Bipasha Paul Shukla

A Source Apportionment Approach to Study the Evolution of Convective Cells: An Application to the Nowcasting of Convective Weather Systems

A new algorithm is developed, based on Source Apportionment (SA) technique for tracking and nowcasting of Mesoscale Convective Systems (MCS) using satellite image sequences. The algorithm does not use the conventional threshold method for identifying convective regions, instead it introduces neighbourhood search criteria to select contiguous pixels. This offers better opportunity for data mining, and inherently provides an automatic method for tracking. The nowcasting scheme is based on growth curve model fitting and extrapolation using time varying coefficients. The methodology proposed in the paper is demonstrated for the life cycle of a MCS. The algorithm is also tested on an ensemble set comprising of thermal infrared image sequences acquired from Kalpana-1 satellite. Error statistics and skill score analysis indicate the operational feasibility of the proposed algorithm.

Prediction of Satellite Image Sequence for Weather Nowcasting Using Cluster-Based Spatiotemporal Regression

The flawed characterization of transitions between different meteorological structures is often regarded as one of the largest sources of error in weather forecasting. This paper attempts to improve upon the satellite-image-based nowcasting capability of models by coupling a clustering technique into a spatiotemporal autoregression method. Experimental results indicate the superiority of clustering-based regression algorithm in terms of statistically significant skill scores. The tests show an improvement in probability of detection with a decrease in false alarm rate as compared to unclassified predictions. The developed model has also been demonstrated to be useful in nowcasting of convective systems.

Extrapolation of Sequence of Geostationary Satellite Images for Weather Nowcasting

In this letter, a novel scheme for image sequence extrapolation is proposed and demonstrated, particularly for the purpose of near-real-time weather nowcasting during satellite launches. The highlight of this model is its ability to produce a sequence of simulated satellite images extended in time scale, which is very important for forecasting the evolution of a meteorological system. For this, three different models based on spatio-temporal autoregressive technique, discrete Fourier transform, and hybrid approach are developed and tested on an extensive data set of satellite image sequences covering different meteorological conditions.

A Novel Neural Network Based Meteorological Image Prediction from a Given Sequence of Images

A novel neural network method to predict the spectral signature in the predicted meteorological image is presented here. Back propagation algorithm has been used in this work. Based on computation cost, three different dimensional feature vectors are provided from two consecutive images as input to neural net for training and testing. Various kinds of testing are made depending upon position of predicted images and input images. We divide the intensity levels of each image by four clusters using \textit{K}-means clustering method and we build different neural nets for each corresponding cluster. Mean Square Error is used to evaluate the performance of the net and PSNR is used to judge the accuracy of predicted image. Results are encouraging.

Automatic smoke detection using satellite imagery: preparatory to smoke detection from Insat-3D

Identification of smoke on satellite imagery is a prerequisite to study and retrieve physical, chemical, and optical properties of smoke and also forms a crucial part in fire‐management systems. Automatic detection of smoke is a challenge in itself, owing to the large overlap in the spectral signatures of smoke and other scene types, and it becomes all the more complex over the Indian region owing to less contrast between background and target. In this study, an algorithm, which is based on multiband thresholding technique, employing a conventional as well as a pseudo‐channel, is developed for the Indian region with the help of radiative transfer simulation and is a preparatory exercise for setting up algorithms for application of INSAT 3D imager data. The algorithm has been executed using MODIS data on agricultural fire spread over north‐western India for the year 2005. The outputs are validated using MODIS AOD and Cloud Fraction products, and the results suggest that the algorithm is able to isolate smoke pixels in the presence of other scene types such as clouds, although it performs better in identifying fresh dense smoke as compared with highly diffused smoke.

INSAT-3D cloud microphysical product: retrieval and validation

ABSTRACT The 6-channel Imager onboard the Indian geostationary satellite Indian National Satellite-3D (INSAT-3D) provides half-hourly multispectral images over the Indian monsoon region. The availability of shortwave infrared (SWIR) (1.6 μm) channel along with visible (0.6 μm) channel observations provide an opportunity to estimate cloud microphysical parameters (CMP) over the India and surrounding regions with high temporal frequency. In this paper, we describe the retrieval and validation of two important CMPs, i.e. cloud optical thickness (COT) and cloud effective radius (CER) over the ocean from INSAT-3D. This is the first time; a CMP product has been made available for INSAT-3D. We describe here the development of the forward model, based on a look-up-table (LUT) approach using Radiative transfer simulations. The inversion is carried out by selecting the vector (CMP) which provides the best match between the observed and simulated radiances. The present retrieval is limited to water clouds over ocean only. The retrieved INSAT-3D CMP were then compared with MODerate Resolution Imaging Spectroradiometer (MODIS) product for the months of January and July 2016. For cloudy month (July), the mean correlation between INSAT-3D and MODIS was 0.73 and 0.47 for COT and CER, respectively. Similarly, for the month of January with less cloud cover, the mean correlation was 0.60 and 0.40 for COT and CER, respectively. INSAT-3D products are available every half hourly in real time through web portal www.mosdac.gov.in and will be valuable for studying short-term variation in cloud-microphysics over the equatorial Indian Ocean.

Prediction of meteorological images based on relaxation labeling and artificial neural network from a given sequence of images

In this paper an algorithm to predict the spectral signature of the short-term evolution of cloud formations using image sequences acquired from ISRO meteorological satellite (Kalpana-1) is described. The proposed algorithm consists of four steps: first step perform image processing activities (thresholding and relaxation); the second step is dedicated to determination of neural net for each pixel in each and every images. The third and fourth step consists of a novel neural network based training and prediction respectively. The main goal of this work is to maximize the prediction accuracy. Various kind of predictions are made depending upon number of feature vectors and number of net used. Mean Square Error is used to evaluate the performance of the neural net and PSNR is used to judge the accuracy of predicted image. Some experimental results obtained by using real image sequences acquired from ISRO meteorological satellite are shown that are extreamly encouraging.