Ana Maria Heuminski de Ávila

Robust Pruning of Training Patterns for Optimum-Path Forest Classification Applied to Satellite-Based Rainfall Occurrence Estimation

The decision correctness in expert systems strongly depends on the accuracy of a pattern classifier, whose learning is performed from labeled training samples. Some systems, however, have to manage, store, and process a large amount of data, making also the computational efficiency of the classifier an important requirement. Examples are expert systems based on image analysis for medical diagnosis and weather forecasting. The learning time of any pattern classifier increases with the training set size, and this might be necessary to improve accuracy. However, the problem is more critical for some popular methods, such as artificial neural networks and support vector machines (SVM), than for a recently proposed approach, the optimum-path forest (OPF) classifier. In this letter, we go beyond by presenting a robust approach to reduce the training set size and still preserve good accuracy in OPF classification. We validate the method using some data sets and for rainfall occurrence estimation based on satellite image analysis. The experiments use SVM and OPF without pruning of training patterns as baselines.

Temperature variation in the 24 hours before the initial symptoms of stroke

UNLABELLED A few studies have performed to evaluate the temperature variation influences over on the stroke rates in Brazil. METHOD 176 medical records of inpatients were analyzed after having had a stroke between 2004 and 2006 at Hospital Israelita Albert Einstein. The temperature preceding the occurrence of the symptoms was recorded, as well as the temperature 6, 12 and 24 hours before the symptoms in 6 different weather substations, closest to their houses in São Paulo. RESULTS Strokes occurred more frequently after a variation of 3 C between 6 and 24 hours before the symptoms. There were most hospitalizations between 23-24 C. CONCLUSION Incidence of stroke on these patients was increased after a variation of 3 masculine Celsius within 24 hours before the ictus. The temperature variations could be an important factor in the occurrence of strokes in this population.

CLEARMiner: a new algorithm for mining association patterns on heterogeneous time series from climate data

Recently, improvements in sensor technology contributed to increasing in spatial data acquisition. The use of remote sensing in many countries and states, where agricultural business is a large part of their gross income, can provide a valuable source to improve their economy. The combination of climate and remote sensing data can reveal useful information, which can help researchers to monitor and estimate the production of agricultural crops. Data mining techniques are the main tools to analyze and extract relationships and patterns. In this context, this paper presents a new algorithm for mining association patterns in Geo-referenced databases of climate and satellite images. The CLEARMiner (CLimatE Association patteRns Miner) algorithm identifies patterns in a time series and associates them with patterns in other series within a temporal sliding window. Experiments were performed with synthetic and real data of climate and NOAA-AVHRR sensor for sugar cane fields. Results show a correlation between agroclimate time series and vegetation index images. Rules generated by our new algorithm show the association patterns in different periods of time in each time series, pointing to a time delay between the occurrences of patterns in the series analyzed, corroborating what specialists usually forecast having the burden of dealing with many data charts.

Análise de zonas homogêneas em séries temporais de precipitação no Estado da Bahia

The aim of this study was to identify rainfall homogeneous areas in the State of Bahia, Brazil and analyze the climatic condi- tions of each area for the period between 1981 and 2010. It was applied a data mining technique, clustering (grouping of data), by using the k-means algorithm for transforming time series of precipitation in five rainfall homogeneous areas, in response to topography, maritime dimension, and weather systems operating in the region of study. Data of average monthly rainfall of 92 meteorological stations were used. The results indicate that the driest areas are situated in the central part of the state, from north to south, mainly in the north with the lowest annual volumes, around 480 mm. The area located in the north of the state contrasts with that one located on the coast, where the largest volumes of annual rainfall of the study were observed (ap- proximately 1.380 mm). The high rainfall variability occurs in almost all areas, especially in two of those of semiarid ones with Coefficients of Variation (CV) reaching 42 and 28%. This characteristic differs from the area belonging to the coastal area, which presents regular rainfall during all the year and a CV of 15%. The rainy and dry seasons are well defined. Precipitation values of the rainy season accounts for about 81% of the annual total, with emphasis on the zones located in the central-west and west of the state with 95 and 96% of the annual total.

Analysis of large scale climate data: how well climate change models and data from real sensor networks agree?

Research on global warming and climate changes has attracted a huge attention of the scientific community and of the media in general, mainly due to the social and economic impacts they pose over the entire planet. Climate change simulation models have been developed and improved to provide reliable data, which are employed to forecast effects of increasing emissions of greenhouse gases on a future global climate. The data generated by each model simulation amount to Terabytes of data, and demand fast and scalable methods to process them. In this context, we propose a new process of analysis aimed at discriminating between the temporal behavior of the data generated by climate models and the real climate observations gathered from ground-based meteorological station networks. Our approach combines fractal data analysis and the monitoring of real and model-generated data streams to detect deviations on the intrinsic correlation among the time series defined by different climate variables. Our measurements were made using series from a regional climate model and the corresponding real data from a network of sensors from meteorological stations existing in the analyzed region. The results show that our approach can correctly discriminate the data either as real or as simulated, even when statistical tests fail. Those results suggest that there is still room for improvement of the state-of-the-art climate change models, and that the fractal-based concepts may contribute for their improvement, besides being a fast, parallelizable, and scalable approach.

