Retrieval-augmented generation (RAG) has become a cornerstone for enhancing large language models (LLMs) with real-time information from the Web, but its performance often heavily depends on the quality of the retrieved documents. Given that RAG systems frequently draw from vast and often noisy Web corpora, ensuring the reliability of retrieved content is paramount. While rerankers improve the factual accuracy of the RAG system by elevating the proportion of ground-truth documents (GD) in high-ranked results, the shifts of document type distributions during reranking remain unclear, hindering the understanding of the reranker's behavior. To bridge this gap, we conduct an empirical study to categorize documents and compare their distribution before and after reranking. We reveal a counterintuitive finding: though rerankers improve the proportion of GD, they also significantly increase the proportion of harmful documents (HD) in top-ranked retrieved documents. It not only narrows the potential context window for ranking the GD higher but also increases the risk of HD misleading the LLMs, potentially leading to the generation and propagation of misinformation across Web platforms. Motivated by this finding, we propose a risk-aware reranking method for RAG with LLMs, which balances the risk and benefit during reranking. Given a query, the RAG framework first retrieves relevant documents. Then, our approach quantifies the potential beneficial and harmful impacts of various documents on the LLMs' generation. To estimate the impacts, we conduct a dual-aspect document impact assessment via information gain, which employs a risk clipping to avoid the numerical fluctuations in the estimation. Finally, we conduct the reranking according to the potential impact of each document, enabling the reranker to significantly reduce the HD proportion. Experiments and analysis across multiple models and datasets, including Wikipedia, web news, and research papers, show the effectiveness of our method. Our code is available at https://github.com/lzz335/hidden_risk_of_reranking.

Large language models (LLMs) demonstrate significant potential in various applications; however, they are susceptible to generating hallucinations, which can lead to the spread of online misinformation. Existing studies address hallucination detection by (1) employing reference-based methods that consult external resources for verification or (2) utilizing reference-free methods that mainly estimate answer uncertainty based on LLM's internal states. However, reference-based methods incur significant costs and can be infeasible for obtaining reliable external references. Besides, existing uncertainty estimation (UE) methods often overlook the impact of scenario backgrounds inherited from the query's lexical resources, leading to noise in UE. In almost all real-world applications, users care about the uncertainty concerning semantics or facts instead of the query's scenario information. Therefore, we argue that mitigating scenario-related noise and focusing on semantic information can yield a more desirable UE. In this paper, we introduce a plug-and-play scenario-independent framework to enhance unsupervised UE in LLMs by removing scenario-related noise and focusing on semantic information. This framework is compatible with most existing UE methods, as it leverages only the existing UE methods' outputs. Specifically, we design a scenario-specific sampling to paraphrase queries, maintaining their common semantics while diversifying the scenario distribution. Subsequently, to estimate the contribution of the common semantics, we design a factor analysis (FA) model to disentangle the UE score obtained from the given UE method into a combination of multiple latent factors, which represent the contribution of the common semantics and scenario-related noise. By solving the FA model, we decompose the impact of the most significant factor to approximate the uncertainty caused by the common semantics, thus achieving scenario-independent UE. Extensive experiments and analysis across multiple models and datasets demonstrate the effectiveness of our approach.

Integrated water vapor (IWV) is a dominant influence element in radiation absorption, energy transfer, and water circulation on both local and global scales. The Sentinel-3 Ocean and Land Color Imager (OLCI) instrument provides operational IWV measurements using a two-band ratio of an IWV absorption channel (O19; 900 nm) and a referenced channel (O18; 885 nm). However, the operational OLCI/Sentinel-3 satellite product does not offer IWV estimates under cloudy-sky conditions, as OLCI-sensed near-infrared data have considerable uncertainties when clouds are in existence. We develop a practical machine learning-based retrieval algorithm to derive IWV estimates from OLCI near-infrared radiance observations under all-sky conditions. The retrieval method utilizes O19 900-nm and O20 940-nm IWV absorption bands as well as O18 885-nm and O21 1020-nm referenced bands, based on both two-band and three-band ratio methods. IWVs from the Global Navigation Satellite System (GNSS) are used as the desired IWV retrievals. The results show that all newly derived IWV retrievals have an excellent agreement with reference IWV from additional GNSS and radiosonde data, regardless of sky weather conditions. The weighted-mean IWV retrievals present the highest performance with GNSS and radiosonde IWV [correlation coefficient: 0.86 and 0.85; root-mean-square error (RMSE): 2.85 and 3.49 mm; and mean bias (MB): −0.14 and −0.99 mm]. The newly retrieved cloudy-sky IWV is comparable to operational clear-sky IWV, denoting the capability and effectiveness of the retrieval algorithm. The retrieval approach exhibits a dependable performance in both spatial and temporal dimensions, which could be employed in other areas and periods.

