Evapotranspiration (ET) is a significant hydrological process. It can be studied and estimated using remote sensing based methods at multiple spatial and temporal scales. Most commonly and widely used remote sensing based methods to estimate actual evapotranspiration (AET) are a) methods based on energy balance equations, b) vegetation coefficient based method and c) contextual methods. These three methods require reflectance and land surface temperature (LST) data measured at optical and thermal portion of the electromagnetic spectrum. However, these data are available only for clear sky conditions and fail to be retrieved under overcast conditions creating gaps in the data, which result in discontinuous of AET product. Moreover, energy balance equation based methods and evaporative fraction (EF) based contextual methods are difficult to apply over overcast conditions. In this context, vegetation coefficient based (Tasumi et al., 2005; Allen et al., 2005) and microwave remote sensing based methods can be applied under cloudy sky conditions (Sun et al., 2012), since microwave radiations can penetrate through clouds, but these data are available at coarse resolution. In the vegetation coefficient method temporal upscaling can be avoided. Therefore in this research vegetation coefficient based method is employed over Cauvery basin to estimate daily AET for clear and cloudy sky conditions. Required critical variables for this method such as reference evapotranspiration (ETo) and vegetation coefficients are obtained using LST and optical vegetation indices for all sky conditions. In this study, all sky conditions refer to both clear and cloudy sky conditions. Most important variable for estimation of ETo using radiation and temperature based models is air temperature (Ta). In this study, for better accuracy of Ta, two satellite based approaches namely, Temperature Vegetation Index (TVX) and Advance Statistical Approaches (ASA) were evaluated. In the TVX approach, in addition to traditional Normalized Difference Vegetation Index (NDVI), other vegetation indices such as Enhanced Vegetation Index (EVI) and Global Vegetation Moisture Index (GVMI) were also examined. In case of ASA, bootstrap technique was used to generate calibration and validation samples and Levenberg Marquardt algorithm was used to find the solution of the models. The better of the Ta results obtained out of these two approaches were employed in the ETo models and are referred as Ta based ETo models. Instead of Ta, processed LST data obtained directly from the satellite (Aqua/Moderate Resolution Imaging Spectroradiometer (MODIS)) was applied in the ETo models and these are referred as LST based ETo models. These Ta and LST based Hargreaves-Samani (H-S), Makkink (Makk) and Penman Monteith Temperature (PMT) models were evaluated by comparing with the FAO56 PM model. Additionally, simple LST based equation (SLBE) proposed by Rivas et al. (2004) was also examined. Required solar radiation (Rs) data for ETo estimation was obtained from Kalpana1/VHRR satellite data. Results implied that, Ta based PMT model performed better than the Ta based H-S, Makk and SLBE with less RMSE, MAPE and MBE values for all land cover classes and for various climatic regions for clear sky conditions. LST based H-S, PMT, Makk and Ta based Makkink advection models predominantly overestimated ETo for the study region. In the case of TVX approach, to estimate maximum Ta (Tmax), GVMI performed better than NDVI and EVI. Nevertheless, TVX approach poorly estimated Tmax in comparison with statistical approach. ASA performed better for both Tmax and minimum Ta. This study demonstrates the applicability of satellite based Ta and ETo models by considering very few variables for clear sky conditions. Spatially distributed vegetation coefficients (Kv) data with high temporal resolution is another important variable in vegetation coefficient method for daily AET estimation and also it is in demand for crop condition assessment, irrigation scheduling, etc. But available Kv models application hinders because of two main reasons i.e 1) Spectral reflectance based Kv accounts only for transpiration factor but not evaporation, which fails to account for total AET. 2) Required optical spectral reflectances are available only during clear sky conditions, which creates gaps in the Kv data. Hence there is a necessity of a model which accounts for both transpiration and evaporation factors and also gap filling method, which produces accurate continuous quantification of Kv values. Therefore, different combinations of EVI, GVMI and temperature vegetation dryness index (TVDI) have been employed in linear and non linear regression techniques to obtain best model. This best Kv model had been compared with Guershman et al. (2009) Kv model. To fill the gaps in the data, initially, temporal fitting of Kv values have been examined using Savitsky-Goley (SG) filter for three years of data (2012 to 2014), but this fails when sufficient high quality Kv values were unavailable. In this