Spatial mapping of rice extent using multi temporal sentinel-1 SAR imagery in southern Telangana zone

Introduction

Rice is the staple food for more than half of the world's population and plays a crucial role in global food security India, the world's second-largest rice producer and consumer of rice, cultivates nearly 45 million hectares of rice annually and provides livelihoods to over 100 million livelihoods (IRRI 2025) [3] . The country contributes approximately 20% of global rice production, highlighting the need for timely and accurate estimation of rice cultivation area. According to the FAO [2] , India's average per capita annual rice consumption is 68.2 kg, exceeding the global average of 50 kg. Rapid population growth and changing dietary preferences are expected to increase global rice demand, with production projected to reach 555 million tonnes by 2035. (Son et al., 2014) [8] . Accurate and reliable information on rice cultivation extent is essential for national and state-level agricultural planning, food security, crop production forecasting, policy formulation and climate resilience strategies at both national and regional scales. Traditionally, the spatial extent of crop acreage has been assessed through field surveys and direct observations. These conventional approaches are limited by the issues of accuracy, efficacy and restricted geospatial coverage. However, these methods are often laborintensive, time-consuming, costly, and constrained by limited spatial coverage and variable accuracy. Recent advances in satellite remote sensing have provided efficient alternatives for large scale crop monitoring (Xue et al. 2008) [13] . Tools such as Remote Sensing (RS), Geographic Information Systems (GIS) and the Global Positioning System (GPS) have been widely adopted for land-resource mapping and agricultural land-use planning. Among these, GIS facilitates the integration, storage, analysis and visualization of multi-source spatial data, thereby supporting informed decision-making in agricultural planning and land-resource management (Reddy et al., 2018) [6] . Optical Remote Sensing is often constrained by persistent cloud cover, during the rainy season, when rice is predominantly cultivated, clouds limiting its effectiveness for timely and accurate crop mapping. In contrast, Multi-temporal Synthetic Aperture Radar (SAR) data provide all-weather, day-and-night imaging capabilities, enabling reliable monitoring of rice growth throughout the cropping season thereby improving accuracy (Setiyono et al., 2017) [7] . High-resolution SAR data further facilitates accurate estimation of rice cultivation at farm and village scales (Qader et al., 2023) [5] . Previous studies have demonstrated that rice area maps generated from SAR data can achieve high classification accuracies, ranging from 88% (Tiwari et al., 2021) [12] to 94.5 (Pazhivelan et al., 2022) [4] . Sentinel-1 operating in the C-band, is particularly well suited for rice monitoring due to its sensitivity to changes in crop structure and surface moisture Variations in VV and VH polarization backscatter effectively capture key rice phenological stages, including flooding, transplanting and tillering-through backscatter changes in VV and VH polarisations. In this context, the present study employs multi-temporal Sentinel-1 SAR imagery to map rice cultivation extent in the Mahbubnagar and Wanaparthy districts of Telangana, demonstrating the potential of SAR data for operational rice area estimation.

Study area

Mahbubnagar district in Telangana, is located at approximately 16.74° North latitude and 78.01° East longitude, with an average elevation of about 498 meters above sea level and Wanaparthy district is situated at around 16.36° North latitude and 78.06° East longitude and lies at an average elevation of approximately 400 meters above sea level. Mahbubnagar's economy is primarily agrarian, supported by red sandy and black cotton soils. The major crops grown include paddy, jowar, maize, cotton, groundnut and castor. Wanaparthy's economy is primarily agrarian, supported by Red Sandy Soils, Sandy loam Soil, loamy sand and Black Cotton Soils. The major crops grown in the district are rice, groundnut, jowar, maize, redgram, castor, cotton and chillies.

Ground truth collection:

Ground truth data was collected from different mandals of the Mahbubnagar and Wanaparthy districts during the field surveys conducted during September. The representative samples of rice, other crops, fallow land and waterbodies were recorded using Google Earth mobile based application. To ensure accurate georeferencing and minimize edge effects, ground truth points were collected at the center of each field.

