| GISdevelopment.net ---> AARS ---> ACRS 1994 ---> Poster Session |
Crop Yield Prediction in
Command Area using Satellite Data
C. S. Muthy, S. Jonna, P. V. Raju, S. Thurivengadachari and K. A. Hakeem
Water Resources Group, National Remote Sensing Agency
Dept of Space, Govt. of India
Hyderabad 500 037, India
C. S. Muthy, S. Jonna, P. V. Raju, S. Thurivengadachari and K. A. Hakeem
Water Resources Group, National Remote Sensing Agency
Dept of Space, Govt. of India
Hyderabad 500 037, India
Abstract
Advance information on crop yield during the season in irrigated command areas is vital to effect the correction measures in problem distributaries and to achieve the optimal utilisaion of irrigation water, in addition to it is importance in efficient post harvest management. The existing procedures for crop yield estimation through crop cutting experiments (CCE), can provide the estimates only at larger area units such as total command area, that too only 2-3 months after only and hence the historic yield information is not available for most of the command areas. In view of these limitations, as it provides information on smaller areal units such as distributary, before the Command Area for yield prediction of paddy crop during rabi 1993-1994 using IRAS and LANDSAT data. NDVI statistics pertaining to each experimental plot of CCE conducted during rabi 1992-93, as represented by a grid of 3x3 pixels has been extracted for different physiologicl stages. Relationship between yield and NDVI at different stages indicates that the correlation is stronger with NDVI at heading stage and with time composited NDVI (TCV). As TCVI during the season takes care of differential crop calendar in the study area, it is found to be more relevant to use is relationship with yield for further estimations. The VI-yield relationship thus derived has been used for estimating the yield at distributary level during rabi 1992-93. The relationship s extended for rabi 1993-94 after taking into account the changes in crop calendar. The estimates are obtained for CCE plots and are validated by comparing with actual yield as recoded in CCE. While the estimates are found to be promising in accuracy with less than 10 per cent deviation from actuals, there is need to refine the yield and NDVI model by integrating the CCE data of two to three seasons so as to get more dependable estimates is subsequent yield predications.
Introduction
In the last few years attention has been paid towards using satellite remote sensing data in crop estimation surveys in view of its advantages over traditional procedures in terms of cost effectiveness and timeliness in the availability of information over larger areas. Crop yield estimation well in advance in the season gained much importance in this direction, as no reliable conventional procedures are available. The existing procedures derive post harvest yield estimates through CCE, over larger administrative units such as district. Even such procedures have not been operationalised on regular basis in irrigated command areas. Hence there is a need for developing an objective, standardized and possibility cheaper and faster methodology for predicting spatial crop yields in irrigated measures in problem areas so as to achieve optimal utilization of irrigation water, in addition to its importance in efficient post harvest management. Remote sensing and crop growth simulation models have become increasingly recognized as potential tools for growth monitoring and yield estimation (Bauman 1992). District level crop yield estimation using satellite data has been operationalised in India (SAC 1990). The estimates of wheat yield obtained through ground surveys have been improved in accuracy through stratification of primary sampling units using TM derived vegetation indexes (Sing et.al 1992). Tennakoon (1992) estimated the yield of rice using TM data by developing the relationship between reflectance values and actual grain yield. Thus several studies have demonstrated the application of satellite data in crop yield estimation. However, use of satellite data for prediction/estimation of crop yield particularly its spatial variability within the command area is new application discussed in this paper.
Study Area
The study is conducted in Bhadra project command area. Bhadra river projected consists of a storage reservoir with a capacity of 2025 million cubic meters a left bank canal and a right bank canal with irrigatable areas of 7031 ha. And 92360 ha. respectively. The command area is divided into three administrative divisions namely Bhadravathi, Malebennur and Devangere. Paddy is the principal crop in both kharif and rabi seasons in the command are. During rabi, the crop is transplanted during February/March and harvested during May/June.
Satellite Data
The information on date of transplantation and date of harvesting over experimental plots of CE conducted during rabi 1992-93 have been analysed to study the general rop calendar to enable selection of satellite data. IRS IA data of 20.02.93 and 14.03.93 have been used for classification of paddy area as these dates represented the standing paddy crp at 15-30 days after transplantation. IRS IA data of 05.04.93 and 27.04.93 and IRS IB data of 16.0.493 and 08.05.93 and have been used for yield estimation as this period represents panicle initiation heading and maturity phases of the crop. During rabi 1993 - 94, IRS-1B data of 19.2.94 and 2.03.94 have been used for crop classification and TM data of 16.4.94 and 2.5.94 and IRS 1B data of 25.04.94 have been used for yield estimation.