regard, three gap filling techniques namely regression, Artificial Neural Networks (ANNs) and interpolation techniques have been analyzed. Microwave polarization difference index (MPDI) has been employed in ANN technique to estimate Kv values under cloudy sky conditions. The results revealed that the combination of GVMI and TVDI using linear regression technique performed better than other combinations and also yielded better results than Guershman et al. (2009) Kv model. Furthermore, the results indicated that SG filter can be used for temporal fitting and for filling the gaps, regression technique can be used as it performed better than other techniques for Berambadi station. Land Surface Temperature (LST) with high spatiotemporal resolution is required in the estimation of ETo to obtain AET. MODIS is one of the most commonly used sensors owing to its high spatial and temporal availability over the globe, but is incapable of providing LST data under cloudy conditions, resulting in gaps in the data. In contrast, microwave measurements have a capability to penetrate under clouds. The current study proposes a methodology by exploring this property to predict high spatiotemporal resolution LST under cloudy conditions during daytime and night time without employing in-situ LST measurements. To achieve this, ANN based models were employed for different land cover classes, utilizing MPDI at finer resolution with ancillary data. MPDI was derived using resampled (from 0.250 to 1 km) brightness temperatures (Tb) at 36.5 GHz channel of dual polarization from Advance Microwave Scanning Radiometer (AMSR)-Earth Observing System and AMSR2 sensors. The proposed methodology was quantitatively evaluated through three performance measures namely correlation coefficient (r), Nash Sutcliffe Efficiency (NSE) and Root Mean Square Error (RMSE). Results revealed that during daytime, AMSR-E(AMSR2) derived LST under clear sky conditions corresponds well with MODIS LST resulting in values of r ranging from 0.76(0.78) to 0.90(0.96), RMSE from 1.76(1.86) K to 4.34(4.00) K and NSE from 0.58(0.61) to 0.81(0.90) for different land cover classes. For night time, r values ranged from 0.76(0.56) to 0.87(0.90), RMSE from 1.71(1.70) K to 2.43(2.12) K and NSE from 0.43 (0.28) to 0.80(0.81) for different land cover classes. RMSE values found between predicted LST and MODIS LST during daytime under clear sky conditions were within acceptable limits. Under cloudy conditions, results of microwave derived LST were evaluated with Ta which indicated that the approach performed well with RMSE values lesser than the results obtained under clear sky conditions for land cover classes for both day and nighttimes. These predicted LSTs can be applied for the estimation of soil moisture in hydrological studies, in climate studies, ecology, urban climate and environmental studies, etc. AET was estimated for all sky conditions using vegetation coefficient method. Essential parameter ETo under cloudy conditions was estimated using LST and Ta based PMT and H-S models and required solar radiation (Rs) in these two models estimated using equation proposed by Samani (2000). In this equation it was found that the differences between LSTmax or Tmax and LSTmin or Tmin could able to capture the variations due to cloudy sky conditions and hence can be used for estimating ETo under cloudy sky conditions. Results revealed that the estimated Rs correlated well with observed Rs for Berambadi station under cloudy conditions for the year 2013. PMT based ETo values were corresponded with observed ETo under cloudy sky condition. The difference between LST and Ta was less during cloudy conditions, therefore LST or Ta can be used as the only input in temperature based PMT model to estimate ETo. AET estimated correlated well with the observed AET values for clear and cloudy sky conditions. In addition, AET estimated using vegetation coefficient method was compared with two source energy balance (TSEB) method developed by Nishida et al. (2003) under clear sky conditions. It was found that the improved vegetation coefficient method performed better than the TSEB method for Berambadi station. Other microwave vegetation indices such as Microwave Vegetation Indices (MVIs) and Emissivity Difference Vegetation Index (EDVI) are available in literature. Therefore in this study, MVIs are used to predict LST under cloudy conditions using proposed methodology to check whether the MVIs could yield better LST values. Results showed that MPDI performed better than MVIs to predict LST under cloudy sky conditions. Furthermore, MPDI obtained using dual polarizations of 37 GHz channel Tb has advantage of having fine spatial resolution compared to MVIs, as it requires Tb of 19 GHz in addition to Tb of 37 GHz channel which is of coarse resolution and therefore uncertainties resulting from re-sampling technique can be minimized. x

