MODIS LAND SURFACE TEMPERATURE DATA RECONSTRUCTION BASED ON THE SCLSTM MODEL

doi:10.3969/j.issn.0253-4967.2023.06.006

Abstract

Abstract:

MODIS land surface temperature(LST)products are of great value in the exchange of atmospheric matter and energy, climate change research, and detection of thermal anomalies as earthquake precursors. However, due to the influence of the cloud, there are a large number of missing values in the MODIS LST data products, limiting its wide application. Therefore, in this study we propose a method of surface temperature reconstruction based on mixed model: SCLSTM(SSA-CLSTM). Compared with traditional methods, this method does not need to establish a complex regression relationship model. In addition, since CNN can fully extract local features of one-dimensional time series data, and LSTM can fully learn long-term time series features of data, the combination of CNN and LSTM is capable of fully learning potential features of data.

Firstly in this study, the trend value of LST time series is extracted by SSA model to fill the missing pixel, and the initial reconstruction of LST is realized. Then, CLSTM(that is, 1DCNN, three-layer stacked LSTM)model is used to learn the local temporal characteristics and long-term dependence of the data, and the iterative prediction of the surface temperature of the missing pixel is realized to complete the fine reconstruction of the data. Based on the experimental results in Hotan region of Xinjiang and Wenchuan region of Sichuan, it can be proved that compared with the other two existing reconstruction methods based on mixed models, the reconstructed return data error is minimum, and the consistency with the original data is the highest. The RMSE of this method can be reduced to 0.712K, the minimum is 0.695K, and the correlation between the reconstructed return data and the original data can reach more than 0.95. In addition, the reliability of the method is further verified by the measured surface temperature data of the meteorological station. In summary, the proposed method provides a new technical means and ideas for deep learn-based reconstruction work, and also provides a solid data foundation for the research of surface processes and seismic thermal anomalies.

Using MODIS MYD11A2 remote sensing data and based on the proposed new method, the reconstruction experiments were carried out in Hotan region of Xinjiang Uygur Autonomous Region and Wenchuan region of Sichuan Province with the strategy of “remove-construction-contrast”, and the results were compared with other two mixed model methods. In addition, the reliability of the reconstruction accuracy of the proposed method is verified based on the 0cm measured surface temperature data of the weather station. The main conclusions are as follows:

(1)Through the reconstruction experiment of LST in Hotan, Xinjiang, it is concluded that compared with the existing two hybrid model reconstruction methods, the new method can better capture the time series features of LST data, so that the reconstructed image can not only better maintain the texture features of the original image, but also improve the accuracy of data reconstruction. The reconstruction error is the smallest among the three methods, RMSE can be reduced to 0.712K, and the correlation with the original data can reach more than 0.95 after reconstruction.

(2)In order to prove the regional applicability of the new method, a reconstruction experiment was carried out in Wenchuan region of Sichuan Province, and the missing values were reconstructed using the proposed method. Through this reconstruction experiment, we found that the reconstruction method in this paper can achieve better data reconstruction effect even in areas with more cloud and fog coverage, poor weather conditions, and complex land cover types, which proves the reliability and regional universality of the method in this paper.

(3)To further verify the reliability of the new method, the accuracy of the new method was evaluated by using the measured 0cm surface temperature data of 6 meteorological stations in Hotan region of Xinjiang. Based on the temporal variation characteristics of the MODIS return data from 2015 to 2019, the return values of 2020 are reconstructed, and the reconstructed results are compared with the measured data. By comparing the 0cm measured data of the meteorological station and the data before and after reconstruction, it can be concluded that the correlation and average deviation of the returned data and measured data after reconstruction based on SCLSTM method are closer to the correlation and average deviation of the original data and measured data. Therefore, the reconstructed data based on the new method can maintain a good consistency with the original data.

(4)By reconstructing the missing value regions of Hotan region of Xinjiang in 2008 and Wenchuan region of Sichuan in 2020, we found that the texture of the images after the supplementary value is fine and natural, without obvious boundary effect. Therefore, it can be proved that the method in this paper can realize the data reconstruction of a large area with missing values.