Pre-processing of SAR data

Sentinel-1A SAR data was pre-processed using SNAP (Sentinel Application Platform) version 12. The large volume of SAR data requires thorough pre processing to reduce noise and error corrections to produce data of adequate quality for subsequent analysis. The pre-processing workflow includes thermal noise removal, radiometric calibration, speckle filtering (using the RefindLee filter) and terrain corrections. Batch processing was adopted to pre-process the SAR data using the batch processing module in the SNAP's Graphical Builder. Thermal noise removal was applied to eliminate random electronic noise generated by the sensor's hardware, improving radiometric accuracy and consistency of the data. Radiometric calibration was performed to derive sigma naught (σ°) backscatter coefficient values, Lee Sigma filter with a 5x5 moving kernel was employed to reduce speckle noise. Geometric distortions caused by terrain were corrected using the Range-Doppler Terrain Correction algorithm with the Shuttle Radar Topography Mission (SRTM) Digital Elevation Model (DEM). Finally, the calibrated backscattered values were converted to decibel (dB) scale. The pre-processed VV and VH polarization images were exported in TIFF/GeoTIFF format and subsequently clipped to the study area, mosaicked, layer stacked to generate a multitemporal dataset covering Mahbubnagar and Wanaparthy districts during the kharif season (June to November).

Image classification

Crop area was estimated using a supervised classification approach, based on the Random Forest (RF) machine learning algorithm. The RF classifier was developed using the threshold values derieved from temporal backscatter signatures and texture metrics, enabling the algorithm to differentiate the various land cover classes including agricultural crops, water bodies, scrub and natural vegetation. The derived decision rules effectively captured the spatial and temporal characteristics that are specific to crop for accurate identification of agricultural areas. The RF model was trained and validated using the ground truth data collected from the study area on the layer stacked temporal Sentinel-1 SAR dataset.

Accuracy assessment

The crop area map generated using the Random Forest (RF) algorithm was applied to the temporal SAR, Sentinel-1A (VV-VH polarization) data, showed better results with ground referenced data, demonstrating the effectiveness of the classification approach. In the Mahbubnagar and Wanaparthy districts, rice emerged as the predominant crop during the kharif season. The classified crop maps were then validated against ground truth data, with 75% of the samples used for model training and the remaining 25% reserved for validation in the study area. Classification performance was evaluated using a confusion (error) matrix, from which Producer's Accuracy, User's Accuracy, Overall Accuracy and the Kappa coefficient were computed.

Results and Discussion

Rice area estimation

The temporal SAR backscatter signature of the rice canopy exhibited distinct seasonal variations corresponding to different crop growth stages. In the vertical-vertical (VV) polarization, the backscatter values ranged from -14.6 dB to -9.1 dB, while in the vertical-horizontal (VH) polarization they varied from -19.71 dB to -16.3 dB throughout the growing season. Both VV and VH polarizations exhibited a characteristic temporal pattern, with a minimum at the transplanting stage due to the specular reflection from the standing water. As the crop developed, the backscatter gradually increased, reaching its maximum during the tillering and flowering stages as a result of increased canopy biomass and volume scattering. Specifically, the VV backscatter during the transplanting stage ranged from -18.0 dB to -12.0 dB, increasing to -13.0 dB to -9.6 dB during the tillering stage. Similarly, the VH backscatter varied from -25.0 dB to -18.0 dB during transplanting and increased to -17.0 dB to -14.0 dB during the tillering stage. This temporal behaviour is consistent with the findings reported by Raman et al. (2019) [10] , confirming the sensitivity of Sentinel-1 SAR backscatter to rice phenological development. The total rice area was estimated as 96,623 ha and 96,272 ha, in Mahbubnagar and Wanaparthy districts, respectively, with a deviation of 4.02% and 2.02% against the statistics reported by the Department of Agriculture (DoA) [1] , Telangana, during the kharif, 2025. Among the mandals, Pangal of Wanaparthy district has recorded the highest area about 12,069 ha followed by Koilkonda (10,183 Accuracy assessment A confusion matrix was generated to evaluate the accuracy of rice area map by using ground reference data collected during field surveys conducted in the kharif 2025 season across the study area. A total 163 validation points, representing rice and non-rice land cover classes, were used for assessment. The Random Forest (RF)-based rice acreage mapping approach using multi-temporal Sentinel-1 SAR data achieved the overall accuracy of 92.02% and with a kappa coefficient of 0.893 indicating strong agreement between classified map and reference data. The user's accuracy for the rice class was 96.92% indicating a high reliability in mapping rice crop while the non-rice class achieved a user's accuracy of 87.49%. These results indicated that rice and non-rice classes were classified with a high degree of reliability and with minimal commission errors. The producer's accuracy, measures the proportion of actual ground reference samples that were correctly identified in the classification was 94.03% for the rice class and 86.12% for the non-rice classes. These values indicate that the RF classifier's efficiency in detected rice cultivated areas with few omission errors. These results are consistent with prior research field. Subbarao et al. (2021) [9] estimated kharif rice acreage in Bhandara district, Maharashtra achieved 91% overall accuracy with a kappa coefficient of 0.80, Tamilmounika et al. (2024) [11] estimated rice area in the Cauvery Delta region of Tamil Nadu achieved overall accuracies of 92.5% for kharif with kappa coefficient of 0.85 respectively, indicating almost perfect agreement. The slightly higher Kappa coefficient obtained in the present study (0.893) further demonstrates the robustness of the proposed RF-based classification approach for rice area estimation using multi-temporal SAR data.