<div>Observations from experiments and post-earthquake surveys have shown that drift is the key parameter for identifying potential damage of a structure during ground motions (Sozen, 1981). These observations suggest that drift should govern seismic design and evaluation of structures.</div><div><br></div><div>In this study, three methods for estimating drift demands were considered: 1) the method proposed by Sozen (2003) referred to in this study as Velocity of Displacement (VOD), 2) the Coefficient Method and 3) Nonlinear Dynamic Analysis (NDA). The reliability of each method was evaluated by comparing estimates of roof and maximum story drift ratios with measurements from 46 reinforced concrete structures with initial periods shorter than 3 seconds.</div><div><br></div><div>Measurements from long-period structures (with periods longer than 3 seconds) were not available. To produce data to evaluate the reliability of the three mentioned methods for</div><div>long-period structures as well as understand the displacement and base-shear response of such structures, seven scaled Multi-Degree-of-Freedom (MDOF) specimens with an initial period of approximately 1.2 seconds were tested with five scaled base motions of varying intensities. Each motion was scaled in time such that its scaled spectral shape near the initial period of the specimen was similar to the spectral shape of the unscaled motions for periods ranging from approximately 1 to 10 seconds. A total of 118 tests were conducted.</div><div><br></div><div>The effect of loading history on drift demands and drift estimates was also evaluated by quantifying changes in drift demands of structures subjected to repeats of the same ground motion. Data from 1) experimental tests of structures subjected to repeated ground motions, and 2) numerical analyses of Single-Degree-of-Freedom (SDOF) oscillators subjected to multiple sequences of ground motions of varying intensities were used.</div><div><br></div><div><div>Based on comparisons of measured and calculated drifts as well as data from the experimental program, the following observations were made:</div></div><div><br></div><div>1) For structures with periods shorter than 3 seconds, all three methods for estimating drift demands produced estimates of both roof and maximum story drifts of similar</div><div>quality despite large differences in the effort required to use each method.</div><div><br></div><div>2) For structures with periods longer than 3 seconds, NDA produced drift estimates close to the mean of measured values while VOD overestimated measured values, on average, by approximately 30%. The Coefficient Method produced estimates that were, on average, smaller than measurements by approximately 40%.</div><div><br></div><div>3) For structures (not susceptible to decay in lateral strength) subjected to sequences of ground motions of similar intensities, the relative increase in drift demands was,</div><div>on average, no more than 20%. Larger increases in drift demands were observed for structures where the first motion (in a pair of repeated motions) was mild enough</div><div>not to cause cracking and/or yielding, and the second motion was preceded by larger intensity motions that did cause cracking and/or yielding.</div><div><br></div><div>4) For test structures with periods longer than 3 seconds, drifts in the nonlinear range of response were generally smaller than linear estimates, and maximum base-shear</div><div>demands were as much as three times those calculated assuming a linear lateral load distribution.</div>

隨著網路的發達,各式各樣的資訊和商品也在網路上充斥著,使用者尋找資訊或是上網購物時,有的網站有推薦系統(recommender system)能提供使用者相關資訊或商品。若推薦系統能夠讓消費者所搜尋的相關資訊或商品能夠符合他們的習性時,便能讓消費者增加對系統的信賴程度,因此系統是否能準確預測出使用者的偏好就成為一個重要的課題。本研究使用兩筆資料,並以相關研究的三篇文獻進行分析和比較。這三篇文獻分別為IRT模型法(IRT model-based method)、相關係數法(correlation-coefficient method)、以及矩陣分解法(matrix factorization)。 在經過一連串的實證分析後,歸納出以下結論: 1. 模型法在預測方面雖然精確度不如其他兩種方法來的好,但是模型有解釋變數之間的關係以及預測機率的圖表展示,因此這個方法仍有存在的價值。 2. 相關係數法容易因為評分稀疏性的問題而無法預測,建議可以搭配內容式推薦系統的運作方式協助推薦。 3. 矩陣分解法在預測上雖然比IRT模型法還好，但分量的數字只是一個最佳化的結果，實際上無法解釋這些分量和數字的意義。 / With the growth of the internet, websites are full of a variety of information and products. When users find the information or surf the internet to shopping, some websites provide users recommender system to find with which related. Hence, whether the recommender system can predict the users' preference is an important topic. This study used two data,which are "Mondo" and "MovieLens", and we used three related references to analyze and compare them. The three references are following: IRT model-based method, Correlation-coefficient method, and Matrix factorization. After the data analysis, we get the following conclusions: 1. IRT model-based method is worse then other methods in predicting, but it can explain the relationship of variables and display the graph of predicting probabilities. Hence this method still has it's value. 2. Correlation-coefficient method is hard to predict because of sparsity. We can connect it with content filtering approach. 3. Although matrix factorization is better then IRT model-based method in predicting, the vectors is a result of optimization. It may be hard to explain the meaning of the vectors.