In summary, the method proposed in this paper provides a new idea and technique for MODIS surface temperature reconstruction work based on deep learning, and also provides a solid data foundation for the ground process research and seismic thermal anomaly information extraction based on MODIS LST.

Key words: Land surface temperature, SSA, CNN, LSTM, MODIS

摘要：

MODIS(Moderate-resolution Imaging Spectroradiometer, 中分辨率成像光谱仪)LST(Land Surface Temperature, 地表温度)产品在大气物质和能量交换、气候变化研究及地震前兆热异常探测等方面具有重要价值。然而, 由于云的遮挡导致MODIS LST数据产品中存在大量空值, 限制了其广泛应用。为此, 文中提出了一种基于混合模型的地表温度重建方法——SCLSTM(即SSA-CLSTM)。与传统方法相比, 该方法无需建立复杂的回归关系模型。此外, 由于CNN(Convolutional Neural Network, 卷积神经网络)能够充分提取一维时间序列数据的局部特征, 而LSTM(Long Short-Term Memory, 长短期记忆)能够充分学习数据的长时间序列特征, 因此将CNN和LSTM结合能够更加充分地学习数据的潜在特征。首先, 使用SSA(Singular Spectrum Analysis, 奇异谱分析)模型提取出地表温度时间序列中的趋势值用于填补缺值像元, 实现地表温度的初步重建。然后, 再利用SCLSTM(即1DCNN-3层堆叠LSTM)模型学习数据的局部时序特征和长期依赖关系, 并实现对缺失像元的地表温度进行迭代预测, 完成数据的精细重建。新疆和田地区和四川汶川地区的实验结果表明, 文中方法与现有其他2种基于混合模型的重建方法相比, 重建后的LST数据误差最小, 与原始数据的一致性最高。其中, 文中方法的RMSE可降至0.712K, AD为0.695K, 重建后的LST数据与原始数据的相关系数可达0.95以上。此外, 气象站的实测地表温度数据也进一步验证了该方法的可靠性。文中所提方法为基于深度学习的LST重建工作提供了一种新的技术手段和思路, 同时也为基于LST的地表过程和地震热异常研究提供了坚实的数据基础。

关键词: 地表温度, SSA, CNN, LSTM, MODIS

SONG Dong-mei, ZHANG Man-yu, SHAN Xin-jian, WANG Bin. MODIS LAND SURFACE TEMPERATURE DATA RECONSTRUCTION BASED ON THE SCLSTM MODEL[J]. SEISMOLOGY AND GEOLOGY, 2023, 45(6): 1349-1369.

宋冬梅, 张曼玉, 单新建, 王斌. 基于SCLSTM模型的MODIS地表温度产品重建方法[J]. 地震地质, 2023, 45(6): 1349-1369.

Figures/Tables 18

Fig. 1 Overview of study area.

Fig. 2 Method showing data reconstruction process in this study.

Fig. 3 Process of preliminary LST data reconstruction based on the SSA method.

Fig. 4 Schematic diagram of the structure of one-dimensional convolutional layers and the mode of kernel slide.

Fig. 5 Schematic diagram of the LSTM model.

Fig. 6 Data reconstruction accuracy of different LSTM layers in SCLSTM model.

Table1 Parameter information of CNN and LSTM model

模型	参数	数值
CNN	卷积层	5
	卷积核大小	1×3
	空洞率	1/2/4/8/16
	通道数	32
LSTM	层数	3
	神经元数	128/64/32
	步长	1
	学习率	0.002

Fig. 7 Convolution operation between kernel weight and local segments of LST time series.

Fig. 8 Process of refined LST data reconstruction with the three-layer superimposed LSTM.

Fig. 9 Process of refined reconstruction of LST time series data through iteration.

Fig. 10 Comparison of reconstruction effects between CNN-LSTM and LSTM.

Fig. 11 Accuracy and efficiency analysis of LST reconstruction in Hotan region, Xinjiang.

Fig. 12 Comparison of reconstruction effects between CNN-LSTM and LSTM.