Base Map Preparation
Using the command area index maps supplied by the field authorities and Survey of India topo sheets in 1:50,000 scale, a base map of the command area is prepared 1:50,000 scale (Raju e al. 1994). The base map is essentially a map of the command area consisting of canal/distributary network, major rivers/streams, reservoirs/waer bodies, settlements, roads railways with each distributary command area delineated. The base map is used to extract distributary wise crop area and average NDVI.
NDVI Generation
Normalized Difference Vegetation Index (NDVI) is calculated from readiometrically noramlised red and infrared reflectance values. NDVI is calculated with IRS data using the equation:
Where,
CH4 = radiance in infra red channel
CH3 = radiance in red channel
NDVI images of paddy area are produced in 8 bits using gain = 400 and offset = 0
Yield Estimation
The existing procedures for crop yield estimation in irrigated command area through CCE can provide estimates only at larger areal units such as total command area, that too only 2-3 months after harvesting. For generating spatial yield information, with CCE, separate sampling design is needed for each areal unit. Which leads to numerous experimental plots making the survey uneconomical an cumbersome. Hence, satellite based crop yield estimation attains grater importance, as it provides information on smaller real units such as distributray, before the harvesting season. Tennakon et. Al (1992) classifieds TM image of six bands covering paddy area according to yield variability s obtained through interviews of cultivators in the study are and obtained a good correlation between actual grain yield and reflectance values in some bands of the images taken during maturity state of rice. The validity of yield data collected in such studies is influenced by subjectivity in response, respondent differences and non response. As a result the variability in yield many not be accounted completely as training areas cover only broad categories of yield. This limitation is overcome through yield modeling experiments which involve establishing a valid relationship between yield and its attributes data collected over experimental plots. Satellite derived crop condition is one such important attribute. The relationship thus derived can be employed for estimating yield at smaller areal units. Several studies in remote sensing applications have proved the reliability of derived relationship between yield and NDVI for se in subsequent estimations (Rasummussan 1992 and SAC 1990).
The data on yield, date of transplantation, date of harvesting and location of different plots of CCE conducted during Rabi of 1992-93 have been analysed. The experimental plots are identified on the satellite image with the help of base map, in the form of 3x3 pixel grids. NDVI statistics from multidate satellite data pertaining to different physiological stages have been extracted for different plots. A time composited NDVI image is also generated using three dates of 5-4-93, 16-4-93, 27-4-93 and 08-05-93 to account for spatial variations in crop calendar, and to arrive at the maximum greenness vale for each paddy pixel which represents just before heading stage. NDVI statistics is extracted from time composited image also for each plot.
Correlation between yield and NDVI
The validity of crop yield models with satellite derived NDVI is determined by the strengths of association between the two variables included in the model. Hence it is essential to have understanding about the correlation existing between yield and NDVI at different phonological stages of crop for selecting appropriate date of pass to include in the model. NDVI statistics extracted from multidate satellite data representing panicle initiation, heading and ripening stages over CCE plots has been correlated with yield data. The results are presented in Table 1
Table 1 Correlation between yield and NDVI at different stages
of paddy crop over CCE plots
The correlation coefficients are found to be statistical significant with the NDVI of 16th April and 27th April which represents 'towards heading phase' of standing paddy crop. The correlations have become weaker with NDVI moving away from heading i.e. at panicle initiation or ripending. Hence, it is appropriate to select the satellite data representing either heading or just before heading phase. However, in practice, as single date may not represent the same physiological phase allover the study are due to differential crop calendar adopted by farmers. As a result the correlation coefficient is not applicable to all over the study area due to differential crop calendar adopted by farmers. Such problems can be overcome through time composition of multidate satellite data representing, in general, panicle initiation heading phases (Thiruvengadachari et al. 1994 and Jonna 1994). As a result maximum NDVI value is considered for each crop pixel, which normally occurs at heading phase of paddy. From Table1, it may be observed that the correlation with TCVI is also high and statistically significant.