Water vapor is an important component in the Earth’s climate and weather processes, underscoring the critical need for precise water vapor data on both regional and global scales. In this paper, an innovative retrieval method is developed to retrieve precipitable water vapor (PWV) using thermal infrared (IR) data from the Moderate Resolution Imaging Spectroradiometer (MODIS) in all weather conditions. Unlike the operational MODIS PWV retrieval algorithm that utilizes 11 IR channels from 4.5 μm to 14.2 μm, two split-window bands at 11 μm and 12 μm as well as one PWV absorption band at 7.2 μm are used. The retrieval approach defines the functional relationship between the retrieved PWV and brightness temperature and multiple dependence elements – latitude, longitude, height, view zenith angle, month, solar zenith angle, and cloud, based on a machine learning method. The retrieval algorithm does not require the first guess and surface emissivity that are traditionally used in satellite IR retrieval methods. The results show that the new PWV retrievals outperform operational PWV data from MODIS thermal IR observations, reducing the root-mean-square error by 26.49% and 12.58% when compared with reference Global Navigation Satellite System (GNSS) and radiosonde PWV, respectively. The retrieval method can reduce the retrieval errors of IR PWV estimates at different clear confidence levels, illustrating the capability of the proposed retrieval approach. This study offers a machine learning-driven retrieval approach to derive all-weather water vapor retrievals using satellite-sensed MODIS thermal IR observations, which could be also applicable to other similar IR sensors.

Precipitable water vapor (PWV) data from satellite-sensed near-infrared (NIR) measurements offer a unique source for monitoring atmospheric water vapor distribution locally and globally. However, the observational quality of satellite-based operational NIR PWV products is considerably affected by the presence of clouds. We develop a spatial and temporal cloud fraction based calibration method (STCFCM) to calibrate satellite-sensed NIR PWV products and improve the PWV accuracy. The STCFCM is built based on Light Gradient Boosting Machine using cloud fraction, together with spatial-temporal fields – latitude, longitude, height, and month. The newly calibrated PWV estimates from the MODIS sensor onboard the Terra satellite show that the STCFCM-estimated PWV data exhibit a better agreement with reference PWV estimates from GNSS and radiosonde observations. The root-mean-square error of MODIS/Terra operational PWV products is reduced by 55.53% from 11.40 mm to 5.07 mm compared to GNSS PWV and 60.74% from 14.67 mm to 5.76 mm compared to radiosonde PWV. The calibrated all-weather PWV estimates outperform operational clear-sky PWV products, highlighting the effectiveness of the STCFCM in calibrating satellite-sensed NIR PWV retrievals, particularly for cloudy sky conditions. The newly developed STCFCM approach is the first one in the research community to calibrate global PWV data from satellite NIR measurements. It is a promising technique to calibrate remote sensing PWV products from other satellite observations under all weather conditions.

