Theses and Dissertations

Permanent URI for this collectionhttp://ir.daiict.ac.in/handle/123456789/1

Browse

Filters

20112

Settings

Has files: NoSubject: search.filters.subject.Mathematical models

Search Results

Now showing 1 - 2 of 2
  • Thumbnail Image
    ItemMetadata only
    Comparative study of single date (LISS-III) and multi-date (AWiFS) satellite data for pre-harvest wheat crop production forecasting of Vidisha district, Madhya Pradesh, India
    (Dhirubhai Ambani Institute of Information and Communication Technology, 2011) Thakar, Kashyap; Sharma, Rajesh; Ghosh, Ranendu
    The study was aimed to learn the techniques for crop production forecasting, using single date LISS-III data and multi date A WiFS data for Vidisha District, Madhya Pradesh, India. The techniques in complete enumeration approach for single date LISS-III is based on supervised classification of multi-date dataset by unsupervised Iterative Self Organizing Data Analysis Technique (ISODATA), labeling of classes based on temporal spectral profiles of wheat, mustard and other competing crops, discrimination of non crop classes by decision rule classifier and decision rule based integration of final classified image. This hybrid classification technique takes advantage of inherent clustering tendency of land use / land cover classes in feature space with temporal dimension added to it in terms of NDVI time series data and it also makes use of signatures of known crop classes for labeling the clusters. A common approach of sample segment includes random stratification of study area, extraction of strata, training site generation followed by supervised classification. Comparative analysis of two types of approaches shows the estimated acreage 194.1 (000’ ha) by complete enumeration approach and 198.1 (000’ ha) from sample segment approach. The acreage estimates using sample segment approach is 1.89 percent more as compared to complete enumeration acreage estimates for LISS-III data. Using A WiFS data the acreage estimates is 203.7 (000’ ha) in the complete enumeration approach and 206.89 (000’ ha) in sample segment approach. Here also it is found that the estimates using sample segment approach is 1.56 percent more as compared to complete enumeration. This comparison shows that the results estimated through sample segment as well as complete enumeration are not having significant difference. Comparing the acreage estimation result of A WiFS data shows 4.77 percent over estimation than LISS–III data. Comparing sample segment approaches results of both sensors; it is found that A WiFS results shows 4.43 percent over estimation than LIS-III results. This comparison shows that higher resolution sensor satellite data have more accuracy than low resolution sensor satellite data because in low resolution possibilities of mixed pixels are more than high resolution data. Yield prediction has been done using the time series analysis. The historical crop statistics have been regressed with time to predict wheat crop yield for the season 2010-11 and R2 was found to be 0.67. Predicted yield using trend model is 1773 kg / ha for Vidisha district. Production was also estimated using Acreage and yield estimates.
  • Thumbnail Image
    ItemMetadata only
    Comparative study of single date (LISS-III) and multi-date (AWiFS) satellite data for pre-harvest mustard and wheat crop production forecasting
    (Dhirubhai Ambani Institute of Information and Communication Technology, 2011) Singh, Mridula; Bairagi, G.D.; Ghosh, Ranendu
    The study presents the learning of methodology or approaches already developed for classifying wheat and mustard crops production forecasting using NDVI time-series derived from multi-date A WiFS data and single date LISS-III data for Rabi season(2011) were used in this study. Multi-date A WiFS data classification technique is based on two stage classification of multi-date dataset by unsupervised Iterative Self Organizing Data Analysis Technique (ISODATA), labeling of classes based on temporal spectral profiles of wheat, mustard and other competing crops, creation of non crop mask by decision rule classifier and decision rule based integration for final classified image. This hybrid classification technique takes advantage of inherent clustering tendency of land use / land cover classes in feature space with temporal dimensions added to it in terms of NDVI time series data and it also makes use of signatures of known crop classes for labeling the clusters. The techniques used in complete enumeration approach for single date LISS III data analysis was based on supervised classification using digital ground truth layer. Sample segment approach includes stratification of study area, extraction of strata, training site generation followed by supervised classification and aggregation for LISS-III data analysis. An inter-sensor comparison of acreage estimates for wheat and mustard crops shows that the acreage estimates through LISS-III single date data having -2.53 percent under estimation when compared with A WiFS multi date estimates for wheat crop, while it was -2.12 for mustard crop. The yield prediction models have been developed based on historical NDVI and Agro- met indices. The NDVI based regression model has R2 0.77 and Agro- met indices model has R2 0.59. The yield predictions using NDVI and Agro- met based yield model were 2627 and 2728 respectively. When the acreage estimates for single date LISS- III data by sample segment and complete enumeration approaches were compared a under estimate of -2.99 and -7.27 was found for wheat and mustard respectively. For multi-date A WiFS data the deviation came out to be -0.087 and -0.82 for respective crops.