Statistical Downscaling Of Precipitation Research Articles

Statistical downscaling of precipitation in northwestern Iran using a hybrid model of discrete wavelet transform, artificial neural networks, and quantile mapping

AbstractDownscaling of daily precipitation from Global Circulation Models (GCMs)-predictors at a station level, especially in arid and semi-arid regions, has remained a formidable challenge yet. The current study aims at proposing a coupled model of Discrete Wavelet Transform (DWT), Artificial Neural Networks (ANNs), and Quantile Mapping (QM) for statistical downscaling of daily precipitation. Given the historic (1978–2005) and future (2006–2100) predictors of eight-selected GCMs under Representative Concentration Pathways (RCPs) 2.6, 4.5, and 8.5, a viable DWT-ANN model was developed for each station. Subsequently, we linked QM to DWT-ANN for bias correction and drizzle effect postprocessing of the DWT-ANN-historic/future projected precipitation. The skill of DWT-ANN-QM was demonstrated using various evaluation metrics, including Taylor diagram, Quantile–Quantile plot, Empirical Cumulative Distribution Function, and wet/dry spell analysis. We appraise the efficacy of the coupled model at 12 weather stations over the Gharehsoo River Basin (GRB) in northwestern Iran. Compared to the observed wet/dry spells, the dry-spells were better simulated via DWT-ANN-QM rather than the wet-spells wherein length and exceedance probability of the spells were overestimated. Results indicated that the future precipitation across the GRB will rise, on average, from 10 to 17% depending on weather station. Seasonal spatial distribution of the middle future (2041–2070) precipitation illustrated that an increase for fall and winter, especially, is expected, whereas the amount of the future spring and summer precipitation is projected to be declined. Having been developed and tested in a semi-arid basin, the efficacy of the coupled model should be further assessed in more humid climates.

Statistical Downscaling of Precipitation in the South and Southeast of Mexico

The advancements in global climate modeling achieved within the CMIP6 framework have led to notable enhancements in model performance, particularly with regard to spatial resolution. However, the persistent requirement for refined techniques, such as dynamically or statistically downscaled methods, remains evident, particularly in the context of precipitation variability. This study centered on the systematic application of a bias-correction technique (quantile mapping) to four designated CMIP6 models: CNRM-ESM2-6A, IPSL-CM6A-LR, MIROC6, and MRI-ESM2-0. The selection of these models was informed by a methodical approach grounded in previous research conducted within the southern–southeastern region of Mexico. Diverse performance evaluation metrics were employed, including root-mean-square difference (rmsd), normalized standard deviation (NSD), bias, and Pearson’s correlation (illustrated by Taylor diagrams). The study area was divided into two distinct domains: southern Mexico and the southeast region covering Tabasco and Chiapas, and the Yucatan Peninsula. The findings underscored the substantial improvement in model performance achieved through bias correction across the entire study area. The outcomes of rmsd and NSD not only exhibited variations among different climate models but also manifested sensitivity to the specific geographical region under examination. In the southern region, CNRM-ESM2-1 emerged as the most adept model following bias correction. In the southeastern domain, including only Tabasco and Chiapas, the optimal model was again CNRM-ESM2-1 after bias-correction. However, for the Yucatan Peninsula, the IPSL-CM6A-LR model yielded the most favorable results. This study emphasizes the significance of tailored bias-correction techniques in refining the performance of climate models and highlights the spatially nuanced responses of different models within the study area’s distinct geographical regions.

On the modern deep learning approaches for precipitation downscaling

Deep Learning (DL) based downscaling has become a popular tool in earth sciences recently. Increasingly, different DL approaches are being adopted to downscale coarser precipitation data and generate more accurate and reliable estimates at local (~few km or even smaller) scales. Despite several studies adopting dynamical or statistical downscaling of precipitation, the accuracy is limited by the availability of ground truth. A key challenge to gauge the accuracy of such methods is to compare the downscaled data to point-scale observations which are often unavailable at such small scales. In this work, we carry out the DL-based downscaling to estimate the local precipitation data from the India Meteorological Department (IMD), which was created by approximating the value from station location to a grid point. To test the efficacy of different DL approaches, we apply four different methods of downscaling and evaluate their performance. The considered approaches are (i) Deep Statistical Downscaling (DeepSD), augmented Convolutional Long Short Term Memory (ConvLSTM), fully convolutional network (U-NET), and Super-Resolution Generative Adversarial Network (SR-GAN). A custom VGG network, used in the SR-GAN, is developed in this work using precipitation data. The results indicate that SR-GAN is the best method for precipitation data downscaling. The downscaled data is validated with precipitation values at IMD station. This DL method offers a promising alternative to statistical downscaling.

