Remote Sensing Applications for Agriculture Monitoring in Northern Eurasia
S. Bartalev1, E. Loupian1, I. Neyshtadt1 and I. Savin2
Space Research Institute, Russian Academy of Sciences, Moscow, Profsoyuznaya str. 84/32, Russia – email@example.com
Institute for the Protection and Security of the Citizen, EC JRC, Ispra, TP 441, 21020 Ispra (VA), Italy – firstname.lastname@example.org
Abstract – Northern Eurasian ecosystems present monitoring. Blue (459-479 nm) and shortwave infrared
challenges to social and ecosystem scientists assessing (1628-1652 nm) spectral channels with 500 m resolution
the human dimensions of land cover and land use have been used for cloud and snow cover detection at the
change. One of the main factors affecting the state of pre-processing stage.
the environment in Northern Eurasia is the change of
area and structure of agricultural lands. Reliable and
up-to-date information on agricultural lands is critical 3. TEST REGIONS
issue of the day and is needed for sustainable socio-
economic development of the region and global climate Rostov oblast and Krasnodarsky kray were selected as test
change modeling. The paper gives an overview of first regions. Total area of both selected test regions is about
results of the monitoring of agricultural lands in Russia 200 th. sq. km. These regions situated in the south of
based on remote sensing data, including the Russia produce about 45% of sunflower croppage and 20%
methodology and set of derived thematic products on of wheat. An extension of developed methods over the all
various aspects of agricultural lands status and main agriculture regions of Russia is planned for the near
Keywords: Agriculture, MODIS, Northern Eurasia, Land- 4. METHODOLOGY
The developed methodology is based on analysis of the
1. INTRODUCTION remote sensing data time-series. Using the time series of
vegetation index pixels can be classified into different
Over the past 25 years the Northern Eurasian region has classes. Every class corresponds to one particular crop or
experienced a dramatic changes in arable lands usage. group of crops.
Since the dissolution of the Soviet Union individual farmers Data processing is divided into two main parts: pre-
got the independence in their agriculture activities. processing chain and thematic processing. At the first step
Nowadays all information concerning arable lands usage in the mask of pixels which are not suitable for thematic
Russia is provided by farmers. Lack of reliable information processing is produced. This mask includes cloud cover,
points at the necessity of development of agriculture snow cover, cloud shadows and pixels with resolution
monitoring system based on remote sensing data, which worse than stated. Algorithms for pre-processing use data
will be independent from the information provided by about reflectance in red and near infrared bands for
farmers. Since 2003 under the treaty with Russian Ministry cloud/snow detection, data about solar illumination and
of Agriculture, Space Research Institute has been instrument viewing geometry for detecting of cloud
developing agriculture monitoring system for main Russian shadows and for removing pixels with insufficient
agriculture regions (Bartalev et al., 2004). This paper resolution. The composite image of the 8 daily products is
discuses methods and first results of arable lands mapping, produced using mentioned above mask. These composite
and classification of crops received with use of the images are suitable for thematic analysis.
Terra/Aqua-MODIS time-series data. Second step is the thematic processing. The Perpendicular
Vegetation Index (PVI) is used for crop growth monitoring.
2. SATTELITE DATA It is calculated in the following way:
Terra/Aqua-MODIS instrument data have been used as the PVI = −0.83* R1 + 0.56 * R 2 − 0.005
primary data source for agriculture monitoring. Spectral Where R1 and R2 are correspondingly the reflectance in red and
bands and spatial resolution fit the needs of agriculture near-infrared channels.
vegetation monitoring. As input data standard daily
MODIS products were used MOD09GHK, MOD09GQK, PVI time series for each pixel, where one point in time
MODMGGAD, and MOD09GST series represents 8 days, were calculated with the aim to
(http://lpdaac.usgs.gov/main.asp). Mentioned products classify annual PVI time series into several representative
provide information about surface reflectance, solar land use classes. Figure 1 shows virtual PVI time series for
illumination and instrument viewing geometry. Red (620- several crops. These time series were obtained combining
670 nm) and near infrared (841-875 nm) spectral channels virtual land use data and MODIS composite images.
with 250 m resolution have been used for vegetation
It is significant to mention that the usage of fixed thresholds
provides much more reliable results since we don’t adjust
classification parameters to various years and regions.
RESULTS and VALIDATION
The following maps were produced using MODIS data
from 2001 to 2004: clean fallow, winter crops, and
sunflower. An arable lands map was compiled based on
crop masks for the years 2001-2004. This map was
obtained by overlapping individual crop maps. Fig. 2 shows
the spatial distribution of arable lands in Rostov oblast.
Figure 1. Different crops have different vegetation index
behavior during the growing season.
Additionally a simple Land Cover map, which includes
several types of land cover: water bodies, built-up areas,
deciduous forests, coniferous forests, grasslands, bare
sands, salt-marshes, and swamps was produced.
The special algorithms to classify three types of agriculture
land use: clean fallow, winter crops and sunflower were
A unique physical feature of clean fallow fields is that fields
are not covered by vegetation during all vegetation season.