Crop yield estimation with Yield -NDVI Model
TCVI is considered for yield model as it accounts for differences in crop calendar in the study area so that the model can be adapted for deriving the estimates over space and time. TCVI and yield over CCE plots are used to develop the yield model. Scatter diagram and regression line between yield and TCVI are shown in Fig 1
Figure 1 Relationship between yield and TCVI over CCE Plots
The estimated regression line is
Yield (kg/ha) = 42.23 TCVI - 3439.05
(R2 = 0.75)
Using the regression equation further yield estimates re obtained for smaller areal units such as distributary command by substituting mean NDVI as input parameter. The paddy yield has also been estimated for every pixel 72.5 mt. size, though in practice the estimates are better performed for any aerial units such as whole or part of distributary command to know the spatial yield variability across the command area. The yield estimates thus obtained have been validated through oral interviews with framers.
Thus, yield model derives a general applicable relationship between yield and NDVI which can be further to obtain estimates all over the study area. However, it may be noted that the reliability of such estimates depends upon the distribution pattern of input data i.e, the data o CCE. If CCE plots are well distributed all over the study area to represent differential crop condition then the resultant yield model would be unbiased. The sampling design adopted in existing directly related to yield, for stratification and subsequent sample selection and hence there is a need to improve the methodology by incorporating satellite derived crop area and crop condition information (NRSA 1993). In view of these limitations, the plots data from improved CCE is desirable for yield model.
Yield Prediction
The paddy yield model, developed based on CCE data and satellite data of 1992-93 rabi season has been used to predict the yield during 1993-94 rabi season. Satellite data of 16th April, 25t April and 02 May 1994 was co registered and time composited. The CCE data for rabi 1993-94 were collected and TCVI statistics was extracted for each plot as was done earlier done. Sing TCVI values, yield estimates are obtained for each plot. The comparison estimated yield and the actual yield obtained CCE is presented in Table 2
Table 2 Validity of yield estimates in 1993-94 Rabi Season
The adaptability of yield model developed with 1992-93 Rabi data to predict the yield of 1993-94 yield has thus been validated. Further, yield map showing pixel wise yield has been generated for both the seasons, using the same model, to study the spatial and temporal variations in yield over the command area, which helps temporal variations in yield over the command area, which helps identification of problem areas for effecting correlation measures.
Conclusions
In the absence of any syntematic procedures, to estimate the average yield of principal crops in irrigated command areas, the satellite data based crop yield models offer tremendous scope in command area management. The spatial yield information derived from yield model is vital to identify the problem areas in the command and to evaluate the performance of irrigation system. Once the model is established after continuous validation with the data of 2-3 seasons, the crop performance may be evaluated for historic seasons for which no yield information is now available, by adopting suitable normalization procedures for satellite data and crop calendar (Jonna et al. 1994). Such information highlights the spatial and temporal variations in crop yield in the command area which is useful in economic evaluation of irrigation projects for decision making in government policy matters.
Acknowledgements
Grateful thanks are due to Prof. B.L. Deekshatulu, Director and Dr. D.P. Rao, Associate Director, National Remote Sensing Agency, Hyderabad, India, for the valuable guidance and supported extended to this study. The valuable co-operation and support extended to this study. The valuable co-operation and support received from Shri M.L Lath, Commissioner (WM), Ministry of Water Resources, Government of India is also gratefully acknowledged. This study would not have been possible without the wholehearted support of Bhadra Project Engineers.
References
Advance information on crop yield during the season in irrigated command areas is vital to effect the correction measures in problem distributaries and to achieve the optimal utilisaion of irrigation water, in addition to it is importance in efficient post harvest management. The existing procedures for crop yield estimation through crop cutting experiments (CCE), can provide the estimates only at larger area units such as total command area, that too only 2-3 months after only and hence the historic yield information is not available for most of the command areas. In view of these limitations, as it provides information on smaller areal units such as distributary, before the Command Area for yield prediction of paddy crop during rabi 1993-1994 using IRAS and LANDSAT data. NDVI statistics pertaining to each experimental plot of CCE conducted during rabi 1992-93, as represented by a grid of 3x3 pixels has been extracted for different physiologicl stages. Relationship between yield and NDVI at different stages indicates that the correlation is stronger with NDVI at heading stage and with time composited NDVI (TCV). As TCVI during the season takes care of differential crop calendar in the study area, it is found to be more relevant to use is relationship with yield for further estimations. The VI-yield relationship thus derived has been used for estimating the yield at distributary level during rabi 1992-93. The relationship s extended for rabi 1993-94 after taking into account the changes in crop calendar. The estimates are obtained for CCE plots and are validated by comparing with actual yield as recoded in CCE. While the estimates are found to be promising in accuracy with less than 10 per cent deviation from actuals, there is need to refine the yield and NDVI model by integrating the CCE data of two to three seasons so as to get more dependable estimates is subsequent yield predications.
