STConvS2S: Spatiotemporal Convolutional Sequence to Sequence Network for weather forecasting

Castro, Rafaela; Souto, Yania Molina; Ogasawara, Eduardo; Porto, Fábio; Bezerra, Eduardo

doi:10.1016/j.neucom.2020.09.060

Cited by 58 publications

(31 citation statements)

References 19 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…However these models require very large datasets with tens of thousands of images, due the data-intensive training process. For this reason, CNNs and ConvLSTMs are mainly applied to data sets with short time intervals of no more than a few minutes between data points, which are typically much larger than data sets with longer time intervals [29][30][31][32][33][34][35][36][37][38][39][40]. For single-output prediction, a wider range of ML tools and time frames have been used, from linear methods in [17,21,41,42], to ensemble methods in [43][44][45], to hybrid methods in [28,[46][47][48], to deep models in [49][50][51][52][53][54][55][56] covering time scales from minutes to years.…”

Section: Literature Review and Scope Of The Researchmentioning

confidence: 99%

Basic Statistical Estimation Outperforms Machine Learning in Monthly Prediction of Seasonal Climatic Parameters

et al. 2021

View full text Add to dashboard Cite

Machine learning (ML) has been utilized to predict climatic parameters, and many successes have been reported in the literature. In this paper, we scrutinize the effectiveness of five widely used ML algorithms in the monthly prediction of seasonal climatic parameters using monthly image data. Specifically, we quantify the predictive performance of these algorithms applied to five climatic parameters using various combinations of features. We compare the predictive accuracy of the resulting trained ML models to that of basic statistical estimators that are computed directly from the training data. Our results show that ML never significantly outperforms the statistical baseline, and underperforms for most feature sets. Unlike previous similar studies, we provide error bars for the relative performance of different predictors based on jackknife estimates applied to differences in predictive error magnitudes. We also show that the practice of shuffling data sequences which was employed in some previous references leads to data leakage, resulting in over-estimated performance. Ultimately, the paper demonstrates the importance of using well-grounded statistical techniques when producing and analyzing the results of ML predictive models.

show abstract

Section: Literature Review and Scope Of The Researchmentioning

confidence: 99%

Basic Statistical Estimation Outperforms Machine Learning in Monthly Prediction of Seasonal Climatic Parameters

et al. 2021

View full text Add to dashboard Cite

show abstract

“…A ConvLSTM [15] improves an FC-LSTM by capturing both spatial and temporal patterns, and was applied to precipitation nowcasting (short period, up to 6 hours). Since then, spatiotemporal climate forecasting has widely mentioned ConvLSTM as a baseline method [24][25][26] or adopted in other model architectures [29][30][31]. Predictive RNN (PredRNN) [24] adds extra connections between adjacent time steps in a core stack of spatiotemporal LSTM (ST-LSTM), which outperforms ConvLSTM in precipitation nowcasting.…”

Section: Related Workmentioning

confidence: 99%

“…MIM captures those non-stationarity properties by adding stationary and non-stationary modules separately inside the ST-LSTM block [24]. Lastly, the most recent work on spatiotemporal climate forecasting is the spatiotemporal convolutional sequence-to-sequence network (STConvS2S) [26]. STConvS2S comprises three components, the sets of convolutional layers responsible for different tasks: the temporal block, which learns the temporal representation of sequential input by two approaches (temporal casual block [32] and temporal reversed block); spatial block, which extracts the spatial features from the temporal block; and temporal generator block, designed to output a sequence of arbitrary length.…”

Section: Related Workmentioning

confidence: 99%

“…In summary, ConvLSTM and spatiotemporal climate forecasting models [24][25][26] consider only sequential input, which is the short sequence that does not cover the periodicity patterns. SSAE [33] focuses on the long sequence that covers the entire season, but the model does not regard changes in multiple periodic cycles.…”

