Balancing prediction accuracy and generalization ability: A hybrid framework for modelling the annual dynamics of satellite-derived land surface temperatures

Liu, Zihan; Zhan, Wenfeng; Lai, Jiameng; Hong, Falu; Quan, Jinling; Bechtel, Benjamin; Huang, Fan; Zhi, Zou

doi:10.1016/j.isprsjprs.2019.03.013

Cited by 52 publications

(48 citation statements)

References 58 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…These ACPs depend on daily LSTs and are not influenced by some gaps due to cloud contamination [12]. These informative parameters can be used to study the LST climatology [12,16] or to downscale LST images [27].…”

Section: A Accuracy Assessments Of the Yycd_acp3 Modelmentioning

confidence: 99%

“…These two models are simple and have a robust physical basis [26]. Later, some researchers proposed a hybrid framework with the related surface parameters, such as air temperature, soil moisture, albedo, and the relative humidity for modeling the annual dynamics of satellite-derived LST [27,28]. Although this framework can achieve higher accuracy, it is complicated because of the input of a large amount of auxiliary data.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Modeling Year-to-Year Variations of Clear-Sky Land Surface Temperature Using Aqua/MODIS Data

Xing

Duan

et al. 2020

IEEE Access

View full text Add to dashboard Cite

Land surface temperature (LST) and its annual or inter-annual variations play an important role in understanding global climate change, urban heat island, and the process of land-atmosphere energy exchange. Many annual temperature cycle (ATC) models [i.e., ATC with three or five parameters (ACP3 or ACP5)] have been proposed to analyze the annual variations of LST in the past decades. In this study, two year-to-year continuous and derivable models (YYCD_ACP3 and YYCD_ACP5 models) were proposed to model several years of ATCs. The fitting results of the YYCD_ACP3 model with global Aqua/MODIS daytime LSTs from 2014 to 2018 show that the YYCD_ACP3 model achieved a good performance in fitting the time-series LSTs with an overall normalized root-mean-square error (NRMSE) of 0.21, coefficient of determination (R 2) of 0.74, and refined index of agreement (d) of 0.85. In addition, the modeling results of ten representative samples covering different climatic conditions and land cover worldwide show that, except for two sites located in tropical and Antarctic, the YYCD_ACP3 model could show a good performance with R 2 greater than 0.6. Although the ACP3 model shows similar performance to the YYCD_ACP3 model, the fitting curve of the YYCD_ACP3 model is continuous and smooth for describing the interannual variations of LST. When the LSTs of 2014-2018 are fitted as a whole by using both models, the YYCD_ACP3 model shows a slightly better performance than that of the ACP3 model. The application of the YYCD_ACP3 model with the global MODIS LSTs from 2003 to 2018 indicates that the results of the YYCD_ACP3 model have the potential to reveal the interannual variations of LST. Therefore, we conclude that the YYCD models are valuable for modeling the variations of LST over several years and can be widely applied.

show abstract

Section: A Accuracy Assessments Of the Yycd_acp3 Modelmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Modeling Year-to-Year Variations of Clear-Sky Land Surface Temperature Using Aqua/MODIS Data

Xing

Duan

et al. 2020

IEEE Access

View full text Add to dashboard Cite

show abstract

“…However, a major shortcoming of TIR measurements is their low tolerance to cloud contamination and atmospheric conditions [16]. This sensitivity to clouds and incapacity to deal with cloudy pixels [17] lead to more than one-half of LST unavailable [2], which greatly challenges practical applications of TIR LST particularly when high spatiotemporal resolution is indispensable [16], [18], [19]. Therefore, accurate and practical methods for reconstructing seamless LST are highly desirable.…”

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

“…Model-driven methods are proposed mainly based on physical (or sometimes statistical) models with capability of delineating spatiotemporal variations of LST to predict missing values under cloudy coverage [18]. Especially, annual temperature cycle (ATC) models based on LST temporal structure are good at estimating missing data even when large spatiotemporal gap existing [13], [18]. However, spatiotemporal interpolation and model simulation methods, with exclusive usage of TIR LST, are only an approximation of the theoretical clear-sky LST rather than an estimation of actual cloudy LST [2], [22].…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Adjustment From Temperature Annual Dynamics for Reconstructing Land Surface Temperature Based on Downscaled Microwave Observations

Chen

Xia

et al. 2020

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

View full text Add to dashboard Cite

Land surface temperature (LST) is crucial to wide varieties of environmental issues, whereas its low tolerance to cloud contamination greatly challenges its applications. Passive microwave (PMW) measurements are employed to retrieve LST due to its great capability of penetrating clouds. Despite of great efforts from previous studies, their further applications are limited by coarse resolution of PMW, uncertainty from cloudy coverage, and less attention on large-scale applications. To address these problems, this paper proposes a method combining adjustment from annual temperature circle to reconstruct land surface temperature based on downscaled microwave observations over mainland China. In order to conduct a comprehensive and solid accuracy evaluation, the proposed method is validated according to three strategies namely comparing with MODIS LST, ground surface temperature (GST), and in-situ observations from the Heihe Watershed Allied Telemetry Experimental Research (HiWATER) respectively. In the strategy 1, mean RMSE of predicted LST is 2.39 K and 1.33 K at day and night. According to the strategy 2, mean RMSE of MODIS, predicted and merged LST is 4.97 K, 5.07 K, and 4.93 K at daytime, and 2.32 K, 4.57 K and 3.39 K at nighttime. In the strategy 3, they are 3.75 K, 3.90 K and 2.87 K at day, and 2.80 K, 3.61 K and 2.79 K at night. Furthermore, adjustments from annual temperature cycle are discussed. The proposed method is promising in future research for reconstructing LST with high spatiotemporal resolution at large scales thanks to its simple process, high accuracy, and seamless reconstruction. Index Terms-land surface temperature; passive microwave; annual temperature cycle; downscaling

show abstract

A Machine Learning-Based Model for Predicting Temperature Under the Effects of Climate Change

Shams¹,

Tarek²,

Elshewey³

et al. 2023

Studies in Big Data

View full text Add to dashboard Cite

Balancing prediction accuracy and generalization ability: A hybrid framework for modelling the annual dynamics of satellite-derived land surface temperatures

Cited by 52 publications

References 58 publications

Modeling Year-to-Year Variations of Clear-Sky Land Surface Temperature Using Aqua/MODIS Data

Modeling Year-to-Year Variations of Clear-Sky Land Surface Temperature Using Aqua/MODIS Data

Adjustment From Temperature Annual Dynamics for Reconstructing Land Surface Temperature Based on Downscaled Microwave Observations

A Machine Learning-Based Model for Predicting Temperature Under the Effects of Climate Change

Contact Info

Product

Resources

About