Use of Climate Forecasts to Soybean Yield Estimates

The soybean is an annual legum that have many industrial, human, and agricultural uses. United States are the main producing and exporters of soybean grain, ranking as the first highest agricultural commodity of this specific agricultural cultural (FAO, 2008). Considering the total production of soybeans by the 20 highest producing countries in 2008 was 35% from US, 26% from Brazil and 20% from Argentine, having equivalent agricultural commodities values. Some studies in agronomic experimental stations suggest that this culture was initially introduced in Bahia State, northeastern of Brazil, 1882. However, only after the 40s in southern of Brazil, the soybean crop became commercial in the country. Nowadays, this culture is considered the most important agricultural commodity in Brazil, and one of its main export products (Esquerdo, 2001). The production of soybean has a great importance for the economy of Brazil. Historical data of soybean harvest for Brazil (IBGE, 2008) show a high correlation between soybean production economic value and productivity of this culture (Cardoso et al., 2010). According to IBGE, the Brazilian production in 2007 was 58 million tons, with Mato Grosso, Parana and Rio Grande do Sul States adding higher crop production. Soybean cultivars can be classified according to the duration of your cycle, being early (75 to 115 days), semi-early (116 to 125 days), medium (126 to 137 days), late medium (138 to 150 days) and late (over 150 days), according to Farias et al. (2000). According to Camargo (1994), the climate is the main factor responsible by annual fluctuations in grain production in Brazil. The occurrence of drought is the main cause of harms (71% of cases), followed by excessive rainfall (22% of cases), hail, frost, pests and diseases (Gopfert et al., 1993). The observations of weather conditions applied to the crop forecast models are useful to provide the most accurate crop simulations, and the importance of solar radiation, precipitation and air temperature variables is stood out (Hoogenboom, 2000). Research has been conducted with the goal of exploring the climate patterns to improve the yield of this crop agriculture. The soybean production can be significantly affected by water

Optimum-Path Forest-Based Rainfall Estimation

Meteorological conditions are crucial for the agricultural production. Rainfall, in particular, can be cited as the most influential by having direct relation with hydric balance. Meteorological satellites that cover the whole earth have been extensively used for the development of statistical and artificial intelligence models for rainfall estimation. However, some of these techniques have flaws and need to be revisited. The Optimum-Path Forest (OPF) classifier is a novel of graph-based approach for supervised pattern recognition that have been demonstrated to be superior than Artificial Neural Networks using Multilayer Perceptron (ANN-MLP) and similar to Support Vector Machines (SVM), but much faster. We introduce here the OPF classifier for rainfall estimation using satellite images and their comparison against ANN-MLP and SVM. Another round of experiments were also executed with different metrics to show the robustness of our image descriptor. We are also the first to derive the OPF classifier complexity analysis.

SART: a new association rule method for mining sequential patterns in time series of climate data

Technological advancement has enabled improvements in the technology of sensors and satellites used to gather climate data. The time series mining is an important tool to analyze the huge quantity of climate data. Here, we propose the Sequential Association Rules from Time series - SART method to mine association rules in time series that keeps the information of time between related events through an overlapped sliding-window approach. Also the proposed method mines association rules, while the previous ones produce frequent sequences, adding the semantic information of confidence, which was not previously defined by sequential patterns. Experiments were conducted with real data collected from climate sensors. The results showed that the proposed method increases the number of mined patterns when compared with the traditional sequential mining, revealing related events that occur over time. Also, the method adds the semantic information related to the confidence and time to the mined patterns.

Rainfall Estimation Using Transductive Learning

Precipitation is a crucial link in the hydrological cycle, and its spatial and temporal variations are enormous. A knowledge of the amount of regional rainfall is essential to the welfare of society. Rainfall can be estimated remotely, either from ground-based weather radars or from satellite. Despite the large amount of available data provided by satellites, most of them are unlabeled, and the acquisition of labeled data for a learning problem often requires a skilled human agent to manually classify training examples. In this paper we introduce the use of semi-supervised support vector machines for rainfall estimation using images obtained from visible and infrared NOAA satellite channels. The semi-supervised learners combine both labeled and unlabeled data to perform the classification task. Two experiments were performed, one involving traditional SVM and other using semi-supervised SVM (S3VM). The S3VM approach outperforms SVM in our experiments, with can be seen as a good methodology for rainfall satellite estimation, due to the large amount of unlabeled data. The accuracies obtained for SVM and S3VM were, respectively, 90.6% and 95.96%.

To be or not to be real: fractal analysis of data streams from a regional climate change model

This paper proposes a new analysis process aimed at discriminating the temporal behavior of the data generated by climate models from the real climate observations gathered from ground-based meteorological stations. Our approach combines fractal data analysis and the monitoring of the real and the model-generated data streams to detect deviations considering the intrinsic correlation among climate time series. Experimental studies showed that our approach can discriminate the data either as real or as generated by a model. Those results suggest that there are yet space to improve the climate change models, and that the fractal-based concepts may contribute in this improvement.