Section: Related Workmentioning

confidence: 99%

“…Although ConvLSTM and other novel spatiotemporal forecasting methods [24][25][26] are suitable for spatiotemporal forecasting, they do not consider periodicity, which is an essential component in climate forecasting. Periodic Convolutional Recurrent Network (Periodic-CRN) [27] has been proposed in a different domain (Taxi transportation) as a spatiotemporal forecasting model that makes use of the data's periodicity.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Soft Periodic Convolutional Recurrent Network for Spatiotemporal Climate Forecast

et al. 2021

View full text Add to dashboard Cite

Many machine-learning applications and methods are emerging to solve problems associated with spatiotemporal climate forecasting; however, a prediction algorithm that considers only short-range sequential information may not be adequate to deal with periodic patterns such as seasonality. In this paper, we adopt a Periodic Convolutional Recurrent Network (Periodic-CRN) model to employ the periodicity component in our proposals of the periodic representation dictionary (PRD). Phase shifts and non-stationarity of periodicity are the key components in the model to support. Specifically, we propose a Soft Periodic-CRN (SP-CRN) with three proposals of utilizing periodicity components: nearby-time (PRD-1), periodic-depth (PRD-2), and periodic-depth differencing (PRD-3) representation to improve climate forecasting accuracy. We experimented on geopotential height at 300 hPa (ZH300) and sea surface temperature (SST) datasets of ERA-Interim. The results showed the superiority of PRD-1 plus or minus one month of a prior cycle to capture the phase shift. In addition, PRD-3 considered only the depth of one differencing periodic cycle (i.e., the previous year) can significantly improve the prediction accuracy of ZH300 and SST. The mixed method of PRD-1, and PRD-3 (SP-CRN-1+3) showed a competitive or slight improvement over their base models. By adding the metadata component to indicate the month with one-hot encoding to SP-CRN-1+3, the prediction result was a drastic improvement. The results showed that the proposed method could learn four years of periodicity from the data, which may relate to the El Niño–Southern Oscillation (ENSO) cycle.

show abstract

Improving the CFSv2 prediction of the Indian Ocean Dipole based on a physical‐empirical model and a deep‐learning approach

Chen

Sun

Wang

et al. 2022

Intl Journal of Climatology

View full text Add to dashboard Cite

Despite significant advances in seasonal climate forecasts, the reliability of both dynamical and empirical models for the Indian Ocean Dipole (IOD) prediction is still limited to a lead time of one season or less. In this study, the skill of the NCEP Climate Forecast System version 2 (CFSv2) for the IOD prediction during the period 1982-2014 is evaluated. The results indicate that the model performance for the IOD prediction is the worst in spring among the four seasons, which is manifested in the fact that a skilful prediction of spring IOD event is limited to a lead time of only about 1-2 months. To improve the forecast of spring IOD events, a physical-empirical (PE) model and a convolutional neural network (CNN) model are established in the present study. The IOD in April-May-June (AMJ) is taken as the predictand, and the CFSv2-predicted sea surface height (SSH) in AMJ and the observed Laptev sea ice in the preceding December are used as the two predictors. The original CFSv2-predicted IOD time series has an insignificant correlation with the observed IOD time series with a temporal correlation coefficient (TCC) of 0.03; the PE model (CNN model) can largely improve the IOD prediction with a TCC of 0.74 (0.77) between the PE-model-predicted (CNN-model-predicted) IOD and the observed IOD during AMJ. Thus, the PE model and the CNN model developed in the present study can be applied to improve the IOD predictions from numerical models in the future.

show abstract

STConvS2S: Spatiotemporal Convolutional Sequence to Sequence Network for weather forecasting

Cited by 58 publications

References 19 publications

Basic Statistical Estimation Outperforms Machine Learning in Monthly Prediction of Seasonal Climatic Parameters

Basic Statistical Estimation Outperforms Machine Learning in Monthly Prediction of Seasonal Climatic Parameters

Soft Periodic Convolutional Recurrent Network for Spatiotemporal Climate Forecast

Improving the CFSv2 prediction of the Indian Ocean Dipole based on a physical‐empirical model and a deep‐learning approach

Contact Info

Product

Resources

About