Logo of nihpaAbout Author manuscriptsSubmit a manuscriptHHS Public Access; Author Manuscript; Accepted for publication in peer reviewed journal;
Remote Sens Environ. Author manuscript; available in PMC 2013 Jun 1.
Published in final edited form as:
PMCID: PMC3298121

Identification of “ever-cropped” land (1984–2010) using Landsat annual maximum NDVI image composites: Southwestern Kansas case study


A time series of 230 intra- and inter-annual Landsat Thematic Mapper images was used to identify land that was ever cropped during the years 1984 through 2010 for a five county region in southwestern Kansas. Annual maximum Normalized Difference Vegetation Index (NDVI) image composites (NDVIann-max) were used to evaluate the inter-annual dynamics of cropped and non-cropped land. Three feature images were derived from the 27-year NDVIann-max image time series and used in the classification: 1) maximum NDVI value that occurred over the entire 27 year time span (NDVImax), 2) standard deviation of the annual maximum NDVI values for all years (NDVIsd), and 3) standard deviation of the annual maximum NDVI values for years 1984–1986 (NDVIsd84-86) to improve Conservation Reserve Program land discrimination.

Results of the classification were compared to three reference data sets: County-level USDA Census records (1982–2007) and two digital land cover maps (Kansas 2005 and USGS Trends Program maps (1986–2000)). Area of ever-cropped land for the five counties was on average 11.8 % higher than the area estimated from Census records. Overall agreement between the ever-cropped land map and the 2005 Kansas map was 91.9% and 97.2% for the Trends maps. Converting the intra-annual Landsat data set to a single annual maximum NDVI image composite considerably reduced the data set size, eliminated clouds and cloud-shadow affects, yet maintained information important for discriminating cropped land. Our results suggest that Landsat annual maximum NDVI image composites will be useful for characterizing land use and land cover change for many applications.

Keywords: Landsat, NDVI, land use change detection, Conservation Reserve Program (CRP)

1. Introduction

The proportion of cropland in the Great Plains of the United States has remained relatively stable since the mid-twentieth century (Cunfer, 2005). Various crop and conservation measures put in place during the Great Depression, and resurrected after 1945, sought limits to land cultivation. Policies were intended both to retire marginal crop land and to reduce surpluses in principal crops like winter wheat (Gardner, 2002; Kirkendall, 1966; Serra et al., 2009; Sullivan, 2004). More recently, growing awareness of the benefits of biodiversity preservation and restoration have also contributed to the design of conservation programs. In general, conservation policies have largely succeeded in restraining production to levels recorded at mid-century, or even reducing them in certain regions (Lubowski et al 2006). Nevertheless the picture of relative stability is complicated by known patterns of shifting cultivation. Never-cropped grasslands are occasionally brought into cultivation and subsequently idled. The Conservation Reserve Program (CRP) has likely amplified this practice since 1985 (Wu, 2000; Wu & Lin, 2010), increasing encroachment on native grassland, as farms respond to short-term market forces.

The biophysical implications of patterns of prior cultivation are important for understanding the health of ecosystem services. Single year assessments offer a reliable assessment of current allocations of cropland, from a compliance perspective, but do not reflect the cumulative levels of disturbance. Only with a multi-year assessment of trends in disturbance, deep time series – of habitat loss and fragmentation, greenhouse gas emissions, soil and water integrity – can move beyond the kind of aggregate estimates of land cultivation that inform global change analysis (Licker et al., 2010; Ramankutty et al., 2008). The physical extent of native habitat is likely much smaller than the conventional, time-specific, monitoring recognizes. The highly resolved estimates provided in this article suggest that the cumulative magnitudes of past disturbance are higher than previously recognized.

Our goal was to identify land that was cropped one or more years during the 27-year time span 1984 through 2010 (“ever-cropped” land) using a time series of intra- and inter-annual Landsat satellite imagery. Use of Landsat time series of this size has been impeded in the past by multiple challenges including imagery costs, accessibility, quality, geo-registration, cloud cover, and hardware and software tools necessary for managing, processing, and analyzing these tremendously large multi-temporal multi-resolution datasets. Advances in computer technology and the new USGS policy of providing no-cost, high-quality, geo-referenced Landsat data offers new opportunities to explore land cover change at unprecedented spatial and temporal scales.

