One-step ahead forecasting of geophysical processes within a purely statistical framework
- 152 Downloads
Abstract
The simplest way to forecast geophysical processes, an engineering problem with a widely recognized challenging character, is the so-called “univariate time series forecasting” that can be implemented using stochastic or machine learning regression models within a purely statistical framework. Regression models are in general fast-implemented, in contrast to the computationally intensive Global Circulation Models, which constitute the most frequently used alternative for precipitation and temperature forecasting. For their simplicity and easy applicability, the former have been proposed as benchmarks for the latter by forecasting scientists. Herein, we assess the one-step ahead forecasting performance of 20 univariate time series forecasting methods, when applied to a large number of geophysical and simulated time series of 91 values. We use two real-world annual datasets, a dataset composed by 112 time series of precipitation and another composed by 185 time series of temperature, as well as their respective standardized datasets, to conduct several real-world experiments. We further conduct large-scale experiments using 12 simulated datasets. These datasets contain 24,000 time series in total, which are simulated using stochastic models from the families of AutoRegressive Moving Average and AutoRegressive Fractionally Integrated Moving Average. We use the first 50, 60, 70, 80 and 90 data points for model-fitting and model-validation, and make predictions corresponding to the 51st, 61st, 71st, 81st and 91st respectively. The total number of forecasts produced herein is 2,177,520, among which 47,520 are obtained using the real-world datasets. The assessment is based on eight error metrics and accuracy statistics. The simulation experiments reveal the most and least accurate methods for long-term forecasting applications, also suggesting that the simple methods may be competitive in specific cases. Regarding the results of the real-world experiments using the original (standardized) time series, the minimum and maximum medians of the absolute errors are found to be 68 mm (0.55) and 189 mm (1.42) respectively for precipitation, and 0.23 °C (0.33) and 1.10 °C (1.46) respectively for temperature. Since there is an absence of relevant information in the literature, the numerical results obtained using the standardized real-world datasets could be used as rough benchmarks for the one-step ahead predictability of annual precipitation and temperature.
Keywords
ARFIMA Benchmarking time series forecasts Machine learning Neural networks Precipitation Random forests Simple exponential smoothing Support vector machines Temperature Univariate time series forecastingAbbreviations
- ACF
AutoCorrelation Function
- AR
AutoRegressive
- ARFIMA
AutoRegressive Fractionally Integrated Moving Average
- ARIMA
AutoRegressive Integrated Moving Average
- ARMA
AutoRegressive Moving Average
- MA
Moving Average
- SARIMA
Seasonal AutoRegressive Integrated Moving Average
Background
Forecasting geophysical variables in various time scales and horizons is useful in technological applications (e.g. Giunta et al. 2015), but a difficult task as well. Precipitation and temperature forecasting is mostly based on deterministic models as the Global Circulation Models (GCMs), which simulate the Earth’s atmosphere using numerical equations; therefore, deviating from traditional time series forecasting, i.e. univariate time series forecasting. This particular deviation has been questioned by forecasting scientists (Green and Armstrong 2007; Green et al. 2009; Fildes and Kourentzes 2011, see also the comments in Keenlyside 2011; McSharry 2011). Traditional time series forecasting can be performed using several classes of regression models, as reviewed in De Gooijer and Hyndman (2006), while the two major classes are stochastic and machine learning. Regression models are in general fast-implemented in contrast to their computationally intensive alternative in precipitation and temperature forecasting, i.e. the GCMs. For their simplicity and easy applicability, the former have been proposed as benchmarks for the latter by Green et al. (2009).
Examples of univariate time series forecasting in geoscience
s/n | Study | Process | Number of original time series | Forecast time scale | Forecast horizon(s) [step(s) ahead] | Univariate time series forecasting method(s) |
---|---|---|---|---|---|---|
1 | Hong (2008) | Precipitation | 9 | Hourly | 1 | (1) Support vector machines (2) Hybrid model, i.e. a combination of recurrent neural networks and support vector machines |
2 | Chau and Wu (2010) | 2 | Daily | 1, 2, 3 | (1) Neural networks (2) Hybrid model, i.e. a combination of neural networks and support vector machines | |
3 | Htike and Khalifa (2010) | 1 | Monthly, biannually, quarterly, yearly | 1 | Neural networks | |
4 | Wu et al. (2010) | 4 | Monthly, daily | 1, 2, 3 | (1) Linear regression (2) k-nearest neighbours (3) Neural networks (4) Hybrid model, i.e. a combination of neural networks | |
5 | Narayanan et al. (2013) | 6 | Yearly | 21 × 3 (months) | AutoRegressive Integrated Moving Average (ARIMA) | |
6 | Wang et al. (2013) | 1 | Monthly | 12 | Seasonal AutoRegressive Integrated Moving Average (SARIMA) | |
7 | Babu and Reddy (2012) | Temperature | 1 | Yearly | 10 | (1) ARIMA (2) Wavelet-based ARIMA |
8 | Chawsheen and Broom (2017) | 1 | Monthly | 121 | SARIMA | |
9 | Lambrakis et al. (2000) | Streamflow or river discharge | 1 | Daily | 1 | (1) Farmer’s model (2) Neural networks |
10 | Ballini et al. (2001) | 1 | Monthly | 1, 3, 6, 12 | (1) AutoRegressive Moving Average (ARMA) (2) Neural networks (3) Neurofuzzy networks | |
11 | Yu et al. (2004) | 2 | Daily | 1 | (1) Support vector machines coupled with an evolutionary algorithm (2) Standard chaos technique (3) Naïve (4) Inverse approach (5) ARIMA | |
12 | Komorník et al. (2006) | 7 | Monthly | 1, 3, 6, 12 | (1) Threshold AutoRegressive (AR) with aggregation operators (2) Logistic smooth transition AR (3) Self-exciting threshold AR (4) Naïve | |
13 | Yu and Liong (2007) | 2 | Daily | 1 | (1) Support vector machines coupled with decomposition (2) Standard chaos technique (3) Naïve (4) Inverse approach (5) ARIMA | |
14 | Koutsoyiannis et al. (2008) | 1 × 12 (months) | Yearly | 1 | (1) Stochastic (2) Analogue method (3) Neural networks | |
15 | Wang et al. (2015) | 3 | Monthly | 12 | SARIMA |
In a somehow different direction, Papacharalampous et al. (2017c) conduct a multiple-case study, i.e. a synthesis of 50 single-case studies, by using monthly precipitation and temperature time series of various lengths observed in Greece. Some important points regarding the comparison of univariate time series forecasting methods and additional concerns introduced when implementing the machine learning ones (hyperparameter optimization and lagged variable selection) in one- and multi-step ahead forecasting are illustrated in the latter study. Nevertheless, only large-scale forecast-producing studies could provide empirical solutions to several problems appearing in the field of (geophysical) time series forecasting. Such studies are rare in the literature. Beyond geoscience, Makridakis and Hibon (2000) use a real-world dataset composed by 3003 time series, mainly originating from the business, industry, macroeconomic and microeconomic sectors, to assess the one- and multi-step ahead forecasting accuracy of 24 univariate time series forecasting methods. In geoscience, on the other hand, there are only four recent studies, all companions of the present, to be subsequently discussed.
