Search

The role of feature-level fusion using the Dempster-Shafer rule and that of the data-level fusion in the Markov random field (MRF) context is studied to obtain an optimally segmented image. Subsequently, segments are validated, and classification accuracy for the test data is evaluated.

To develop an operational methodology for estimating soil moisture and crop biophysical parameters and to generate a crop cover map, backscattering signatures of vegetation canopies are investigated using multitemporal Radarsat synthetic aperture radar (SAR) data over a predominantly cotton-growing area in India during low to peak crop growth stage. A simple parameterization of the water-cloud model with volumetric soil moisture content ([m.sub.v]) and leaf area idex (LAI) is used to simulate the microwave backscattering coefficient ([[sigma].sup.0]), as it is found to be a good candidate for operational purposes as demonstrated by several workers in past. The influence of crop height (H), LAI, and [m.sub.v] on [[sigma].sup.0] is investigated during peak crop growth stage. A linear relationship between LAI and crop height is derived semiempirically, and a linear zone is chosen for analysis during the peak croup-growing stage. Estimation of average volume fraction of leaves ([[bar.V].sub.l]) and attenuation factor (L) by two different approaches is discussed: 1) using linear relationship between LAI versus crop height and 2) from the water-cloud model parameter (K) estimation by iterative minimum least square error approach. It is observed that model-estimated parameters agree well with the measured values within an acceptable error limit. At lower soil moisture, [m.sub.v] [congruent to] 0.02 ([cm.sup.3] x [cm.sup.-3]), the dynamic range of [[sigma.sup.0] is found to be about +5 dB for 0-70 cm of crop height but monotonously decreases to null at a transition point, having [m.sub.v] [approximately equal to] 0.38 ([cm.sup.3] x [cm.sup.-3]). A positive correlation is found between backscattering coefficient and crop height till this transition point but shows a negative correlation beyond that, signifying the predominant attenuation by vegetation over soil. Differential moisture sensitivity (d[[sigma].sup.0]/d[m.sub.v]) of the backscattering coefficient decreases by half from 20.55 dB/([cm.sup.3] x [cm.sup.-3]) for dry and bare-field conditions to 10.68 dB/([cm.sup.3] x [cm.sup.-3]) for wet and crop-covered fields ([m.sub.v] = 0.38 [cm.sup.3] x [cm.sup.-3], H = 70 cm), whereas differential crop height sensitivity (d[[sigma].sup.0]/dH) varies from 0.22-0.03 dB/cm for bare-field conditions to crop-covered fields with crop height 70 cm. It is found that the percentage of relative error is smallest (2.27 %) for LAI and attenuation factor estimation using the value of [[bar.V].sub.l], from LAI models, whereas it is 4.25% when estimating from the attenuation coefficient (K) from the model. Index Terms--Crop attenuation, differential moisture and crop height sensitivity, leaf area index (LAI), synthetic aperture radar (SAR) backscatter, water-cloud model.

Retrieval of transplantation date (TP) using RISAT-1 medium resolution SCANSAR data for operational pre-harvest crop monitoring has been demonstrated. The methodology used 3-date HH-polarized RISAT data capturing the onset of transplantation (minimal backscatter) and rises as crop grows, for the country scale within 2–4 days of planting and was found to closely agree to ground conditions. Transplantation date ranged from June in north to October in Tamil Nadu. This variation significantly reduced to a month in the summer paddy due to irrigation. Transplantation patterns were different in both the seasons' paddy. Validation exercise for TP in four major states confirmed the feasibility of retrieval datasets for transplantation date monitoring as it matched within 2–4 days in north, south India and MP but 6–8 days in Assam. The technique has reached a level of maturity and is routinely used for monitoring the transplantation date by the Department of Agriculture, India. [ABSTRACT FROM AUTHOR]

This study presents a methodology to classify rice cultural types based on water regimes using multi-temporal synthetic aperture radar (SAR) data. The methodology was developed based on the theoretical understanding of radar scattering mechanisms with rice crop canopy, considering crop phenology and variation in water depth in the rice field, emphasizing the sensitivity of SAR to crop geometry and water. The logic used was the characteristic decrease in SAR backscatter that is associated with the puddled or transplanted field due to specular reflection for little exposure of crop, with increase in backscatter as the crop growth progresses due to volume scattering. Besides, the multiple interactions between SAR and vegetation/water also lead to an increase in backscatter as the crop growth progresses. Classification thresholds were established based on the information provided by each pixel in each image, the pixel's typical temporal behaviour due to crop phenology and changing water depth in rice field and their corresponding SAR signature. Based on this logic, the study site (i.e. South 24 Paraganas district, West Bengal) was classified into three major rice cultural types, namely shallow water rice (SWR; 5 cm ≤ water depth ≤ 30 cm), intermediate water rice (IWR; 30 cm ≤ water depth ≤ 50 cm) and deep water rice (DWR; water depth > 50 cm) during the kharif season. These three types represent most of the traditional rice-growing areas of India. The methodology was validated with the field data collected synchronously with the satellite passes. Classification results showed an overall accuracy of 98.5% (95.5% kappa coefficient) compared with a maximum-likelihood classifier (MLC) with an overall accuracy of 95.5% (84.2% of kappa coefficient) with 95% confidence interval. The relationship between field parameters, especially exposed plant height and water depth with SAR backscatter, was explored to design empirical models for each of the three rice classes. Significant relationships were observed in all the rice classes (coefficient of determination, R 2, value more than 0.85) even though they had similar growth profiles but varied with water depth. The two main conclusions drawn from this study are (i) the importance of multi-temporal SAR data for the classification of rice culture types based on water regimes and (ii) the advantages and flexibility of the knowledge-based classifier for classification of RADARSAT-1 data. However, being empirical, the approach needs modification according to the current rainfall pattern and rice-growing practice. [ABSTRACT FROM AUTHOR]