Conference PaperPDF Available

Extracting Trusted Pixels from Historical Cropland Data Layer Using Crop Rotation Patterns: A Case Study in Nebraska, USA

Authors:

Abstract and Figures

It is still a challenge to generate the timely crop cover map at large geographic area due to the lack of reliable ground truths at early growing season. This paper introduces an efficient method to extract “trusted pixels” from the historical Cropland Data Layer (CDL) data using crop rotation patterns, which can be used to replace the actual ground truth in the crop mapping and other agricultural applications. A case study in the Nebraska state of USA is demonstrated. The common crop rotation patterns of four major crop types, corn, soybeans, winter wheat, and alfalfa, are compared and analyzed. The experiment results show a considerable number of pixels in CDL following the certain crop sequence during the past decade. Each observed crop type has at least one reliable crop rotation pattern. Based on the reliable crop rotation patterns, a great proportion of pixels can be correctly mapped a year ahead of the release of current-year CDL product. These trusted pixels can be potentially used to label training samples for crop type classification at early growing season.
Content may be subject to copyright.
Extracting Trusted Pixels from Historical Cropland
Data Layer Using Crop Rotation Patterns: A Case
Study in Nebraska, USA
Chen Zhang†‡ , Liping Di∗† , Li Lin , Liying Guo
Center for Spatial Information Science and Systems, George Mason University, Fairfax, VA 22030, USA
Department of Geography and Geoinformation Sciences, George Mason University, Fairfax, VA 22030, USA
Email: {czhang11, ldi*, llin2, lguo2}@gmu.edu
Abstract—It is still a challenge to generate the timely crop
cover map at large geographic area due to the lack of reliable
ground truths at early growing season. This paper introduces an
efficient method to extract “trusted pixels” from the historical
Cropland Data Layer (CDL) data using crop rotation patterns,
which can be used to replace the actual ground truth in the crop
mapping and other agricultural applications. A case study in
the Nebraska state of USA is demonstrated. The common crop
rotation patterns of four major crop types, corn, soybeans, winter
wheat, and alfalfa, are compared and analyzed. The experiment
results show a considerable number of pixels in CDL following
the certain crop sequence during the past decade. Each observed
crop type has at least one reliable crop rotation pattern. Based on
the reliable crop rotation patterns, a great proportion of pixels
can be correctly mapped a year ahead of the release of current-
year CDL product. These trusted pixels can be potentially used
to label training samples for crop type classification at early
growing season.
Index Terms—Crop rotation, Crop Mapping, Land use classi-
fication, Cropland Data Layer
I. INTRODUCTION
Ground truth is a crucial component of agricultural land
use classification and modeling. An accurate crop cover map
is commonly produced based on the reliable ground truth
information. For example, Cropland Data Layer (CDL) is an
annual product providing a massive amount of field-level (30-
meter spatial resolution) land use information with overall 95%
accuracy for the entire Contiguous United States (CONUS)
[1], which has been used in many studies such as LULC
change [2], flood mapping [3]–[5], national-scale cultivated
area estimation [6], and crop time series modeling [7]. The
procedure of CDL is based on a large number of reliable
ground truths and the surveys from the farm agency. However,
even though the internal use may be as early as mid to
late growing season, the current-year CDL product is usually
released for public use in the early of following year. A
common issue for the in-season crop mapping is the lack of
ground truth at early growing season. Thus an efficient way to
gather ground truth information for the in-season agricultural
applications is needed.
As a common agricultural practice, crop rotation has been
widely used around the world since thousand years ago.
corresponding author
The long-term crop rotation can affect crop yield as well
as soil quality, such as fertility and soil physical/chemical
properties [8]–[11]. Meanwhile, crop rotation information was
used to support crop mapping. In [12], an approach for remote
sensing-based regional crop rotation mapping was proposed.
In [13], a Markov-based model of crop rotation was presented
which is able to predict the early crop map without the actual
remote sensing imagery. CropRota, a crop rotation modeling
framework, was implemented to support agricultural land use
assessment and management [14]. Based on the information
of cropping sequence, we may find some reliable crop rotation
patterns from the historical CDL time series. Then using these
patterns to predict the crop type of cropland that follows the
regular planting cycles. Once the latest CDL product becomes
available, a map for the “trusted pixels” of CDL can be created
even before the beginning of a growing season. These pixels
can be used to make up the lack of ground truth at early
growing season.
This study introduces an innovative method to extract
trusted pixels from the historical CDL data based on crop
rotation. Several common crop rotation patterns of major
crop types in the Nebraska state of USA are investigated
and evaluated. The accuracy and amount of trusted pixels
for each crop rotation pattern is compared and analyzed.
Section II introduces the data, study area, and method of
trusted pixel extraction. Section III assesses the common crop
rotation patterns in the study area and illustrates the map of
trusted pixels. Section IV discusses the experiment results.
Conclusions and future works are given in the section V.
