S. Panigrahy, M. Chakraborty and J. S. PARIHAR
Agricultural Resources Group (RESA) Space Applications Centre (ISRO) Ahmedabad 380 053, India
Satellite remote sensing based crop inventory and monitoring has the advantage of synoptic, spatial and temporal coverage of an area. The precision of such procedures has improved over the years and now it meets the requirement of timely gathering of routine information on crop prospects. Crop Acreage and Production Estimation (CAPE) is an on-going project in India, which uses optical remote sensing data to forecast production of major cereals, oilseeds, and fibre crops (SAC, 1995). Procedures have been developed for use of single date data from Indian Remote Sensing Satellites (IRS), acquired at peak vegetative growth of the crop, to estimate the crop acreage. Attempts have been made to develop vegetation index based yield models to forecast the crop yield. These have been tested at a large number of sites, over the years and found to perform satisfactorily (Navalgund et al, 1991). After the launch of IRS-1C and 1D satellites, multi-temporal monitoring of crops at regional scales using data from WiFS sensor, also became feasible. This has been successfully demonstrated for making national level, multiple forecast of wheat crop in India.
However, multiple forecasting of crops grown in monsoon season (Kharif), like rice have not been very successful. This happens entirely due to non-availability of temporal, cloud free data. Studies of weather data and Landsat data acquired for south and south east Asia indicate that, there are fewer images available during the entire 120 day growth period of the crop, with virtually no images during the early vegetative stage (Currey et. al., 1987). Similar observations have been made in a study using NOAA-AVHRR data acquired during kharif season in India. It was found that during the months of August and September large parts of country remains cloud covered, limiting the availability of data from optical remote sensing sensors (Bhatt, 1996).
2.0 Importance of Rice Crop for India
Rice is the major food grain of world and south and south east Asia in particular. Rice is grown in almost all states in India. Kharif or rainy season is the main crop growing period in India. Rice being climatically most adaptable cereal, various types of land management systems for rice cultivation exists, these are otherwise known as cultural types. There are two predominant cultural types, the lowlands or wetlands and the uplands. The crop establishment method, the depth and duration of standing water in the fields during crop growth period and the per cent plant cover vary with the cultural practice. Lowland is the predominant cultural type in South and South-East Asia. Shallow and intermediate type of lowland rice lands are predominant in India.
Green revolution in India was the result of substantial increase in production of cereals, particularly wheat and rice. To meet the demand of growing population and to provide food security to its people in the new millennium, it is proposed to aim at doubling the food production in next ten year. Currently a variable rate of growth for different food items has been visualised. Among the cereals, rice and wheat will continue to dominate among various crops. These crops were grown in very vast regions in the country and there adoption in the non-traditional areas has been possible. Though wheat has been introduced in the parts of eastern India, but the increase in rice acreage and production in the north western parts of the country, is noteworthy. This is entirely due to adaptability of rice crop to wider range of agro-climatic conditions. Thus, rice is emerging as the principal food grain of future and management of rice crop production could emerge the key area of management in agriculture.
3.0 Early Experience in use of Sar Data for Rice Crop Inventory
Operational use of SAR data for rice area monitoring called for critical evaluation of accuracy of classification in various rice growing environments. The possibility of examining space borne radar data for large area agricultural application was realised with the successful launch of ERS-1 Synthetic Aperture Radar (SAR) in 1992. Studies were carried out to understand the signature of rice crop at different stages of growth. Early efforts were directed towards studying the classification accuracy of areas belonging to different cultural types as well as early and late sown rice. Developments in this phase are briefly described here.
