Monidipa Das

Deep-STEP: A Deep Learning Approach for Spatiotemporal Prediction of Remote Sensing Data

With the advent of advanced remote sensing technologies in past few decades, acquiring higher resolution satellite images has become easier and cheaper in recent days. However, on the other hand, it has offered a big challenge to the remote sensing community in smart image interpretation from such huge volume of data. Deep learning, which offers efficient algorithms for extracting multiple levels of feature abstractions, may be suitable to serve the purpose. This letter presents a deep learning approach (Deep-STEP) for spatiotemporal prediction of satellite remote sensing data. The proposed learning architecture is derived from a deep stacking network, consisting of a stack of multilayer perceptron, each of which models the spatial feature of the associated region at a particular time instant. The proposed method has been demonstrated on normalized difference vegetation index (NDVI) data sets, derived from satellite remote sensing imagery, containing several thousands to millions of pixels/records. The experimental results (related to NDVI prediction) reveal that the proposed architecture exhibits fairly satisfactory performance with promising learning capabilities.

A probabilistic approach for weather forecast using spatio-temporal inter-relationships among climate variables

Weather forecast is one of the major services provided by the meteorological departments. It has huge impact on the global economy, agriculture, industry, transport and so on. Weather attributes (climate variables), like air temperature, pressure, precipitation, humidity etc. are meteorological variables which depend both on the associated region (or space) and time. They are also highly inter-related to one another in spatio-temporal scale. Therefore, the analysis of these spatio-temporal inter-relationships can be helpful for forecasting weather of any region for any point of time. While there exist several approaches to weather forecast, there is only little work that deals with such spatio-temporal inter-relationships among multiple climate variables. This paper presents a probabilistic approach based on fuzzy Bayesian network (FBN) to forecast the weather condition. The approach first predicts the spatio-temporal interrelationships among different climate variables. Then the predicted relationships are utilized to forecast the weather condition of the particular region. To deal with uncertainty and imprecision present in data, the proposed weather-forecast approach uses the principles of a newly defined FBN, named as NFBN. The proposed approach has been evaluated with data sets from Fetch-Climate Explorer of Microsoft Research, and found to perform better than several existing forecasting techniques.

semBnet: A semantic Bayesian network for multivariate prediction of meteorological time series data

Abstract Meteorological time series prediction plays a significant role in short-term and long-term decision making in various disciplines. However, it is a challenging task involving several issues. Sometimes, the available domain knowledge may help in dealing with certain issues in this regard. This work proposes a multivariate prediction approach based on a variant of semantic Bayesian network, termed as semBnet. The key objective of semBnet is to incorporate the spatial semantics as a form of domain knowledge, in standard/classical Bayesian network (SBN), and thereby improving the accuracy of meteorological prediction. It has been shown that compared to SBN, the proposed semBnet is less prone to parameter value uncertainty. Empirical studies on multivariate prediction of Temperature, Humidity, Rainfall and Soil moisture demonstrate the superiority of proposed approach over linear statistical models (e.g. ARIMA, spatio-temporal ordinary kriging (ST-OK)), and non-linear prediction techniques based on ANN, SBN, hierarchical Bayesian autoregressive model (HBAR) etc. Most significantly, compared to SBN, the proposed semBnet shows average 24% improvement in mean absolute percentage error of prediction.

A Probabilistic Nonlinear Model for Forecasting Daily Water Level in Reservoir

Accurate prediction and monitoring of water level in reservoirs is an important task for the planning, designing, and construction of river-shore structures, and in taking decisions regarding irrigation management and domestic water supply. In this work, a novel probabilistic nonlinear approach based on a hybrid Bayesian network model with exponential residual correction has been proposed for prediction of reservoir water level on daily basis. The proposed approach has been implemented for forecasting daily water levels of Mayurakshi reservoir (Jharkhand, India), using a historic data set of 22 years. A comparative study has also been carried out with linear model (ARIMA) and nonlinear approaches (ANN, standard Bayesian network (BN)) in terms of various performance measures. The proposed approach is comparable with the observed values on every aspect of prediction, and can be applied in case of scarce data, particularly when forcing parameters such as precipitation and other meteorological data are not available.

FORWARD: A Model for FOrecasting Reservoir WAteR Dynamics Using Spatial Bayesian Network (SpaBN)

Natural systems, like the hydrological, climatological, atmospheric, or any other environmental processes, are extremely complex as well as dynamic in nature. It is therefore difficult to forecast, analyze, and quantify these processes by using simple empirical equations. Modeling and forecasting of reservoir water dynamics are not exceptions in this respect, as these involve various challenges due to the effect of meteorological factors, natural processes of stream flow, climatic change, and so on. The intent of our present work is to propose a novel forecasting model, FORWARD, that handles some of these issues in complex reservoir dynamics. FORWARD is based on a variant of spatial Bayesian network (SpaBN), having inherent capability of modeling impact of spatial variability of meteorological factors over the river catchment. The forecasting efficiency of FORWARD has been compared with four other linear and non-linear techniques based on six different statistical performance measures. The experimental results show the superiority of FORWARD over the other techniques. Though FORWARD has been demonstrated with respect to a case study on forecasting reservoir live capacity, the model possesses a generic structure that can also be applied in other domains by introducing minimal augmentation.