社會安全制度，以社會保險及公共救助為主體，兩者之中尤以社會保險為骨幹，社會保險通常扮演著重要的角色。當中一項重要的功能即為所得（財富）重分配功能，亦即政府借助社會保險之力，達成安定經濟社會與改善國民所得分配不均，以達公平之目標。 本文在實證方法上採用「吉尼係數法」與「變異係數法」來計算社會保險的所得重分配效果。利用民國八十五年至民國九十一年行政院主計處「中華民國臺灣地區家庭收支調查報告」之調查資料，探討臺灣地區所得分配不均度上升的原因是否來自於城鄉差異，其次是社會保險政策對於平衡城鄉差距是否有助益。 為了衡量社會保險的所得重分配效果是否會因城鄉發展程度之不同而有所差異，將臺灣地區內之城市分為都市、城鎮及鄉村三級，其分層標準係依照行政院主計處「中華民國臺灣地區家庭收支調查報告」之標準分類。本研究以城鄉別與社會保險為研究主軸，探討臺灣社會保險的所得重分配效果是否在不同城鄉間會有所影響。 綜合研究結果及分析，對於民國八十五至九十一年社會保險實施的所得重分配效果所得到的結論為：1.臺灣地區自民國八十五年後無論是區分層級或整體所得分配效果上的吉尼係數均有逐漸縮小的趨勢，代表政府對於平均所得分配之努力是有所成效的。2.在吉尼係數法下，除了「都市層」外，社會保險實施後「城鎮層」、「鄉村層」與整體所得分配效果的吉尼係數值均高於較社會保險實施前，顯示社會保險政策在平衡城鄉所得差異上的力量似乎薄弱了些。3.在變異係數法下，無論是分層效果或是整體效果實施社會保險後整體的所得分配平均化力量均減弱，故社會保險政策在平均所得分配的效果上似乎沒有達到預期的成效。4.綜合上述兩種方法，除了吉尼係數法下的「都市層」有達成社會保險的所得重分配效果外，吉尼係數法與變異係數法的其他層級和整體效果分析均顯示出實施社會保險未達成所得重分配的效果。 / Social insurance and public rescue are two main components of social security system. Especially, social insurance is also the skeleton of social security system, which has many important functions, one of which is improving the inequity of people’s income assignment. It means that the government redistributes people’s income through social insurance to achieve the goal of equity and further to stabilize economic society. This article uses the data of "Republic of China Taiwan area family budget survey reported", which comes from 1996 to 2002 Directorate－General of Budget, Accounting and Statistics, Executive Yuan, R.O.C.(Taiwan), as investigation material. We calculate the income redistribution effect of social insurance by means of "Gini Coefficient method" and "Coefficient of Variation method". This article has two issues, one of which discusses whether the income inequality in Taiwan does come from the difference between city and countryside. The other one is the benefit of social security policy to balance of disparity of city and countryside. In order to assess whether the income redistribution effects of social security has the difference between cities, we divide the cities in Taiwan into three groups: metropolis, countries and villages, according to standard classification of the investigation material. We use difference between cities and social insurance as two axes of our study to evaluate the effect of income redistribution between different cities. To the effect of social insurance on income redistribution from 1996 to 2002, our study has following findings. First, regardless of classification or summation analysis, the Gini coefficient of income redistribution was gradually reducing from 1996 to 2002. This means that income redistribution policy of government is effective. Second, in Gini Coefficient method, country group and village group had higher Gini Coefficient than before executing social insurance policy. The conclusion shows the influence of social insurance was still not efficient. Third, in Coefficient of Variation method, classification and summation analysis both revealed income redistribution was weaker than before executing social insurance policy, so the policy did not achieve the expected effect. From the above findings, although the metropolis group in Coefficient method did improve income redistribution, other analysis did not achieve the goal of income redistribution.