Search

Although there is an increasing interest in index-based insurances in many developing countries, crop data scarcity hinders its implementation by forcing insurers to charge higher premiums. Expert knowledge has been considered a valuable information source to augment limited data in insurance pricing. This article investigates whether the use of expert knowledge can mitigate model risk which arises from insufficient statistical data. We adopt the Bayesian framework that allows for the combination of scarce data and expert knowledge, to estimate the risk parameter and buffer load. In addition, a benchmark for the evaluation of expert information is created by using a richer dataset generated from resampling. We find that expert knowledge reduces the parameter uncertainty and changes the insurance premium in the correct direction, but that the effect of the correction is sensitive to different strike levels of insurance indemnity.

Summary and conclusions Rapidly growing countries, and particularly those emerging from rural to urban, frequently lack the resources to adequately monitor and forecast the impacts of land development on bio-logic, hydrologic, and social resources. Devel-opment often leads to more intensive land use and related increase in the generation of pollu-tants. Modeling these changes is critical to evalu-ate emerging land use and potential problems in water quality in the watershed. The objective of this study was to assess the nature and emergence of spatial patterns in land use and population density on the water quality of streams in the Chugoku district of Japan. The study employed artificial neural network (ANN) technique to assess the relationship be-tween the total phosphorus in river water and the land use in 21 river basins in the district. The results suggest that the ANN could be a viable tool to relate the land use characteristics to the water quality. The developed model was able to reasonably estimate the Total Phosphorus (TP) in the stream water. The study also focused on the evaluation of the uncertainties associated with water quality predictions from the ANN model using the Monte Carlo framework. The results of the uncertainty analysis indicated that the ANN model predictions are statistically similar to the characteristics of the measured TP values in the 21 watersheds. It was observed from the study that any reduction in forested area or increase in agricultural land in the watersheds may cause an increase in TP concentrations in the stream water, and therefore appropriate watershed management practices should be followed before making any land use change in the Chugoku district.

Applications of process-based morphodynamic models are often constrained by limited availability of data on bed composition, which may have a considerable impact on the modeled morphodynamic development. One may even distinguish a period of “morphodynamic spin-up” in which the model generates the bed level according to some ill-defined initial bed composition rather than describing the realistic behavior of the system. The present paper proposes a methodology to generate bed composition of multiple sand and/or mud fractions that can act as the initial condition for the process-based numerical model Delft3D. The bed composition generation (BCG) run does not include bed level changes, but does permit the redistribution of multiple sediment fractions over the modeled domain. The model applies the concept of an active layer that may differ in sediment composition above an underlayer with fixed composition. In the case of a BCG run, the bed level is kept constant, whereas the bed composition can change. The approach is applied to San Pablo Bay in California, USA. Model results show that the BCG run reallocates sand and mud fractions over the model domain. Initially, a major sediment reallocation takes place, but development rates decrease in the longer term. Runs that take the outcome of a BCG run as a starting point lead to more gradual morphodynamic development. Sensitivity analysis shows the impact of variations in the morphological factor, the active layer thickness, and wind waves. An important but difficult to characterize criterion for a successful application of a BCG run is that it should not lead to a bed composition that fixes the bed so that it dominates the “natural” morphodynamic development of the system. Future research will focus on a decadal morphodynamic hindcast and comparison with measured bathymetries in San Pablo Bay so that the proposed methodology can be tested and optimized.