Papacharalampous et al. (2017a) compare 11 stochastic and nine machine learning univariate time series forecasting methods in multi-step ahead forecasting of geophysical processes and (empirically) prove that stochastic and machine learning methods can perform equally well. The comparisons are conducted using 24,000 simulated time series of 110 values, 24,000 simulated time series of 310 values and 92 mean monthly time series of streamflow with varying lengths, as well as 18 metrics. These 20 methods are also found to collectively compose a representative sample set, i.e. exhibiting a variety of forecasting performances with respect to the different metrics. Alongside with this study, Papacharalampous et al. (2017b) investigate the error evolution in multi-step ahead forecasting when adopting this specific set of methods. The tests are performed on 6000 simulated time series of 150 values, 6000 simulated time series of 350 values and the streamflow dataset used in Papacharalampous et al. (2017a). Some different behaviours are revealed within these experiments, also suggesting the fact that one- and multi-step ahead forecasting are different problems to be examined for the same methods. Moreover, Tyralis and Papacharalampous (2017) focus on random forests, a well-known machine learning algorithm, with the aim to improve its one-step ahead forecasting performance by conducting experiments on 16,000 simulated and 135 annual temperature time series of 101 values. Finally, Papacharalampous et al. (2018) investigate the multi-step ahead predictability of monthly precipitation and temperature by applying seven automatic univariate time series forecasting methods to a sample of 1552 monthly precipitation and 985 monthly temperature time series of 480 values.
Herein, we examine the fundamental problem of one-step ahead forecasting, also complementing the results of the four above-mentioned studies. In more detail, we expand the former of these studies by exploring the one-step ahead forecasting properties of its methods, when applied to geophysical time series. Emphasis is put on the examination of two real-world datasets, a precipitation dataset and a temperature dataset, together containing 297 annual time series of 91 values. These datasets are examined in both their original and standardized forms. We further perform experiments using 24,000 simulated time series of 91 values. These experiments complement the real-world ones by allowing the examination of a large variety of process behaviours, while they are also controlled to some extent, facilitating generalizations and increasing the understanding on the examined problem. The number of forecasts produced using these real-world and simulated datasets are 47,520 and 2,130,000, respectively, i.e. the largest among its companion studies. Our aim is twofold, to provide generalized results regarding one-step ahead forecasting within a purely statistical framework [justified, for example, in Hyndman and Athanasopoulos (2013)] in geoscience and hopefully to establish the results obtained by the examination of the standardized real-world datasets as rough benchmarks for the one-step ahead predictability of annual precipitation and temperature. The establishment of forecasting benchmarks is meaningful, especially for the one-step ahead attempts, as the latter constitute the most simple ones and their accuracy can be quantified using a single metric, i.e. the absolute error.
Data and methods
Datasets of this study (part 1): real-world datasets
s/n | Abbreviated name | Process | Type | Primal dataset | R algorithm | Number of time series |
---|---|---|---|---|---|---|
1 | PrecDat | Precipitation | Original | Peterson and Vose (1997) | 112 | |
2 | TempDat | Temperature | Lawrimore et al. (2011) | 185 | ||
3 | StandPrecDat | Precipitation | Standardized | PrecDat | mleHK {HKprocess} | 112 |
4 | StandTempDat | Temperature | TempDat | 185 |
Datasets of this study (part 2): simulated datasets
s/n | Abbreviated name | Process | Parameter(s) | R algorithm | Number of time series |
---|---|---|---|---|---|
5 | SimDat_1 | AR(1) | φ_{1} = 0.7 | arima.sim {stats} | 2000 |
6 | SimDat_2 | AR(1) | φ_{1} = −0.7 | ||
7 | SimDat_3 | AR(2) | φ_{1} = 0.7, φ_{2} = 0.2 | ||
8 | SimDat_4 | MA(1) | θ_{1} = 0.7 | ||
9 | SimDat_5 | MA(1) | θ_{1} = −0.7 | ||
10 | SimDat_6 | ARMA(1,1) | φ_{1} = 0.7, θ_{1} = 0.7 | ||
11 | SimDat_7 | ARMA(1,1) | φ_{1} = −0.7, θ_{1} = −0.7 | ||
12 | SimDat_8 | ARFIMA(0,0.30,0) | fracdiff.sim {fracdiff} | ||
13 | SimDat_9 | ARFIMA(1,0.30,0) | φ_{1} = 0.7 | ||
14 | SimDat_10 | ARFIMA(0,0.30,1) | θ_{1} = −0.7 | ||
15 | SimDat_11 | ARFIMA(1,0.30,1) | φ_{1} = 0.7, θ_{1} = −0.7 | ||
16 | SimDat_12 | ARFIMA(2,0.30,2) | φ_{1} = 0.7, φ_{2} = 0.2, θ_{1} = −0.7, θ_{2} = −0.2 |
Figure 1 also presents the histograms of the Hurst parameter maximum likelihood estimates (Tyralis and Koutsoyiannis 2011) of the formed real-world time series. These estimates are of importance within this study for two reasons: (1) we implement a univariate time series forecasting method (see later on in this section) that takes advantage of this information under the established assumption of long-range dependence, (2) we standardize the original real-world time series using the mean and standard deviation maximum likelihood estimates (estimated simultaneously with the Hurst parameter) of the Hurst–Kolmogorov process. The standard deviation estimates would be considerably different if we modelled the time series using independent normal variables (Tyralis and Koutsoyiannis 2011). For consistency purposes with respect to the real-world datasets of the present study (but also to approximate the typical length of annual geophysical time series), the simulated time series are of 91 values as well. They originate from the families of AutoRegressive Moving Average (ARMA(p,q)) and AutoRegressive Fractionally Integrated Moving Average (ARFIMA(p,d,q)), the definitions of which can easily be found in the literature, for example in Wei (2006), pp 6–65, 489–494. The simulations are performed with mean 0 and standard deviation of 1. Hereafter, to specify a used R algorithm, we state its name accompanied by the name of the R package, denoted with {}. All algorithms are used with predefined values, unless specified differently.
