Advances in Ahmed Applied Agricultural Science and Sajjad / Advances in Applied Agricultural Science 03 (2015), 03: 16-25 Volume 03 (2015), Issue 03, 16-25 Journal homepage: www.aaasjournal.com
ISSN: 2383-4234 Research Article
Crop acreage estimation of Boro Paddy using Remote Sensing and GIS Techniques: A Case from Nagaon district, Assam, India Raihan Ahmed 1 and Haroon Sajjad 1* 1
Department of Geography, Faculty of Natural Sciences, Jamia Millia Islamia, New Delhi, India.
ARTICLE INFO Article history: Received: August 12, 2014 Revised: September 19, 2014 Accepted: October 06, 2014 Available online March 23, 2015
Keywords: Crop acreage NDVI RVI Supervised Classification Remote Sensing GIS
*
ABSTRACT Agriculture plays an important role in Indian economy. Comprehensive, reliable and timely information on agriculture resource is very much necessary for a country like India for making decisions for all agricultural related problems. The study makes an attempt to estimate the crop acreage of Boro paddy with the help of remote sensing and GIS techniques. Boro paddy of Nagaon district is unique crop because it is not dependable on rainfall and it is also risk free from flood hazards. The application of NDVI, RVI and supervised classification were used in this study to examine the variation of these models in acreage estimation. The remote sensing data for this study were collected from USGS earth explorer. Survey of India topographical sheet was used for extracting district area mask. Statistical data of Assam agricultural department were used to analyze the percentage of deviation. The result of this study showed that supervised classification proved to be more accurate model than NDVI and RVI. Supervised classification model validated the spectral signature with GPS survey data for better accuracy of the acreage estimation.
Corresponding Author; Β© 2015 AAAS Journal. All rights reserved.
E. Mail:
[email protected]
A
griculture is the backbone of Indian
necessary for acquiring reliable agricultural information
economy. Agriculture and allied sector
since market economies have turned globalized. Remote
contribute 19.5 per cent of GDP and
sensing has shown great potential in agricultural land use
provide support to more than 50% of population
mapping and monitoring due to its advantages over
(Ministry of Finance, 2013-14). Comprehensive, reliable
traditional procedures in terms of cost effectiveness and
and timely information on agriculture resource is very
timeliness in the availability of information over larger
much necessary for a country like India for making
areas. Space-borne remotely sensed data being repetitive
decisions by all stakeholders. It has also become
and multi-spectral in nature present an ideal choice for
16
Ahmed and Sajjad / Advances in Applied Agricultural Science 03 (2015), 03: 16-25
monitoring
and
assessing
agricultural
(summer rice) is unique. Cultivation is practiced in the
resources. The information on crop acreage estimation
month of November-December and harvesting before
is of paramount significance on agricultural statistical
monsoon i.e. the month of May-June. Boro yields high
system, if area has a strong inter-annual variability while
production and this crop is risk free from agricultural
yield remains relatively stable. Timely and reliable
related hazards (flood, drought, etc.). Therefore, it helps
information on crop area is of great importance to
to understand the pattern of agriculture and give a
planners and policy makers for efficient and timely
valuable result for future decision making for farmers of
agricultural
the study area
development
dynamic
and
making
important
decisions with respect to procurement, storage, public distribution, export, import and other related issues. However, with more emphasis on local area planning, there is further need for estimating crop area with respect to different varieties grown in the area, irrigation availability, the soil type, etc. which can go a long way in rapid development of the region (Goswami S. B. et al. 2012).
Scholars have conducted various studies on crop acreage estimation. Mauraya A. K. (2011) mapped areas of soybean field with the help of MODIS (TERRA) satellite data and GIS database and Dadhwal V. K. et al. (2002) explained the basic concept of crop inventory. They presented the usefulness and uniqueness of optical and microwave data for crop identification in operational scenario in Rabi (winter season) and Kharif (summer
Crop acreage estimation procedure consists of two
seasons). Wu Bingfang et al. (2003) estimated crop
approaches i.e. complete enumeration and sample
proportion using cluster sampling and crop type
segment approach. In the complete enumeration
proportions of different crop types using transect
approach, the study area (Blocks or district boundary) is
sampling and GVG survey system. Y. Jia et al. (2013)
super-imposed on the satellite data and all the pixels
explained the use of microwave remote sensing data on
within this boundary area analyzed. This method is
estimating crop planting area. Dutta S. et al. (1994)
suitable for small area. Estimation of crop acreages for
calculated the accuracy assessment in cotton acreage
large areas like state requires handling of a very large
estimation using Indian remote sensing satellite data. In
volume of data and larger efforts are needed in
this paper an attempt has been made to estimate crop
conducting ground truth exercise for data collection. To
acreage under Boro paddy using normalized difference
overcome this problem, stratification sampling technique
vegetation index NDVI), ratio vegetation index (RVI)
based on sample segment procedure has been developed
and supervised classification approaches and to assess
under Crop Acreage and Production Estimation Project
the suitability of best method of crop acreage estimation
(Sajjad H. et al. 2013). The Nagaon district is a paddy
for agricultural land use planning.