Conclusion

The present study demonstrated the effectiveness of multi temporal Sentinel-1A SAR data combined with the Random Forest (RF) classification approach for mapping of kharif rice areas in Mahbubnagar and Wanaparthy districts of Telangana. The proposed methodology achieved an overall accuracy of 92.06% and a kappa coefficient of 0.893, confirming the reliability of SAR based rice area estimation under monsoon conditions. These results highlight the potential of SAR based mapping as an alternative to conventional ground survey methods, particularly where Optical data availability is limited during the kharif season due to cloud cover. The study demonstrates that multi-temporal SAR data provide a dependable and operational alternative to conventional groundbased surveys and Optical Remote Sensing for crop acreage estimation during the kharif season. The proposed approach can support timely crop monitoring, agricultural planning and policy formulation and has the potential to be extended to other ricegrowing regions with similar agro-climatic conditions.

Acknowledgement

AcknowledgementI sincerely thank to the RS and GIS Lab of Professor Jayashankar Telangana Agricultural University for providing the necessary facilities, resources and continuous support throughout the research work.

References

1. Department of Agriculture, Government of Telangana. Season and crop coverage report Vanakalam -2025. 2025 Sep 10. https://agri.telangana.gov.in
2. Food and Agriculture Organization. FAOSTAT food supply database. Food and Agriculture Organization of the United Nations; 2016. https://www.fao.org/faostat/en/
3. Pazhanivelan S, Geethalakshmi V, Tamilmounika R, Sudarmanian NS, Kaliaperumal R, Ramalingam K, et al. Spatial rice yield estimation using multiple linear regression analysis, semi-physical approach and assimilating SAR satellite derived products with DSSAT crop simulation model. Agronomy. 2022;12(9):2008.
4. Qader SH, Utazi CE, Priyatikanto R, Najmaddin P, HamaAli EO, Khwarahm NR, et al. Exploring the use of Sentinel2 datasets and environmental variables to model wheat crop yield in smallholder arid and semi-arid farming systems. Science of The Total Environment. 2023;869:61716.
5. Reddy GO. Spatial data management, analysis, and modeling in GIS: principles and applications. In: Geospatial Technologies in Land Resources Mapping, Monitoring and Management. 2018;21:127-142.
6. Setiyono TD, Quicho ED, Gatti L, Campos-Taberner M, Busetto L, Collivignarelli F, et al. Spatial rice yield estimation based on MODIS and Sentinel-1 SAR data and Oryza crop growth model. Remote Sensing. 2018;10(2):293.
7. Son NT, Chen CF, Chen CR, Minh VQ, Trung NH. A comparative analysis of multitemporal MODIS EVI and NDVI data for large-scale rice yield estimation. Agricultural and Forest Meteorology. 2014;197:52-64.
8. Subbarao NVT, Mani JK, Shrivastava A, Srinivas K, Varghese AO. Acreage estimation of kharif rice crop using Sentinel-1 temporal SAR data. Spatial Information Research. 2021;29(4):495-505.
9. Raman MG, Kaliaperumal R, Pazhanivelan S, Kannan B. Rice area estimation using parameterized classification of sentinel 1A SAR data. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. 2019;42:141-147.
10. Tamilmounika R, Muthumanickam D, Pazhanivelan S, Ragunath KP, Kumaraperumal R, Sivamurugan AP, et al. Quantifying methane emissions in major rice growing areas of Tamil Nadu using remote sensing and land surface temperature model. Plant Science Today. 2024;11:01-08.
11. Tiwari SK, Rao MP, Rao GP. Paddy crop acreage assessment using Sentinel-1 (C band) SAR data in Andhra Pradesh state, India. Journal of Crop and Weed. 2021;17(2):10-18.
12. Xue Y, Li Y, Guang J, Zhang X, Guo J. Small satellite remote sensing and applications-history, current and future. International Journal of Remote Sensing. 2008;29(15):43394372.