A cost-efficient approach for measuring Moran's index of spatial autocorrelation in geostationary satellite data

Spatial autocorrelation (SA), describing correlation of a particular feature/phenomenon with itself across space, is one of the major properties of any spatial data. Among the various measures of SA proposed till date, the Morans index (I) is the most common as well as significant one. However, measuring Morans I, which needs to deal with spatial weight between each pair of spatial data objects, becomes almost unfeasible in case of large-scale raster data, like geostationary satellite data, containing several millions of pixels. This paper proposes a method based on the Hadoop MapReduce framework for computing Morans I in large-scale raster data. The main contribution of the work lies in the implementation of the Mapper and Reducer processes for a cost effective estimation of Morans I, considering both rook case and queen case of spatial contiguity. The key feature of these algorithms is an efficient manipulation of the spatial weight matrix, and thereby reducing the overall memory and time requirement. The experimentation shows a promising result in this regard.

Detection of climate zones using multifractal detrended cross-correlation analysis: A spatio-temporal data mining approach

There has been a significant change in climate throughout the last few decades, resulting into the phenomenon of global warming with all its adverse effects on human life and activities. In this context, detection of climate zone is an important issue, since this may help to avert, or to take adequate measures against, any unprecedented natural calamity. Most of the existing works for this purpose are limited only to the independent study of different climate variables featuring a climate zone. In this paper, we have described a novel approach based on Multifractal Detrended Cross-correlation Analysis (MF-DXA) between each pairs of such climate variables of interest. In this approach, the spatio-temporal pattern of any location, as determined by the multifractal correlation study, has been exploited by a K-means based clustering technique, which can accurately detect various climate zones over a large region. The approach has been evaluated with the daily time series data of the year 2013 for land surface temperature and precipitation rate, collected from 73 different locations over the entire Eastern and North-Eastern region of India. The high resemblance of the identified climate zones with the World Map of Köppen-Geiger climate classification proves the accuracy and efficacy of the proposed approach.

Short-term prediction of land surface temperature using multifractal detrended fluctuation analysis

Nature changes herself continuously at every moment. Hence, the prediction of any natural phenomenon (like weather), with hundred percent accuracy, is an extremely challenging task. But in spite of all uncertainties, there exists a rhythm and intrinsic regularity in all natural events. This paper presents a novel approach to predict the land surface temperature (LST) of a particular region using the theory of fractals. Although there exist several approaches for temperature prediction, there is only little work that captures the past regularities in the system dynamics while doing the prediction. In this paper we have described a prediction framework which at first captures the regularities in the dynamics of the LST series by estimating its generalized multifractal dimensions using Multifractal Detrended Fluctuation Analysis (MF-DFA). Then the prediction is performed on the basis of these captured regularities in system dynamics. The proposed approach has been evaluated with the LST data sets (of 60 years) collected from FetchClimate Explore of Microsoft Research. The results show that the proposed approach predicts the LST more accurately than several other existing prediction techniques.

Data-driven approaches for meteorological time series prediction: A comparative study of the state-of-the-art computational intelligence techniques

Abstract With the proliferation of sensor generated weather data, the data-driven modeling for prediction of meteorological time series has gained increasing research interest in current years. The recent advancement in machine learning and artificial intelligence paradigm has made such data analysis process more effective, flexible and sound. This paper attempts to provide a comparative study of the state-of-the art computational intelligence (CI) techniques, which have been successfully applied for meteorological time series prediction purpose. The study has been carried out considering eleven distinct variants of CI techniques, especially based on artificial neural network (ANN), fuzzy logic, Bayesian network (BN) and other probabilistic models. Further, one more hybrid CI technique (SpaFBN), derived from the existing approaches, has been proposed in the present work. All these CI techniques have been empirically studied with respect to a multivariate meteorological time series prediction problem, in comparison with three benchmark statistical approaches. Overall, the experimental results demonstrate the superiority of the BN-based models in meteorological prediction. The presently proposed spatial fuzzy Bayesian network (SpaFBN) is also found to be an effective tool, especially for predicting humidity and precipitation rate time series. Moreover, the proposed SpaFBN is a generic CI technique which can be applied for predicting spatial time series from the domains other than meteorology as well.

A Deep-Learning-Based Forecasting Ensemble to Predict Missing Data for Remote Sensing Analysis

The problem of missing data in remote sensing analysis is manifold. The situation becomes more serious during multitemporal analysis when data at various a-periodic timestamps are missing. In this work, we have proposed a deep-learning-based framework (Deep-STEP_FE) for reconstructing the missing data to facilitate analysis with remote sensing time series. The idea is to utilize the available data from both earlier and subsequent timestamps, while maintaining the causality constraint in spatiotemporal analysis. The framework is based on an ensemble of multiple forecasting modules, built upon the observed data in the time-series sequence. The coupling between the forecasting modules is accomplished with the help of dummy data, initially predicted using the earlier part of the sequence. Then, the dummy data are progressively improved in an iterative manner so that it can best conform to the next part of the sequence. Each of the forecasting modules in the ensemble is based on Deep-STEP, a variant of the deep stacking network learning approach. The work has been validated using a case study on predicting the missing images in normalized difference vegetation index time series, derived from Landsat-7 TM-5 satellite imagery over two spatial zones in India. Comparative performance analysis demonstrates the effectiveness of the proposed forecasting ensemble.