High-latitude regions are frequently challenging for observing precipitable water vapor (PWV) from satellite near-infrared (NIR) channels because of the high surface reflectance and large solar zenith angle. Satellite-observed NIR PWV retrievals also present much higher uncertainties under cloudy sky conditions than under clear sky conditions. In this work, we propose an improved neural network-based algorithm for the first time to retrieve PWV from Moderate Resolution Imaging Spectroradiometer (MODIS) NIR all-weather observations in high-latitude areas. The retrieval algorithm is developed based on ground-based Global Navigation Satellite System (GNSS)-sensed PWV estimates, together with ERA5-based first-guess PWV as well as several dependence factors associated with NIR PWV retrievals. The results show that the newly retrieved PWV estimates remarkably outperform operational MODIS-derived NIR PWV products, exhibiting an all-weather reduction in root-mean-square error (RMSE) of 78.32% from 7.15 to 1.55 mm compared with GNSS-observed reference PWV and 81.38% from 7.52 to 1.40 mm compared with radiosonde-observed reference PWV. The observational accuracy of all-weather PWV retrievals is also comparable to that of PWV retrievals under clear sky conditions, denoting the capability and effectiveness of the retrieval method. The retrieval approach presents much larger RMSE reductions compared to previous algorithms that do not use first-guess water vapor, implying that the addition of first-guess PWV contributes to generating improved PWV estimates from satellite-sensed NIR measurements. While the retrieval method is developed for deriving MODIS NIR water vapor in high-latitude regions, it also has significant potential to be applicable to other satellite sensors as well as other worldwide regions.

The knowledge of atmospheric water vapor distribution is vital to our understanding of weather and climate. In this article, we propose a new calibration scheme based on machine learning to enhance the observational performance of official all-weather precipitable water vapor (PWV) data records from thermal infrared (IR) measurements of the Moderate Resolution Imaging Spectroradiometer (MODIS) sensor. The calibration scheme takes several influence factors into consideration, which are linked with the performance of satellite-retrieved IR PWV measurements. The ground-based water vapor data, acquired from 214 Global Positioning System (GPS) sites across China in 2016, are regarded as reference PWV to train the machine learning based calibration approaches. The evaluation result during 2017-2019 across China shows that the calibrated MODIS IR all-weather PWV product agrees better with GPS-retrieved reference PWV observations, with R2of 0.88-0.94, root-mean-square-error (RMSE) of 2.79-4.08 mm, and mean bias of 0.16-0.52 mm. The RMSE between water vapor measurements from MODIS and GPS can be reduced by 41.74%, 45.76%, 44.29%, and 49.04% in confident-clear, probably-clear, probably-cloudy, and confident-cloudy conditions, respectively. Our methods, developed based on the new calibration scheme, could be a promising tool to the calibration of other satellite-derived IR all-weather water vapor products, which could be also extended to other regions or time periods.

A back propagation neural network (BPNN)-based calibration method is developed to enhance the all-weather quality of official integrated water vapor (IWV) products from near-infrared (NIR) observations of the Ocean and Land Color Instrument (OLCI) onboard Sentinel-3A and Sentinel-3B satellites. The model utilizes multiple variables that link with the derivation of satellite NIR IWV products, including official OLCI NIR IWV, latitude, land quality and science flag, month, and solar zenith angle. The in situ IWV observations, collected from 100 Global Positioning System (GPS) stations, are employed as the desired IWV estimates. The model is evaluated using one-year water vapor data at additional 114 GPS stations and 97 radiosonde stations from June 1, 2019 to May 31, 2020 in China and its surrounding regions. The results show that the BPNN-based calibration model reduces the root-mean-square error (RMSE) of official OLCI NIR all-weather IWV products by 24.52% from 3.10 to 2.34 mm for Sentinel-3A and by 25.00% from 3.44 to 2.58 mm for Sentinel-3B, when compared with in situ GPS-observed reference IWV. When compared with in situ radiosonde-observed IWV, the RMSE reduces 21.01% from 4.57 to 3.61 mm and 20.81% from 5.19 to 4.11 mm for Sentinel-3A and Sentinel-3B, respectively.