3.1 Sar Signature in Relation to Rice Growing Environment
The study of SAR data showed that all low land rice irrespective of their cultural type viz. shallow, intermediate and deep water rice, exhibited a characteristic temporal backscatter profile in the SAR data. All the rice fields showed a distinct decrease in backscatter in the data which corresponded to the transplanted fields of rice. Very low backscatter was observed from rice fields during the early vegetative stage, irrespective of the cultural type. Surface scattering from the field water does not contribute much to the backscatter and the crop cover being very low, the volume scattering from the canopy was also assumed to be low at this stage of growth. Maximum contrast of rice fields was observed during this period. The field ridges, trees lining the boundaries were very bright due to corner reflection effect. The field boundaries, canals lined with trees and small drainage channels could be seen, very prominently. Rice showed the largest dynamic range of backscatter during the early stages of crop growth. In the subsequent dates, a considerable increase in backscatter was observed from all types of rice fields and it peaked around 60-80 days after transplantation. The contrast of rice fields decreased sharply and field boundaries were then no longer visible. As the crop reached near maturity, the backscatter increased significantly and the separability between classes decreased sharply. Thus, multi-temporal SAR data has emerged as a good source of remotely sensed data. The typical signature of rice crop in multi date SAR images was gainfully employed in it classification achieving over all identification and classification accuracy of more than 90 per cent (Chakraborty et al., 1997).
3.2 Rice Crop Classification with Sar Data
Preliminary analysis of C band 23 degree incidence angle SAR data of ERS were carried out during 1992-93 for rice, sugarcane, cotton and groundnut grown under different agro-climatic condition in India (SAC, 1995). These studies showed that the wetland cultivation practice of rice exhibits unique temporal backscatter in temporal ERS SAR data. Low backscatter characterised freshly transplanted rice fields due to specular returns from water in puddled fields and it increased steadily with the crop growth. Multidate data acquired at critical bio-window of the crop growth was used to identify and classify rice fields with high accuracy (Patel et. al. 1995). Similar results have been reported from other Asian rice growing areas (Kuroso et. al. 1994, ESA, 1995). Multi-date SAR data available from RADARSAT was critically examined for classification accuracy of rice crop representing varying rice culture types.
3.3 Importance of Dates Of Acquisition of Multi-Temporal Sar Data
Data acquisition period in relation to crop growth was found critical to obtain high classification accuracy. A set of combination were tried to evaluate the classification accuracy achieved. Among the set of 2,3 and 4 acquisition data sets, three date data was found to result in higher than 95 percent classification accuracy, which was considered optimum for this purpose. While selecting the dates of acquisition, it was observed that, the data acquired at field preparation stage i.e. puddling/transplanting stage was essential for rice crop estimation. The 35 day repeat cycle of ERS SAR was found to be a constraint in the selection of optimum dates of data acquisition, coinciding with proper bio-windows.
4.0 Large Area Rice Crop Inventory
Systematic efforts on district level rice acreage estimation using multi-date ERS SAR data for selected districts of West Bengal and Orissa during 1994-95 kharif season (Panigrahy et al., 1997). Boundary mask approach with complete enumeration of data was adopted in this case. During the year 1996-97 study was extended to larger areas with selection of districts in the states of Assam, West Bengal, Orissa and Tamil Nadu. In this case a sampling based approach was adopted for selection of multi date ERS- SAR data for digital classification.
The scope of utilising SAR data widened with the launch of RADARSAT in 1995. Investigations were carried out under the project entitled “RADARSAT data evaluation for crop identification and characterization”, under Application Development and Research Opportunity (ADRO) programme (ADRO project ID 349) of RADARSAT sponsored by RADARSAT International (RSI) and Canada Centre of Remote Sensing (CCRS), Canada. In this study RADARSAT S1, S5, S6 ,S7 and ScanSAR Narrow data were used to investigate rice signature and classification (Panigrahy et al 1999). The analysis showed that rice crop retains its unique temporal backscatter profile in SAR data irrespective of incidence angle. Classification accuracy was more than 90 per cent in all the beam mode data investigated. However, the shallow incidence angle (S6, S7) were found to be less sensitive to wind induced roughness than the steep angle beam (S1). This increased the separability of water and rice field in shallow angle data. ScanSAR Narrow (SN2) data acquired at early stages of crop planting resulted 95 per cent accuracy for puddled fields in the irrigated lowland rice area. It also showed the possibility of deriving information on sowing period, growing environment etc. Due to the large swath and shallow incidence angle, SN2 data was adjudged suitable for large area monitoring.