Fig. 13 Accuracy and efficiency analysis of LST reconstruction in Wenchuan region, Sichuan.

Table2 The spatial locations of the meteorological stations

站点	北纬/(°)	东经/(°)
皮山	37.37	78.17
墨玉	37.1	79.38
和田	37.08	79.56
洛浦	37.05	80.10
策勒	37.01	80.48
于田	36.51	81.39

Fig. 14 Reliability analysis of measured LST based on meteorological station.

Fig. 15 Comparison of the original and reconstructed LST images for the Hotan region in Xinjiang.

Fig. 16 Comparison of the original and reconstructed LST images for the Wenchuan region in Sichuan.

References 41

[1]	陈赐阳, 陈德旺. 2021. 基于CNN-LSTM的锂电池剩余寿命(RUL)间接预测研究[J]. 电源技术, 45(5): 589—594.
	CHEN Ci-yang, CHEN De-wang. 2021. Research on indirect prediction of lithium battery RUL based on CNN-LSTM[J]. Power Supply Technology, 45(5): 589—594 (in Chinese).
[2]	陈浩文. 2020. 基于奇异谱分析与神经网络的中期负荷预测方法研究[D]. 青岛: 华北电力大学.
	CHEN Hao-wen. 2020. Research on medium load forecast method based on Singular Spectrum Analysis and Neural Network[D]. School of Electrical and Electronic Engineering. North China Electric Power University, Qingdao (in Chinese).
[3]	陈顺云, 马瑾, 刘培洵, 等. 2014. 利用卫星遥感热场信息探索现今构造活动: 以汶川地震为例[J]. 地震地质, 36(3): 775—793. doi: 10.3969/j.issn.0253-4967.2014.03.018.
	CHEN Shun-yun, MA Jin, LIU Pei-xun, et al. 2014. Exploring the current tectonic activity with satellite remote sensing thermal information: A case of the Wenchuan earthquake[J]. Seismology and Geology, 36(3): 775—793 (in Chinese).
[4]	陈顺云, 马瑾, 刘培洵, 等. 2009. 中国大陆地表温度年变基准场研究[J]. 地球物理学报, 52(9): 2273—2281.
	CHEN Shun-yun, MA Jin, LIU Pei-xun, et al. 2009. A study on the normal annual variation field of land surface temperature in China[J]. Chinese Journal of Geophysics, 52(9): 2273—2281 (in Chinese).
[5]	陈顺云, 马瑾, 刘培洵, 等. 2004. 中国地表亮度温度年变基准场[J]. 地震地质, 26(3): 528—538. doi: 10.3969/j.issn.0253-4967.2004.03.017.
	CHEN Shun-yun, MA Jin, LIU Pei-xun, et al. 2004. Normal annual variation field of land surface brightness temperature in China[J]. Seismology and Geology, 26(3): 528—538 (in Chinese).
[6]	倪楠. 2020. 中国降水的时空变化特征研究[D]. 北京: 对外经济贸易大学.
	NI Nan. 2020. Spatial and temporal variation of precipitation in China[D]. University of International Business and Economics, Beijing (in Chinese).
[7]	苏扬, 吴鹏海, 程洁, 等. 2022. AMSR-E地表温度数据重建深度学习方法[J]. 遥感学报, 26(4): 739—751.
	SU Yang, WU Peng-hai, CHENG Jie, et al. 2022. Research on deep learning methods for AMSR-E land surface temperature data reconstruction[J]. Journal of Remote Sensing, 26(4): 739—751 (in Chinese).
[8]	闫章美, 周德成, 张良侠. 2021. 我国3大城市群地区城市和农业用地地表热环境效应对比研究[J]. 生态学报, 41(22): 8870—8881.
	YAN Zhang-mei, ZHOU De-cheng, ZHANG Liang-xia. 2021. Contrasting surface thermal environmental effects of urban and agricultural lands in three major urban agglomerations in China[J]. Acta Ecologica Sinica, 41(22): 8870—8881 (in Chinese).
[9]	臧琳. 2017. 基于红外异常的发震信号识别与危险性评估[D]. 青岛: 中国石油大学(华东).
	ZANG Lin. 2017. Seismic signals identification and risk assessment based on thermal infrared anomalies[D]. China University of Petroleum(East China), Qingdao (in Chinese).
[10]	张军, 覃志豪, 刘梅, 等. 2011. 利用空间插值法估算云覆盖像元地表温度的可行性研究[J]. 地理与地理信息科学, 27(6): 45—49.
	ZHANG Jun, QIN Zhi-hao, LIU Mei, et al. 2011. Estimating of land surface temperature under the cloud cover with spatial interpolation[J]. Geography and Geo-Information Science, 27(6): 45—49 (in Chinese).