Introduction
In the last few years attention has been paid towards using satellite remote sensing data in crop estimation surveys in view of its advantages over traditional procedures in terms of cost effectiveness and timeliness in the availability of information over larger areas. Crop yield estimation well in advance in the season gained much importance in this direction, as no reliable conventional procedures are available. The existing procedures derive post harvest yield estimates through CCE, over larger administrative units such as district. Even such procedures have not been operationalised on regular basis in irrigated command areas. Hence there is a need for developing an objective, standardized and possibility cheaper and faster methodology for predicting spatial crop yields in irrigated measures in problem areas so as to achieve optimal utilization of irrigation water, in addition to its importance in efficient post harvest management. Remote sensing and crop growth simulation models have become increasingly recognized as potential tools for growth monitoring and yield estimation (Bauman 1992). District level crop yield estimation using satellite data has been operationalised in India (SAC 1990). The estimates of wheat yield obtained through ground surveys have been improved in accuracy through stratification of primary sampling units using TM derived vegetation indexes (Sing et.al 1992). Tennakoon (1992) estimated the yield of rice using TM data by developing the relationship between reflectance values and actual grain yield. Thus several studies have demonstrated the application of satellite data in crop yield estimation. However, use of satellite data for prediction/estimation of crop yield particularly its spatial variability within the command area is new application discussed in this paper.
Study Area
The study is conducted in Bhadra project command area. Bhadra river projected consists of a storage reservoir with a capacity of 2025 million cubic meters a left bank canal and a right bank canal with irrigatable areas of 7031 ha. And 92360 ha. respectively. The command area is divided into three administrative divisions namely Bhadravathi, Malebennur and Devangere. Paddy is the principal crop in both kharif and rabi seasons in the command are. During rabi, the crop is transplanted during February/March and harvested during May/June.
Satellite Data
The information on date of transplantation and date of harvesting over experimental plots of CE conducted during rabi 1992-93 have been analysed to study the general rop calendar to enable selection of satellite data. IRS IA data of 20.02.93 and 14.03.93 have been used for classification of paddy area as these dates represented the standing paddy crp at 15-30 days after transplantation. IRS IA data of 05.04.93 and 27.04.93 and IRS IB data of 16.0.493 and 08.05.93 and have been used for yield estimation as this period represents panicle initiation heading and maturity phases of the crop. During rabi 1993 - 94, IRS-1B data of 19.2.94 and 2.03.94 have been used for crop classification and TM data of 16.4.94 and 2.5.94 and IRS 1B data of 25.04.94 have been used for yield estimation.
Base Map Preparation
Using the command area index maps supplied by the field authorities and Survey of India topo sheets in 1:50,000 scale, a base map of the command area is prepared 1:50,000 scale (Raju e al. 1994). The base map is essentially a map of the command area consisting of canal/distributary network, major rivers/streams, reservoirs/waer bodies, settlements, roads railways with each distributary command area delineated. The base map is used to extract distributary wise crop area and average NDVI.
NDVI Generation
Normalized Difference Vegetation Index (NDVI) is calculated from readiometrically noramlised red and infrared reflectance values. NDVI is calculated with IRS data using the equation:
| NDVI = | (CH4 -
CH3) --------- (CH4 + CH3) |
Where,
CH4 = radiance in infra red channel
CH3 = radiance in red channel
NDVI images of paddy area are produced in 8 bits using gain = 400 and offset = 0
Yield Estimation
The existing procedures for crop yield estimation in irrigated command area through CCE can provide estimates only at larger areal units such as total command area, that too only 2-3 months after harvesting. For generating spatial yield information, with CCE, separate sampling design is needed for each areal unit. Which leads to numerous experimental plots making the survey uneconomical an cumbersome. Hence, satellite based crop yield estimation attains grater importance, as it provides information on smaller real units such as distributray, before the harvesting season. Tennakon et. Al (1992) classifieds TM image of six bands covering paddy area according to yield variability s obtained through interviews of cultivators in the study are and obtained a good correlation between actual grain yield and reflectance values in some bands of the images taken during maturity state of rice. The validity of yield data collected in such studies is influenced by subjectivity in response, respondent differences and non response. As a result the variability in yield many not be accounted completely as training areas cover only broad categories of yield. This limitation is overcome through yield modeling experiments which involve establishing a valid relationship between yield and its attributes data collected over experimental plots. Satellite derived crop condition is one such important attribute. The relationship thus derived can be employed for estimating yield at smaller areal units. Several studies in remote sensing applications have proved the reliability of derived relationship between yield and NDVI for se in subsequent estimations (Rasummussan 1992 and SAC 1990).