Univariate time series forecasting methods of this study (part 1): stochastic methods
s/n | Abbreviated name | Category | R algorithm(s) | Implementation notes |
---|---|---|---|---|
1 | Naïve | Simple | ||
2 | RW | rwf {forecast} | drift = TRUE | |
3 | ARIMA_f | AutoRegressive Integrated Moving Average (ARIMA) | Arima {forecast}, forecast {forecast} | Arima {forecast}: include.mean = TRUE, include.drift = FALSE, method = ”ML” |
4 | ARIMA_s | Arima {forecast}, simulate {stats} | ||
5 | auto_ARIMA_f | auto.arima {forecast}, forecast {forecast} | ||
6 | auto_ARIMA_s | auto.arima {forecast}, simulate {stats} | ||
7 | auto_ARFIMA | AutoRegressive Fractionally Integrated Moving Average (ARFIMA) | arfima {forecast}, forecast {forecast} | arfima {forecast}: estim = ”mle” |
8 | BATS | State space | bats {forecast}, forecast {forecast} | |
9 | ETS_s | ets {forecast}, simulate {stats} | ||
10 | SES | Exponential smoothing | ses {forecast} | |
11 | Theta | thetaf {forecast} |
Univariate time series forecasting methods of this study (part 2): machine learning methods
s/n | Abbreviated name | Category | Model structure information | R algorithm(s) | Implementation notes | |
---|---|---|---|---|---|---|
Hyperparameter optimized using grid search (grid values) | Lagged variable selection procedure (see Table 6) | |||||
12 | NN_1 | Neural networks | Single hidden layer multilayer perceptron | CasesSeries {rminer}, fit {rminer}, lforecast {rminer}, nnet {nnet} | Number of hidden nodes (0, 1, …, 15) | 1 |
13 | NN_2 | 2 | ||||
14 | NN_3 | nnetar {forecast} | 3 | |||
15 | RF_1 | Random forests | Breiman’s random forests algorithm with 500 grown trees | CasesSeries {rminer}, fit {rminer}, lforecast {rminer}, randomForest {randomForest} | Number of variables randomly sampled as candidates at each split (1, …, 5) | 1 |
16 | RF_2 | 2 | ||||
17 | RF_3 | 3 | ||||
18 | SVM_1 | Support vector machines | Radial basis kernel “Gaussian” function, C = 1, epsilon = 0.1 | CasesSeries {rminer}, fit {rminer}, lforecast {rminer}, ksvm {kernlab} | Sigma inverse kernel width (2^{ n }, n = − 8, − 7, …, 6) | 1 |
19 | SVM_2 | 2 | ||||
20 | SVM_3 | 3 |
Lagged variable selection procedures adopted for the machine learning methods of Table 5
s/n | Time lags | R algorithm |
---|---|---|
1 | The corresponding to an estimated value for the AutoCorrelation Function (ACF) | acf {stats} |
2 | The corresponding to a statistical important estimated value for the ACF. If there is no statistical important estimated value for the ACF, the corresponding to the largest estimated value | acf {stats} |
3 | According to nnetar {forecast}, i.e. the time lags 1, …, n, where n is the number of AutoRegressive (AR) parameters that are fitted to the time series data | ar {stats} |
Error metrics and accuracy statistics of this study
s/n | Abbreviated name | Full name | Category | Values | Optimum value |
---|---|---|---|---|---|
1 | E | Error | Error metrics | (−∞, +∞) | 0 |
2 | AE | Absolute error | [0, +∞) | 0 | |
3 | PE | Percentage error | (−∞, +∞) | 0 | |
4 | APE | Absolute percentage error | [0, +∞) | 0 | |
5 | MdoAE | Median of the absolute errors | Accuracy statistics | [0, +∞) | 0 |
6 | MdoAPE | Median of the absolute percentage errors | [0, +∞) | 0 | |
7 | LRC | Linear regression coefficient | (−∞, +∞) | 1 | |
8 | R2 | Coefficient of determination | [0, 1] | 1 |
Experiments of this study
s/n | Abbreviated name | Category | Dataset (see Table 3) | Metrics (see Table 7) | |
---|---|---|---|---|---|
1 | RWE_1i | Real-world | PrecDat | 1, 2, 7–20 | 1–8 |
2 | RWE_2i | TempDat | |||
3 | RWE_3i | StandPrecDat | 1, 2, 7–20 | 1, 2, 5, 7, 8 | |
4 | RWE_4i | StandTempDat | |||
5 | SE_1i | Simulation | SimDat_1 | 1–6, 8–20 | 1, 2, 5, 7, 8 |
6 | SE_2i | SimDat_2 | |||
7 | SE_3i | SimDat_3 | |||
8 | SE_4i | SimDat_4 | |||
9 | SE_5i | SimDat_5 | |||
10 | SE_6i | SimDat_6 | |||
11 | SE_7i | SimDat_7 | |||
12 | SE_8i | SimDat_8 | 1, 2, 7–20 | ||
13 | SE_9i | SimDat_9 | |||
14 | SE_10i | SimDat_10 | |||
15 | SE_11i | SimDat_11 | |||
16 | SE_12i | SimDat_12 |
Part of the time series used within each experiment according to the i value
s/n | i | Data points of each time series used for the model-fitting (required for all models) and model-validation (required for the machine learning models) | Data points of each time series used for model-testing |
---|---|---|---|
1 | a | 1, 2, 3, …, 50 | 51 |
2 | b | 1, 2, 3, …, 60 | 61 |