Submitted by mcsantos@lnec.pt (mcsantos@lnec.pt) on 2010-12-28T16:29:58Z No. of bitstreams: 1 jp leitão et al 2010-resumo.pdf: 23818 bytes, checksum: c3208049613835730cfa10b343a1b9ec (MD5) Approved for entry into archive by rmatos@lnec.pt(rmatos@lnec.pt) on 2011-01-05T10:59:10Z (GMT) No. of bitstreams: 1 jp leitão et al 2010-resumo.pdf: 23818 bytes, checksum: c3208049613835730cfa10b343a1b9ec (MD5) Made available in DSpace on 2011-01-05T10:59:10Z (GMT). No. of bitstreams: 1 jp leitão et al 2010-resumo.pdf: 23818 bytes, checksum: c3208049613835730cfa10b343a1b9ec (MD5) Previous issue date: 2010-10 Made available in DSpace on 2014-10-20T12:58:26Z (GMT). No. of bitstreams: 3 jp leitão et al 2010-resumo.pdf.txt: 1595 bytes, checksum: bd60384fbe272e474ceb272cac0947c8 (MD5) license.txt: 1835 bytes, checksum: e3e7e4a2a1954a7496e0b0a11a2f4deb (MD5) jp leitão et al 2010-resumo.pdf: 23818 bytes, checksum: c3208049613835730cfa10b343a1b9ec (MD5) Previous issue date: 2010-10 Made available in DSpace on 2017-04-12T14:52:52Z (GMT). No. of bitstreams: 3 jp leitão et al 2010-resumo.pdf: 23818 bytes, checksum: c3208049613835730cfa10b343a1b9ec (MD5) license.txt: 1835 bytes, checksum: e3e7e4a2a1954a7496e0b0a11a2f4deb (MD5) jp leitão et al 2010-resumo.pdf.txt: 1595 bytes, checksum: bd60384fbe272e474ceb272cac0947c8 (MD5) Previous issue date: 2010-10 6 4 2246-2255pp Vol 62 No 10 DHA/NES Water Science & Technology - WST

25 pages, 8 figures, 3 tables., [EN] In recent decades many authors have adopted the Penman- Monteith formula as the standard way to estimate reference evapotranspiration from climate data. The main drawback associated with the Penman-Monteith method is the relatively high data demand: temperature, solar radiation, relative humidity and wind speed are the minimum inputs required by the formula. In the Spanish Pyrenees historical databases usually consist of temperature and rainfall only, although an improvement in data acquisition is expected in terms of parameters monitored. Under situations of data scarcity, some authors recommend the use of less data intensive methods, such as the empirical Hargreaves equation. Other authors suggest that it is better to estimate the missing parameters and to apply the Penman-Monteith equation. This paper presents a study on the accuracy of the Penman-Monteith method for a situation where some parameters have to be estimated from available temperature, and wind speed data must be replaced by a constant value. The results have been compared with the information available for a location in the central Spanish Pyrenees (period 1999-2003) where parameters required by the Penman-Monteith method have been monitored. A comparison is then made with the Hargreaves model in order to assess the most appropriate method for calculation of the reference evapotranspiration. The results indicate that the use of the Penman-Monteith formula results in errors in reference evapotranspiration ETo estimation of different magnitude and sign throughout the year. However, in general, it offers a more accurate estimation of reference evapotranspiration than the Hargreaves formula., [ES] En los últimos años, diferentes autores han adoptado el método de Penman-Monteith para estimar la evapotranspiración potencial mediante datos climáticos. El principal inconveniente de este método es que precisa de un gran número de variables para ser calculado: temperatura, radiación solar, humedad relativa y velocidad del viento. En el Pirineo español las bases de datos climáticas a menudo solamente muestran disponibilidad de datos térmicos y pluviométricos. Bajo estas carencias, algunos autores recomiendan el uso de métodos que precisan de menos datos, como es el caso de la ecuación empírica de Hargreaves. Otros autores sugieren que resulta preferible estimar los datos no disponibles necesarios para aplicar la ecuación de Penman –Monteith a partir de los datos disponibles de temperatura y velocidad del viento. Los resultados de ambos procedimientos han sido comparados con la información disponible en una estación del Pirineo central español (1999-2003) donde están disponibles los parámetros para aplicar la ecuación de Penman-Monteith. Se han comparado los resultados con el modelo de Hargreaves para valorar el método más apropiado de cálculo. Los resultados muestran errores estacionales muy diferentes mediante la fórmula de Penman-Monteith, sin embargo, este procedimiento muestra una aproximación mejor para obtener la evapotranspiración potencial que aplicando la fórmula de Hargreaves., This work has been supported by the research projects CGL2008-01189/BTE, CGL2006-11619/HID and CGL2008-1083/CLI financed by the Spanish Commission of Science and Technology and FEDER, EUROGEOSS (FP7- ENV-2008-1-226487) and ACQWA (FP7-ENV-2007-1- 212250) financed by the VII Framework Programme of the European Commission, “Las sequías climáticas en la cuenca del Ebro y su respuesta hidrológica” and “La nieve en el Pirineo aragonés: Distribución espacial y su respuesta a las condiciones climática” Financed by “Obra Social La Caixa” and the Aragón Government and “Programa de grupos de investigación consolidados” financed by the Aragón Government.