This feature underlies the classification method. The
1 Figure 2. White – arable lands in Rostov oblast.
characteristic f = ∑ pvii , where I – is the set of
N i∈ I
composite images from the 1 of March to the 1 of A comparison of these maps with official statistical data
September for the current year was used for the pixels and field observation data was done for validation of the
classification. Further all pixels were classified using one quality of the results. The statistical information about
threshold value for all territory and for all years. sown areas for each crop for every sub-region was used as
Winter crops detection algorithm is based on prior official data. Additionally, field observation records for
knowledge about winter crops development. Temporal about 200 fields with information about cultivated crop
dynamics of winter crops can be characterized by two main were available for validation of the results. An example of
periods: the sowing period, the growing period. For each comparison of the classification results with field data is
pixel we calculate PVI for two principal time frames: 1st of shown at fig.3.
September – 5th of October, and 1st of November – 31st of
December. Then this two-dimensional attribute space was
classified into “winter crops” and “other” classes using
predefined threshold values. Used threshold values are the
same for all years, but different for Rostov and Krasnodar
regions. Using two different threshold values was
necessary due to geographical differences between the
Sunflower is one of the main crops in the concerned Winter crops, 2003 year Sunflower, 2003 year
Clean fallow, 2003 year
regions. Air temperature data was used to define two – Ground truth data
principal time frames responsible for sowing and flowering – Remote sensing classification
of the crop. The relation between cumulative air Figure 3. An example of the results validation based on
temperature and dates of sowing and flowering was ground truth data for one collective farm in Rostov region.
established based on expert analysis of local agronomical
practice. For these dates we derived The PVI values from
composite images were derived for these dates, and two- Table A presents a relative error between official data and
dimensional attribute space was constructed. Threshold remote sensing classification at the level of administrative
classification was performed with fixed thresholds to region. A considerable discrepancy can be caused by a
distinguish “sunflower” pixels. Threshold values were number of factors. First of all there is a doubt about
fixed for all years and regions. In more detail algorithms accuracy of official data. Additionally, discrepancy in
are presented in Bartalev et al., 2004. winter crop acreage estimation can be caused by improper
Table A. Relative error between official data and remote
sensing classification results.
Type/Year 2001 2002 2003 2004 With the advent of new generation of remote sensing
Winter Crops – 0,38 0,04 – systems like MODIS, new challenges arise to develop
Sunflower 0,24 0,17 -0,23 – methods of compiling of land use maps for various
Krasnodarskij kray applications. The approach to obtain annual land use maps
Type/Year 2001 2002 2003 2004 for agriculture applications is under consideration in given
Winter Crops – 0,28 0,19 0,28 paper. In summary, this study has demonstrated the
Sunflower 0,10 -0,14 -0,07 – potential of MODIS data and reliability of the crop
identification algorithms for the operative mapping of
different crops within large territories. In the near future
Table B represents comparison between the results of
the investigations shall be concentrated on producing of
classification and official statistical data at sub-region level.
arable lands map for all main Russian agriculture regions,
In this case a correlation coefficient was selected as a
and on extension of crop masks to neighboring regions.
divergence characterizing value.
Table B. Correlation between remote sensing classification 6. REFERENCES
results and official statistical data on crop acreage for sub-
Agriculture, hunting and forestry in Russia, Rosstat, 478p.,
Moscow, 2004. (In Russian)
Bartalev S.A., Loupian E.A., Neyshtadt I.A. et al., Remote
estimation of arable lands parameters with MODIS data, ,
Type/Year 2001 2002 2003 2004
Proceedings “Current aspects of remote sensing of earth
Fallow 0,83 0,87 0,84 – from space” conference, Moscow, 2004a. (In Russian, In
Winter Crops – 0,85 0,81 – Press)
Sunflower 0,76 0,80 0,84 – Bartalev S.A., Burcev M.A., Ershov D.V. et al., Automated
Krasnodarskij kray receiving, processing and distribution system for agriculture
Type/Year 2001 2002 2003 2004 monitoring, Proceedings “Current aspects of remote sensing
Winter Crops – 0,90 0,86 0.90 of earth from space” conference, Moscow, 2004b. (In
Sunflower 0,83 0,88 0,74 – Russian, In Press)
Earth Observation System (EOS) Data Products Handbook,
High values of the correlation coefficient proof in general NASA Goddard Space Flight Center, Greenbelt Maryland,
the reliability of the applied algorithms. Vol. I, 258 p., 2003.
Justice C.O., Townshend J.R.G., Vermote E.F. et al., “An
The validation of the whole arable lands acreage estimation overview of MODIS Land data processing and product
has a particular interest. A comparison with official data at status”, Remote Sensing of Environment, vol 83, p.p. 3-15.,
sub-region level demonstrates its substantial agreement 2002.
(Fig. 4). Sowing area, croppage and yield of crops and perennial
plants, Krasnodarstat, Krasnodar, 2002, 2003, 2004. (In
Sowing area, croppage and yield of crops in farms of all
categories, Rostovoblstat, Rostov-on-Don, 2001, 2002,
2003. (In Russian)
Zoidze E.K. and Ovcharenko L.I., Comparison assessment
of agricultural climate potential of Russian Federation
territory and of its agricultural resources degrees used by
crops, Hydrometeoizdat, 74 p., Saint-Petersburg, 2000. (In
Figure 4. Official statistical data about arable lands acreage
for sub-regional level versus results of the satellite