vegetation which was assigned a value of 0.0. Although we allowed labels to imply the same fractional cover values, we did not allow vegetation-dominant labels to have higher fractions of crops than crop-dominated labels. For instance, other vegetation with crops could not have a higher fractional crop cover than crop/other vegetation mosaic.

To perform the calibration of fractional crop cover values for the six labels, we evaluated all possible combinations of fractional cover (in intervals of 0.05) and calculated the aggregate area of croplands over each political unit for which inventory data are available. The total amount of cropland for each political unit implied by the satellite classification was compared to agricultural inventory data using a simple linear regression. We selected the set of fractional crop cover values that yielded the best correlation coefficient when restricting the slopes to range from 0.9 to 1.1. (Nigeria was a significant outlier in Africa, and we left it out of the calibration. In Eurasia, the Russian Federation, which has a crop area an order of magnitude larger than the other countries, dominated the solution. Hence we left out Russia during the calibration.) Table 2 and Figure 1 in Ramankutty and Foley (1998) show the results of the calibration procedure. The calibration procedure was performed separately for each continent because the original satellite-based land cover data was produced separately for each continent. Finally, we aggregated the 1 km resolution fractional cover maps into a 5 min resolution grid using a simple area-weighted averaging procedure.

To estimate historical crop cover change, we first compiled an extensive database of historical croplands at the national and subnational (state, province, etc.) level (henceforth referred to as political unit level) (see Appendix A of Ramankutty and Foley (1999) for details). The data was collected for 339 countries at the national level and for 8 countries at the subnational level, consistent with present-day political boundaries. With the exception of the Russian Federation, we have subnational information for most of the large countries, or countries with extensive croplands. The inventory data is obtained at 5-10 year intervals in the best situation, and often at much wider time intervals. We linearly interpolate in between data to obtain annual values. Often the data need adjustments for consistency, and these are described in Appendix A of Ramankutty and Foley (1999).

The 1992 croplands data set was then used as an initial condition for a simple land cover change model, which runs backward in time-generating historical land cover maps, using the historical crop inventory data as a constraint for each political unit (see figure in section 9.4). In other words, the land cover change model is merely a simple algorithm for spatially distributing the historical cropland inventory data within each political unit. However, there is insufficient global scale data of historical land use and land cover change to calibrate an elaborate model, much less to validate it. Hence, in our judgment, the simplest possible approach is most appropriate (employing the principle of Occam's razor). Our basic assumption was that within each political unit, the cropland pattern of 1992 represents the historical spatial patterns. The historical inventory data provides the temporal information needed to describe the differences in cropland area among the political units. Within each political unit, for each year in the past, we adjusted the spatial crop cover pattern of 1992 so that the cropland total for that unit matches the historical inventory data. This assumption will cause problems in large countries with no subnational information, but for several large countries (with the exception of the Russian Federation), we have subnational cropland inventory. The model simulations begin with the initial conditions for 1992 and simulate the crop cover backward in time, annually, until 1700. The reconstruction procedure is outlined in more detail in Appendix B of Ramankutty and Foley (1999).

5. EQUIPMENT

5.1 Instrument Description

The croplands data set used the 1 km resolution land cover classification data developed at the EROS data center (Loveland et al., 2000). The base satellite data for the 1 km land cover product was monthly composites of the AVHRR NDVI measurements over the April 1992-March 1993 period.

5.1.1 Platform (Satellite, Aircraft, Ground, Person)

Not applicable to this data set.

5.1.2 Mission Objectives

Not applicable to this data set.

5.1.3 Key Variables

Not applicable to this data set.

5.1.4 Principles of Operation

Not applicable to this data set.

5.1.5 Instrument Measurement Geometry

Not applicable to this data set.

5.1.6 Manufacturer of Instrument

Not applicable to this data set.

5.2 Calibration

5.2.1 Specifications

5.2.1.1 Tolerance

Not applicable.

5.2.2 Frequency of Calibration

Not applicable.

5.2.3 Other Calibration Information

Not applicable.

6. PROCEDURE

6.1 Data Acquisition Methods

The satellite-derived land cover classification data set was obtained using the world wide web from the following web site http://edcdaac.usgs.gov/glcc/glcc.html. The historical census data were obtained from various sources world wide web, published census reports, publications by historical geographers, etc. The various sources are listed in Appendix A of the publication, Ramankutty and Foley (1999).