This paper presents a new method to identify ever-cropped land using intra- and inter-annual Landsat image time series. We developed the method using 230 intra- and inter-annual Landsat TM images (Path 30 Row 34) from 1984 to 2010, the most comprehensive Landsat image time series used in a land cover classification effort that the authors are aware of. The method was developed and applied to a five-county region in southwestern Kansas and the results compared to three reference data sets: County-level U.S. Agricultural Census records, 2005 Kansas land use map, and U.S. Geological Survey Trends Project maps.

2. Methods

2.1 Study area

The study area covers five counties (Ford, Gray, Haskell, Seward, and Stevens) in southwestern Kansas (10,145 km2) centering on latitude 37.534N longitude 100.583W. Average annual precipitation (1971–2000) in this region ranges from 460 mm in the west to 610 mm in the east. The natural vegetation is primarily prairie grassland where short grass prairie (dominant genera: grama (Bouteloua), buffalograss (Buchloë)) and sandsage prairie (Andropogon, Artemisia, Calamovilfa) dominates the western region (Stevens, Seward, Haskell and western Gray Counties) and tall grass prairie (bluestem prairie (Andropogon, Panicum, Sorghastrum)) with scattered patches of sandsage prairie along river banks dominates the eastern region (eastern Gray and Ford Counties) (Küchler, 1974). At the beginning of the century, only 37 percent of the land area in the five county region (10,008 km2) was reported as land in farms and a mere 7.3 percent of the total land area was reported as ‘improved’ in the 1900 census of agriculture (Gutmann, 2005). A little over a generation later, these landscapes were transformed unrecognizably. Nearly 73.5 percent of the total land area in the five counties was reported as cropland in the 1925 census (Gutmann, 2005). The proportions have remained very similar since. Reported cropland reached a peaked of 79.3 percent of the total land area in 1940, and then fluctuated for the remainder of the century between 75 and 80 percent of total land area. By 2005, farm land comprised nearly all of the land area in the region, and land in farms remains primarily cropped land (79.7%, including CRP land) with grassland dominating the remaining land (19.3%) (Peterson et al., 2008). The major crops grown in this region since the 1980’s have been winter wheat (grown as crop/fallow annual rotation), corn, and sorghum (USDA NASS, 2011).

2.2 Landsat image selection and pre-processing

Landsat Thematic Mapper (TM) images (Path 30 Row 34) between March 1 (DOY 60) and October 31 (DOY 304) for all years 1984 through 2010 with a quality flag of nine and at least partially cloud-free were downloaded from the U.S. Geological Survey (USGS) Global Visualization Viewer (GloVis) website (http://glovis.usgs.gov/; accessed January 2011). 239 images constituted the original data set with a median of 9 images per year (range 2 to 12 images per year). Bands 3 (red visible) and 4 (near-infrared) were radiometrically corrected to at-sensor reflectance to correct for seasonal reflectance variances caused by sun angle and distance (Chander et al., 2009) and reprojected to Albers Conical Equal Area using cubic convolution resampling at 30 meter spatial resolution. The Normalized Difference Vegetation Index (NDVI) was calculated for each image date ((B4-B3)/(B4+B3)). The NDVI, originally developed by Rouse et al. (1973), represents a measure of canopy ‘greenness’ where values below 0.2 are generally non-vegetated surfaces such as bare soil, clouds, cloud shadows, and dense green vegetation canopies are generally greater than 0.6 (Gamon et al., 1995; Wang et al., 2005).

NDVI images for all dates within a given year were layer stacked into a single file (i.e., all 1984 NDVI images in one file, all 1985 NDVI images in a second file, etc.). Each of the annual NDVI image stacks was then visually inspected using a rotation of color-band combinations to identify images that were misregistered. Nine of the 239 Landsat images were found to be mis-registered and removed from the annual NDVI image stacks resulting in a final count of 230 images used in the analysis. An annual maximum NDVI image (NDVIann-max) was then generated for each year by taking the maximum NDVI value that occurred in the time series. Since clouds and cloud shadows have generally low NDVI values (NDVI < 0.2), selection of the maximum NDVI value minimized cloud/cloud shadow features (Figure 1).