II. ME TH OD S
A. Study Area
This paper demonstrates a case study in Nebraska, a mid-
western state in the United States which lies in the western
part of the Corn Belt region. It is one of the top agriculture
production state in the United States with agricultural land of
45.2 million acres, which takes 91% of the state’s total land
area. The crop cover map and the land use information of
the study area in 2017 are shown in Figure 1. The statistics
show that corn and soybean are two dominant crop types in
the Nebraska state. Besides, winter wheat and alfalfa account
for 2% of total land area each. Other crop types, such as
Preprint submitted to 8th International Conference on Agro-Geoinformatics (July 2019)
TABLE I
EXA MPL ES O F CROP ROTATIO N PATTER NS F OR COR N
Cycle Pattern 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018
1-year A C C C C C C C C C C C C Corn
2-year A-B C NC C NC C NC C NC C NC C NC Corn
3-year A-B-B C NC NC C NC NC C NC NC C NC NC Corn
A-A-B C C NC C C NC C C NC C C NC Corn
A-B-A C NC C C NC C C NC C C NC C Corn
sorghum, dry beans, millet, and oats, account for a very small
proportion of all croplands. In this paper, we will focus on the
four dominant crop types (corn, soybean, winter wheat, and
alfalfa) in Nebraska.
Fig. 1. Statistics of major agricultural land use classes of Nebraska, USA, in
2017 (data from USDA NASS)
B. Data
CDL is the major source used in this study. It is the annual
product that covers the entire CONUS at 30-meter spatial
resolution from 2008 to present and some states from 1997
to 2007. In this study, we adopt the 16 years of CDL data
(2002-2017) for the study area. Figure 2 shows the map of
2017 CDL data for Nebraska state.
Fig. 2. The 2017 Cropland Data Layer of Nebraska, USA
C. Trusted Pixel Extraction
Crop rotation is the practice of growing a sequence of
different types of crops on the same cropland. Based on the
crop rotation patterns, the crop type of the subsequent year can
be predicted. Table I summarizes five common crop rotation
patterns and demonstrates how to predict the crop type from
the historical crop rotation information. It is marked with “C”
if the crop type of the pixel for the corresponding year is corn,
or “NC” if it is non-corn.
The “trusted pixels” refers pixels that consistently following
some certain crop rotation patterns in the historical CDL data.
Technically, the crop type in the subsequent growing season
is predicted by maximizing a priori probability based on the
historical crop rotation information. In this paper, the accuracy
and amount of pixels for four major crop types in Nebraska
(corn, soybean, winter wheat, and alfalfa), which are derived
from each crop rotation pattern, are assessed and compared.
According to the planting cycle, the five crop rotation
patterns described in the table can be divided into three
types: monocropping (1-year) pattern, alternate cropping (2-
year) pattern, and three-year planting cycle crop rotation
pattern. Specifically, the monocropping refers to the practice
of continuously growing a single crop on the same land unit.
The alternate cropping follows the two-year planting cycle that
growing a specific crop type on the same land every other year.
Similar with the alternate cropping, the three-year planting
cycle crop rotation is the practice of continuously growing the
specific crop type once (or twice) every three years.
To assess the reliability of the crop rotation pattern, a new
metric, Trusted Pixel Accuracy (TPA), is developed. TPA
measures the agreement of trusted pixels with the original
CDL pixels, which can be expressed as:
T rustedP ixelAccur acy =
GoodT rustedP ixel
T rustedP ixel (1)
where the “Good Trusted Pixel” is the amount of correctly
predicted trusted CDL pixels and the “Trusted Pixel” is the
amount of total trusted pixels. The value of TPA is ranging
from 0 to 1.
III. EXP ER IM EN TS
A. Assessment of Crop Rotation Patterns
This section presents a group of experiments to assess
common crop rotation patterns in Nebraska. The procedure of
assessment includes the following steps: (1) extracting pixels
that follow the specific crop rotation pattern from the historical
CDL data; (2) calculating the TPA and amount of the pixels;
(3) analyzing the impact of the change of the historical CDL
record on the result; and (4) determining the reliable crop
rotation patterns with high TPA.
(a) Corn (b) Soybeans (c) Winter Wheat (d) Alfalfa
Fig. 3. Change of trusted pixel accuracy and amount for the monocropping pattern
(a) Corn (b) Soybeans (c) Winter Wheat (d) Alfalfa
Fig. 4. Change of trusted pixel accuracy and amount for the alternate cropping pattern
1) Monocropping Pattern: Monocropping pattern extracts
pixels that have the constant crop type in the past few years.
Figure 3 presents the change of TPA and amount of pixels
derived from the monocropping pattern. The vertical axis of
each graph represents the TPA (right) and pixel amount (left).
The horizontal axis of each graph represents the length of
the referenced historical CDL data. For each crop type, we
compared the result of 2015 to 2017 using the moving window
with length from 1-year to 12-year. It can be observed from
the result that the change of corn is regular. With the increase
of CDL history, the corn’s TPA keeps going up. The result of
other crop types is not as satisfactory as corn. Some croplands
continuously growing soybean and winter wheat, yet many of
them will be replaced by other crop types at any time. The
result of alfalfa is unique. When using the 2-year historical
CDL data as reference, the alfalfa’s TPA reaches over 80%,
which is higher than the result based on the longer CDL
record. This result means a lot of croplands in Nebraska state
growing alfalfa for three consecutive years.
2) Alternate Cropping Pattern: Alternate cropping is the
most widely used crop rotation pattern in not only Nebraska
state but many other states in the Corn Belt region of mid-
western United States. Take corn as an example, if it is
constantly growing in the planting cycle of “corn, non-corn”,
we regard these pixels as the trusted pixel that following the
alternate cropping pattern. Figure 4 shows the change of the
TPA and amount of the trusted pixels that derived from the
alternate cropping pattern. For each crop type, we compared
the result of 2015 to 2017 using the moving window with
length from 2-year to 12-year. We can find out a great number
of croplands are following the alternate cropping pattern. After
two planting cycles (4 years), the TPA of corn and soybean
can achieve over 90% and keep going up with the increase
of the moving window of historical CDL record. After three
planting cycles (6 years), the trusted pixel of winter wheat
can reach over 80% agreement with the original CDL data.