6.2 Spatial Characteristics

6.2.1 Spatial Coverage

The data coverage is global.

6.2.2 Spatial Resolution

The data are provided in a equal-angle latitude/longitude grid with a resolution of 0.5 X 0.5 degree.

6.3 Temporal Characteristics

6.3.1 Temporal Coverage

From 1700 to 1992.

6.3.2 Temporal Resolution

Every 50 years from 1700 to 1850, and every 10 years thereafter.

7. OBSERVATIONS

7.1 Field Notes

Not applicable to this data set.

8. DATA DESCRIPTION

8.1 Table Definition with Comments

8.2 Type of Data

8.3 Sample Data Record

Not available at this revision.

8.4 Data Format

ASCII.

8.5 Related Data Sets

Another historical land cover/land use data set has been developed by Klein Goldewijk (2000).

9. DATA MANIPULATIONS

9.1 Formulas

9.1.1 Derivation Techniques/Algorithms

Ramankutty and Foley (1998) derived a global crop cover map for 1992 by calibrating the cropland categories of the 1km-resolution land cover data set of Loveland et al. (2000) against cropland inventory data for 1992. Ramankutty and Foley (1999) derived a historical croplands data set from 1700 to 1992, by extrapolating the 1992 croplands data into the past using historical cropland inventory data (see 9.2.1 for more details).

9.2 Data Processing Sequence

9.2.1 Processing Steps and Data Sets

See section 4 above, Theory of algorithm/measurements, and appendices of Ramankutty and Foley (1999).

9.2.2 Processing Changes

Not applicable to this data set.

9.3 Calculations

Not applicable to this data set.

9.3.1 Special Corrections/Adjustments

9.4 Graphs and Plots

10. ERRORS

10.1 Sources of Error

Errors in both the satellite-derived land cover classification data set, and errors/biases in the cropland inventory data will affect the quality of this historical croplands data set.

10.2 Quality Assessment

A systematic quality assessment has not been performed.

10.2.1 Data Validation by Source

10.2.2 Confidence Level/Accuracy Judgment

10.2.3 Measurement Error for Parameters and Variables

10.2.4 Additional Quality Assessment Applied

11. NOTES

11.1 Known Problems with the Data

In particular, we are aware of errors in the inventory data for China, Nigeria, and the Former Soviet Union that are yet to be resolved. Lack of subnational inventory data for the Russian Federation affects the quality of this data set in that region. The presence of extensive cultivation in Patagonia, and strips of cultivation in the Sahara likely reflects misclassifications in the satellite data.

11.2 Usage Guidance

Use only for continental-to-global scale studies over the timescale of decades to centuries.

11.3 Other Relevant Information

12. REFERENCES

12.1 Satellite/Instrument/Data Processing Documentation

Not applicable to this data set.

12.2 Journal Articles and Study Reports

Klein Goldewijk, K., Estimating global land use change over the past 300 years: the HYDE database, Global Biogeochemical Cycles, Vol 15 (2), 417-434, 2000.

Loveland, T.R., B.C. Reed, J.F. Brown, D.O. Ohlen, J. Zhu, L. Yang, and J.W. Merchant, Development of a Global Land Cover Characteristics Database and IGBP DISCover from 1-km AVHRR Data, International Journal of Remote Sensing, 21 (no. 6/7), 1303-1330, 2000.

Ramankutty, N., and J.A. Foley, Characterizing patterns of global land use: An analysis of global croplands data, Global Biogeochemical Cycles, 12, 667-685, 1998.

Ramankutty, N., and J.A. Foley, Estimating historical changes in global land cover: Croplands from 1700 to 1992, Global Biogeochemical Cycles, 13, 997-1027, 1999.

13. DATA ACCESS

13.1 Contacts for Archive/Data Access Information

13.2 Archive Identification

13.3 Procedures for Obtaining Data

13.4 Archive/Status/Plans

14. OUTPUT PRODUCTS AND AVAILABILITY

14.1 Tape Products

14.2 Film Products

14.3 Other Products

15. GLOSSARY OF ACRONYMS

AVHRR Advanced Very High Resolution Radiometer

NDVI Normalized Diffrence Vegetation Index

SLCR Seasonal Land Cover Regions