3 | c | 1, 2, 3, …, 70 | 71 |
4 | d | 1, 2, 3, …, 80 | 81 |
5 | e | 1, 2, 3, …, 90 | 91 |
The only assumption of our methodological approach concerns the application of the auto_ARFIMA method within the real-world experiments and is that the annual precipitation and temperature variables can be sufficiently modelled by the normal distribution. This assumption is rather reasonable (implied by the Central Limit Theorem; Koutsoyiannis 2008, chapter 2.5.6) and could hardly harm the results. In general, such fundamental assumptions are preferable to the introduction of extra parameters, e.g. to using the Box-Cox transformation to normalize the data. The rest of the methods are non-parametric and, thus, not affected by the possible non-normality. To take advantage of some well-known theoretical properties, in the SE_1i–SE_7i simulation experiments the ARIMA_f and ARIMA_s methods are given the same AutoRegressive (AR) and Moving Average (MA) orders used in the respective simulation process, while d is set 0. These two methods, as well as the simple, auto_ARIMA_f, auto_ARIMA_s and auto_ARFIMA methods serve as reference points within our approach. In particular, ARIMA_f, auto_ARIMA_f and auto_ARFIMA are theoretically expected to be the most accurate within our simulation experiments [for an explanation see Papacharalampous et al. (2017a), chapter 2], while BATS is also expected to perform well in these experiments, since it comprises an ARMA model. In summary, the experiments are controlled to some extent, while their components (datasets, methods and metrics) are selected to provide a multifaceted approach to the problem of one-step ahead forecasting in geoscience.
Results and discussion
In this section, we summarize the basic quantitative and qualitative information gained from the experiments of the present study, while the total amount is available in the Additional files 1, 2, 3, 4, 5, 6 and 7. We further discuss the findings and explicate their contribution in light of the literature.
Experiments using the precipitation datasets
Minimum, maximum and mean values of the MdoAE within the experiments using the PrecDat dataset
Minimum (mm) | Maximum (mm) | Mean (mm) | |
---|---|---|---|
RWE_1a | 111 (RF_1) | 172 (NN_1) | 135 |
RWE_1b | 68 (SVM_1) | 146 (ETS_s) | 91 |
RWE_1c | 91 (SVM_3) | 171 (ETS_s) | 119 |
RWE_1d | 143 (BATS) | 189 (RF_2) | 162 |
RWE_1e | 98 (Theta) | 150 (NN_1) | 122 |
Minimum, maximum and mean values of the MdoAPE within the experiments using the PrecDat dataset
Minimum | Maximum | Mean | |
---|---|---|---|
RWE_1a | 0.12 (RF_1) | 0.21 (RW) | 0.16 |
RWE_1b | 0.09 (SVM_1) | 0.18 (ETS_s) | 0.12 |
RWE_1c | 0.12 (SVM_3) | 0.21 (NN_1) | 0.15 |
RWE_1d | 0.15 (BATS) | 0.22 (NN_1) | 0.17 |
RWE_1e | 0.12 (Theta) | 0.18 (NN_1) | 0.15 |
Minimum, maximum and mean values of the MdoAE within the experiments using the StandPrecDat dataset
Minimum | Maximum | Mean | |
---|---|---|---|
RWE_3a | 0.70 (RF_1) | 1.22 (NN_1) | 0.92 |
RWE_3b | 0.55 (SVM_2) | 0.95 (ETS_s) | 0.69 |
RWE_3c | 0.72 (BATS) | 1.42 (NN_1) | 0.86 |
RWE_3d | 0.99 (Theta) | 1.42 (ETS_s) | 1.14 |
RWE_3e | 0.69 (Theta) | 1.07 (ETS_s) | 0.89 |
Experiments using the temperature datasets
Minimum, maximum and mean values of the MdoAE within the experiments using the TempDat dataset
Minimum (°C) | Maximum (°C) | Mean (°C) | |
---|---|---|---|
RWE_2a | 0.42 (NN_3) | 0.72 (NN_1) | 0.51 |
RWE_2b | 0.23 (Theta) | 0.54 (NN_1) | 0.32 |
RWE_2c | 0.38 (BATS) | 0.66 (RW) | 0.47 |
RWE_2d | 0.78 (RW) | 1.10 (NN_3) | 1.01 |
RWE_2e | 0.38 (Theta) | 0.62 (ETS_s) | 0.46 |
Minimum, maximum and mean values of the MdoAPE within the experiments using the TempDat dataset
Minimum | Maximum | Mean | |
---|---|---|---|
RWE_2a | 0.04 (Theta) | 0.06 (NN_1) | 0.04 |
RWE_2b | 0.02 (auto_ARFIMA) | 0.05 (ETS_s) | 0.03 |
RWE_2c | 0.03 (SVM_1) | 0.06 (RW) | 0.04 |
RWE_2d | 0.07 (Naïve) | 0.08 (NN_1) | 0.08 |
RWE_2e | 0.03 (RF_1) | 0.05 (NN_1) | 0.04 |
Minimum, maximum and mean values of the MdoAE within the experiments using the StandTempDat dataset
Minimum | Maximum | Mean | |
---|---|---|---|
RWE_4a | 0.61 (BATS) | 0.93 (ETS_s) | 0.71 |
RWE_4b | 0.33 (Theta) | 0.73 (NN_1) | 0.47 |
RWE_4c | 0.56 (SES) | 0.96 (ETS_s) | 0.69 |
RWE_4d | 1.20 (NN_1) | 1.46 (Theta) | 1.36 |
RWE_4e | 0.48 (Theta) | 0.82 (ETS_s) | 0.61 |
Experiments using the simulated datasets
- (1)
The E values are approximately symmetric around 0 (mean value of the simulations).