Alfalfa, however, does not follow the alternate cropping pattern
since most alfalfa pixels are following the 3-consecutive-year
monocropping pattern.
3) Three-year Planting Cycle Crop Rotation Pattern: In
a three-year planting cycle, a certain crop could be grown
once or twice every three years. These patterns are denoted as
ABB”, “ABB”, and “ABA”, where “A” represents the crop
of interest (e.g. corn), “B” represents the crop other than the
crop of interest (e.g. non-corn). Figure 5 shows the change
of TPA and amount of trusted pixels extracted by the “ABB”
crop rotation pattern. We compared the result of each crop
type from 2015 to 2017 using the moving window with length
from 3-year to 12-year. The result shows a certain amount
of corn, soybean, and winter wheat pixels follow the “ABB”
pattern. After three planting cycles (9 years), corn and soybean
can achieve the TPA of over 80%. The TPA of winter wheat
is a little bit lower but keeps going up with the increase
of referenced CDL data. Alfalfa does not follow the “ABB”
pattern.
The results of the “AAB” pattern (Figure 6) indicate a
certain amount of corn pixels consistently follow the “AAB”
pattern. After three planting cycles (9 years), a small number
of soybean pixels achieve 80% of TPA. Winter wheat and
alfalfa do not follow “ABA” pattern.
The results of the “ABA” pattern (Figure 7) suggest some
corn pixels continuously follow the “ABA” pattern, which can
reach TPA of 80% after three planting cycles (9 years). A
few alfalfa pixels follow this pattern as well, but the result
varies between years. Soybean and winter wheat do not follow
ABA” pattern.
(a) Corn (b) Soybeans (c) Winter Wheat (d) Alfalfa
Fig. 5. Change of trusted pixel accuracy and amount for the “ABB” crop rotation pattern.
(a) Corn (b) Soybeans (c) Winter Wheat (d) Alfalfa
Fig. 6. Change of trusted pixel accuracy and amount for the “AAB” crop rotation pattern.
(a) Corn (b) Soybeans (c) Winter Wheat (d) Alfalfa
Fig. 7. Change of trusted pixel accuracy and amount for the “ABA” crop rotation pattern.
B. Mapping Trusted Pixels
Based on the above analysis, we can observe that each crop
type has been following one or more reliable crop rotation
patterns in Nebraska. In this study, a crop rotation pattern
will be regarded as the “reliable crop rotation pattern” if its
TPA is higher than 80%. Table II summarizes the minimum
planting cycle that required for each crop rotation pattern to
reach high TPA, where “C”, “NC”, “A”, “S”, “NS”, “W”,
and “NW” represent “corn”, “non-corn”, “alfalfa”, “soybean”,
“non-soybean”, “winter wheat”, and “non-winter wheat”, re-
spectively. The percentage of TPA and Proportion in CDL is
the mean value of three target years (2015-2017).
Once the latest CDL data is released to public, the current-
year trusted pixel can be created based on these reliable
crop rotation patterns without any current-year ground truth
information. Figure 8 show the 2018 trusted pixel map of
Nebraska derived by February 2018 using the CDL data from
2009 to 2017.
IV. DISCUSSION
It can be seen from the experiment result in Section
III-A that each crop type follows one or more crop rotation
patterns. The total amount of trusted pixels for each crop
rotation pattern keeps going down with the increase of the
referenced historical CDL record. This change is caused by
the following reasons: (1) many cropland units broke the
common crop rotation patterns at some stage of the historical
agricultural practice; (2) the pixel information of some regions
is inaccurate or missing in the historical CDL data; (3) the
cropland follows some other crop rotation patterns which are
not covered in this study.
Technically, “AAB” pattern and “ABA” pattern have the
same planting cycle (twice in three years), except for 1-
year time lag. However, the experiment result shows the
TPA of “AAB” crop rotation pattern is higher than “ABA”
crop rotation pattern. This difference could be explained by
the experiment results of monocropping pattern and alternate
cropping pattern. Take corn for instance, it can be found from
the first column in Figure 3 that only 40% of corn pixels
do not change their crop type in the next year. On the other
hand, the first column in Figure 4 indicates that over 80% of
corn pixels follow the alternate cropping pattern and change
the crop type after one alternate cropping planting cycle. It
can be concluded from the historical CDL data that the crop
type of most pixels in Nebraska is more likely to be changed
in the following growing season. In the long term, “ABA”
TABLE II
SUMMARY OF RELIABLE CROP ROTATIO N PATTER NS IN NEBRASKA, U SA
Planting Cycle Historical Crop Rotation Record Subsequent Crop TPA Proportion in CDL
1-year C-C-C-C-C-C Corn 83.09% 11.08%
A-A Alfalfa 82.94% 54.85%
2-year C-NC Corn 82.91% 46.80%
S-NS-S-NS Soybean 87.02% 46.55%
W-NW-W-NW-W-NW Winter Wheat 80.53% 17.90%
3-year C-NC-NC-C-NC-NC-C-NC-NC Corn 88.27% 0.75%
C-C-NC Corn 83.86% 14.04%
Fig. 8. The 2018 trusted pixel map of Nebraska, USA (produced by February 2018)
pattern and “AAB” pattern are equivalent with each other. In
the short term, the subsequent growing crop of the “ABA”
pattern follows the crop type of “A”, which is same as the
previous year. However, most land units follow the alternate
cropping pattern and keep changing the crop type year after
year, which can explain the accuracy difference between crop
rotation pattern of “AAB” and crop rotation pattern of “ABA”.