dependable district of Assam. Paddy is cultivated in all three cultivation season. Therefore it is a significant task
Study area
to calculate acreage of paddy for this district. By the
The Central Assam District of Nagaon is one of the
estimation of crop acreage the status of paddy cultivation
largest districts of Assam (Figure 1). It sprawls across
of Nagaon district can better be analyzed. For this
nearly 4000 km2 of fertile alluvial plains and thickly
acreage estimation process the selection of Boro paddy
forested hills. Nagaon extends from 25045' to 26045' 17
Ahmed and Sajjad / Advances in Applied Agricultural Science 03 (2015), 03: 16-25
Fig. 1. Location Map of Study area
Fig. 2. Flow diagram
18
Ahmed and Sajjad / Advances in Applied Agricultural Science 03 (2015), 03: 16-25 North latitude and 92025β to 93020β East longitude. The
the models of vegetation indices and supervised
district is bounded by Sonitpur district and the river
classification is the method of land use and land cover
Brahmaputra in the north, West Karbi Anglong and
classification based on ground truth. The results obtained
North Cachar Hills in the south and East Karbi Anglong,
through all the models were compared with surveyed
Golaghat district in the east and Morigaon district in the
statistical data for evaluating the potential of each model.
west. The river Brahmaputra flows along the northern periphery of the district. Agriculture is the backbone of its economy providing livelihood to above 70% of the total population. Rice is the staple food of the inhabitants
The reflective and absorptive properties of vegetation with respect to red and near-infrared radiation are much different than that of soil. Active vegetation uses light energy available in the atmosphere as an ingredient in the
and paddy is the principal crop of the district.
photosynthesis process. Therefore, the normalized vegetation index and ratio vegetation index were calculated by using following equations:
Materials and Methods ππ·ππΌ = Database and Methodology Single date Landsat 5 TM digital data coinciding with
π
ππΌ =
ππΌπ
βπ
ππ β¦ β¦ β¦ β¦ β¦ β¦ β¦ β¦ β¦ β¦ (1) ππΌπ
+π
ππ
NIR β¦ β¦ β¦ β¦ β¦ β¦ β¦ β¦ β¦ β¦ β¦ β¦ β¦ (2) Red
maximum crop growth (March, 2009) has been used for contrast stretching =
identification and estimation of crop acreage. Several studies have made use of Landsat satellite data for crop
π·π(ππππ’π‘)βπ·π(πππ) π·π(πππ₯)βπ·π(πππ)
Γ
255 β¦ β¦ β¦ β¦ β¦ β¦ (3)
acreage estimation in meeting more accuracy level (Hanuschak et al. 1982; Dadhwal V. K. et al., 2002; Wu Bingfang et al., 2003). The administrative boundary of
Where, NIR is the reflectance of NIR radiation and
the study area was demarcated from Survey of India Topographical sheets on the scale of 1:50000 and
Red is the reflectance of visible red radiation
overlaid on the image to extract the entire pixels of the study area. The statistical data of crop were collected from
Directorate
of
Economic
and
Statistics,
Government of Assam. Surveyed and GPS data were collected for ground truth requirements.
Accuracy assessment of supervised classification was carried out using error matrix. The matrix was calculated on category by category basis. The overall accuracy was computed by dividing the total number of correctly classified pixels by the total number of reference pixels.