Wavelet-based predictor screening for statistical downscaling of precipitation and temperature using the artificial neural network method

Abstract One of the challenging issues in statistical downscaling of climate models is to select dominant large-scale climate variables (predictors). Correlation-based methods have been revealed to be efficacious to select the predictors; however, traditional correlation analysis has shown limited ability due to the nonstationary and nonlinear nature of climatic time series. Hence, in this study, Wavelet Coherence Transform (WTC) was employed to assess the high common powers and the multi-scale correlation between two time series (i.e., predictand and predictor) as a function of time and frequency. To this end, a coefficient correlation (CC) and a wavelet-based method were used for predictor screening and the results were compared in statistical downscaling. To apply the wavelet-based method, Continuous Wavelet Transform (CWT) was utilized to identify the potent periodicity in the time series of predictands. WTC was applied to determine the coherence between predictors and predictands in the potent periodicities, and Scale Average (SA) wavelet coherency was applied to rank them. In order to implement statistical downscaling, the ANN model was developed. In this study, three climate models including BNU-ESM Can-ESM5, and INM-CM5 have been used. The projection of the future climate based on the ANN downscaling revealed that precipitation will undergo a 7.1–28.92% downward trend, while the temperature will experience a 2.25–4.21 °C increase.

Statistical downscaling of precipitation using inclusive multiple modelling (IMM) at two levels

Abstract Topical research on hydrological behaviour of climate change in terms of downscaling of monthly precipitation is investigated in this paper by formulating an inclusive multiple modelling (IMM) strategy. IMM strategies manage multiple models at two levels and the paper uses statistical downscaling model, Sugeno fuzzy logic and support vector machine at Level 1 and feeds their outputs to a neuro-fuzzy model at Level 2. In the downscaling stage, large-scale NCEP (National Centres for Environmental Prediction)/NCAR (National Centre for Atmospheric Research) data are used for a station with local data record from 1961 to 2005 for training and testing Level 1 models. The results are found to be ‘fit-for-purpose’, but the variations between them signify some room for improvements. The model at Level 2 combines outputs of those at Level 1 and produces Level 2 results, which improve compared with those at the Level 1 models in terms of dispersion of residual errors. In this way, IMM provides a more defensible modelling strategy for application in the projection stage. The comparison between observed and projected precipitation indicates that precipitation will be likely to reduce compared with observed precipitation in cold seasons (October–February), but the projected precipitation will be likely to increase slightly in wet seasons (April and May).

An Uncertainty-Based Regional Comparative Analysis on the Performance of Different Bias Correction Methods in Statistical Downscaling of Precipitation

Statistical downscaling of General Circulation Models (GCM) simulations is widely used for projecting precipitation at different spatiotemporal scales. However, the downscaling process is linked with different source of uncertainty including structural/parametric uncertainty of the model and output uncertainty. This research proposes a novel framework to assess the parametric uncertainty of downscaling model, and used this framework to assess the performance of different bias correction methods linked to the regression-based statistical downscaling model. The used downscaling framework in the current paper is Statistical Downscaling Model (SDSM). The conventional bias correction method linked with SDSM is the Variance InFlation method (VIF), this paper substitutes this method with three different bias correction methods including Local Intensity Scaling (LOCI), Power Transformation (PT), and Quantile Mapping (QM) to assess the associated parametric and global uncertainty of each method in different climate by using a new approach. The proposed method is applied to six different stations located in Iran and United States with different climate status. Average Relative Interval Length (ARIL), P-level, and Normalized Uncertainty Efficiency (NUE) are used as uncertainty indicators to evaluate the results. Results represent that in every assessed climate class, LOCI, and PT, work better than conventional VIF in both amount and occurrence modules of SDSM framework. More precisely, LOCI works better in station that has wet summer, while PT performs well in the stations where there is no or very limited precipitation in summer. Substituting LOCI with VIF, result in increasing the value of NUE by at least factor of 3 in occurrence and amount model which means the significant reduction in structural uncertainty. Also applying PT in arid regions improves the NUE indicator at least by factor 2 in occurrence and amount model and by factor 3 in output uncertainty assessment, and results in less parametric and output uncertainty. Results illustrate the important role of bias correction approaches in reducing structural, and output uncertainty and improving the statistical efficiency of the downscaling model.

Challenges and potential solutions in statistical downscaling of precipitation

Downscaling is an effective technique to bridge the gap between climate model outputs and data requirements of most crop and hydrologic models for assessing local and site-specific climate change impacts, especially on future food security. However, downscaling of temporal sequences, extremes in daily precipitation, and handling of nonstationary precipitation in future conditions are considered common challenges for most statistical downscaling methods. This study reviewed the above key challenges in statistical downscaling and proposed potential solutions. Ten weather stations located across the globe were used as proof of concept. The use of a stochastic Markov chain to generate daily precipitation occurrences is an effective approach to simulate the temporal sequence of precipitation. Also, the downscaling of precipitation extremes can be achieved by adjusting the skewness coefficient of a probability distribution, as they are highly correlated. Nonstationarity in precipitation downscaling can be handled by adjusting parameters of a probability distribution according to future precipitation change signals projected by climate models. The perspectives proposed in this study are of great significance in using climate model outputs for assessing local and site-specific climate change impacts, especially on future food security.