We developed six machine learning based calibration models to improve the all-weather accuracy of precipitable water vapor (PWV) product from near-infrared (NIR) observations of the Moderate Resolution Imaging Spectroradiometer (MODIS) instrument, i.e. MOD05 PWV. The six machine learning approaches are: Back Propagation Neural Network (BPNN), Gradient Boosting Decision Tree (GBDT), Generalized Regression Neural Network (GRNN), K-Nearest Neighbor (KNN), Multilayer Perceptron Neural Network (MLPNN), and eXtreme Gradient Boosting (XGBoost). The input of the models included MOD05 PWV, latitude, longitude, elevation, cloud, season, and solar zenith angle, in association with the quality of the MOD05 PWV product. PWV data measured from in-situ 453 Global Positioning System (GPS) stations in Australia in 2017 were utilized as the target water vapor data for model training. The validation results in Australia in 2018–2019 indicate that the models can significantly improve the all-weather quality (increased R2, reduced root-mean-square error (RMSE), and reduced mean bias (MB)) of the MOD05 PWV product, exhibiting a reduction in RMSE of 53.33% for BPNN, 55.25% for GBDT, 53.24% for GRNN, 37.81% for KNN, 54.98% for MLPNN, and 55.16% for XGBoost. The R2was 0.58 ~ 0.79 and the MB was 0.12 mm ~ 0.72 mm, much better than the MOD05 all-weather PWV product (R2= 0.26 and MB = -1.75 mm). Different from previous studies that focused on clear-sky conditions only, this work is the first one to enhance the quality of official MODIS NIR PWV products under all weather conditions, reducing the impact of clouds on PWV products.

Precipitable water vapor (PWV) product, obtained from near-infrared (NIR) measurements of the Medium Resolution Spectral Imager (MERSI) from the FengYun-3A (FY-3A) spacecraft, has not been used in weather forecasting and climate monitoring so far because of its degraded accuracy. In this research, four machine learning based correction approaches are for the first time developed to adjust the long-term observation accuracy of the official MERSI/FY-3A NIR all-weather water vapor product from 2010 through 2017 in China considering multiple influence factors - MERSI/FY-3A NIR PWV, latitude, longitude, month, and cloud. In addition to four machine learning models, the conventional Multiple Parameter Quadratic (MPQ) regression method is also utilized for inter-comparison. The in-situ PWV estimates, acquired from 100 Global Positioning System (GPS) stations in 2010-2017 across China, are utilized as reference PWV in model training. The validation results obtained from comparison with PWV from the other 114 GPS stations during 2010-2017 in China indicate that the methods notably enhance the long-term performance of FY-3A MERSI NIR water vapor observations under all weather conditions, reducing root-mean-squares error (RMSE) by 57.62-71.61%. The calibrated MERSI NIR PWV, calculated using machine learning models, performs better than the conventional MPQ-estimated PWV. After the calibration, the new MERSI NIR water vapor estimates show a performance comparable to other satellite-observed NIR PWV products. The enhanced MERSI/FY-3A NIR all-weather PWV data can complement other water vapor data in the FY-3 series to ensure a continuous long-term data record and benefit the FY-3 user community.

Satellite-based conically scanning microwave radiometers are capable of making precipitable water vapor (PWV) observations over the vast ocean regions. In this study, PWV from five on-orbit satellite-based microwave radiometers (SMWRs), i.e., special sensor microwave imager/sounder (SSMIS) F16, SSMIS F17, SSMIS F18, Advanced Microwave Scanning Radiometer 2 (AMSR2), and global precipitation measurement (GPM) microwave imager (GMI), are evaluated by shipborne global navigation satellite system (GNSS) PWV during a 77-day cruise across the Pacific Ocean from June 01, 2017, to August 16, 2017. This cruise crossed about 90° in latitude (from ~40° S to ~50° N) and about 174° in longitude (from ~72° W to ~114° E). The shipborne GNSS PWV is first compared with PWV derived from the European Centre for Medium-Range Weather Forecasts (ECMWF) Reanalysis v5 (ERA5) products. The comparison results show that the root mean square error (RMSE) between shipborne GNSS PWV and ERA5 PWV is 2.1 kg/m2. The shipborne GNSS PWV is then used to evaluate the PWV derived from five SMWRs. The statistical results show that the PWV from all SMWR has a good agreement with shipborne GNSS PWV. The PWV RMSEs of SSMIS F16, SSMIS F17, SSMIS F18, AMSR2, and GMI evaluated by shipborne GNSS PWV are 2.0, 2.0, 1.8, 1.5, and 1.7 kg/m2, respectively. In addition, statistical results indicate that SSMIS F16, SSMIS F17, SSMIS F18, AMSR2, and GMI overestimate PWV with respect to GNSS by 1.1, 0.5, 0.5, 0.4, and 0.4 kg/m2, respectively.