- (2)
The results may vary significantly across the simulation experiments using different simulated datasets and across the different time series within a specific experiment depending on the forecasting method.
- (3)
Consequently, the relative performance of the forecasting methods may also vary significantly across the simulation experiments using different simulated datasets.
- (4)
On the contrary, the relative performance of the forecasting methods is slightly affected by the length of the time series for the experiments of the present study. The same has been found to mostly apply to the multi-step ahead forecasting performance of the same methods in Papacharalampous et al. (2017a) for two other time series lengths.
- (5)
Some forecasting methods are more accurate than others. The best-performing methods are ARIMA_f, auto_ARIMA_f, auto_ARFIMA, BATS, SES and Theta. This good performance of the former four methods when applied to ARMA and ARFIMA processes is expected from theory, while the Theta forecasting method has also performed well in the M3-Competition (Makridakis and Hibon 2000) and is expected to have a similar performance with SES (Hyndman and Billah 2003). The five above-mentioned forecasting methods are all stochastic.
- (6)
All the machine learning methods except for NN_1 (mostly NN_3 and SVM_3) are comparable to the best-performing methods, as it has also been found to apply in the experiments of Papacharalampous et al. (2017a, b). Likewise, in Tyralis and Papacharalampous (2017), random forests are competitive with the ARFIMA and Theta benchmarks.
- (7)
The simple methods are competitive in specific simulation experiments, as suggested for specific cases in Cheng et al. (2017), Makridakis and Hibon (2000) and Papacharalampous et al. (2017a) as well. Nevertheless, they also stand out because of their bad performance in other simulation experiments.
- (8)
Most of the far outliers are produced by neural networks.
Minimum, maximum and mean values of the MdoAE within the simulation experiments
Minimum | Maximum | Mean | |
---|---|---|---|
SE_1i | 0.68 (ARIMA_f | SE_1a) | 1.05 (NN_1 | SE_1a) | 0.80 |
SE_2i | 0.67 (ARIMA_f | SE_2c) | 1.82 (RW | SE_2e) | 0.95 |
SE_3i | 0.65 (ARIMA_f | SE_3c) | 1.04 (NN_1 | SE_3a) | 0.81 |
SE_4i | 0.67 (ARIMA_f | SE_4c) | 1.21 (ETS_s | SE_4a) | 0.84 |
SE_5i | 0.66 (ARIMA_f | SE_5e) | 1.48 (RW | SE_5c) | 0.90 |
SE_6i | 0.68 (ARIMA_f | SE_6b) | 1.20 (ETS_s | SE_6d) | 0.89 |
SE_7i | 0.66 (auto_ARIMA_f | SE_7d) | 2.91 (RW | SE_7e) | 1.22 |
SE_8i | 0.67 (auto_ARFIMA | SE_8c) | 1.02 (NN_1 | SE_8a) | 0.77 |
SE_9i | 0.67 (auto_ARFIMA | SE_9d) | 1.05 (NN_1 | SE_9b) | 0.80 |
SE_10i | 0.67 (auto_ARFIMA | SE_10e) | 1.22 (RW | SE_10e) | 0.83 |
SE_11i | 0.68 (Theta | SE_11e) | 1.10 (NN_1 | SE_11a) | 0.77 |
SE_12i | 0.69 (auto_ARFIMA | SE_12b) | 1.06 (NN_1 | SE_12a) | 0.78 |
Minimum, maximum and mean values of the MdoAE for each method within the simulation experiments
Minimum | Maximum | Mean | |
---|---|---|---|
Naïve | 0.68 (SE_3c) | 2.88 (SE_7a) | 1.12 |
RW | 0.69 (SE_9c) | 2.91 (SE_7e) | 1.13 |
ARIMA_f | 0.65 (SE_3c) | 0.72 (SE_7a) | 0.69 |
ARIMA_s | 0.91 (SE_2a) | 1.04 (SE_3a) | 0.96 |
auto_ARIMA_f | 0.66 (SE_7d) | 0.75 (SE_6c) | 0.70 |
auto_ARIMA_s | 0.91 (SE_4c) | 1.02 (SE_3d) | 0.97 |
auto_ARFIMA | 0.67 (SE_10e) | 0.73 (SE_10d) | 0.69 |
BATS | 0.67 (SE_3c) | 0.76 (SE_6c) | 0.71 |
ETS_s | 0.93 (SE_3d) | 2.11 (SE_7e) | 1.14 |
SES | 0.66 (SE_3c) | 1.52 (SE_7e) | 0.83 |
Theta | 0.66 (SE_3c) | 1.57 (SE_7a) | 0.84 |
NN_1 | 0.90 (SE_7e) | 1.16 (SE_7a) | 1.01 |
NN_2 | 0.72 (SE_8c) | 0.89 (SE_5b) | 0.79 |
NN_3 | 0.69 (SE_8c) | 0.84 (SE_6c) | 0.74 |
RF_1 | 0.71 (SE_8c) | 1.08 (SE_6a) | 0.82 |
RF_2 | 0.72 (SE_8c) | 1.04 (SE_6c) | 0.83 |
RF_3 | 0.72 (SE_3c) | 0.98 (SE_6c) | 0.80 |
SVM_1 | 0.71 (SE_8e) | 1.23 (SE_7a) | 0.86 |
SVM_2 | 0.68 (SE_8c) | 1.01 (SE_7a) | 0.81 |
SVM_3 | 0.68 (SE_8c) | 0.92 (SE_6c) | 0.76 |
Conclusions
The simulation experiments reveal the most and least accurate methods for long-term one-step ahead forecasting applications, also suggesting that the simple methods may be competitive in specific cases. Furthermore, the relative performance of the forecasting methods is slightly affected by the time series length for the simulation experiments of this study (using time series of 51, 61, 71, 81, 91 values), while it strongly depends on the process. Also importantly, the experiments using the original real-world time series result to minimum and maximum medians of the absolute errors of 68 and 189 mm for precipitation, and 0.23 and 1.10 °C for temperature respectively. Additionally, the experiments using the standardized real-world time series suggest that the minimum and maximum medians of the absolute errors are 0.55 and 1.42 for precipitation, and 0.33 and 1.46 for temperature respectively. These latter numerical results could be used as a rough upper boundary for the one-step ahead predictability of annual precipitation and temperature.