The methodology for crop acreage estimation at district
The producer accuracies were calculated by dividing the
level requires more attention in comparison to the village
number of correctly classified pixels in each category by
and block level study. For acreage estimation of Nagaon
the number of training sets pixels of that category (the
district, we analyzed normalized difference vegetation
column total). User accuracies were computed by
index (NDVI), ratio vegetation index (RVI) and
dividing the number of correctly classified pixels in each
supervised classification techniques. NDVI and RVI are
category by the total number of pixels classified in that
19
Ahmed and Sajjad / Advances in Applied Agricultural Science 03 (2015), 03: 16-25
Table 1. Area under different vegetation classes calculated from NDVI image Vegetation Classes
Range of Gray Area of Sub-divisions (in Hectares) Values Nagaon Hojai 0 - 150 89572.9 70709.21 No Vegetation 151 - 200 71341.5 57622.23 Forest Paddy 201 - 255 30645.9 6897.14 Source: Authorsβ calculation based on NDVI Image
Kaliabor 27751.95 41168.78 5654.79
Table 2. Area under different vegetation classes calculated from RVI image Vegetation Classes
Range of Gray Values 0 - 79 No Vegetation 80 - 169 Forest Paddy 170 - 255 Source: Authorsβ calculation based on RVI image
Area of Sub-divisions (in Hectares) Nagaon Hojai 96753.2 75080.6 66044.8 54092.13 29124.3 6973.13
Kaliabor 32364.69 37356.54 5784.51
Table 3. Area under different land use and land cover Sl. No.
Land Use, Land Cover Classes
Area of Sub-divisions (in hectares) Nagaon Hojai 1 Paddy 32108.2 6396.4 2 28515.7 38277.0 Forest 3 Fallow Land 54130.9 62903.0 4 Sand 2084.85 00 5 Water body 3213.63 3344.13 6 Built-up 72769.8 24224.9 Source: Authorsβ calculation based on supervised classification
category (the row total).
Kaliabor 6983.44 26080.7 26882.7 1239.84 1355.49 13363.5
estimation. In NDVI and RVI models the gray values of all pixels were classified into high, medium and low on
Relative deviation was computed to assess the accuracy of acreage estimation by NDVI, RVI and supervised classification from the actual data. It also helped to assess the suitability of models for crop acreage
the basis of digital number (DN) by density slicing method. On the other hand, supervised classification was performed based on the pixel brightness of image and GPS data.
estimation. Acreage estimation by NDVI, RVI and Supervised classification
Results and Discussion NDVI, RVI and supervised classification models were used to estimate Boro crop acreage. These models were then compared to select the best model for acreage
NDVI is based on the function of near infrared and red bands since vegetations are more sensitive in the near infrared band and healthy vegetation reflects more in comparison to the unhealthy or stressed vegetation (Mroz and Sobieraj, 2004). Ratio vegetation index (RVI) 20
Ahmed and Sajjad / Advances in Applied Agricultural Science 03 (2015), 03: 16-25
Fig. 3. NDVI, RVI and Supervised Classification of Nagaon sub-division
Table 4. Accuracy assessment of Nagaon sub-division Classes
Paddy
Water body 0 213 23 0 0 0 236 90.25%
Sand
Fallow land 15 0 3 340 21 0 379 89.70%
Paddy 325 0 Water body 0 29 Sand 0 245 Fallow land 16 5 Built up 0 0 Forest 21 0 Column Total 362 279 Producer 89.77% 87.81% accuracy * Overall accuracy calculated from diagonal total i.e. 1660
21
Built up
Forest
0 0 0 26 273 0 299 91.30%
19 0 0 0 0 264 283 93.28%
Row Total 359 242 271 387 294 285 1838
User accuracy 90.52% 88.01% 90.40% 87.85% 92.85% 92.63% *90.31%
Ahmed and Sajjad / Advances in Applied Agricultural Science 03 (2015), 03: 16-25
is based on the reflectance of red and near infrared band
hectares in Hojai sub division and 5654 heaters in
where vegetation is more sensitive. Pixels in NDVI and
Kaliabor sub division (Table 1). Similarly the area was
RVI images were classified on the basis of their gray
estimated using RVI model. The area estimated by this
values by density slicing method. The values were
model was of the order of 29124 hectares in Nagaon sub
further enhanced by the contrast stretching. Gray values
division, 6973 hectares in Hojai and 5784 hectares in
provide information of particular vegetation during
Kaliabor sub division (Table 2). Crop acreage by
particular time period of time. The higher reflectance of
supervised classification was estimated 32108 hectares
gray values in NDVI and RVI images was designated as
in Nagaon sub division, 6396 hectares in Hojai and 6983
paddy because paddy growth was maximum during
hectares in Kaliabor sub division (Table 3). Relative
March. Moderate reflectance in gray values was
deviation of estimated area from surveyed data was
identified as forest while low reflectance in gray values
calculated to assess the suitability of best model.