From the experiment result in Section III-B, we can see a
huge volume of trusted pixels can be automatically extracted
from the historical CDL data using reliable crop rotation
patterns. The TPA can be further improved by extracting
trusted pixels from the long-term reference data. As the trade-
off, the proportion of the trusted pixel will go down while
the increasing of the TPA. For major crop types (corn and
soybean) in the Nebraska state, the TPA of all reliable crop
rotation patterns would be stabilized at 90%-95% based on the
historical CDL record of 10-years or longer.
V. CONCLUSIONS AND FUTURE WO RK S
This paper presented an efficient method to extract trusted
pixels from the historical CDL data using crop rotation
patterns. Several common crop rotation patterns, including
monocropping, alternate cropping, and 3-year panting cycle
crop rotation, were investigated. We found seven reliable crop
rotation patterns of four major crop types in the Nebraska state
of USA. The trusted pixels derived by the reliable crop rotation
patterns can reach 90%-95% agreement with the CDL data,
which can be potentially used to replace the actual ground truth
in the crop mapping and other agricultural applications at early
growing season. In the future, we will investigate more crop
rotation patterns for more diverse crop types over the entire
CONUS. Moreover, the feasibility of in-season crop mapping
using the trusted pixels will be explored.
ACK NOW LE DG EM EN T
This study is supported by a grant from U.S. National
Science Foundation (grant#: CNS-1739705, PI: Dr. Liping Di).
REFERENCES
[1] C. Boryan, Z. Yang, R. Mueller, and M. Craig, “Monitoring US agricul-
ture: The US Department of Agriculture, National Agricultural Statistics
Service, Cropland Data Layer Program,” Geocarto International, vol. 26,
no. 5, pp. 341–358, 2011.
[2] T. J. Lark, R. M. Mueller, D. M. Johnson, and H. K. Gibbs, “Mea-
suring land-use and land-cover change using the U.S. department of
agriculture’s cropland data layer: Cautions and recommendations,In-
ternational Journal of Applied Earth Observation and Geoinformation,
vol. 62, pp. 224–235, Oct. 2017.
[3] L. Lin, L. Di, E. G. Yu, L. Kang, R. Shrestha, M. S. Rahman, J. Tang,
M. Deng, Z. Sun, C. Zhang, and L. Hu, “A review of remote sensing
in flood assessment,” in 2016 5th International Conference on Agro-
Geoinformatics, 2016.
[4] L. Lin, L. Di, J. Tang, E. Yu, C. Zhang, M. S. Rahman, R. Shrestha, and
L. Kang, “Improvement and Validation of NASA/MODIS NRT Global
Flood Mapping,” Remote Sensing, vol. 11, no. 2, p. 205, Jan. 2019.
[5] R. Shrestha, L. Di, E. G. Yu, L. Kang, Y. Shao, and Y. Bai, “Regression
model to estimate flood impact on corn yield using MODIS NDVI and
USDA cropland data layer,Journal of Integrative Agriculture, vol. 16,
no. 2, pp. 398–407, Feb. 2017.
[6] L. King, B. Adusei, S. V. Stehman, P. V. Potapov, X.-P. Song, A. Krylov,
C. Di Bella, T. R. Loveland, D. M. Johnson, and M. C. Hansen, “A
multi-resolution approach to national-scale cultivated area estimation of
soybean,” Remote Sensing of Environment, vol. 195, pp. 13–29, Jun.
2017.
[7] D. P. Roy and L. Yan, “Robust Landsat-based crop time series mod-
elling,” Remote Sensing of Environment, Jul. 2018.
[8] J. H. Edwards, D. L. Thurlow, and J. T. Eason, “Influence of Tillage
and Crop Rotation on Yields of Corn, Soybean, and Wheat,Agronomy
Journal, vol. 80, no. 1, pp. 76–80, Jan. 1988.
[9] H. H. Janzen, C. A. Campbell, S. A. Brandt, G. P. Lafond, and
L. Townley-Smith, “Light-Fraction Organic Matter in Soils from Long-
Term Crop Rotations,Soil Science Society of America Journal, vol. 56,
no. 6, pp. 1799–1806, Nov. 1992.
[10] D. L. Karlen, E. G. Hurley, S. S. Andrews, C. A. Cambardella, D. W.
Meek, M. D. Duffy, and A. P. Mallarino, “Crop Rotation Effects on Soil
Quality at Three Northern Corn/Soybean Belt Locations,” Agronomy
Journal, vol. 98, no. 3, pp. 484–495, May 2006.
[11] L. L. Van Eerd, K. A. Congreves, A. Hayes, A. Verhallen, and D. C.
Hooker, “Long-term tillage and crop rotation effects on soil quality,
organic carbon, and total nitrogen,” Canadian Journal of Soil Science,
vol. 94, no. 3, pp. 303–315, May 2014.
[12] G. Waldhoff, U. Lussem, and G. Bareth, “Multi-Data Approach for
remote sensing-based regional crop rotation mapping: A case study for
the Rur catchment, Germany,International Journal of Applied Earth
Observation and Geoinformation, vol. 61, pp. 55–69, Sep. 2017.
[13] J. Osman, J. Inglada, and J.-F. Dejoux, “Assessment of a Markov
logic model of crop rotations for early crop mapping,” Computers and
Electronics in Agriculture, vol. 113, pp. 234–243, Apr. 2015.