Figure 1
Comparison of 1) Landsat TM natural color images (left), 2) the corresponding NDVI images (right), and 3) the maximum 2008 NDVI image (lower right). NDVI values for clouds and cloud shadows are relatively low (dark: tones) as compared to green vegetation. ...

2.3 Characteristics of ever cropped and never cropped lands using annual maximum NDVI values

The NDVI annual maximum composite images were then stacked into a single 27-layer file and characteristics of land cover change evaluated using time series plots for selected sample sites (Figure 2). Initial exploratory analysis of the 27-year maximum annual NDVI (NDVIann-max) time series revealed land that was never cropped had relatively stable NDVIann-max values over the entire 27-year period where values typically ranged from 0.3 to 0.6 (Figure 2, lower right). In contrast, cropped land generally had very high NDVI values (> 0.7) at least once over the 27-year period (Figure 2, upper right and lower left), and/or the maximum annual NDVI values were highly variable from year-to-year (e.g., annual crop-fallow rotations typical for winter wheat) (Figure 2, upper right). High variability in NDVIann-max values during the 1980’s (NDVIsd84-86), were also typical for land that had been converted from cropped land to grassland (CRP land) (Figure 2, upper left). Low NDVIsd values represented land where peak annual greenness values were fairly stable throughout the 27-year period, such as for continuously irrigated cropped land or never cropped land (Figure 2, lower left and lower right).

Figure 2
Landsat color composite image of maximum 27-year NDVI (NDVImax) value and standard deviation of annual maximum NDVI (NDVIsd) value with time series graphs showing the annual maximum NDVIann-max values for four land cover types. Dark green tones are typical ...

2.4 Classification

Bivariate density-sliced histogram plots were used to explore and identify parameters that would maximize the separation of ever-cropped land from never-cropped land (Figure 3). Our initial analysis revealed that two parameters would provide good discrimination between ever- and never-cropped land: 1) the maximum 27-year NDVI value (NDVImax) and 2) the standard deviation of NDVIann-max (NDVIsd). A threshold classification method was selected because of the high distinction between ever-cropped land and never-cropped land readily visible in the bivariate histograms. Initial tests revealed that better discrimination was achieved by setting the thresholds using bivariate histograms at the county-level as opposed to setting one single threshold for the entire study area (Figure 3). Threshold values ranged from 0.61 to 0.71 for NDVImax and ranged from 0.074 to 0.096 for NDVIsd (Table 1).

Figure 3
Bivariate density-sliced histogram plots of the maximum 27-year (1984–2010) NDVI value [NDVImax, x-axis) and the standard deviation of the annual maximum NDVI values (NDVIsd, y-axis) for all counties combined and each individual county. Red colors ...
Table 1
Threshold values for NDVImax, NDVIsd, and NDVIsd84-86 used in the classification of each county in the study area.

Preliminary classification results revealed that using only the NDVImax and NDVIsd parameters was not effectively identifying CRP lands. CRP land is important for our study as it represents land that was originally cropped but was converted at some point to grassland. CRP land is land that the U.S. government pays land owners not to plant crops on for a period of 10 to 15 years in an effort to improve environmental quality, such as to reduce soil erosion and improve water quality and wildlife habitat. The Food Security Act of 1985 established the current CRP. The largest increase in CRP land enrollments occurred in the first few years of the program (1986 to 1990) in Kansas (Figure 4) (USDA, 2011). As of 2010, 1.13 million hectares are enrolled the CRP in Kansas with the largest proportion occurring in western Kansas. 10.0% of land in our five county study area was classified as CRP (Peterson et al. 2008). Many of these CRP fields were cropped for only the first few years during the entire 27-year period. High inter-annual variability of the NDVIann-max values was typical of these fields during the years 1984–1986, indicating the land was primarily crop/fallow annual rotation. We incorporated a third parameter (NDVIsd84-86) representing the variability of maximum annual NDVI values for the years 1984–1986. Threshold values for the NDVIsd84-86 parameter were set to the same threshold values applied for NDVIsd.

Figure 4
Cumulative Conservation Reserve Program (CRP) land 1986 through 2010 for Kansas (left) and for each county in the study area (right).

