بررسی قابلیت داده‌های چندزمانة ENVISAT-ASAR به‌منظور تفکیک زمانی و مکانی مزارع برنج

Introduction
Rice is one of the most important crops in tropical areas, specifically in Southeast Asia. The specific climatology with always cloudy conditions; temporal variability in sowing, growing, and harvesting of rice crops and also different cultivation timetables, make the necessity of using multi-temporal RADAR and Synthetic Aperture Radar (SAR) images for obtaining quantitative and qualitative information from rice fields in tropical areas. The objective of this research is the evaluation of the ability of multi-temporal ENVISAT-ASAR datasets for temporal and spatial discrimination of rice fields, therefore, Palolo Valley, Central Sulawesi, Indonesia as the area of study was selected. The evaluated images were acquired in multi-polarimetric mode (HH, HV or VV, VH) from 4th February to 28th July 2004. These datasets includes two raw or primary datasets: Co-Polarized (HH, VV) and Cross-Polarized (HV, VH) and five derivative or secondary datasets: Mean Texture Co-Polarized, Mean Texture Cross-Polarized, Monthly Subtraction, Polarized Subtraction and Normalized Polarized Subtractions. Here the advantages of the relationship between rice growth process and the changes in the SAR backscatter coefficients using the multi-temporal ENVISAT-ASAR datasets are taken. After doing the preprocessing and the processing stages, the obtained results were evaluated using high resolution Quickbird/MS satellite imagery and available field works information. The obtained results for spatial discrimination of rice fields show the highest accuracy for Co-Polarized dataset with an overall accuracy (88.05%) and Kappa coefficient
(0.867), while Normalized Polarization Subtraction dataset offered the lowest amount of accuracy with an overall accuracy (80.23%) and Kappa coefficient (0.781). The results of temporal discrimination of rice fields show a good correspondence between the rice growth and SAR backscatter coefficients.

Materials and methods:
This study is carried out in one of the most important rice growing areas in Sulawesi, Indonesia where is located within the Sunda Islands The Lore-Lindu National Park, covering an area of 229.000 hectares. This area is placed in the centre of the study region of the STORMA project. For the purpose of the study, the area is divided into two classes, where 28% is rice and 72% is non-rice fields. ENVISAT/ASAR satellites images for six months, from 4th February to 28th July 2004 have been utilized. All polarization modes of the imagery (HH, HV, VV, VH) have been investigated. Due to the lack of precise information about the timetable planting of paddy patches we only used the backscatter change responses of different rice fields during the six months for this work. Data analysis procedures are included in two main pre-processing and processing stages. As the original SAR data has many radiometric and geometric distortions, the basic and elementary pre-processing steps are carried out. Here, after visual comparison of four de-speckling techniques included Frost filter, Lee filter, Lee filter and Time Series filter in different polarizations (HH, VV, VH, HV), the Time Series filter as the best one has been selected. Geometric correction and radiometric calibration using 1:50,000 DEM have been done which is based on the SARscape algorithm with backward geocoding methodology. In order to discriminate the rice fields, seven separate data sets have been evaluated.Two of them are directly from the original imagery; two of them are based on mean texture parameters obtained from two original data sets; other two data sets are computed based on polarization subtraction; the monthly subtraction; and normalized polarization subtracted.

Result and discussion:
By the help of Quickbird/MS satellite imagery nine training areas from nine visual separable rice fields have been chosen and using maximum likelihood classifier the whole data sets were classified separately. Finally, the overall accuracy and kappa coefficients are calculated. The final results show the co-polarization with discrimination accuracy about 88% could give the best results and the normalized co-polarization with the accuracy around 80% gave the weakest results.

Conclusion
The main ambition for this work was based on the multi-temporal Envisat/ASAR satellite imagery to discriminate rice fields. Despite the fact that all data sets used show somewhat similar results, the original Co-Polarized (HH, VV) data set gave the best results. Having the border of rice fields, Multi-temporal SAR data can help to get better results of temporal rice changes. Another result from this work is that texture derivatives of polarized SAR data do not significantly improve the results for the discrimination of rice fields from other crops. Multi-temporal SAR observations are able to characterize the phenological development of rice crops but for enhanced detection of growth stages and better mapping and monitoring of cropping systems, a higher temporal resolution would be desirable.