4.1 Operational use of Sar Data for Rice Inventory
In-season rice crop monitoring was attempted in Assam, Orissa, West Bengal and Tamil Nadu states during 1998-99 kharif season using RADARSAT ScanSAR Narrow beam data. Two date data acquired early in the season were used to assess the possible crop prospects based on puddling and transplanting activities. Three and four date data were used to estimate the acreage. A stratified random sampling approach was used to analyse sample segments and estimate the area. An automated software -‘SARCROPS’ was developed for this purpose (Chakraborty, 1999). The objective of developing this software was to use a standardised and uniform technique for all study states, as well as fast and efficient data processing. The software consisted of three sub-modules- SAR image and ancillary header data extraction, optimal speckle suppression and conversion of digital numbers to backscatter. Besides this attempts were made to use the satellite ephemeral data to improve the map-to-image transformation model. This software improved the efficiency and timeliness of analysis This is of particular significance for the number of participants from the state user agencies like Department of Agriculture, State Remote Sensing Centre etc., who participate in the implementation of the project. A decision rule based classification algorithm was developed and used to delineate rice areas. The temporal data provided the required basis to categorise rice crop based on its planting date and growth characteristic.
During 1999-2000 kharif season – the project was extended to 13 states in India these states contribute around 92 per cent to national rice production. ScanSAR narrow beam-2 data acquired from pre-field preparation to about 45 days after transplanting was used. Multi-date registered data set of minimum three dates enabled estimation of rice acreage by middle of September, which is otherwise not possible with remotely sensed data.
4.2 Monitoring Progress in Planting of Rice
In India planting period of rice crop has a large spread. This leads to a situation where fields could be found having different stages of crop at any given time. Still in the large part of rice growing areas a definite crop calender is followed, any major deviation from this results in loss of production as well as affects the crop rotation practice. Use of multi date SAR data in the early part of crop growing season has been found useful to detect and monitor prolonged flooding of rice fields in parts of Assam state. This resulted in non-availability of seedlings by the time fields were available for planting. This type of information has potential use in, deciding about the area which can be covered by early planting of winter pulses and oilseeds.
It has also been possible to detect delay in planting caused due to late arrival and insufficient rains. In the state of Orissa, north-eastern part of state received low rains which was clearly observed in RADARSAT data of July and August months of 1998. Monitoring of the area till middle of September indicated that large tract of agricultural fields remained unplanted during the season.
4.3 Rice Crop Growth Modelling
The temporal backscatter of rice fields exhibited different pattern for differently managed system. The irrigated well managed rice fields showed a steady increase in initial growth which saturated early in season due to dense canopy cover. In poorly managed intermediate areas with slow crop growth, the profiles showed negligible increase during initial growth phase. Fields flooded after a initial crop establishment period exhibited a different profile. This indicates the possibility of modelling for crop growth assessment using shallow incidence angle SAR data.
4.4 Assessment of Damage Due to Flooding / Sea Serge
The state of Orissa was hit by a high magnitude cyclone (Super Cyclone) with wind speed exceeding 250 km/hour. This was associated with more than 400 mm of rain in a day and also sea surge. Large tract of land was submerged in the Mahanadi delta, predominantly rice growing area. Rice is planted in this region during the month of August and harvested in November and December months. Thus the rice crop was at early stage of reproductive phase or nearing maturity. ScanSAR data of November 2 and 4 was the only usable data available for assessing the areas affected by floods at the time. These were used in conjunction with the ScanSAR narrow beam 2 data of July and August months. The flood area mask generated with the November data was overlayed on the rice area map created earlier. It was found that out of 1107.4 thousand ha rice 316..36 thousand ha i.e. 28 percent of rice grown in the 8 coastal districts of Orissa state was submerged for a period of 4 days or more. This would lead to loss of production of 426.9 thousand tonnes of rice due to inundation alone.
5.0 Characterization of Rice Cropping Systems
The north eastern and eastern region belonging to the traditional rice area of India has a great potential for higher production due to the rich and favourable growing environment. A cropping system approach has been envisaged during the coming decade by the Ministry Of Agriculture, Govt. of India to increase the cropping intensity as well as yield. The analysis using multi-date SN2 data acquired from July to October months, highlighted the peculiarity and constraints of rice cropping system of this area. This information can be used to plan and adopt better productive cropping pattern in the region. The following are some of the observations made after the analysis of this data.