Classification was performed using an object-based image classification approach (Definiens AG, 2009). Five layers were used in the classification process: the 2001 U.S. Geological Survey (USGS) National Land Cover Dataset (NLCD) general land cover map, a county boundary file, and the three images for the parameters NDVImax, NDVIsd, and NDVIsd84-86. The NLCD c2001 land cover map was used to mask out built environments (e.g., urban, roads), forested land, and water. Land classified as barren, shrubland, grassland, hay/pasture, or cropped land in the NLCD map (NLCD codes = 52, 71, 72, 81, 82 respectively) was reclassified into either “ever-cropped” or “never-cropped” land based on the NDVImax, NDVIsd, and NDVIsd84-86 threshold values (Table 1). The images were first segmented using parameters set at scale 30, shape 0.2 and compactness 0.8. A rule-based classification approach was then applied using the threshold values. The first rule delineated the region in the feature space images that was primarily never-cropped land using the NDVImax and NDVIsd parameters. For example, regions in Ford County where NDVImax values were less than 0.71 and NDVIsd values were less than 0.096 were identified as never-cropped land and all other regions were classified as ever-cropped land. A second rule was applied using the NDVIsd84-86 parameter to re-classify the areas likely to be CRP land into ever-cropped land. In Ford County, NDVIsd84-86 values greater than 0.096 would be re-classified from never-cropped to ever-cropped land.

2.5 Comparison to reference data

Results were compared to three reference data sets: County-level U.S. Agricultural Census records, 2005 Kansas land use map, and USGS Trends Project maps. USDA Agricultural Census records were used to derive total area cropped for the years 1982, 1987, 1992, 1997, 2002, and 2007 within each county (Gutmann, 2005). Maximum area ever cropped within each county was determined by summing the area of cropland harvested, failed cropland, and fallow or idle cropland for each of the census years and then determining the maximum value that occurred over the time span.

A digital land cover map produced by the Kansas Applied Remote Sensing Program (KARS) of the Kansas Biological Survey (Peterson et al. 2008) was compared to our results at the pixel level. The 2005 Kansas map was generated using multi-temporal Landsat TM imagery from 2004 and 2005 with an overall accuracy of 90.7% when compared to ground reference data (interpreted aerial photographs and ancillary databases). Comparison of our results to a map generated with circa 2005 imagery has obvious limitations due to land use change. However, land that was classified as cropped land or CRP land in the Kansas 2005 map should be classified as ever-cropped land in our map. The 2005 Kansas map consisted of eleven land use/land cover classes. Land classified as cropland (69.6%), grassland (19.3%), and CRP (10.0%) land were evaluated individually with the remaining classes (e.g., urban woodland, water) lumped into an “other” class for comparison to our results.

We also compared our results to the USGS Trends Project maps at the pixel level. The Trends Project produced land use change maps for sample blocks (10 km by 10 km) spanning across the US using Landsat imagery and aerial photographs for target years 1986, 1993, and 2000 (Loveland et al., 2002). Six Trends blocks either partially or fully intersected the study area representing a total area of 469.2 km2. Land classified in the Trends maps as water, developed, mechanical disturbance, mine/quarry, barren, forest, and wetland were grouped and re-classified into an “other” class. Land classified as cropped in any Trends map was re-classified as cropped land (85.7% of total area in the Trends maps). Land that was classified as grass or shrub in all maps in the Trends time series was re-classified as never cropped land (13.6%).

3. Results and discussion

3.1 Comparison of results to the 2005 Kansas map

Overall agreement between our results and the 2005 Kansas map was 91.9% with individual counties ranging from 88.2% (Seward County) to 96.3% (Haskell County) (Table 2). Omission and commission errors were low for the ever-cropped land class (3.0% and 6.7%) yet higher for the never-cropped land class (29.8% and 15.2%). At the county level, difference of area classified as cropped land (including CRP) in the 2005 Kansas map and land classified as ever-cropped was very low, varying from only 0.1% (Ford County) to 1.6% (Gray County) with average 0.2%.

Table 2
Contingency matrices comparing the Kansas 2005 map to our results. Counts are in hectares.