In recent decades many authors have adopted the Penman- Monteith formula as the standard way to estimate reference evapotranspiration from climate data. The main drawback associated with the Penman-Monteith method is the relatively high data demand: temperature, solar radiation, relative humidity and wind speed are the minimum inputs required by the formula. In the Spanish Pyrenees historical databases usually consist of temperature and rainfall only, although an improvement in data acquisition is expected in terms of parameters monitored. Under situations of data scarcity, some authors recommend the use of less data intensive methods, such as the empirical Hargreaves equation. Other authors suggest that it is better to estimate the missing parameters and to apply the Penman-Monteith equation. This paper presents a study on the accuracy of the Penman-Monteith method for a situation where some parameters have to be estimated from available temperature, and wind speed data must be replaced by a constant value. The results have been compared with the information available for a location in the central Spanish Pyrenees (period 1999-2003) where parameters required by the Penman-Monteith method have been monitored. A comparison is then made with the Hargreaves model in order to assess the most appropriate method for calculation of the reference evapotranspiration. The results indicate that the use of the Penman-Monteith formula results in errors in reference evapotranspiration ETo estimation of different magnitude and sign throughout the year. However, in general, it offers a more accurate estimation of reference evapotranspiration than the Hargreaves formula., En los últimos años, diferentes autores han adoptado el método de Penman-Monteith para estimar la evapotranspiración potencial mediante datos climáticos. El principal inconveniente de este método es que precisa de un gran número de variables para ser calculado: temperatura, radiación solar, humedad relativa y velocidad del viento. En el Pirineo español las bases de datos climáticas a menudo solamente muestran disponibilidad de datos térmicos y pluviométricos. Bajo estas carencias, algunos autores recomiendan el uso de métodos que precisan de menos datos, como es el caso de la ecuación empírica de Hargreaves. Otros autores sugieren que resulta preferible estimar los datos no disponibles necesarios para aplicar la ecuación de Penman –Monteith a partir de los datos disponibles de temperatura y velocidad del viento. Los resultados de ambos procedimientos han sido comparados con la información disponible en una estación del Pirineo central español (1999-2003) donde están disponibles los parámetros para aplicar la ecuación de Penman-Monteith. Se han comparado los resultados con el modelo de Hargreaves para valorar el método más apropiado de cálculo. Los resultados muestran errores estacionales muy diferentes mediante la fórmula de Penman-Monteith, sin embargo, este procedimiento muestra una aproximación mejor para obtener la evapotranspiración potencial que aplicando la fórmula de Hargreaves.