In Assam, only 15-20 per cent area was found to be transplanted by August 05. Large area flooding was observed in Barak valley as early as by July 12. The fields were submerged for more than a month in many parts of the state. Crop was sown as late as September in this area. Flood hazard probability was high in Barpeta, Nalbari, Lakhimpur and Sonitpur districts in Brahmputra valley. Flood in early July and August though does not affect the standing rice crop in most of the districts, however, it has adverse affect on the seed beds raised for nursery. The inundation of fields up to August delays the transplanting. This will have two effects – reduction in critical growth period, causing large reduction in yield, and delay in harvesting of rice, which will affect the sowing of mustard crop, the next immediate crop of the valley. The analysis indicates that the agro-ecology of the valley renders the present cropping pattern a vulnerable system. Flood is an annual phenomenon in early and mid monsoon period in one or other parts of the valley. The calendar of Sali rice, the major crop of the area coincide with it. This prevents achieving even the minimum targeted production of the crop. In spite of favourable climate, fertile soils and the sincere efforts made by the Department of Agriculture, the productivity has stagnated and the average yield of the state is only 1250 kg/ha, one of the lowest in the country.
In the West Bengal state, more than 60 per cent areas were sown by July end, with normal onset of monsoon. Sowing is earlier in North Bengal area compared to South Bengal. Damage due to submergence of crop early in the season exits in Maldah and Murshidabad area. The alluvial zone of South Bengal region showed very good crop management practice. Bardhaman district showed the best managed crop growing environment in the state with almost no hazards of flood or moisture stress.
In Orissa, onset of monsoon and amount of rainfall was found to control the rice sowing almost in all the districts. Even in the coastal region, with irrigation facility, the crop sowing/transplanting continues as late as, September first week. Delayed transplanting and moisture stress, late in the season are the two major constraints observed in this area.
The advantage of radar remote sensing for rice crop lies in its independence from cloud cover and solar illumination. Sensitivity of SAR to canopy geometry and moisture is promising not only for crop discrimination but also to model crop growth and condition.
The wet land practice of rice field preparation was found to have unique temporal signature in C band SAR data of ERS and RADARSAT. The shallow incidence angle and HH polarisation of RADARSAT was found to be less sensitive to roughness compared to the steep angle and VV polarisation of ERS SAR. This increased the separability of water from rice field. More than 95 per cent classification accuracy for rice was achieved using three date data of RADARSAT ScanSAR Narrow B data.
ScanSAR Narrow B data with its large area coverage and shallow angle is found suitable to carry out large area rice crop monitoring. It is possible to monitor the rice lands just after puddling to peak vegetative stage. It is feasible to identify constraints of rice production in terms of flood hazard, delayed sowing, moisture stress.
Thus, in addition to crop acreage, it was feasible to derive information on progress of transplanting, anomaly in crop growth, extent and duration of flooding etc. The damage caused to rice area due to total submergence as a result of the Super Cyclone which hit the state during October 29-30, 1999 could be assessed using ScanSAR data acquired during the first week of November.
The authors are grateful to Dr. K. Kasturirangan, Chairman, Indian Space Research Organization, for kind approval to carry out the investigations and pilot studies on use of SAR data for crop inventory. The results reported in the paper are the outcome of vigorous scientific investigations by the scientists at Space Applications Centre (ISRO), State Remote Sensing Applications Centres and other participating state agencies of the state concerned. This has lead to develop understanding, demonstrate and validate the use of SAR data for survey and monitoring of rice crop and its growing environment. Initial thrust to these studies was provided by ADRO Programme of RADARSAT International and Canada Centre for Remote Sensing.
- Bhatt, N.B., 1996, State level probability of cloud free days over India using remotely sensed data, Scientific Note: RSAG/SAC/CAPE-II/SN/96, p 19.
- Bush, T. T. and F. T. Ulaby, 1978, An evaluation of radar as a crop classifier, Remote Sensing of Environment, 7, 15?36.