The data on yield, date of transplantation, date of harvesting and location of different plots of CCE conducted during Rabi of 1992-93 have been analysed. The experimental plots are identified on the satellite image with the help of base map, in the form of 3x3 pixel grids. NDVI statistics from multidate satellite data pertaining to different physiological stages have been extracted for different plots. A time composited NDVI image is also generated using three dates of 5-4-93, 16-4-93, 27-4-93 and 08-05-93 to account for spatial variations in crop calendar, and to arrive at the maximum greenness vale for each paddy pixel which represents just before heading stage. NDVI statistics is extracted from time composited image also for each plot.
Correlation between yield and NDVI
The validity of crop yield models with satellite derived NDVI is determined by the strengths of association between the two variables included in the model. Hence it is essential to have understanding about the correlation existing between yield and NDVI at different phonological stages of crop for selecting appropriate date of pass to include in the model. NDVI statistics extracted from multidate satellite data representing panicle initiation, heading and ripening stages over CCE plots has been correlated with yield data. The results are presented in Table 1
| S.No. | Date of satellite overpass | Approximate Phonological stage | Correlation between yield and NDVI |
| 1 | 05-04-93 | Panicle initiation heading | 0.59 |
| 2 | 16.04.93 | Panicle initiation heading | 0.85 |
| 3 | 27.04.93 | Heading | 0.94 |
| 4 | 08.05.93 | Ripending | 0.44 |
| 5 | Time composited VI | Just before heading or heading | 0.88 |
The correlation coefficients are found to be statistical significant with the NDVI of 16th April and 27th April which represents 'towards heading phase' of standing paddy crop. The correlations have become weaker with NDVI moving away from heading i.e. at panicle initiation or ripending. Hence, it is appropriate to select the satellite data representing either heading or just before heading phase. However, in practice, as single date may not represent the same physiological phase allover the study are due to differential crop calendar adopted by farmers. As a result the correlation coefficient is not applicable to all over the study area due to differential crop calendar adopted by farmers. Such problems can be overcome through time composition of multidate satellite data representing, in general, panicle initiation heading phases (Thiruvengadachari et al. 1994 and Jonna 1994). As a result maximum NDVI value is considered for each crop pixel, which normally occurs at heading phase of paddy. From Table1, it may be observed that the correlation with TCVI is also high and statistically significant.
Crop yield estimation with Yield -NDVI Model
TCVI is considered for yield model as it accounts for differences in crop calendar in the study area so that the model can be adapted for deriving the estimates over space and time. TCVI and yield over CCE plots are used to develop the yield model. Scatter diagram and regression line between yield and TCVI are shown in Fig 1
Figure 1 Relationship between yield and TCVI over CCE Plots
The estimated regression line is
(R2 = 0.75)
Using the regression equation further yield estimates re obtained for smaller areal units such as distributary command by substituting mean NDVI as input parameter. The paddy yield has also been estimated for every pixel 72.5 mt. size, though in practice the estimates are better performed for any aerial units such as whole or part of distributary command to know the spatial yield variability across the command area. The yield estimates thus obtained have been validated through oral interviews with framers.
Thus, yield model derives a general applicable relationship between yield and NDVI which can be further to obtain estimates all over the study area. However, it may be noted that the reliability of such estimates depends upon the distribution pattern of input data i.e, the data o CCE. If CCE plots are well distributed all over the study area to represent differential crop condition then the resultant yield model would be unbiased. The sampling design adopted in existing directly related to yield, for stratification and subsequent sample selection and hence there is a need to improve the methodology by incorporating satellite derived crop area and crop condition information (NRSA 1993). In view of these limitations, the plots data from improved CCE is desirable for yield model.