As mentioned in the Methods section, comparison between a map produced using only 2005 Landsat imagery to our map generated with 1984–2010 imagery is limited. At a minimum, the ever-cropped class should include land classified as cropped land or CRP in the 2005 Kansas map. Comparison between the two maps was evaluated at the pixel level by generating a “difference” map with five classes: 2005 Kansas cropland/ever-cropped land, 2005 Kansas grassland/never-cropped land, 2005 Kansas cropland/never-cropped land, 2005 Kansas grassland/ever-cropped land, and other (Figure 5). Most of the discrepancies were in larger landscape patches or fields, as opposed to individual pixels or field edges.

Figure 5
Difference map for the Kansas 2005 map compared to our results for the entire study area with a close-up of the map near Hugoton, Stevens County, Kansas. Annual maximum NDVI time series graphs are shown for areas labeled as “a” and “b” ...

Discrepancies between areas classified as grassland in the 2005 Kansas yet classified as ever-cropped land in our map (Figure 5, blue colors on the Difference map) were primarily due to the different time periods of the Landsat imagery used in developing the maps (2005 imagery for the Kansas map versus 1984 through 2010 imagery for our map). For example, the NDVIann-max time series for the area labeled as “a” in Figure 5 (close-up map) had relatively low NDVI maximum values for the year 2005 (typical of grassland vegetation) yet during the years 1994–2003, the NDVIann-max values clearly indicated the land was cropped as demonstrated by the crop/fallow annual rotation pattern. In this case, both maps were correct.

Discrepancies between land classified as cropped or CRP in the 2005 Kansas map yet classified as never-cropped land in our analysis (Figure 5, red colors on the Difference map) were related mainly to CRP land. CRP land represented 10.0% of the total area classified as cropped land in the Kansas 2005 map. 79.6% of the area classified as CRP land was correctly classified as ever-cropped land in our map. Examination of the NDVIann-max time series for land that was classified as CRP land in the Kansas map yet classified as never-cropped land in our map showed that many of these areas had typical grassland patterns throughout the 27-year time period (Figure 5, lower right time series graph labeled “b”) indicating the land may have been cropped prior to 1984 and had remained in grassland for the years 1984–2010. Collection and analysis of imagery prior to 1984 (e.g., Landsat MSS) would be needed to identify pre-1984 cropped land. In other areas, the NDVIann-max time series indicated the land was originally grassland, then cropped for a few years (e.g., 1990–1993) and then converted back to grassland. The NDVIann-max, NDVIsd, and NDVIsd84-86 were all relatively low, hence the land was classified as never-cropped. Including additional multi-year NDVIann-max variation parameters, similar to the NDVIsd84-86, may improve the identification of these areas (e.g., NDVIsd87-90, NDVI sd91-95).

3.2 Comparison to Trends project maps

Overall agreement between our results and the Trends maps (1986, 1993, and 2000) was very high with an overall agreement of 97.2% (Table 3). Omission and commission errors were less than 10.3%. Differences covered only 2.7% of the land area and were represented as fields or field edges as opposed to scattered pixels, similar to the Kansas 2005 map comparison. Discrepancies between land classified as grassland in all three of the Trends maps and land classified as ever-cropped land in our map were primarily due to the different image data sets used in the classification (Figure 6, blue colors on the Difference map). Closer evaluation of the NDVIann-max time series for these areas revealed that most appeared to have been cropped at some point over the 27-year period suggesting our map was correct. For example, in Figure 6, the NDVIann-max time series for area “a” shows a typical crop/fallow rotation pattern for the years 1984–1987. Our image data set that spanned consecutive years allowed for better identification of cropped lands in a crop/fallow annual rotation because of the high inter-annual variability recorded in the continuous time series.

Figure 6
Difference map for the USGS Trends maps compared to our results. Annual maximum NDVI time series graphs (NDVIann-max) are shown for areas labeled as “a” and “b” on the close-up map. The red vertical lines on the time series ...
Table 3
Contingency matrix comparing the USGS Trends maps to our results. Counts are in hectares.