[11]	Arslan N, Sekertekin A. 2019. Application of Long Short-Term Memory neural network model for the reconstruction of MODIS Land Surface Temperature images[J]. Journal of Atmospheric and Solar-Terrestrial Physics, 194: 1—9.
[12]	Chen S, Liu P, Feng T, et al. 2020. Exploring changes in land surface temperature possibly associated with earthquake: Case of the April 2015 Nepal M_W7.9 earthquake[J]. Entropy, 22(4): 1—23. DOI URL
[13]	Cui J Y, Zhang M Y, Song D M, et al. 2022. MODIS land surface temperature product reconstruction based on the SSA-BiLSTM Model[J]. Remote Sensing, 14(4): 1—16. DOI URL
[14]	Fan X M, Liu H G, Liu G H, et al. 2014. Reconstruction of MODIS land-surface temperature in a flat terrain and fragmented landscape[J]. International Journal of Remote Sensing, 35(23): 7857—7877. DOI URL
[15]	Fu P, Xie Y, Weng Q, et al. 2019. A physical model-based method for retrieving urban land surface temperatures under cloudy conditions[J]. Remote Sensing of Environment, 230(26): 1—12.
[16]	Hochreiter S, Schmidhuber J. 1997. Long short-term memory[J]. Neural Computation, 9(8): 1735—1780. DOI PMID
[17]	Kilibarda M, Hengl T, Heuvelink G B M, et al. 2014. Spatio-temporal interpolation of daily temperatures for global land areas at 1km resolution[J]. Journal of Geophysical Research: Atmospheres, 119(5): 2294—2313. DOI URL
[18]	Kou X, Jiang L, Bo Y, et al. 2016. Estimation of land surface temperature through blending MODIS and AMSR-E data with the Bayesian maximum entropy method[J]. Remote Sensing, 8(2): 1—17. DOI URL
[19]	Malamiri G R H, Rousta I, Olafsson H, et al. 2018. Gap-filling of MODIS time series land surface temperature(LST)products using singular spectrum analysis(SSA)[J]. Atmosphere, 9(9): 1—19.
[20]	Mao K B, Yuan Z J, Zuo Z Y, et al. 2019. Changes in global cloud cover based on remote sensing data from 2003 to 2012[J]. Chinese Geographical Science, 29(2): 306—315. DOI
[21]	Moncet J L, Liang P, Galantowicz J F, et al. 2011. Land surface microwave emissivities derived from AMSR-E and MODIS measurements with advanced quality control[J]. Journal of Geophysical Research: Atmospheres, 116(16): 1—20.
[22]	Neteler M. 2010. Estimating daily land surface temperatures in mountainous environments by reconstructed MODIS LST data[J]. Remote Sensing, 2(1): 333—351. DOI URL
[23]	Nguyen O V, Kawamura K, Trong D P, et al. 2015. Temporal change and its spatial variety on land surface temperature and land use changes in the Red River Delta, Vietnam, using MODIS time-series imagery[J]. Environmental Monitoring and Assessment, 187(7): 1—11. DOI URL
[24]	Prigent C, Jimenez C, Aires F. 2016. Toward “all weather, ” long record, and real-time land surface temperature retrievals from microwave satellite observations[J]. Journal of Geophysical Research: Atmospheres, 121(10): 5699—5717. DOI URL
[25]	Shwetha H R, Kumar N D. 2016. Prediction of high spatio-temporal resolution land surface temperature under cloudy conditions using microwave vegetation index and ANN[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 117: 40—55. DOI URL
[26]	Song D M, Xie R H, Zang L, et al. 2018. A new algorithm for the characterization of thermal infrared anomalies in tectonic activities[J]. Remote Sensing, 10(12): 1—33. DOI URL