Yield Prediction
The paddy yield model, developed based on CCE data and satellite data of 1992-93 rabi season has been used to predict the yield during 1993-94 rabi season. Satellite data of 16th April, 25t April and 02 May 1994 was co registered and time composited. The CCE data for rabi 1993-94 were collected and TCVI statistics was extracted for each plot as was done earlier done. Sing TCVI values, yield estimates are obtained for each plot. The comparison estimated yield and the actual yield obtained CCE is presented in Table 2
| Plot | Estimated yield (Kgs/ha) | Actual yield (Kgs/ha) | % dev. from act. yield |
| 1. | 4922 | 4810 | 2.33 |
| 2. | 5345 | 5195 | 2.89 |
| 3. | 5001 | 5421 | 7.75 |
| 4. | 5852 | 6471 | 9.75 |
| 5. | 4669 | 4949 | 5.66 |
| 6. | 4796 | 4757 | 0.82 |
| 7. | 5852 | 6032 | 2.98 |
| 8. | 5809 | 6331 | 8.25 |
| 9. | 5598 | 5456 | 2.60 |
The adaptability of yield model developed with 1992-93 Rabi data to predict the yield of 1993-94 yield has thus been validated. Further, yield map showing pixel wise yield has been generated for both the seasons, using the same model, to study the spatial and temporal variations in yield over the command area, which helps temporal variations in yield over the command area, which helps identification of problem areas for effecting correlation measures.
Conclusions
In the absence of any syntematic procedures, to estimate the average yield of principal crops in irrigated command areas, the satellite data based crop yield models offer tremendous scope in command area management. The spatial yield information derived from yield model is vital to identify the problem areas in the command and to evaluate the performance of irrigation system. Once the model is established after continuous validation with the data of 2-3 seasons, the crop performance may be evaluated for historic seasons for which no yield information is now available, by adopting suitable normalization procedures for satellite data and crop calendar (Jonna et al. 1994). Such information highlights the spatial and temporal variations in crop yield in the command area which is useful in economic evaluation of irrigation projects for decision making in government policy matters.
Acknowledgements
Grateful thanks are due to Prof. B.L. Deekshatulu, Director and Dr. D.P. Rao, Associate Director, National Remote Sensing Agency, Hyderabad, India, for the valuable guidance and supported extended to this study. The valuable co-operation and support extended to this study. The valuable co-operation and support received from Shri M.L Lath, Commissioner (WM), Ministry of Water Resources, Government of India is also gratefully acknowledged. This study would not have been possible without the wholehearted support of Bhadra Project Engineers.
References
- Bauman, B.A.M, 1992, Linking physical remote sensing models with crop growth simulation models applied for sugarbeet, International Journal of Remote Sensing, 14, pp2565-2581
- Jpnna.S, 1994, Compensating for variations due to differences in crop calendar in satellite evaluation of irrigation system performance in command areas, communicated to Indian Journal of Remote Sensing.
- Jonna.S., Chari.S.T., Raju P.V and Murthy C.S 1994, Satellite data normalization for change detection studies in irrigated command areas, to be presented in ICORG, Dec (3-6), JNTU, Hyderabad, INDIA.
- NRSA., 1993 Remote Sensing Data in crop cutting experiments, Technical, Report, NRSA, India.
- Raju P.V, MurthyC.S Jonna S. and Thrivengadachari.S. 1994 Integration of Cadastral and topomaps for monitoring and evaluation of an irrigated command area, paper sent to INCA synopsis, to be held at Banglroe, India, Dec 1994.
- Rasumussen M.S 1992, Assessment of millets yields and production in northern Burkinor Faso using integral NDVI from the AVHRR, International Journal of Remote Sensing,18, pp3431-3442.
- SAC, 1990 Status report on crop acreage and production estimation, RSAM/SAC/CAPE/SR/25/90, ISRO, INDIA.
- Singh.R., goyal R.C Saha S.K. and Chhikara R.S 1992, use of satellite spectral data in crop yield estimation surveys, International Journal of Remtoe Sensing, 14, pp2583-2592.
- Tennakoon S.B. Murthy V.V.N and Euimoh A.a 1992 Estimation of cropped area and grain yield of rice using remote data, International Journal of Remote Sensing 13, pp 427-439.
- Thiruvengadachari.S., Murthy.C.S, Jonna.S, Raju P.V, Hakeem.K.A. 1994, Paddy yield estimation using satellite data-Bhadra Project Command area in Karnataka State, Project Report, Aug 1994, NRSA, Hyderabad.