A new algorithm to retrieve water vapor from Moderate Resolution Imaging Spectroradiometer (MODIS) near-infrared (NIR) channels using the ensemble-based empirical regression model, which was developed based on the North Hemisphere (western North America) data, was for the first time applied and validated to the South Hemisphere, mainly the Australia and its surrounding regions. By employing the empirical regression algorithm to retrieve water vapor from MODIS Level 1 reflectance data, the wet bias of MODIS product has been significantly reduced. Validation against global positioning system (GPS) water vapor observations over the period January 1, 2017 to December 31, 2019 in and around Australia shows that the root mean square error (RMSE) of water vapor data obtained from MODIS/Terra has reduced by 58.53% from 5.712 to 2.369 mm when using two-channel ratio transmittance and has reduced by 56.14% to 2.505 mm when using three-channel ratio transmittance. For the data obtained from MODIS/Aqua, the RMSE has reduced by 49.17% from 5.170 to 2.628 mm using two-channel ratio transmittance and has reduced by 46.60% to 2.761 mm using three-channel ratio transmittance, respectively. In addition, validations of the retrieved water vapor results over such a large research area (0°-55°S in latitude and 95°-180°E in longitudes) also show no temporal or spatial dependence, implying that the algorithm is homogeneous, accurate, and robust.

The integrated water vapor (IWV) products collected from June 1, 2019 to May 31, 2020 from the ocean and land color instrument (OLCI) sensor, onboard the Sentinel-3 satellites, are evaluated against reference water vapor data estimated from ground-based 214 global positioning system (GPS) stations in the Mainland China. This is the first time to thoroughly evaluate the quality of Sentinel-3 OLCI IWV products byin situGPS-measured IWV data from such a large spatial coverage as China. The validation results show that, under cloud-free conditions, the OLCI IWV measurements agree very well with the ground-based GPS water vapor data, with a root-mean-square error (RMSE) of 3.03 mm for Sentinel-3A satellite and 3.13 mm for Sentinel-3B satellite. The dependence of OLCI IWV on various parameters was also analyzed. Analysis showed that the accuracy of inland OLCI IWV products was superior to that in coastal areas and that OLCI tended to overestimate IWV value in lower elevation and underestimate IWV value in higher elevation. The accuracy of OLCI IWV measurements increased as IWV decreased. Solar zenith angle analysis showed that the OLCI IWV product had a higher accuracy at a larger solar zenith angle. In spring and winter, the OLCI IWV observations had higher accuracy than those in summer and autumn. OLCI IWV tended to underestimate IWV value in most land cover types. Except for the polar climatic zone, the Sentinel-3 OLCI IWV products tended to overestimate IWV value. The validation results against previous studies were also discussed in this work.

The Moderate Resolution Imaging Spectroradiometer (MODIS) sensor can observe precipitable water vapor (PWV) at near-infrared (NIR) bands. In this paper, we proposed a novel Back Propagation Neural Network (BPNN) based water vapor retrieval algorithm to enhance the all-weather retrieval accuracy of the PWV estimation from MODIS NIR observations. The input of the model includes the transmittance, latitude, longitude, elevation, season, cloud, and solar zenith angle information. The water vapor data collected from in-situ 453 Global Positioning System (GPS) sites in Australia in 2017 were employed as the output PWV for the training of the BPNN method. The performance of the retrieval approach was evaluated utilizing reference GPS-measured PWV data during 2018–2019 over Australia, independent of the 2017 training data. The results indicate that the algorithm can notably enhance the accuracy of MODIS NIR PWV retrieval under all weather conditions as well as under each type of weather condition. The BPNN-retrieved weighted mean PWV data calculated with 2-channel ratio approach had the best retrieval accuracy, reducing root-mean-square error (RMSE) by 57.26% from 10.95 mm to 4.68 mm for all-weather conditions and 47.49% from 5.58 mm to 2.93 mm for confident-clear conditions. The new all-weather PWV estimates present a retrieval accuracy superior to official MODIS NIR confident-clear PWV product, illustrating the effectiveness of the retrieval algorithm. This algorithm has been well evaluated in Australia and its performance in other global regions will be investigated in the future, while the MODIS official products are applicable to the global regions.