We subsequently state the limitations of this study and some future directions. The provided empirical solution to the problem of one-step ahead forecasting in geoscience is rather qualitative than quantitative, while the experiments using standardized precipitation and temperature data have offered rough benchmarks only. In the future more real-world data could be used to develop improved benchmarks for assessing the respective predictabilities. It would be of interest to further investigate how these predictabilities depend on the location from which the data originate. In this case, more stations spanning around the globe would be required. Moreover, a direct and large-scale comparison, set on a common base (if this is feasible), between deterministic and statistical approaches to forecasting geophysical processes would be useful and interesting. Another limitation of this study is related to the adopted modelling approach, i.e. the data-driven one, according to which the selection of the model does not depend on the properties of the examined process and, therefore, the latter are mostly not investigated. Furthermore, the improvement of the performance of the machine learning models requires extensive comparisons between different procedures of hyperparameter optimization and lagged variable selection. Finally, future research could focus on the examination of the respective predictabilities, when using exogenous predictor variables as well, while a definitely worth-stated future direction is related to the adoption of probabilistic forecasting methods, instead of the point forecasting ones.
Notes
Authors’ contributions
GP and HT worked on the analyses equally and to all of their aspects. GP, HT and DK discussed the results and contributed in the writing of the main manuscript. All authors read and approved the final manuscript.
Acknowledgements
We thank the Editor Bellie Sivakumar and two anonymous reviewers, whose comments have substantially improved the quality of this paper.
The analyses and visualizations have been performed in R Programming Language (R Core Team 2017) by using the contributed R packages forecast (Hyndman and Khandakar 2008, Hyndman et al. 2017), fracdiff (Fraley et al. 2012), gdata (Warnes et al. 2017), ggplot2 (Wickham 2016), HKprocess (Tyralis 2016), kernlab (Karatzoglou et al. 2004), knitr (Xie 2014, 2015, 2017), nnet (Venables and Ripley 2002), randomForest (Liaw and Wiener 2002), readr (Wickham et al. 2017) and rminer (Cortez 2010, 2016).
We acknowledge the Asia Oceania Geoscience Society (AOGS) for providing the publication cost. A preliminary research by Papacharalampous et al. (2017d) was presented in the 14th AOGS Annual Meeting.
Competing interests
The authors declare that they have no competing interests.
Availability of data and materials
This is a fully reproducible research paper; all the codes and data, as well as their outcome results, are available in the Additional files (Papacharalampous and Tyralis 2018). The sources of the real-world datasets are Lawrimore et al. (2011) and Peterson and Vose (1997).
Funding
This research has not received any funding.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary material
References
- Armstrong JS, Fildes R (2006) Making progress in forecasting. Int J Forecast 22(3):433–441. https://doi.org/10.1016/j.ijforecast.2006.04.007 CrossRefGoogle Scholar
- Assimakopoulos V, Nikolopoulos K (2000) The theta model: a decomposition approach to forecasting. Int J Forecast 16(4):521–530. https://doi.org/10.1016/S0169-2070(00)00066-2 CrossRefGoogle Scholar
- Babu CN, Reddy BE (2012) Predictive data mining on average global temperature using variants of ARIMA models. In: Proceeding of 2012 international conference on advances in engineering, science and management (ICAESM)Google Scholar
- Ballini R, Soares S, Andrade MG (2001) Multi-step-ahead monthly streamflow forecasting by a neurofuzzy network model. In: IFSA World Congress and 20th NAFIPS International Conference, pp 992–997. https://doi.org/10.1109/nafips.2001.944740
- Chau KW, Wu CL (2010) A hybrid model coupled with singular spectrum analysis for daily rainfall prediction. J Hydroinform 12(4):458–473. https://doi.org/10.2166/hydro.2010.032 CrossRefGoogle Scholar
- Chawsheen TA, Broom M (2017) Seasonal time-series modeling and forecasting of monthly mean temperature for decision making in the Kurdistan Region of Iraq. J Stat Theory Pract 11(4):604–633. https://doi.org/10.1080/15598608.2017.1292484 CrossRefGoogle Scholar
- Chen XY, Chau KW, Busari AO (2015) A comparative study of population-based optimization algorithms for downstream river flow forecasting by a hybrid neural network model. Eng Appl Artif Intell 46(Part A):258–268. https://doi.org/10.1016/j.engappai.2015.09.010 CrossRefGoogle Scholar
- Cheng KS, Lien YT, Wu YC, Su YF (2017) On the criteria of model performance evaluation for real-time flood forecasting. Stoch Environ Res Risk Assess 31(5):1123–1146. https://doi.org/10.1007/s00477-016-1322-7 CrossRefGoogle Scholar
- Cortez P (2010) Data mining with neural networks and support vector machines using the R/rminer tool. In: Perner P (ed) Advances in data mining. Applications and theoretical aspects. Springer, Heidelberg, pp 572–583. https://doi.org/10.1007/978-3-642-14400-4_44 CrossRefGoogle Scholar
- Cortez P (2016) rminer: data mining classification and regression methods. R package version 1.4.2. https://CRAN.R-project.org/package=rminer