[14] M. Sch¨
onhart, E. Schmid, and U. A. Schneider, “CropRota – A crop
rotation model to support integrated land use assessments,” European
Journal of Agronomy, vol. 34, no. 4, pp. 263–277, May 2011.
... In Osman et al. (2015), a Markov logic networks-based model for early crop mapping was proposed, which can predict the crop type from the historical crop rotation at the accuracy of 60% before the growing season. Meanwhile, several models based on the field-level crop sequences can be potentially used in the prediction of future crop planting information (Aurbacher and Dabbert, 2011;Xiao et al., 2014;Zhang et al., 2019a). Based on the previous studies, a question arises: "can we predict the future crop planting from the historical crop planting maps?" ...
Article
Full-text available
An accurate crop planting map can provide essential information for decision support in agriculture. The method of post-season and in-season crop mapping has been widely studied in the land use and land cover community. However, it remains a challenge to predict the spatial distribution of crop planting before the growing season. This paper is the first attempt to use machine learning approach on the prediction of field-level annual crop planting from historical crop planting maps. We present an end-to-end machine learning framework for crop planting prediction using Cropland Data Layer (CDL) time series as reference data and multi-layer artificial neural network as prediction model. The proposed framework was first tested at Lancaster County of Nebraska State, then scaled up to the U.S. Corn Belt. According to the experiment results from 53 Agricultural Statistics Districts, we found the machine-learned crop planting map was expected to reach 88% agreement with the future CDL. Meanwhile, the crop acreage estimates derived from the machine-learned prediction were highly correlated (R2 > 0.9) with the crop acreage estimates of CDL and official statistics by the U.S. Department of Agriculture National Agricultural Statistics Service. This study provides a low-cost and efficient way to predict annual crop planting map, which can be used to support many agricultural applications and decision makings before the beginning of a growing season.
... Hao et al. (2016) explored the crop classification based on the previous-year crop knowledge. Zhang et al. (2019a) produced a crop cover map of Nebraska State based on the common crop rotation patterns of corn, soybeans, winter wheat, and alfalfa. They further implemented a crop sequence-based machine learning framework for prediction of crop cover maps (Zhang et al., 2019b). ...
Article
Full-text available
As the most widely used crop-specific land use data, the Cropland Data Layer (CDL) product covers the entire Contiguous United States (CONUS) at 30-meter spatial resolution with very high accuracy up to 95% for major crop types (i.e., Corn, Soybean) in major crop area. However, the quality of early-year CDL products were not as good as the recent ones. There are many erroneous pixels in the early-year CDL product due to the cloud cover of the original Landsat images, which affect many follow-on researches and applications. To address this issue, we explore the feasibility of using machine learning technology to refine and correct misclassified pixels in the historical CDLs in this study. An end-to-end deep learning-based framework for restoration of misclassified pixels in CDL image is developed and tested. By feeding the CDL time series into the artificial neural network, a crop sequence model is trained and the misclassified pixels in an original CDL map can be restored. In the experiment with the 2005 CDL data of the State of Illinois, the misclassified pixels over Agricultural Statistics Districts (ASD) #1760 were corrected with a reasonable accuracy (> 85%). The findings suggest that the proposed method provides a low-cost and reliable way to refine the historical CDL data, which can be potentially scaled up to the entire CONUS.
Article
Full-text available
Space-based crop identification and acreage estimation have played a significant role in agricultural studies in recent years, due to the development of Remote Sensing technology. The Cropland Data Layer (CDL), which was developed by the U.S. Department of Agriculture (USDA), has been widely used in agricultural studies and achieved massive success in recent years. Although the CDL’s accuracy assessments report high overall accuracy on various crops classifications, misclassification is still common and easy to discern from visual inspection. This study is aimed to identify and resolve inaccurate crop classification in CDL. A decision tree method was employed to find questionable pixels and refine them with spatial and temporal crop information. The refined data was then evaluated with high-resolution satellite images and official acreage estimates from USDA. Two validation experiments were also developed to examine the data at both the pixel and county level. Data generated from this research was published online in two repositories, while both applications allow users to download the entire dataset at no cost.
Article
Synthetic Aperture Radar (SAR) Sentinel-1 (S1) surface soil moisture (SSM) retrieval is promising but challenging in canopy areas. Additional data from optical-thermal infrared (TIR) sensors (e.g., Moderate Resolution Imaging Spectroradiometer, MODIS) have potential to compensate for SSM monitoring from single S1 product, suggesting their synergistic use for SSM retrieval in vegetated terrains. A physical, straightforward and non-calibration synergy method of SAR S1 and optical-TIR MODIS was proposed in this study to enhance SSM estimation in an agro-pastoral transitional zone. Change detection for S1 backscatter (σ°) and simplified triangle method representing land surface temperature (LST) – normalized difference vegetation index (NDVI) space (Ts-VI) for MODIS, were combined under a synergistic framework regarding vegetation conditions. The synergy method performance was evaluated against in-situ soil moisture sites for different phenological stages. Correlation analysis results indicated that S1 σ° is highly sensitive to SSM in non-growing season, whereas MODIS LST is negatively linked to root-zone soil moisture during summer crop growth as the seasonal vegetation effect that can be explained via the observed trapezoidal Ts-VI shape. For the synergy, S1 is mainly involved in during non-growing period, while major contribution in growing period was from MODIS, which accounted for 90% improvements in correlation (R, 0.29 – 0.55) and 30% for Root-Mean-Square Error (RMSE, 0.093 – 0.065 m³m⁻³) during a four-month crop growth as compared to single S1 product. For entire period, the synergy method outperformed single S1 over all sites, where the average R and RMSE improved 30% (0.46 – 0.60) and 14% (0.070 – 0.060 m³m⁻³), respectively, with highest values (R = 0.73 and RMSE = 0.039 m³m⁻³) observed in a maize site. This synergy suggests a suitability of integrating optical-TIR data to improve SAR SSM estimation in agricultural lands, especially over the crop growth period when water is a major limiting factor.