Jason-3 is equipped with the Advanced Microwave Radiometer-2 (AMR-2) to account for the zenith wet delay (ZWD) caused by the troposphere in the altimeter signal, from which the precipitable water vapor (PWV) can be deduced. In order to investigate the accuracy of PWV from Jason-3 AMR-2 on a global scale, we adopted PWV observations from 263 radiosonde stations and 103 Global Navigation Satellite System (GNSS) stations as reference PWV. These reference PWVs are recorded during Jason-3 cycles 0-119 and are globally distributed in coastal and island regions. Over 60 000 Jason-3 PWV versus radiosonde PWV comparison points and over 380 000 Jason-3 PWV versus GNSS PWV comparison points are used in this study. For GNSS PWV, two PWV height reduction methods (Kouba empirical method and European Centre for Medium-Range Weather Forecasts (ECMWF) method) are used to reduce the PWV from height of station to sea level. The comparison results indicate that the root-mean-square error (RMSE) of Jason-3 PWV evaluated using radiosonde PWV is 3.4 kg/m2. Jason-3 PWV has an RMSE of 3.0 kg/m2with GNSS PWV derived using ECMWF PWV height correction, while the RMSE between Jason-3 PWV and GNSS PWV derived using Kouba PWV height correction is 3.1 kg/m2. In addition, the accuracy of Jason-3 PWV increases when the latitude of its footprints or the distance from its footprints to land increases.

A new algorithm of retrieving atmospheric water vapor from MODIS near-infrared (IR) (NIR) data by using a regression fitting method based on Global Positioning System (GPS)-derived water vapor is developed in this work. The algorithm has been used to retrieve total column water vapor from Moderate Resolution Imaging Spectroradiometer (MODIS) satellites both Terra and Aqua under cloud-free conditions from solar radiation in the NIR channels. Water vapor data estimated from GPS observations recorded from 2003 to 2017 by the SuomiNet GPS network over the western North America are used as ground truth references. The GPS stations were classified into six subsets based on the surface types adopted from MCD12Q1 IGBP legend. The differences in surface types are considered in the regression fitting procedure, thus different regression functions are trained for different surface types. Thus, the wet bias in the operational MODIS water vapor products has been significantly reduced. Water vapor retrieved from each of the three absorption channels and the weighted water vapor of combined three absorption channels are analyzed. Validation shows that the weighted water vapor performs better than the single-channel results. Compared to the MODIS/Terra water vapor products, the RMSE has been reduced by 50.78% to 2.229 mm using the two-channel ratio transmittance method and has been reduced by 53.06% to 2.126 mm using the three-channel ratio transmittance method. Compared to the MODIS/Aqua water vapor products, the RMSE has been reduced by 45.54% to 2.423 mm using the two-channel ratio transmittance method and has been reduced by 45.34% to 2.432 mm using the three-channel ratio transmittance method.

A novel algorithm for water vapor retrieval from Moderate Resolution Imaging Spectroradiometer (MODIS) near-infrared (NIR) channels is proposed in this research. In contrast to conventional retrieval algorithms based on radiative transfer methods, this algorithm uses the empirical regression functions to calculate precipitable water vapor (PWV). In this article, water vapor data observed from January 1, 2003, to December 31, 2017, from 464 GPS stations situated in western North America serve as reference data to determine the relationship between the transmittance of the water vapor absorption channels and atmospheric water vapor content. The model is trained on different subsets of the training data through the bootstrap resampling method. Validation results against PWV observations during the period 2010-2017 from five globally distributed GPS stations illustrate that the algorithm can significantly improve the accuracy of MODIS NIR water vapor data, with root-mean-square error (RMSE) reduction of 22.48% from 7.670 to 5.946 mm for two-channel ratio method and 21.69% from 7.670 to 6.006 mm for three-channel ratio method for MODIS/Terra satellite data, and RMSE reduction of 16.42% from 7.191 to 6.010 mm and 15.26% from 7.191 to 6.094 mm for PWV derived from two-channel and three-channel ratio methods from Aqua, respectively, for MODIS/Aqua satellite data.