- De Gooijer JG, Hyndman RJ (2006) 25 years of time series forecasting. Int J Forecast 22(3):443–473. https://doi.org/10.1016/j.ijforecast.2006.01.001 CrossRefGoogle Scholar
- De Livera AM, Hyndman RJ, Snyder RS (2011) Forecasting time series with complex seasonal patterns using exponential smoothing. J Am Stat Assoc 106(496):1513–1527. https://doi.org/10.1198/jasa.2011.tm09771 CrossRefGoogle Scholar
- Fildes R, Kourentzes N (2011) Validation and forecasting accuracy in models of climate change. Int J Forecast 27(4):968–995. https://doi.org/10.1016/j.ijforecast.2011.03.008 CrossRefGoogle Scholar
- Fraley C, Leisch F, Maechler M, Reisen V, Lemonte A (2012) fracdiff: fractionally differenced ARIMA aka ARFIMA(p,d,q) models. R package version 1.4-2. https://CRAN.R-project.org/package=fracdiff
- Gholami V, Chau KW, Fadaee F, Torkaman J, Ghaffari A (2015) Modeling of groundwater level fluctuations using dendrochronology in alluvial aquifers. J Hydrol 529(Part 3):1060–1069. https://doi.org/10.1016/j.jhydrol.2015.09.028 CrossRefGoogle Scholar
- Giunta G, Salerno R, Ceppi A, Ercolani G, Mancini M (2015) Benchmark analysis of forecasted seasonal temperature over different climatic areas. Geosci Lett 2:9. https://doi.org/10.1186/s40562-015-0026-z CrossRefGoogle Scholar
- Green KC, Armstrong JS (2007) Global warming: forecasts by scientists versus scientific forecasts. Energy Environ 18(7):997–1021. https://doi.org/10.1260/095830507782616887 CrossRefGoogle Scholar
- Green KC, Armstrong JS, Soon W (2009) Validity of climate change forecasting for public policy decision making. Int J Forecast 25(4):826–832. https://doi.org/10.1016/j.ijforecast.2009.05.011 CrossRefGoogle Scholar
- Hong WC (2008) Rainfall forecasting by technological machine learning models. Appl Math Comput 200(1):41–57. https://doi.org/10.1016/j.amc.2007.10.046 Google Scholar
- Htike KK, Khalifa OO (2010) Rainfall forecasting models using focused time-delay neural networks. In: Proceeding of 2010 international conference on computer and communication engineering (ICCCE). https://doi.org/10.1109/iccce.2010.5556806
- Hyndman RJ, Athanasopoulos G (2013) Forecasting: principles and practice. OTexts: Melbourne, Australia. http://otexts.org/fpp/
- Hyndman RJ, Billah B (2003) Unmasking the Theta method. Int J Forecasting 19(2):287–290. https://doi.org/10.1016/S0169-2070(01)00143-1 CrossRefGoogle Scholar
- Hyndman RJ, Khandakar Y (2008) Automatic time series forecasting: the forecast package for R. J Stat Softw 27(3):1–22. https://doi.org/10.18637/jss.v027.i03 CrossRefGoogle Scholar
- Hyndman RJ, O’Hara-Wild M, Bergmeir C, Razbash S, Wang E (2017) forecast: forecasting functions for time series and linear models. R package version 8.2. https://CRAN.R-project.org/package=forecast
- Karatzoglou A, Smola A, Hornik K, Zeileis A (2004) kernlab—an S4 package for kernel methods in R. J Stat Softw 11(9):1–20CrossRefGoogle Scholar
- Keenlyside NS (2011) Commentary on “Validation and forecasting accuracy in models of climate change”. Int J Forecast 27(4):1000–1003. https://doi.org/10.1016/j.ijforecast.2011.07.002 CrossRefGoogle Scholar
- Komorník J, Komorníková M, Mesiar R, Szökeová D, Szolgay J (2006) Comparison of forecasting performance of nonlinear models of hydrological time series. Phys Chem Earth Parts A/B/C 31(18):1127–1145. https://doi.org/10.1016/j.pce.2006.05.006 CrossRefGoogle Scholar
- Koutsoyiannis D (2008) Probability and statistics for geophysical processes. National Technical University of Athens, Athens. https://doi.org/10.13140/RG.2.1.2300.1849/1 Google Scholar
- Koutsoyiannis D, Yao H, Georgakakos A (2008) Medium-range flow prediction for the Nile: a comparison of stochastic and deterministic methods. Hydrol Sci J 53(1):142–164. https://doi.org/10.1623/hysj.53.1.142 CrossRefGoogle Scholar
- Lambrakis N, Andreou AS, Polydoropoulos P, Georgopoulos E, Bountis T (2000) Nonlinear analysis and forecasting of a brackish karstic spring. Water Resour Res 36(4):875–884. https://doi.org/10.1029/1999WR900353 CrossRefGoogle Scholar
- Lawrimore JH, Menne MJ, Gleason BE, Williams CN, Wuertz DB, Vose RS, Rennie J (2011) An overview of the Global Historical Climatology Network monthly mean temperature data set, version 3. J Geophys Res. https://doi.org/10.1029/2011JD016187 Google Scholar
- Liaw A, Wiener M (2002) Classification and regression by randomForest. R News 2(3):18–22Google Scholar
- Makridakis S, Hibon M (2000) The M3-competition: results, conclusions and implications. Int J Forecast 16(4):451–476. https://doi.org/10.1016/S0169-2070(00)00057-1 CrossRefGoogle Scholar
- McSharry PE (2011) Validation and forecasting accuracy in models of climate change: comments. Int J Forecast 27(4):996–999. https://doi.org/10.1016/j.ijforecast.2011.07.003 CrossRefGoogle Scholar
- Narayanan P, Basistha A, Sarkar S, Kamna S (2013) Trend analysis and ARIMA modelling of pre-monsoon rainfall data for western India. C R Geosci 345(1):22–27. https://doi.org/10.1016/j.crte.2012.12.001 CrossRefGoogle Scholar
- Papacharalampous GA, Tyralis H (2018) One-step ahead forecasting of geophysical processes within a purely statistical framework: supplementary material. figshare. https://doi.org/10.6084/m9.figshare.5357359.v1
- Papacharalampous GA, Tyralis H, Koutsoyiannis D (2017a) Comparison of stochastic and machine learning methods for the multi-step ahead forecasting of hydrological processes. Preprints. https://doi.org/10.20944/preprints201710.0133.v1 Google Scholar