Areas in red in the Difference map in Figure 6 were classified as cropped in at least one of the Trends map years, yet classified as never-cropped land in our map. Some evidence was noted in the time series graphs that cropping may have occurred in these areas. The time series graph “b” for one of these areas indicated crop/fallow cycles during the years 1984–1992. We included the NDVIsd84-86 parameter for distinguishing CRP lands that were cropped during the 1984–1986 time period as the vast majority of the CRP land was converted during the years 1986–1990 (Figure 4). As mentioned above, additional multi-year NDVIann-max variation parameters, such as NDVIsd87-90, may help improve the identification of these cropped lands.

3.3 Comparison of results to Agricultural Census county-level data

Our results showed that the area classified as ever-cropped land was higher in all counties as compared to Agricultural Census area estimates. Area of land classified as ever-cropped for the five counties was on average 11.8 % higher than Census estimates (maximum area of land ever cropped during the period 1982 to 2007). Differences ranged from 8.1% (Haskell County) to 16.5% (Seward County) (Figure 7). De Wit and Clevers (2004) have found overestimates sometimes occur from land cover at the edge of crop fields being classified as cropped land when the pixels actually contain a mixture of land cover types (e.g., cropped land and grassland). The higher estimates could also be a result of setting the NDVImax, NDVIsd, and/or NDVIsd84-86 threshold values too low. The differences may also be due to underestimates in the Census. The county-level Census data are increasingly subject to error from non-response and incompleteness of mailing lists (www.agcensus.usda.gov/Help/FAQs/2002_Census/index2.asp; accessed March 2011). In 2002 approximately 2.8 M questionnaires were mailed with only an 88.0% response rate (2002 Census of Agriculture). Nevertheless, comparison of our ever-cropped map to both the 2005 Kansas map and Trends maps indicate that there is high agreement with other recent analyses, and that the time series method presented here represents an important advance in accounting for past disturbance. Without this temporal dimension, baseline estimates of land cover change remain inaccurate.

Figure 7
Comparison of area estimates from USDA Agricultural Census data and land classified as ever-cropped land. The Census values represent the maximum area ever cropped for census years 1982, 1987, 1992,1997, 2002, and 2007.

3.4 Summary

The initial goal of our project was to develop a method to identify cropped and non-cropped land using individual target years (e.g., c1995, c2005) for ten sample sites within the Great Plains region. The new Landsat image policy implemented by USGS in 2009 provided an opportunity to explore the potential for identifying land that had ever been cropped over a continuous time period, spanning the entire Landsat TM archive. We developed and tested a novel approach using a dense time series of intra- and inter-annual Landsat TM data in southwestern Kansas. Further testing in other regions in the Great Plains is necessary to determine if the approach is flexible enough to accommodate the wide variability in crop management and climates. Never-cropped land in our study area had relatively low NDVIann-max values and low NDVIsd values, compared to cropped lands, enabling good separation of the two classes. In other areas the distinction may not be as clear. Misclassifications will occur in cases where never-cropped land consistently reaches very high NDVImax values from year-to-year resulting in high NDVIann-max values and low NDVIsd values. Also, non-irrigated crops planted every year in regions where vegetation is constrained by moisture, temperature, or poor soil conditions will result in relatively low NDVImax values and low NDVIsd values and therefore be misclassified as never-cropped land.

We evaluated the potential for using Landsat Multispectral Scanner (MSS) imagery in the analysis to extend the time series back to 1972, which may have allowed for better identification of CRP land without the need for the second rule based on the NDVIsd84-86 parameter. However, we found that many of the MSS images were misregistered and would require extensive manual effort to correct which was not feasible to complete within the time constraints of our project. The USGS also notes that variations in the geometric quality of MSS imagery requires manual processing which limits its usefulness in time series studies (http://landsat.usgs.gov/NewMSSProduct.php; accessed April 2011). Incorporation of Landsat ETM+ imagery in deriving the annual maximum NDVI composite image, including slc-off images, is also a possibility.