The retrieval accuracies of three typical near-infrared (NIR) precipitable water vapor (PWV) products are thoroughly discussed in this article. The NIR PWV data are obtained from three satellite sensors: the Moderate-Resolution Imaging Spectroradiometer (MODIS)/Terra, the medium-resolution imaging spectrometer (MERIS)/Envisat, and the medium-resolution spectral imager (MERSI)/FY-3A. Collocated Global Positioning System (GPS) PWV data from GPS network are employed as the reference data set because of its high precision in water vapor measurement. Relative difference and root-mean-square (rms) difference are computed for “Clear,” “Cloudy,” and “All Weather” categories for each NIR water vapor product. The results reveal that PWV derived from NIR sensors tend to underestimate the water vapor values with the existence of cloud, as NIR signals cannot penetrate the cloud. Under “Clear” condition, the overall rms for remote sensors are close to the expected goal accuracies, namely, with root-mean-square-error (RMSE) of 5.480 mm for MODIS/Terra, 3.708 mm for MERIS/Envisat, 8.644 mm for MERSI/FY-3A. MERIS/Envisat has the highest PWV retrieval accuracy, while the MODIS/Terra PWV product has the best correlation with GPS PWV (R2is 0.951). The MODIS/Terra tends to overestimate PWV value, while MERSI/FY-3A tends to underestimate the PWV value. Moreover, a comprehensive comparison of seasonal variation and wet/dry variation for each NIR PWV product is also performed in this study. The results indicate that the RMSE increases significantly under wet conditions (PWV larger than 20 mm) than under dry conditions (PWV smaller than 20 mm) for all remote sensing PWV products.

Water vapor, one of the green house gases, plays many important roles in global environmental change. For instance, it has an impact on global water and energy circulation, heat transmission and climate of the world [1-3]. Additionally, water vapor is an important parameter in atmosphere correction of remote sensing [2]. The distribution of atmospheric water vapor varies significantly in both space and time domains. Specifically, water vapor in tropical area is 10 times more than that near the poles and changes temporally from milliseconds to decades due to air currents and temperature change [3]. Therefore, large-scale, long-time, and accurate water vapor retrievals are necessary in weather prediction, climate monitoring, hydrological and energy interchange.

Tropospheric path delay is an important error source in range measurements of many Earth observation systems. In this paper, the accuracies of 9 zenith hydrostatic delay (ZHD) and 18 zenith wet delay (ZWD) models are assessed using benchmark values derived from 10 years (2003–2012) of radiosonde data recorded at 92 stations in the China region. Our study confirms that ZHD can be well modeled with an accuracy of several millimeters by using surface meteorological observations. ZHD derived from the European Center for Medium-Range Weather Forecasts (ECMWF) has the best agreement of 2.8 mm with the radiosonde data in the China region, while the Baby ZHD model achieves the second best with an accuracy of 6.0 mm. All of the ZWD models can only estimate the ZWD with an accuracy of a few centimeters. ECMWF can provide ZWD estimation with the best accuracy of 21.4 mm, followed by the Baby semiempirical, Hopfield, Goad and Goodman, Askne and Nordius, Saastamoinen, Callahan, and Berman 74 ZWD models whose errors are below 40 mm. We find that in the China region all of the ZWD models perform better in winter than in summer and have higher accuracy in high latitudes than low latitudes. The performances of the 18 ZWD models are further validated in a Global Positioning System (GPS) precise point positioning (PPP) computation at 6 GPS stations in China. The PPP results also confirm that ECMWF is the best model. Considering its performance and simplicity, we conclude that Saastamoinen is the optimal ZWD model for the China region.

Traditional level set methods usually require repeated tuning of parameters, which is quite laborious and thus limits their applications. In order to simplify the parameter setting, this letter presents a fast level set algorithm that is a further extension of the original Chan-Vese model. For computational efficiency, we start by initializing the level set function in our algorithm as a binary step function rather than the often used signed distance function. Then, we eliminate the curvature-based regularizing term that is commonly used in traditional models. Thus, we can use a relatively larger time step in the numerical scheme to expedite our model. Furthermore, to keep the evolving level curves smooth, we introduce a Gaussian kernel into our algorithm to convolve the updated level set function directly. Finally, compared with other existing popular algorithms in an experiment of recognizing building roofs from high spatial resolution panchromatic images, the proposed model is much more computationally efficient while object recognition performance is comparable to other popular models.