The rapid advancement of artificial intelligence in domains such as natural language processing has catalyzed AI research across various fields.This study introduces a novel strategy,the AutoKeras-Knowledge Distillati...The rapid advancement of artificial intelligence in domains such as natural language processing has catalyzed AI research across various fields.This study introduces a novel strategy,the AutoKeras-Knowledge Distillation(AK-KD),which integrates knowledge distillation technology for joint optimization of large and small models in the retrieval of surface temperature and emissivity using thermal infrared remote sensing.The approach addresses the challenges of limited accuracy in surface temperature retrieval by employing a high-performance large model developed through AutoKeras as the teacher model,which subsequently enhances a less accurate small model through knowledge distillation.The resultant student model is interactively integrated with the large model to further improve specificity and generalization capabilities.Theoretical derivations and practical applications validate that the AK-KD strategy significantly enhances the accuracy of temperature and emissivity retrieval.For instance,a large model trained with simulated ASTER data achieved a Pearson Correlation Coefficient(PCC)of 0.999 and a Mean Absolute Error(MAE)of 0.348 K in surface temperature retrieval.In practical applications,this model demonstrated a PCC of 0.967 and an MAE of 0.685 K.Although the large model exhibits high average accuracy,its precision in complex terrains is comparatively lower.To ameliorate this,the large model,serving as a teacher,enhances the small model's local accuracy.Specifically,in surface temperature retrieval,the small model's PCC improved from an average of 0.978 to 0.979,and the MAE decreased from 1.065 K to 0.724 K.In emissivity retrieval,the PCC rose from an average of 0.827 to 0.898,and the MAE reduced from 0.0076 to 0.0054.This research not only provides robust technological support for further development of thermal infrared remote sensing in temperature and emissivity retrieval but also offers important references and key technological insights for the universal model construction of other geophysical parameter retrievals.展开更多
Accurate predictions of sea surface temperature(SST)are crucial due to the significant impact of SST on the global ocean-atmospheric system and its potential to trigger extreme weather events.Many existing machine-lea...Accurate predictions of sea surface temperature(SST)are crucial due to the significant impact of SST on the global ocean-atmospheric system and its potential to trigger extreme weather events.Many existing machine-learning-based ssT predictions adapt the traditional iterative point-wise prediction mechanism,whose predicting horizons and accuracy are limited owing to the high sensitivity to cumulative errors during iterative predictions.Therefore,this paper proposes a novel granulation-based long short-term memory(LsTM)-random forest(RF)combination model that can fully capture the feature dependencies involved in the fluctuation of SsT sequences,reduce the cumulative error in the iteration process,and extend the prediction horizons,which includes two sub-models(adaptive granulation model and hybrid prediction model).They can restack the one-dimensional ssT time-series into multidimensional feature variables,and achieve a strong forecasting ability.The analysis shows that the proposed model can achieve more accurate prediction-hours in nearly all prediction ranges from 1 to 125 h.The average prediction error of the proposed model in 25-125 h is 0.07 K,similar to that(0.067 K)in the first 24 h,which exhibits a high generalization performance and robustness and isthus a promising platform for the medium-and long-term forecasting of hourly SSTs.展开更多
基金supported by the Key Project of Natural Science Foundation of Ningxia Department of Science and Technology(No.2024AC02032)Fengyun Satellite Application Pilot Program Development and Application of Fengyun all-weather Land Surface Temperature Spatiotemporal Fusion Dataset of China Meteorological Administration(FY-APP-2022.0205).
文摘The rapid advancement of artificial intelligence in domains such as natural language processing has catalyzed AI research across various fields.This study introduces a novel strategy,the AutoKeras-Knowledge Distillation(AK-KD),which integrates knowledge distillation technology for joint optimization of large and small models in the retrieval of surface temperature and emissivity using thermal infrared remote sensing.The approach addresses the challenges of limited accuracy in surface temperature retrieval by employing a high-performance large model developed through AutoKeras as the teacher model,which subsequently enhances a less accurate small model through knowledge distillation.The resultant student model is interactively integrated with the large model to further improve specificity and generalization capabilities.Theoretical derivations and practical applications validate that the AK-KD strategy significantly enhances the accuracy of temperature and emissivity retrieval.For instance,a large model trained with simulated ASTER data achieved a Pearson Correlation Coefficient(PCC)of 0.999 and a Mean Absolute Error(MAE)of 0.348 K in surface temperature retrieval.In practical applications,this model demonstrated a PCC of 0.967 and an MAE of 0.685 K.Although the large model exhibits high average accuracy,its precision in complex terrains is comparatively lower.To ameliorate this,the large model,serving as a teacher,enhances the small model's local accuracy.Specifically,in surface temperature retrieval,the small model's PCC improved from an average of 0.978 to 0.979,and the MAE decreased from 1.065 K to 0.724 K.In emissivity retrieval,the PCC rose from an average of 0.827 to 0.898,and the MAE reduced from 0.0076 to 0.0054.This research not only provides robust technological support for further development of thermal infrared remote sensing in temperature and emissivity retrieval but also offers important references and key technological insights for the universal model construction of other geophysical parameter retrievals.
基金supported by Second Tibetan Plateau Scientific Expedition and Research Program(STEP)-‘Dynamic monitoring and simulation of water cycle in Asian water tower area’[grant number 2019QZKK0206]Open Fund of the State Key Laboratory of Remote Sensing Science[grant number OFSLRSS202201]+1 种基金Ningxia Science and Technology Department Flexible Introduction talent project[grant number 2021RXTDLX14]Fengyun Application Pioneering Project[grant number FY-APP-2022.0205].
文摘Accurate predictions of sea surface temperature(SST)are crucial due to the significant impact of SST on the global ocean-atmospheric system and its potential to trigger extreme weather events.Many existing machine-learning-based ssT predictions adapt the traditional iterative point-wise prediction mechanism,whose predicting horizons and accuracy are limited owing to the high sensitivity to cumulative errors during iterative predictions.Therefore,this paper proposes a novel granulation-based long short-term memory(LsTM)-random forest(RF)combination model that can fully capture the feature dependencies involved in the fluctuation of SsT sequences,reduce the cumulative error in the iteration process,and extend the prediction horizons,which includes two sub-models(adaptive granulation model and hybrid prediction model).They can restack the one-dimensional ssT time-series into multidimensional feature variables,and achieve a strong forecasting ability.The analysis shows that the proposed model can achieve more accurate prediction-hours in nearly all prediction ranges from 1 to 125 h.The average prediction error of the proposed model in 25-125 h is 0.07 K,similar to that(0.067 K)in the first 24 h,which exhibits a high generalization performance and robustness and isthus a promising platform for the medium-and long-term forecasting of hourly SSTs.