Much time and effort can be expended trying to select cloud-free images (Homer et al., 2001; Huang et al., 2009). We found that the annual maximum NDVI composite image provided a cloud-free representation of the landscape eliminating the time consuming task of identifying cloud-free images or performing cloud and cloud shadow classification. Use of this method for reducing/removing cloud contamination in satellite data sets has been applied to AVHRR and MODIS image data sets since the 1980’s (Justice et al., 1985; Holben, 1986). Although high-speed computing resources and internet access are needed to download and process the large number of images needed for this method, much of the processing was automated, allowing for fairly rapid image processing. An automated process for identifying the NDVImax and NDVIsd threshold values would be useful for mapping large geographical regions (Maxwell et al., 2004; Sezgin et al., 2004).

Annual maximum NDVI image composites may be useful for identifying specific years when land use changes occur, such as conversion of cropped land to CRP or non-irrigated to irrigated. Characterization of forest land changes may also benefit from this approach (Cohen et al., 2010; Huang et al., 2010; Kennedy et al., 2010).

4. Conclusion

Advances in hardware and software technology, and the recent no-cost Landsat data policy, open up new opportunities to derive information from the vast archive of Landsat imagery. We used 230 intra- and inter-annual Landsat TM images spanning 1984 through 2010 to generate 27 annual maximum NDVI composite images (NDVIann-max). The annual NDVIann-max time series eliminated cloud and cloud shadow effects, and allowed for visualization and analysis of inter-annual cropland dynamics such as crop/fallow rotation, changes in cropping practices (e.g., crop/fallow to irrigated), and land use conversion (e.g., cropped land to CRP). Effective discrimination of ever-cropped land from never-cropped land resulted using three parameters derived from the 27-year NDVIann-max composite images: 1) maximum NDVI value that occurred over the entire 27 year time span (NDVImax), 2) standard deviation of the annual maximum NDVI values for all years (NDVIsd), and 3) standard deviation of the annual maximum NDVI values for years 1984–1986 (NDVIsd84-86) to improve Conservation Reserve Program land discrimination. Our study suggests that Landsat annual maximum NDVI image composites may be useful for many applications.


  • 230 Landsat images were used to identify ever-cropped land (1984–2010).
  • We compared the results to USDA Census records and two digital land use maps.
  • Area of ever-cropped land was 11.8 % higher than Census record estimates.
  • Agreement with the two digital land use maps was 91.9% and 97.2%.


Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.