- Chakraborty, M., and S. Panigrahy, 1996, Evaluation of four per?pixel classifiers using ERS?1 SAR data for classification of rice crop. in Proceedings of Indo? US symposium (abstracts), IIT, Bombay, October, 1996.
- Chakraborty, M., S. Panigrahy and S. A. Sharma, 1997, Discrimination of rice crop grown under different cultural practices using temporal ERS?1 SAR data. ISPRS Photogramm.& Remote Sensing, 52:183?191.
- Chakraborty, M., 1999, SARCROPS – Procedure and Technique for the use of Synthetic Aperture Radar (SAR) Data for Rice Crop Inventory, Scientific Note: RSAM/SAC/FASAL-TD/TR/04/99, p 22.
- Currey, B., A. S. Fraser and K. L. Bardsley, 1987, How useful is Landsat monitoring. Nature, 328:587?590.
- ESA, 1995. Satellite Radar in Agriculture, Experience with ERS?1, SP?1185, ESA publi cations, pp66.
- Kuroso, T., T. Suitz, M. Fujita, K. Chiba and T. Moriya, 1993, Rice crop monitoring with ERS?1 SAR ? A first year result, Proceedings of 2nd ERS?1 Symposium, Ham burg, Germany, October 11?14, ESA SP?361, Vol.1, pp.97?102.
- Navalgund, R.R., J.S., Parihar, Ajai and P.P.Nageswara Rao, (1991). Crop inventory using Remotely Sensed data, Current Science, vol.61 nos. 3 & 4, pp 162-171.
- Panigrahy, S., M. Chakraborty, S. A. Sharma, N. Kundu, S. C. Ghose and M. Pal, 1997, Early estimation of rice acre using temporal ERS?1 synthetic aperture radar data?a case study for Howrah and Hughly districts of West Bengal, India. Int. J. Rem. Sens., 18:1827?1833.
- Panigrahy, S., K.R. Manjunath, M. Chakraborty, N. Kundu and J. S. Parihar, 1999, Evaluation of RADARSAT Standard Beam data for identification of potato and rice crops in India. ISPRS Journal of Photogrammetry & Remote Sensing 54 (1999): 254 – 262.
- Patel, N. K, T. T. Medhavy, C. Patnaik and A. Hussain, 1995. Multi temporal ERS?1 SAR data for identification of rice crop. Journal of Indian Society of Remote Sensing, 23, 33?39.
- SAC,1995, Manual for crop production forecasting using spaceborne remotely sensed data, a joint project of Department of Space and Ministry of Agriculture, Govt. of India, Technical Note, RSAM/SAC/CAPE?II/TN/46/95, Space Applications Centre, Ahmedabad,
Table 1 Rice area allocation by cultural type in India and important states (Huke, 1986)
Country/state Dry land Deep water Shallow Intermediate Irrigated India 5973 2434 12677 4470 11134 UP 714 555 1884 524 982 BIHAR 531 672 1698 847 1440 ORISSA 691 150 1743 886 893 WB 883 677 1685 853 984
Table-2: Achievements/targets(Mt) and growth rates(%) for food production in India (MOA 1998)
Item IX Plan X Plan Ten Years Food grains 234.0 (3.26) 287.9 (4.23) 300.0 (4.45) Rice 99.0 (4.02) 124.2 (4.64) 130.0 (4.53) Wheat 83.0 (3.68) 104.1 (4.65) 109.0 (5.08) Coarse cereals 35.5 (0.70) 40.0 (2.43) 41.1 (2.80) Pulses 16.5 (2.67) 17.9 (3.26) 20.0 (4.32) Oilseeds 30.0 (6.07) 42.7 (6.99) 45.0 (6.62) Sugarcane 336.0 (3.91) 435.2 (6.67) 495.0 (6.64) Fruits & Veg. 179.0 (8.75) 307.2 (11.39) 342.0 (10.33) Milk 87.0 (5.10) 121.5 (6.92) 130.0 (6.23) Egg (M No.) 3500 (5.10) 4928.6 (7.00) 5300.0 (6.29) Fish 6.9 (5.70) 9.1 (5.70) 9.6 (5.67) Meat 4.4 (in1996)