was designated as no vegetation area (Figure 3, 4 and 5). Boro is transplanted during the month of December and
Validation of the supervised classification
no rainfall occurs during this month. Further, it requires
Accuracy assessments of supervised classification for all
low watering for growth. Therefore, its cultivation is
three sub-divisions are calculated on the basis of
carried out near water bodies so that it may get required
producer, user and overall accuracy. The error occurred
moisture. The chlorophyll in the canopy of paddy
during sample pixel selection is examined through this
remains higher than the other vegetation. Hence the
accuracy assessment. The accuracy assessment Table 4,
contrast in gray values of paddy and vegetation is clearly
5 and 6 show the percentage of accuracy for
seen in NDVI and RVI images.
classification of Nagaon, Hojai and Kaliabor sub-
Gaussian Maximum likelihood classifier method was also applied for estimating Boro crop acreage. On the basis of the reflectance samples pixels were selected from the image. Some pixels were wrongly classified due to the similarity of brightness value. These errors were minimized by the recoding of pixel of such locations on the basis of GPS data. Five distinct classes of level I were identified as vegetation, built-up, fallow land, sand and water body. The vegetation class was divided into two sub-classes of paddy and other vegetation based on their reflectance and ground truth (See Figure 3, 4 and 5).
divisions respectively. Overall accuracy of land use/ land cover classification of Nagaon sub division was 90.31%. User and producer accuracy level of paddy area under this classification was calculated as. 90.52% and 89.77% respectively (Table 4).Overall accuracy, user and producer accuracies in the case of Hojai sub-division were calculated as 88.8%, 91.88% and 92.10% respectively (Table 5) while overall accuracy, user accuracy and producer accuracy in Kaliabor sub-division were calculated as89.6%, 89.1% and 90.3%. Thus it can be safely inferred that the pixels classified as paddy showed closer to accuracy.
Crop acreage of Boro was estimated through gray values of NDVI and RVI images. The acreage of Boro estimated in three sub-divisions of Nagaon district from NDVI model was 30645 hectares in Nagaon sub division, 6897
Validation of NDVI, RVI and supervised classification based on surveyed data Estimated area by different models and relative deviation 22
Ahmed and Sajjad / Advances in Applied Agricultural Science 03 (2015), 03: 16-25
Fig. 4. NDVI, RVI and Supervised Classification of Hojai sub-division
Table 5. Accuracy assessment of Hojai sub-division Data
Paddy
Water body 0 50 0 1 2 53 94.33%
Built up
Paddy 385 0 Water body 0 0 Built up 0 255 Fallow land 2 42 Forest 31 8 Column Total 418 305 Producer 92.10% 83.60% Accuracy * Overall accuracy calculated from diagonal total i.e. 1432
Fallow land
Forest
4 2 43 222 1 272 81.61%
30 2 10 1 520 563 92.36%
23
Row Total 419 54 308 268 562 1611
User Accuracy 91.88% 92.59% 82.79% 82.83% 92.52% *88.88%
Ahmed and Sajjad / Advances in Applied Agricultural Science 03 (2015), 03: 16-25
Table 6. Accuracy assessment of Kaliabor sub-division Classes
Paddy
Water body 0 198 20 0 0 0 218 90.82%
Sand
Paddy 205 0 Water body 0 21 Sand 0 255 Fallow land 13 0 Forest 9 0 Built up 0 6 Column Total 227 282 Producer 90.30% 90.42% Accuracy * Overall accuracy calculated from diagonal total i.e. 1349
Fallow land 22 0 0 346 15 14 397 87.15%
Forest 3 0 0 12 189 0 204 92.64%
Built up 0 0 4 17 0 156 177 88.13%
Row Total 230 219 279 388 213 176 1505
User Accuracy 89.13% 90.41% 91.39% 89.17% 88.73% 88.63% *89.63%
Table 7. Relative deviation (RD) of estimated Boro acreage from surveyed data in Nagaon district sub-divisions District subdivisions
Nagaon Hojai Kaliabor
Paddy area in hectare (Surveyed data) 31765 5845 6610
Estimated Paddy area under NDVI (in hectares)
Estimated Paddy area under RVI (in hectares)
Area
RD (%)
Area
RD (%)
Estimated Paddy area under supervised classification (in hectares) Area RD (%)
30645.9 6897.14 5654.79
-3.65 +15.24 -16.88
29124.3 6973.13 5784.51
-9.06 +16.17 -14.27
32108.2 6396.4 6983.44
+1.06 +8.62 +5.34
of all sub-divisions of the district is given in Table 7. The
deviation of the estimated area by supervised
table revealed that the relative deviations of NDVI and
classification is found lowest in all the sub divisions.