  • Census of Agriculture (Issued 2004). U.S. Department of Agriculture, United States Summary and State Data. Vol. 1. National Agricultural Statistics Service; 2002. (Geographic Area Series Part 51, AC-02-A-51).
  • Cohen WB, Yang Z, Kennedy R. Detecting trends in forest disturbance and recovery using yearly Landsat time series: 2. TimeSync – tools for calibration and validation. Remote Sensing of Environment. 2010;114:2911–2924.
  • Cunfer G. On the Great Plains: agriculture and environment. Texas A&M University Press; 2005.
  • de Wit AJW, Clevers JGPW. Efficiency and accuracy of per-field classification for operational crop mapping. International Journal of Remote Sensing. 2004;25:4091–4112.
  • Gamon JA, Field CB, Goulden ML, Griffin KL, Hartley AE, Joel G, Penuelas J, Valentini R. Relationships between NDVI, canopy structure, and photosynthesis in three Californian vegetation types. Ecological Applications. 1995;5:28–41.
  • Gardner BL. American agriculture in the twentieth century: how it flourished and what it cost. Harvard University Press; 2002.
  • Gutmann MP. Great Plains Population and Environment Data: Agricultural Data, , 1870–2007 [UNITED STATES] [Computer file]. ICPSR04254-v1. Ann Arbor, MI: University of Michigan [producer]; Ann Arbor, MI: Inter-university Consortium for Political and Social Research [distributor]; 2005. 2005-06-22. [Cross Ref]
  • Holben BN. Characteristics of maximum-value composite images from temporal AVHRR data. International Journal of Remote Sensing. 1986;7:1417–1434.
  • Homer C, Dewitz J, Fry J, Coan M, Hossain N, Larson C, Herold N, McKerrow A, VanDriel JN, Wickham J. Completion of the 2001 National Land Cover Database for the conterminous United States. Photogrammetric Engineering and Remote Sensing. 2007;73:337–341.
  • Huang C, Goward SN, Masek JG, Thomas N, Zhu Z, Vogelmann JE. An automated approach for reconstructing recent forest disturbance history using dense Landsat time series stacks. Remote Sensing of Environment. 2010;114:183–198.
  • Justice CO, Townshend JRG, Holben BN, Tucker CJ. Analysis of the phenology of global vegetation using meteorological satellite data. International Journal of Remote Sensing. 1985;6:1271–1318.
  • Kennedy RE, Yang Z, Cohen WB. Detecting trends in forest disturbance and recovery using yearly Landsat time series: 1. LandTrendr – temporal segmentation algorithms. Remote Sensing of Environment. 2010;114:2897–2910.
  • Kirkendall RS. Social scientists and farm politics in the age of Roosevelt. University of Missouri Press; 1966.
  • Küchler AW. A new vegetation map of Kansas. Ecology. 1974;55:596–604.
  • Licker R, Johnston M, Foley JA, Barford C, Kucharik CJ, Monfreda C, Ramankutty N. Mind the gap: how do climate and agricultural management explain the ‘yield gap’ of croplands around the world? Global Ecology and Biogeography. 2010;19:769–782.
  • Loveland TR, Sohl TL, Stehman SV, Gallant AL, Sayler KL, Napton DE. A strategy for estimating the rates of recent United States land-cover changes. Photogrammetric Engineering & Remote Sensing. 2002;68:1091–1099.
  • Lubowski RN, Bucholtz S, Claassen R, Roberts MJ, Cooper JC, Gueorguieva A, Johansson R. United States Department of Agriculture, Economic Research Service. Environmental Effects of Agricultural Land-Use Change: The Role of Economics and Policy. 2006.
  • Maxwell SK, Nuckols JR, Ward MH, Hoffer RM. An automated approach to mapping corn from Landsat imagery. Computers and Electronics in Agriculture. 2004;43:43–54.
  • Peterson DL, Egbert SL, Price KP, Martinko EA. Identifying historical and recent land-cover changes in Kansas using post-classification change detection techniques. Transactions of the Kansas Academy of Science. 2004;107:105–118.
  • Peterson DL, Whistler JL, Lomas JM, Dobbs KE, Jakubauskas ME, Egbert SL, Martinko EA. Kansas Biological Survey. University of Kansas; 2008. 2005 Kansas land cover patterns Phase I: Final Report. KBS Report 150; p. 29.
  • Ramankutty N, Evan AT, Monfreda C, Foley JA. Farming the planet: 1. Geographic distribution of global agricultural lands in the year 2000. Global Biogeochem Cycles. 2008;22:GB1003.
  • Rouse JW, Haas RH, Schell JA, Deering DW. Monitoring vegetation systems in the Great Plains with ERTS. Proceedings of the Third ERTS Symposium, NASA SP-351; 1973. pp. 309–317.
  • Serra T, Zilberman D, Gil JM, Featherstone A. The effects of decoupling on land allocation. Applied Economics. 2009;41:2323–2333.
  • Sezgin M, Sankur B. Survey over image thresholding techniques and quantitative performance evaluation. Journal of Electronic Imaging. 2004;13:146–165.
  • Sullivan PJ. The Conservation Reserve Program economic implications for rural America. U.S. Dept. of Agriculture, Economic Research Service; 2004.
  • USDA Farm Service Agency. Conservation Reserve Program: Annual summary and enrollment statistics – FY. Washington, DC: USDA Farm Service Agency; 2010.
  • USDA Farm Service Agency. [accessed July 2011.];CRP Enrollment and Rental Payments by State, 1986–2010. 2011 Website: www.fsa.usda.gov/FSA/
  • USDA National Agricultural Statistics Service (NASS) [accessed April 2011];Agricultural Statistics. 2011 Website: http://www.nass.usda.gov/
  • Wang J, Rich PM, Price KP, Kettle WD. Relations between NDVI, grassland production, and crop yield in the Central Great Plains. Geocarto International. 2005;20:5–11.
  • Wu JJ. Slippage Effects of the Conservation Reserve Program. American Journal of Agricultural Economics. 2000;82:979–992.
  • Wu JJ, Lin H. The Effect of the Conservation Reserve Program on Land Values. Land Economics. 2010;86:1–21.
PubReader format: click here to try


Related citations in PubMed

See reviews...See all...


Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...