Article
Full-text available
The United States Department of Agriculture (USDA) Cropland Data Layer (CDL) provides spatially explicit information about crop production area and has served as a prevalent data source for characterizing cropland change in the U.S. in the last decade. Understanding the accuracy of the CDL is paramount because of the reliance on it for management and policy making. This study examined the characteristics of the CDL from 2007 to 2017 using comparisons to other USDA datasets. The results showed when examining the cropland area for the same year, the CDL produced comparable trends with other datasets (R2 > 0.95), but absolute area differed. The estimated area of cropland changes from 2007 to 2012, 2008 to 2012 and 2012 to 2017 varied from weak to moderate correlation between the CDL and the tabular data (R2 = 0.005~0.63). Differences in area of cropland change varied widely between data sources with the CDL estimating much larger change area. A series of image processing techniques designed to improve the confidence in cropland change estimated using the CDL reduced the area of estimated cropland change. The techniques also, unexpectedly, lowered the correlation in change estimated between the CDL and the tabular datasets. Estimated land cover change area varied widely based on analyses applied and could reverse from increasing to declining area in cropland. Further analyses showed unlikely change scenarios when comparing different year combinations. The authors recommend the CDL only be used for land cover change analysis if the error can be estimated and is within change estimates.
Article
Training samples is fundamental for crop mapping from remotely sensed images, but difficult to acquire in many regions through ground survey, causing significant challenge for crop mapping in these regions. In this paper, a transfer learning (TL) workflow is proposed to use the classification model trained in contiguous U.S.A. (CONUS) to identify crop types in other regions. The workflow is based on fact that same crop growing in different regions of world has similar temporal growth pattern. This study selected high confidence pixels across CONUS in the Cropland Data Layer (CDL) and corresponding 30-m 15-day composited NDVI time series generated from harmonized Landat-8 and Sentinel-2 (HLS) data as training samples, trained the Random Forest (RF) classification models and then applied the models to identify crop types in three test regions, namely Hengshui in China (HS), Alberta in Canada (AB), and Nebraska in USA (NE). NDVI time series with different length were used to identify crops, the effect of time-series length on classification accuracies were then evaluated. Furthermore, local training samples in the three test regions were collected and used to identify crops (LO) for comparison. Results showed that overall classification accuracies in HS, AB and NE were 97.79%, 86.45% and 94.86%, respectively, when using TL with NDVI time series of the entire growing season for classification. However, LO could achieved higher classification accuracies earlier than TL. Because the training samples were collected across USA containing multiple growth conditions, it increased the potential that the crop growth environment in test regions could be similar to those of the training samples; but also led to situation that different crops had similar NDVI time series, which caused lower TL classification accuracy in HS at early-season. Generally, this study provides new options for crop classification in regions of training samples shortage.
Article
Full-text available
Google Earth Engine (GEE) is an ideal platform for large-scale geospatial agricultural and environmental modeling based on its diverse geospatial datasets, easy-to-use APIs, rich reusable library, and high-performance computational capacity. However, using GEE to prepare agricultural land use data for geospatial agricultural and environment modeling requires not only the programming skills of GEE APIs but also the knowledge of the data. This paper presents a toolkit AgKit4EE to facilitate the use of the Cropland Data Layer (CDL) products over GEE platform. The toolkit contains a variety of frequently used functions for use of CDL products including crop sequence modeling, crop frequency modeling, confidence layer modeling, and land use change analysis. The experimental results suggest that the toolkit can significantly reduce the workload for modelers who perform geospatial agricultural and environmental modeling with CDL data as well as developers who build the GEE-enabled cyberinfrastructure for agricultural land use modeling of the conterminous United States.
Article
Full-text available
The remote-sensing based Flood Crop Loss Assessment Service System (RF-CLASS) is a web service based system developed and managed by the Center for Spatial Information Science and Systems (CSISS). The system uses Moderate Resolution Imaging Spectroradiometer (MODIS)-based flood data, which was implemented by the Dartmouth Flood Observatory (DFO), to provide an estimation of crop loss from floods. However, due to the spectral similarity between water and shadow, a noticeable amount of false classification of shadow can be found in the DFO flood products. Traditional methods can be utilized to remove cloud shadow and part of mountain shadow. This paper aims to develop an algorithm to filter out noise from permanent mountain shadow in the flood layer. The result indicates that mountain shadow was significantly removed by using the proposed approach. In addition, the gold standard test indicated a small number of actual water surfaces were misidentified by the proposed algorithm. Furthermore, experiments also suggest that increasing the spatial resolution of the slope helped reduce more noise in mountains. The proposed algorithm achieved acceptable overall accuracy (>80%) in all different filters and higher overall accuracies were observed when using lower slope filters. This research is one of the very first discussions on identifying false flood classification from terrain shadow by using the highly efficient method.