RVI from the surveyed data are negatively deviated in
This shows that supervised classification coupled with
Nagaon
ground truth data may be used as potential model to
sub
division
i.e.
-3.65%
and
-9.06%
respectively. On the other hand supervised classification
estimate crop acreage quite independently.
is positively deviated from surveyed data i.e. +1.06%. Relative deviations of NDVI and RVI are higher in Hojai due to high intensity of forest cover i.e. +15.24% and
Conclusion
+16.17% respectively. Forest cover showed similar pixel reflectance as that of paddy and hence high rate of
This paper analyzed the results of one season remote
deviation from actual paddy area was occurred here.
sensing based sub division level of boro paddy acreage
Deviation of estimated paddy acreage from surveyed
estimation in Nagaon district. Single date remote sensing
data was +8.26% in this sub-division.
digital data from Landsat TM sensor coinciding with maximum crop growth of paddy crop was analyzed for
Relative deviation of estimated acreage of paddy from surveyed data in NDVI and RVI models was negative i.e. -16.88% and -14.27% respectively. In supervised classification the deviation was +5.34%.
Relative
acreage estimation using NDVI, RVI and maximum likelihood supervised classification and ground truth validation. The performance of paddy crop acreage estimation was evaluated by comparing relative 24
Ahmed and Sajjad / Advances in Applied Agricultural Science 03 (2015), 03: 16-25
deviations of NDVI, RVI and supervised classification
of USDA's statistical reporting service (1972-1982).
with surveyed data. Analysis of models used in this study
Proceedings of 8th International Symposium Machine
showed variation in results. Supervised classification
Processing Remotely Sensed Data with Special Emphasis on Crop Inventory and Monitoring. LARS/Purdue
showed lowest deviation in the estimated area from the
University, WL, Indiana.
surveyed data. Thus supervised classification with validation of GPS data can effectively provide accurate
6. Jia Y & Yu F. (2013). Research on estimating crop planting area by integrating the optical and microwave
estimation of crop acreage for better agricultural land use
remote sensing data. International Archives of the
planning.
Photogrammetry.
Remote
Sensing
and
Spatial
Information Sciences. XL-7/W1.
Acknowledgement
7. Mauraya A K. (2011). Estimation of Acreage & Crop Production through Remote Sensing & GIS Technique.
We sincerely thank the anonymous reviewers for their constructive comments and suggestions to improve the
Geospatial World Forum. Hyderabad. India. 8. Ministry of Finance (2013-14). Mid-Year Economic
overall quality of the manuscript.
Analysis, Department of Economic Affairs, Economic Division, Government of India. 9. Mroz M, Sobieraj A. (2004). Comparison of several vegetation indices calculated on the basis of a seasonal
References
SPOT XS time series, and their suitability for land cover and agricultural crop identification. Technical Sciences.
1. Dadhwal V K, Singh R P, Dutta S & Parihar J S. (2002).
7: 40-66.
Remote sensing based crop inventory: A review of Indian experience. Tropical Ecology 43(1): 107-122.
10. Sajjad H, Siddiqui M.A and Siddiqui L. (2013).
2. Directorate of Economic and Statistics (2009-10). Sub
Preparing Crop Inventory and Detecting Land Use
Division/ District wise area, average yield and production,
Changes: Shift in Methodological Paradigm, Proceeding
Guwahati, Government of Assam.
of the National Symposium on Paradigm Shift in Geography,
3. Dutta S, Sharma S A, Khera A P, Ajai, Yadav M, Hooda
(Edited-
Qureshi.
M.
H),
Manak
Publications.
R S, Mothikumar K E & Manchanda M L. (1994). Accuracy assessment in cotton acreage estimation using
11. Wu Bingfang & Li Qiangzi. (2003). Crop area estimation
Indian remote sensing satellite data. ISPRS Journal of
using remote sensing on two stage stratified sampling.
Photogrammetry and Remote Sensing 49: 21-26.
Institute of Remote Sensing Applications, Chinese Academy of Sciences, Beijing, China. Commission VI,
4. Goswami S B, Saxena A & Bairagi G D. (2012). Remote
WG VII/3.
Sensing and GIS based wheat crop acreage estimation of Indore district, M.P. International Journal of Emerging Technology and Advanced Engineering, 2(3): 200-203. 5. Hanuschak G A, Allen R D & Wigton W H. (1982). Integration of landsat data into the crop estimation program
25