Evaluation Procedures for Forecasting with Spatiotemporal Data

Oliveira, Mariana; Torgo, Luı́s; Costa, Vítor Santos

doi:10.3390/math9060691

Cited by 16 publications

(12 citation statements)

References 33 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…First, we examined root mean square error (RMSE), mean absolute difference (MAD) of predicted normalized yield, and Deviance Information Criterion (DIC). There is agreement that random cross-validation is optimistic and block cross-validation is pessimistic, so we examined both (Oliveira et al, 2021). Two cross-validations were tested: a five-fold random cross-validation, and a block cross-validation, using 6 contiguous space-time blocks (dividing the dataset into two contiguous space-blocks and three contiguous time-blocks).…”

Section: Discussionmentioning

confidence: 99%

A Comparison of Yield Prediction Approaches Using Long-Term Multi-Crop Site-Specific Data

Casanova,

Heineck,

Huggins

2024

Journal of the ASABE

View full text Add to dashboard Cite

Highlights Long-term site-specific multiple-crop yield data is used in two modeling approaches. Different sources of data, including soil properties, topography, weather, and multispectral data, are tested. Linear modeling and Bayesian hierarchical modeling (BHM) are evaluated by examining predictions of relative yield. BHM with spatio-temporal effects provides the best estimates, handles missing data, and provides uncertainty estimates in time and space. Abstract. Growers in the inland Pacific Northwest face numerous challenges in managing cropping systems. Climate variability, soil degradation, and topography all lead to significant spatial and temporal variability in yield. Often, yield modeling approaches such as deep learning can be “black boxes” or suffer from parameter uncertainty and instability, as with process-based crop models. To explore an alternative, we examined nearly two decades of crop rotation data from the R.J. Cook Agronomy Farm, along with soil properties, topography, weather, and multispectral data. We then tested two modeling approaches to estimate yield: linear modeling (LM) and Bayesian hierarchical modeling (BHM). We found BHM with spatial and temporal random effects performed best in predicting relative yield, both using soil variables as predictors or remotely sensed data. Since the BHM approach handles missing data, offers the possibility of farmer knowledge to be incorporated into prior probabilities, and gives uncertainty, this methodology lends itself well to decision support tools and on-farm study design. Keywords: Bayesian, Long-term, Modeling, Remote sensing, Yield.

show abstract

Section: Discussionmentioning

confidence: 99%

A Comparison of Yield Prediction Approaches Using Long-Term Multi-Crop Site-Specific Data

Casanova,

Heineck,

Huggins

2024

Journal of the ASABE

View full text Add to dashboard Cite

show abstract

“…Dealing with forecasting (Oliveira et al., 2021), we do a temporal split for the evaluation. We use years 2009–2018 for training and 2019 for validation.…”

Section: Methodsmentioning

confidence: 99%

Wildfire Danger Prediction and Understanding With Deep Learning

Kondylatos

Prapas

Ronco

et al. 2022

Geophysical Research Letters

View full text Add to dashboard Cite

Climate change exacerbates the occurence of extreme droughts and heatwaves, increasing the frequency and intensity of large wildfires across the globe. Forecasting wildfire danger and uncovering the drivers behind fire events become central for understanding relevant climate‐land surface feedback and aiding wildfire management. In this work, we leverage Deep Learning (DL) to predict the next day's wildfire danger in a fire‐prone part of the Eastern Mediterranean and explainable Artificial Intelligence (xAI) to diagnose model attributions. We implement DL models that capture the temporal and spatio‐temporal context, generalize well for extreme wildfires, and demonstrate improved performance over the traditional Fire Weather Index. Leveraging xAI, we identify the substantial contribution of wetness‐related variables and unveil the temporal focus of the models. The variability of the contribution of the input variables across wildfire events hints into different wildfire mechanisms. The presented methodology paves the way to more robust, accurate, and trustworthy data‐driven anticipation of wildfires.

show abstract

“…To evaluate the performance of the models, the common methods in the literature on spatio‐temporal analysis include the time‐wise holdout methods (i.e., withhold some observations from the last part of the time series as the testing set and train the model on the remaining observations) (Oliveira et al, 2021; Walker et al, 2022) and the ‘target‐oriented’ cross‐validation (CV) strategies (i.e., variants of CV that deal with either spatial dimensional or temporal dimension or both, which includes leave‐location‐out CV, leave‐time‐out CV, and leave‐location‐and‐time‐out CV) (Arowosegbe et al, 2022; Gao et al, 2019; Meyer et al, 2018). For the doubly nested spatial structure of the cost data (patients nested within hospitals that are nested within health service areas), we adopt a modified leave‐location‐out CV to evaluate the performance of the four spatio‐temporal regression methods.…”

Section: Analysis Of the Alcohol‐related Inpatient Hospital Costsmentioning

confidence: 99%

Bayesian Spatio-Temporal Modeling for the Inpatient Hospital Costs of Alcohol-Related Disorders

Härdle

et al. 2022

Journal of the Royal Statistical Society Series A: Statistics in Society

View full text Add to dashboard Cite

Understanding how health care costs vary across different demographics and health conditions is essential to developing policies for health care cost reduction. It may not be optimal to apply the conventional mean regression due to its sensitivity to the high level of skewness and spatio‐temporal heterogeneity presented in the cost data. To find an alternative method for spatio‐temporal analysis with robustness and high estimation efficiency, we combine information across multiple quantiles and propose a Bayesian spatio‐temporal weighted composite quantile regression (ST‐WCQR) model. An easy‐to‐implement Gibbs sampling algorithm is provided based on the asymmetric Laplace mixture representation of the error term. Extensive simulation studies show that ST‐WCQR outperforms existing methods for skewed error distributions. We apply ST‐WCQR to investigate how patients' characteristics affected the inpatient hospital costs for alcohol‐related disorders and identify areas that could be targeted for cost reduction in New York State from 2015 to 2017.

show abstract

Evaluation Procedures for Forecasting with Spatiotemporal Data

Cited by 16 publications

References 33 publications

A Comparison of Yield Prediction Approaches Using Long-Term Multi-Crop Site-Specific Data

A Comparison of Yield Prediction Approaches Using Long-Term Multi-Crop Site-Specific Data

Wildfire Danger Prediction and Understanding With Deep Learning

Bayesian Spatio-Temporal Modeling for the Inpatient Hospital Costs of Alcohol-Related Disorders

Contact Info

Product

Resources

About