Article
Full-text available
Reliable satellite monitoring of agriculture is often difficult because surface variations occur rapidly compared to the cloud-free satellite observation frequency. Harmonic time series models, i.e., superimposed sequences of sines and cosines, have an established provenance for fitting satellite vegetation index time series to coarse resolution satellite data, but their application to medium resolution Landsat data for crop monitoring has been limited. Non-linear harmonic models have been shown to perform well over agricultural sites using single-year Moderate Resolution Imaging Spectroradiometer (MODIS) time series, but have not been explored with Landsat data. The 2017 availability of Landsat Analysis Ready Data (ARD) over the United States provides the opportunity to investigate the utility of temporally rich Landsat data for 30 m pixel-level crop monitoring. In this paper, the capability of 5- and 7-parameter linear harmonic models and a 5-parameter non-linear harmonic model applied to a year of Landsat 5 Thematic Mapper (TM) and Landsat 7 Enhanced Thematic Mapper Plus (ETM+) ARD is investigated. The analysis is undertaken over six sites, each defined by a 5000 × 5000 30 m pixel ARD tile, that together include the major conterminous United States (CONUS) crops identified by inspection of the United States Department of Agriculture (USDA) Cropland Data Layer (CDL). The model fits are evaluated as the root mean square difference (RMSD) between the fitted and the observed Landsat data. Considering locations with at least 21 annual Landsat observations, the 7-parameter linear harmonic model (tile mean crop NDVI RMSD values ranging from 0.052 to 0.072) and the 5-parameter non-linear harmonic model (tile mean crop NDVI RMSD values ranging from 0.054 to 0.074) are shown to be able to fit annual Landsat NDVI time series for most CONUS crops, whereas the 5-parameter linear harmonic model cannot (tile mean crop NDVI RMSD values ranging from 0.072 to 0.099). If there are between 15 and 20 annual Landsat observations, the 5-parameter non-linear harmonic model is recommended for fitting annual NDVI crop time series, and if there are ≥21 observations, then either the 5-parameter non-linear or the 7-parameter linear model can be used. The 7-parameter model had marginally smaller mean NDVI RMSD values but larger standard deviations than the 5-parameter non-linear model, likely due to the relative robustness of the non-linear model to over-fitting and oscillations. None of the models could reliably fit crops with multiple stages, such as alfalfa, that are insufficiently sampled using combined Landsat 5 TM and Landsat 7 ETM+ time series. Given the utility of the growing season peak NDVI for crop yield applications, the date and magnitude of the model fitted peak NDVI are compared to quantify model reporting differences. The differences between the 7-parameter linear and the 5-parameter non-linear harmonic models are not large. For each ARD tile, the mean absolute differences in the estimated peak NDVI days varied from <2 days in the northern ARD tiles, which had short growing seasons and similar crops, to less than a week for the other tiles except for nearly 10 days for the California tile that had longer growing seasons and more diverse crops including crops with multiple stages. The paper concludes with a discussion and recommendations for future research.
Article
Full-text available
Monitoring agricultural land is important for understanding and managing food production, environmental conservation efforts, and climate change. The United States Department of Agriculture's Cropland Data Layer (CDL), an annual satellite imagery-derived land cover map, has been increasingly used for this application since complete coverage of the conterminous United States became available in 2008. However, the CDL is designed and produced with the intent of mapping annual land cover rather than tracking changes over time, and as a result certain precautions are needed in multi-year change analyses to minimize error and misapplication. We highlight scenarios that require special considerations, suggest solutions to key challenges, and propose a set of recommended good practices and general guidelines for CDL-based land change estimation. We also characterize a problematic issue of crop area underestimation bias within the CDL that needs to be accounted for and corrected when calculating changes to crop and cropland areas. When used appropriately and in conjunction with related information, the CDL is a valuable and effective tool for detecting diverse trends in agriculture. By explicitly discussing the methods and techniques for post-classification measurement of land-cover and land-use change using the CDL, we aim to further stimulate the discourse and continued development of suitable methodologies. Recommendations generated here are intended specifically for the CDL but may be broadly applicable to additional remotely-sensed land cover datasets including the National Land Cover Database (NLCD), Moderate Resolution Imaging Spectroradiometer (MODIS)-based land cover products, and other regional, national, and global land cover classification maps.
Article
Full-text available
Spatial land use information is one of the key input parameters for regional agro-ecosystem modeling. Furthermore, to assess the crop-specific management in a spatio-temporal context accurately, parcel-related crop rotation information is additionally needed. Such data is scarcely available for a regional scale, so that only modeled crop rotations can be incorporated instead. However, the spectrum of the occurring multiannual land use patterns on arable land remains unknown. Thus, this contribution focuses on the mapping of the actually practiced crop rotations in the Rur catchment, located in the western part of Germany. We addressed this by combining multitemporal multispectral remote sensing data, ancillary information and expert-knowledge on crop phenology in a GIS-based Multi-Data Approach (MDA). At first, a methodology for the enhanced differentiation of the major crop types on an annual basis was developed. Key aspects are (i) the usage of physical block data to separate arable land from other land use types, (ii) the classification of remote sensing scenes of specific time periods, which are most favorable for the differentiation of certain crop types, and (iii) the combination of the multitemporal classification results in a sequential analysis strategy. Annual crop maps of eight consecutive years (2008–2015) were combined to a crop sequence dataset to have a profound data basis for the mapping of crop rotations. In most years, the remote sensing data basis was highly fragmented. Nevertheless, our method enabled satisfying crop mapping results. As an example for the annual crop mapping workflow, the procedure and the result of 2015 are illustrated. For the generation of the crop sequence dataset, the eight annual crop maps were geometrically smoothened and integrated into a single vector data layer. The resulting dataset informs about the occurring crop sequence for individual areas on arable land, so that crop rotation schemes can be derived. The resulting dataset reveals that the spectrum of the practiced crop rotations is extremely heterogeneous and contains a large amount of crop sequences, which strongly diverge from model crop rotations. Consequently, the integration of remote sensing-based crop rotation data can considerably reduce uncertainties regarding the management in regional agro-ecosystem modeling. Finally, the developed methods and the results are discussed in detail.
