Mapping Specific Crops and Their Penology Approach

MAPPING SPECIFIC CROPS AND THEIR PHENOLOGY – MULTISENSOR AND TEMPORAL APPROACH

Background

India is an agrarian economy and about 60% of its population is directly or indirectly involved in agriculture. With more than a billion mouths to feed and the population increasing at an alarming rate, it needs to monitor manage and estimate its agricultural produce/resources to ensure national food security.

With the advancement of remote sensing in the last few decades , it has been increasingly seen the use of remotely sensed images to identify and map specific crop in agricultural field.(Kumar and Roy, 2011).This database of remotely sensed images can be further used in many agricultural applications that includes estimating acreage of a specific crop, yield estimation , cropping pattern ,etc.(Panigrahy et al, 2009).The results of remotely sensed data providing acreage and spatial distribution of individual crop can be of immense help to policy makers, scientists , agricultural policy makers, businessmen, etc.

Traditionally remotely sensed data has been used extensively to extract thematic class information. Multi-spectral classification of remotely sensed images has been applied using different methods like hard classification, fuzzy and hybrid approaches using ancillary information. The results of these classifications techniques are unique classes according to some statistically determined rules. (Jenson,1996). These classes contain only hard information class ( one pixel is assigned to one class only).

But a single-date data may not be useful to discriminate specific crop. This may be because the spectral response recorded by a sensor may be similar with crops of other types or might overlap with dissimilar classes depending on the agricultural practices followed by farmer, different crop varieties occurring contiguously in the vicinity, etc. Thus different crops may show similar spectral response in a single date imagery. Therefore multi-date/temporal images can allow differentiation between crops or different classes. (Semwal, 2010) and ( Simonneaux and Francois , 2003).

Since every crop has its own unique phenological cycle, it has been found that identification of specific crop using temporal remote sensing data is useful.(Li et.al , 2010).So we need to select proper time for acquiring remote sensed images of proper growth stages of crop to prepare a phenological cycle that can help us in discriminating it from other contiguously occurring crops in the vicinity. (Phenology refers to the occurrence of the important growth stages of the plant/crop in time).

Number of times while acquiring temporal data every few days interval, data may not be available due to cloud cover or non-availability of satellite at the specified time of crop growth. A multi-sensor approach has been found to be useful in providing solution to this problem. (McNairn et al, 2005)

Another problem encountered in using moderate and coarse resolution data is the occurrence of mixed pixels. In hard classification techniques each pixel is assigned to a single class only. But while dealing with coarse resolution data, more often it has been seen to deal with mixed pixels. Mixed pixels occur when a pixel is not entirely covered by a single class but is composed of proportions of two or more classes. It occurs at the boundaries of two or more classes or when the size of the object is less than the spatial resolution of the sensor or when there is a zone of transition between two classes. The problem of mixed pixels has been traditionally tackled by using Linear Mixture Model (LMM) and/or Fuzzy based techniques. Zadeh(1965) introduced the “Fuzzy Set” theory. These techniques deal with sub pixel classification by assigning membership values or proportions to different classes with in a pixel. Its value ranges from 0.0 to 1.0

So a multi-source temporal data approach can be used for identification of specific crop by modelling the crop’s spectral response over time. Therefore by relating the observed spectral response to the expected spectral response of the crop we can assign labels to different crops (Ali, 2002).

The last step is the accuracy assessment of Soft Classified outputs. In hard classification accuracy assessment an error matrix is built for classified classes and corresponding reference class. The sum of the no. of pixels in the diagonal of the matrix provides the overall accuracy of the classification technique and the individual element of the element of the diagonal provides specific class accuracy. Since in fuzzy classification the outputs are fractional images the traditional error matrix for hard classification cannot be used. In the past the fuzzy outputs have been classified using the Fuzzy Error Matrix (FERM). (Binaghi, et al, 1999). A sub pixel confusion uncertainty matrix (SCM) for assessing soft classifiers has also been proposed that reports the confusion intervals in the form of a centre value plus-minus maximum error. (Silvan-cardenas and Wang, 2007). Since in mixed pixels the single pixel contains different proportions of two or more classes a sub-pixel confusion matrix (SCM) technique has also been previously used(Kumar and Dadhwal,2010).

Research Objectives:

1. To identify the no. of temporal data sets required from multi-sensors for creating crop phenology map
2. To create a crop phenology map for identification of specific crop
3. To handle mixed pixels for accurate crop area estimation
4. Image to image based accuracy assessment of fuzzy output classes

Research questions:

1. How to identify and map specific crop?
2. How many time-series data for mapping specific crop is required?
3. How effective the multi-sensor, temporal data is to identify the specific crop?
4. What is the effectiveness of image to image based accuracy assessment method?

DATA:

IRS P6 Data

Indian Remote Sensing (IRS)P6 (RESOURCESAT-1)was launched in October 2003. The sensors on-board this satellite are : LISS III and LISS IV and an advanced Wide Field Scanner (AWiFS). The LISS 3 is multispectral and operates in the visible to mid-infrared region and has a spatial resolution of 23.5m and its swath width is 140 km. The spatial resolution of LISS IV 5.8m.

The instrument swath width of multispectral (visible to near-infrared) is 23.5km and for the panchromatic (monochrome) mode it is 70 km. The spatial resolution of AWiFS is 60m and swath width is 740 km. Radiometric resolution of LISS III and LISS IV is 7 bits, while the AWiFS has 10-bit.This orbit of the satellite is polar, circular and sun-synchronous with an apogee of 800-km, and has a 24-day repeat cycle for the LISS 3 sensor and the repeat cycle for LISS IV is 5 days.

LISS 3

LISS 4

AWiFS

Source: http://geo.arc.nasa.gov/sge/health/sensor/sensors/irsp6.html#spec

(Visited on 16/05/2011)

RESOURCESAT- 2 :

It was launched in April 2011. It is intended to continue the data provide by RESOURCESAT- 1 along with some improved features. It has improved LISS IV multi spectral swath from 23 km to 70 km and improved radiometric accuracy from 7 bits to 10 bits for LISS III and 10 bits to 12 bits for AWiFS.

Source: http://www.isro.org/satellites/resourcesat-2.aspx( Visited on 16/05/2011)

ASTER :

ASTER is the successor of OPS or JERS-1.

Spectral Coverage 0.52-0.86µm 1.60-2.43µm 8.125-11.65µm

Source: http://www.satimagingcorp.com/satellite-sensors/aster.html (Visited on 16/05/2011)

Methodology:

The methodology consists of acquisition of multi sensor and temporal data. Then pre-processing of data such as radiometric corrections, geo-referencing, atmospheric corrections etc. has to be done. Then identification of pure-pixels or end members for generating crop phenology map is done. This crop phenology map is then used for identification of the specific crop. Since the data we are using is coarse resolution the problem of mixed pixels is handled through fuzzy techniques. At last the output is validated using image to image accuracy assessment method (Sub-pixel confusion uncertainty matrix), where the soft classified output of coarse resolution AWiFS and LISS III/ASTER is compared with fine resolution LISS IV data which is used as reference.

September October November December

Preprocessing (Geo referencing, Atmospheric corrections, etc.)

And preparing crop phenology map

EXPECTED RESULTS:

Specific crop is identified using temporal data sets and fuzzy classification techniques.