1 Introduction||2 Background and Motivation||3 A Method for Sensor-Based Activity Recognition in Missing Data Scenario||4 Activity Recognition Using LoRaWAN （Long Range Wide Area Network） for Nursing Care||5 Integrating Activity Recognition and Nursing Care Records||6 Exploring Human Activities Using eSense Earables||7 A Bayesian Approach for Quantifying Data Scarcity when Modeling Human Behavior via Inverse Reinforcement Learning||8 Discussion||9 Conclusion and Future Work, In this thesis, we proposed novel methods to analyze daily life activities of elderly people and quantifying data scarcity for modeling human behavior inside nursing care facilities in order to improve the overall care system. We addressed the existing challenges related to data collection, complex activity understanding and activity recognition by introducing different computational approaches that recover missing data and utilize spatial and temporal contexts of the data to achieve higher recognition accuracy in various modalities. As well as, in this thesis, we proposed a specific sample size determination method as a precursor to build more accurate models of human behavior. These proposed methods will be immensely helpful for the development of human activity recognition systems, mainly for elderly persons and nursing care facilities to improve human behavior understanding in real-life settings. The outline of the thesis is as follows: In chapter 1 and 2, we briefly introduce the thesis work and literature review. In chapter 3, we proposed a method which can improve activity recognition while having missing data without any data recovery. It is not explored or solved by others without any data imputation techniques. In our proposed approach, we explicitly induce different percentages of missing data randomly in the raw sensor data to train the model with missing data. Learning with missing data reinforces the model to regulate missing data during the classification of various activities that have missing data in the test module. This method can effectively improve the recognition accuracy from 80.8% to 97.5% in a developed synthetic dataset. Afterwards, we tested our approach with activities from two challenging benchmark datasets with a combination of 21 features. We achieved a better result of handling different levels of missing data in the dataset without any data imputation techniques. In chapter 4, we introduce activity recognition using LoRaWAN that aim to sustain connections among IoT devices over a long distance. We explored low-power wide-area network （LPWAN） technology for sensing experiment to know the possibility of using LoRaWAN protocol and network for activity recognition. As well as, investigate the number of sensor nodes connected with a single gateway which have an impact on the performance of sensors ultimate data sending capability in terms of data loss. For human activity recognition, we have achieved recognition accuracy of 94.44% by Linear Discriminant Analysis （LDA）, 84.72% by Random Forest （RnF） and 98.61% by K-Nearest Neighbor （KNN） from collected data. Later, in a real nursing care setting experiment, 42 LoRaWAN sensors environmental sensors data is collected to know the data loss ratio. We observe 5% data loss happened by the sensors with a single gateway. In a simulated environment, we checked the activity recognition performance with 5%, 30%, 50% and 80% data loss environment and have found recognition accuracy of 81.94% LDA, 80.55% RnF and 91.66% by KNN while 5% data are lost. Through our proposed framework it can open a new opportunity to significantly increase the sensing range in nursing care center by LoRaWAN. In chapter 5, we assess the daily life activity data obtained during the 4-months experiments at a nursing care center. Through this work, we tried to know whether the daily life nursing care activity data are dependent on subject （e.g., staff or target resident）, or date, and whether the obtained data are meaningful and informative for activity understanding. We collected 38,076 activity labels, 46,803 record details, and 2834 hours of sensor data during this experiment. From this data, we revealed the varieties and dependency of activities and care details which can be a measure of any healthcare outcome. These findings can be essential elements for activity recognition, having many intra-class relationships among activities in nursing care center. In chapter 6, we explore to recognize different types of head and mouth related activities of elderly people which can help to measure the health status. We propose an activity recognition framework to collect head and mouth related behavioral activities （e.g., head nodding, headshaking, eating and speaking） along with other regular activities. We have found that the accuracy is 93.34% by SVM, 91.92% by RnF, 91.64% by KNN, and 93.76% by CNN. In chapter 7, we proposed a sample size determination method based on uncertainty quantification （UQ） for a specific Inverse Reinforcement Learning （IRL） model of human behavior. The main insight behind our method is that the probability of model parameters given training data can be updated from prior to posterior through Bayesian inference. For illustration, we provided an example with a specific hypothetical cost scenario and decision-making rule for MS （Multiple Sclerosis） behavioral modeling, under which our method indicated 935 samples is optimal. Chapter 8 offers a brief discussion of the thesis. In Chapter 9, we conclude the thesis with conclusion and some future work issues. This thesis contribution can help to create various tools to aid stakeholders, such as domain experts and end users, in exploring human behavior understanding. This work will become particularly important with the rise of behavior-aware user interfaces that automatically reason and act in response to people’s behaviors in almost every aspect of their lives., 九州工業大学博士学位論文 学位記番号：工博甲第522号 学位授与年月日：令和3年3月25日, 令和2年度