Article
Full-text available
Flood events and their impact on crops are extremely significant scientific research issues; however, flood monitoring is an exceedingly complicated process. Flood damages on crops are directly related to yield change, which requires accurate assessment to quantify the damages. Various remote sensing products and indices have been used in the past for this purpose. This paper utilizes the moderate resolution imaging spectroradiometer (MODIS) weekly normalized difference vegetation index (NDVI) product to detect and further quantify flood damages on corn within the major corn producing states in the Midwest region of the US. County-level analyses were performed by taking weighted average of all pure corn pixels (>90%) masked by the United States Department of Agriculture (USDA) Cropland Data Layer (CDL). The NDVI-based time-series difference between flood years and normal year (median of years 2000–2014) was used to detect flood occurrences. To further measure the impact of the flood on corn yield, regression analysis between change in NDVI and change in corn yield as independent and dependent variables respectively was performed for 30 different flooding events within growing seasons of the corn. With the R² value of 0.85, the model indicates statistically significant linear relation between the NDVI and corn yield. Testing the predictability of the model with 10 new cases, the average relative error of the model was only 4.47%. Furthermore, small error (4.8%) of leave-one-out cross validation (LOOCV) along with smaller statistical error indicators (root mean square error (RMSE), mean absolute error (MAE), and mean bias error (MBE)), further validated the accuracy of the model. Utilizing the linear regression approach, change in NDVI during the growing season of corn appeared to be a good indicator to quantify the yield loss due to flood. Additionally, with the 250 m MODIS-based NDVI, these yield losses can be estimated up to field level.
Conference Paper
Full-text available
Flood is defined as water that temporarily submerges land. In the United States, flood is being considered as an event only if it last over 72 hours, but the time frame is much longer in the Europe. The information of water bodies is an important source when examining floods. However, this data is not always available or accurate by traditional survey since water surface changes frequently. Remote Sensing provides an alternative method to traditional survey with lower cost and much frequent revisit cycles. Some sensors have very fine spatial resolution but relative long revisit circle; while some fine temporal resolution satellites are not able to provide spatial information in detail. This paper provides a brief review and comparison of major optical and Radar satellite sensors which are currently adopted in flood assessment. Result indicated that both optical sensors and Radar systems play important roles in flood assessment. Although many agencies use Radar or optical data independently when ding flood assessment, the potential of combining data from both systems is huge.
Article
Full-text available
Detailed and timely information on crop area, production and yield is important for the assessment of environmental impacts of agriculture, for the monitoring of the land use and management practices, and for food security early warning systems. A machine learning approach is proposed to model crop rotations which can predict with good accuracy, at the beginning of the agricultural season, the crops most likely to be present in a given field using the crop sequence of the previous 3–5 years. The approach is able to learn from data and to integrate expert knowledge represented as first-order logic rules. Its accuracy is assessed using the French Land Parcel Information System implemented in the frame of the EU’s Common Agricultural Policy. This assessment is done using different settings in terms of temporal depth and spatial generalization coverage. The obtained results show that the proposed approach is able to predict the crop type of each field, before the beginning of the crop season, with an accuracy as high as 60%, which is better than the results obtained with current approaches based on remote sensing imagery.
Article
Fig. 1. Overall flowchart of a multi-resolution approach for national-scale soybean area estimation in US and Argentina.
Article
Long-term studies allow for quantification of the effects of crop production practices, such as tillage and crop rotation, on soil quality and soil C and N stores. In two experiments at Ridgetown, ON, we evaluated the long-term (11 and 15 yr) effect of tillage system and crop rotation on soil quality via the Cornell Soil Health Assessment (CSHA) at 0-15 cm and soil organic C (SOC) and total N at 5-, 10-, and 20-cm increments to 120 cm depth. The CSHA soil quality score and SOC and total N were higher with no-till (NT) than fall moldboard plough with spring cultivation (conventional tillage, CT) and rotations with winter wheat [soybean-winter wheat (S-W) and soybean-winter wheat-corn (S-W-C)] compared with rotations without winter wheat. In both long-term trials, NT had ca. 21 Mg ha(-1) more or 14% higher SOC than CT in the 0- to 100-cm soil profile, a trend which contrasts previous research in eastern Canada. Thus, the two long-term trial results at Ridgetown suggest that to improve soil quality and storage of C and N, growers on clay loam soil in southwestern Ontario should consider adopting NT production practices and including winter wheat in the rotation.
Article
The objective of the study was to determine the influence of agronomic variables on soil light-fraction (LF) organic matter content. Soils from three long-term crop rotation studies in Saskatchewan, Canada, were analyzed for LF content and composition. The experiments, established at Indian Head, Melfort and Scott included wheatbased rotations. Within each site, the LF content was generally highest in treatments with continuous cropping or perennial forages and lowest in those with a high frequency of summer fallow. Fertilizer application generally favored LF accumulation. Differences in LF content were attributed to variable residue inputs and rates of substrate decomposition. The LF content is a sensitive indicator of the effects of cropping but, because of its transient nature, probably reflects primarily short-term effects. -from Authors