[27]	Sun L, Chen Z, Gao F, et al. 2017. Reconstructing daily clear-sky land surface temperature for cloudy regions from MODIS data[J]. Computers & Geosciences, 105(8): 10—20. DOI URL
[28]	Tan J L, Che T, Wang J, et al. 2021. Reconstruction of the daily MODIS land surface temperature product using the two-step improved similar pixels method[J]. Remote Sensing, 13(9): 1—27. DOI URL
[29]	Volkan Y, Senyurek, Masudul H, et al. 2020. A CNN-LSTM neural network for recognition of puffing in smoking episodes using wearable sensors[J]. Biomedical Engineering Letters, 10(2): 195—203. DOI PMID
[30]	Wu L Z, Kong C, Hao X H, et al. 2020. A short-term load forecasting method based on GRU-CNN hybrid neural network model[J]. Mathematical Problems in Engineering, 220: 1—10.
[31]	Wu P, Yin Z, Yang H, et al. 2019. Reconstructing geostationary satellite land surface temperature imagery based on a multiscale feature connected convolutional neural network[J]. Remote Sensing, 11(3): 1—18. DOI URL
[32]	Wu P, Yin Z, Yang H, et al. 2021. Spatially continuous and high-resolution land surface temperature product generation: A review of reconstruction and spatiotemporal fusion techniques[J]. IEEE Geoscience and Remote Sensing Magazine, 9(3): 112—137.
[33]	Xie F, Fan H. 2021. Deriving drought indices from MODIS vegetation indices(NDVI/EVI)and Land Surface Temperature(LST): Is data reconstruction necessary?[J]. International Journal of Applied Earth Observation and Geoinformation, 11: 1—16. DOI URL
[34]	Yang Y, Wang J. 2021. Forecasting wavelet neural hybrid network with financial ensemble empirical mode decomposition and MCID evaluation[J]. Expert Systems with Applications, 166: 1—14.
[35]	Zeng C, Long D, Shen H, et al. 2018. A two-step framework for reconstructing remotely sensed land surface temperatures contaminated by cloud[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 141: 30—45. DOI URL
[36]	Zhang Q, Cheng J. 2020. An empirical algorithm for retrieving land surface temperature from AMSR-E data considering the comprehensive effects of environmental variables[J]. Earth and Space Science 7: 325—346.
[37]	Zhang X, Zhou J, Liang S, et al. 2020. Estimation of 1-km all-weather remotely sensed land surface temperature based on reconstructed spatial-seamless satellite passive microwave brightness temperature and thermal infrared data[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 167: 321—344. DOI URL
[38]	Zhang X, Zhou J, Liang S, et al. 2021. A practical reanalysis data and thermal infrared remote sensing data merging(RTM)method for reconstruction of a 1-km all-weather land surface temperature[J]. Remote Sensing of Environment, 260: 1—21.
[39]	Zhao B, Mao K B, Cai Y L, et al. 2020. A combined Terra and Aqua MODIS land surface temperature and meteorological station data product for China from 2003 to 2017[J]. Earth System Science Data, 12(4): 2555—2577. DOI URL
[40]	Zhao W, Duan S B. 2020. Reconstruction of daytime land surface temperatures under cloud-covered conditions using integrated MODIS/Terra land products and MSG geostationary satellite data[J]. Remote Sensing of Environment, 247: 1—15.
[41]	Zhou J, Zhang X, Zhan W, et al. 2017. A thermal sampling depth correction method for land surface temperature estimation from satellite passive microwave observation over barren land[J]. IEEE Transactions on Geoscience & Remote Sensing, 55(8): 4743—4756.