- Papacharalampous GA, Tyralis H, Koutsoyiannis D (2017b) Error evolution in multi-step ahead streamflow forecasting for the operation of hydropower reservoirs. Preprints. https://doi.org/10.20944/preprints201710.0129.v1 Google Scholar
- Papacharalampous GA, Tyralis H, Koutsoyiannis D (2017c) Forecasting of geophysical processes using stochastic and machine learning algorithms. Eur Water 59:161–168Google Scholar
- Papacharalampous GA, Tyralis H, Koutsoyiannis D (2017d) Large scale simulation experiments for the assessment of one-step ahead forecasting properties of stochastic and machine learning point estimation methods. Asia Oceania Geosciences Society (AOGS) 14th Annual Meeting, Singapore. http://www.itia.ntua.gr/en/docinfo/1719/
- Papacharalampous GA, Tyralis H, Koutsoyiannis D (2018) Predictability of monthly temperature and precipitation using automatic time series forecasting methods. Acta Geophys. https://doi.org/10.1007/s11600-018-0120-7 Google Scholar
- Peterson TC, Vose RS (1997) An Overview of the Global Historical Climatology Network temperature database. B Am Meteorol Soc. 78:2837–2849. https://doi.org/10.1175/1520-0477(1997)078<2837:AOOTGH>2.0.CO;2
- R Core Team (2017) R: a language and environment for statistical computing. R Foundation for Statistical Computing, ViennaGoogle Scholar
- Remesan R, Mathew J (2015) Hydrological data driven modelling. Springer International Publishing, New York. https://doi.org/10.1007/978-3-319-09235-5 CrossRefGoogle Scholar
- Sivakumar B (2017) Chaos in hydrology: bridging determinism and stochasticity. Springer, New York. https://doi.org/10.1007/978-90-481-2552-4 CrossRefGoogle Scholar
- Taormina R, Chau KW (2015) Data-driven input variable selection for rainfall–runoff modeling using binary-coded particle swarm optimization and extreme learning machines. J Hydrol 529(Part 3):1617–1632. https://doi.org/10.1016/j.jhydrol.2015.08.022 CrossRefGoogle Scholar
- Tyralis H (2016) HKprocess: Hurst–Kolmogorov process. R package version 0.0-2. https://CRAN.R-project.org/package=HKprocess
- Tyralis H, Koutsoyiannis D (2011) Simultaneous estimation of the parameters of the Hurst–Kolmogorov stochastic process. Stoch Environ Res Risk Assess 25(1):21–33. https://doi.org/10.1007/s00477-010-0408-x CrossRefGoogle Scholar
- Tyralis H, Koutsoyiannis D (2014) A Bayesian statistical model for deriving the predictive distribution of hydroclimatic variables. Clim Dyn 42(11–12):2867–2883. https://doi.org/10.1007/s00382-013-1804-y CrossRefGoogle Scholar
- Tyralis H, Koutsoyiannis D (2017) On the prediction of persistent processes using the output of deterministic models. Hydrol Sci J 62(13):2083–2102CrossRefGoogle Scholar
- Tyralis H, Papacharalampous G (2017) Variable selection in time series forecasting using random forests. Algorithms 10(4):114. https://doi.org/10.3390/a10040114 CrossRefGoogle Scholar
- Venables WN, Ripley BD (2002) Modern applied statistics with S, 4th edn. Springer-Verlag, New York. https://doi.org/10.1007/978-0-387-21706-2 CrossRefGoogle Scholar
- Wang S, Feng J, Liu G (2013) Application of seasonal time series model in the precipitation forecast. Math Comput Model 58(3–4):677–683. https://doi.org/10.1016/j.mcm.2011.10.034 CrossRefGoogle Scholar
- Wang W, Chau K, Xu D, Chen XY (2015) Improving forecasting accuracy of annual runoff time series using ARIMA based on EEMD decomposition. Water Resour Manag 29(8):2655–2675. https://doi.org/10.1007/s11269-015-0962-6 CrossRefGoogle Scholar
- Warnes GR, Bolker B, Gorjanc G, Grothendieck G, Korosec A, Lumley T, MacQueen D, Magnusson A, Rogers J et al (2017) gdata: various R programming tools for data manipulation. R package version 2.18.0. https://CRAN.R-project.org/package=gdata
- Wei WWS (2006) Time series analysis, univariate and multivariate methods, 2nd edn. Pearson Addison Wesley, BostonGoogle Scholar
- Wickham H (2016) ggplot2: elegant graphics for data analysis, 2nd edn. Springer International Publishing, Cham. https://doi.org/10.1007/978-3-319-24277-4 CrossRefGoogle Scholar
- Wickham H, Hester J, Francois R, Jylänki J, Jørgensen M (2017) readr: read rectangular text data. R package version 1.1.1. https://CRAN.R-project.org/package=readr
- Wu CL, Chau KW, Fan C (2010) Prediction of rainfall time series using modular artificial neural networks coupled with data-preprocessing techniques. J Hydrol 389(1–2):146–167. https://doi.org/10.1016/j.jhydrol.2010.05.040 CrossRefGoogle Scholar
- Xie Y (2014) knitr: a comprehensive tool for reproducible research in R. In: Stodden V, Leisch F, Peng RD (eds) Implementing reproducible computational research. Chapman and Hall/CRC, Boca RatonGoogle Scholar
- Xie Y (2015) Dynamic documents with R and knitr, 2nd edn. Chapman and Hall/CRC, Boca RatonGoogle Scholar
- Xie Y (2017) knitr: a general-purpose package for dynamic report generation in R. R package version 1.17. https://CRAN.R-project.org/package=knitr
- Yu X, Liong SY (2007) Forecasting of hydrologic time series with ridge regression in feature space. J Hydrol 332(3–4):290–302. https://doi.org/10.1016/j.jhydrol.2006.07.003 CrossRefGoogle Scholar
- Yu X, Liong SY, Babovic V (2004) EC-SVM approach for real-time hydrologic forecasting. J Hydroinform 6(3):209–223Google Scholar
Copyright information
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.