Multi-site statistical downscaling of precipitation using generalized hierarchical linear models: a case study of the imperilled Lake Urmia basin

ABSTRACT A downscaling model capable of explaining the temporal and spatial variability of regional hydroclimatic variables is essential for reliable climate change studies and impact assessments. This study proposes a novel statistical approach based on generalized hierarchical linear model (GHLM) to downscale precipitation from the outputs of general circulation models (GCMs) at multiple sites. GHLM partitions the total variance of precipitation into within- and between-site variability allowing for transferring information between sites to develop a regional downscaling model. The methodology is demonstrated by downscaling precipitation using the outputs of eight GCMs in Lake Urmia basin in northwestern Iran. Multi-model ensemble simulations are merged and bias-corrected using Bayesian model averaging and equidistant quantile mapping, respectively. The results of this study show projected declining trends in precipitation resulting in approximately 11.2% and 15.3% decrease during 2060–2080 compared to the historical period of 1985–2005 considering representative concentration pathways (RCPs) 4.5 and 8.5, respectively.

Pros and cons of using wavelets in conjunction with genetic programming and generalised linear models in statistical downscaling of precipitation

Among the regression techniques used in building statistical downscaling models, genetic programming (GP) which mimics Darwin’s theory of biological evolution possesses several pros such as it evolves explicit linear or non-linear relationships while identifying optimum predictors, and it discards irrelevant and redundant information in predictors. However, GP is known to simulate unphysically large outliers of predictands. In statistical downscaling, decomposition of predictand and predictor data into number of different time-frequency components with wavelet transform and modelling each component separately should better simulate the time-frequency properties of the predictand, in theory. Therefore, it is important to investigate pros and cons of using GP with wavelet transform in building downscaling models. In this study, wavelet and non-wavelet-based precipitation downscaling models were developed employing GP and generalised liner models (GLM) for 50 stations located in wet and dry climate, with 20CR and NCEP/NCAR reanalysis data, for the investigation of the above matter. It was found that regardless of the mother wavelet, vanishing moment and climate regime, with the increase in decomposition level, the wavelet-based downscaling models developed with GLM tended to show a distinct deterioration in performance in both calibration and validation unlike the wavelet-based downscaling models developed with GP. This was because GP is able to discard unnecessary/redundant information flowing into the model with the increase in the decomposition level through evolution. Furthermore, it was found that when GP is coupled with wavelet transform, the simulation of unphysically large values of the predictand increases significantly.

Statistical Downscaling of Precipitation and Temperature Using Gene Expression Programming

Statistical Downscaling of Precipitation and Temperature Using Gene Expression Programming

Statistical downscaling of precipitation using machine learning techniques

Statistical models were developed for downscaling reanalysis data to monthly precipitation at 48 observation stations scattered across the Australian State of Victoria belonging to wet, intermediate and dry climate regimes. Downscaling models were calibrated over the period 1950–1991 and validated over the period 1992–2014 for each calendar month, for each station, using 4 machine learning techniques, (1) Genetic Programming (GP), (2) Artificial Neural Networks (ANNs), (3) Support Vector Machine (SVM), and (4) Relevance Vector Machine (RVM). It was found that, irrespective of the climate regime and the machine learning technique, downscaling models tend to better simulate the average (compared to other statistics) and under-estimate the standard deviation and the maximum of the observed precipitation. Also, irrespective of the climate regime and the machine learning technique, at the majority of stations downscaling models showed an over-estimating trend of low to mid percentiles (i.e. below the 50th percentile) of precipitation and under-estimating trend of high percentiles of precipitation (i.e. above the 90th percentile). The over-estimating trend of low to mid percentiles of precipitation was more pronounced at stations located in dryer climate, irrespective of the machine learning technique. Based on the results of this investigation the use of RVM or ANN over SVM or GP for developing downscaling models can be recommended for a study such as flood prediction which involves the consideration of high extremes of precipitation. Also, RVM can be recommended over GP, ANN or SVM in developing downscaling models for a study such as drought analysis which involves the consideration of low extremes of precipitation. Furthermore, it was found that irrespective of the climate regime, the SVM and RVM-based precipitation downscaling models showed the best performance with the Polynomial kernel.

Quantification of Uncertainty Propagation Effects during Statistical Downscaling of Precipitation and Temperature to Hydrological Modeling

To understand the water balance and other environmental impacts under climate change condition, hydrological models are used to simulate the hydrological cycle and predict future scenarios by using general circulation models (GCMs) outputs. Due to the mismatch of the spatial resolution, different downscaling techniques are usually applied to GCMs outputs to generate higher resolution data for the use with the hydrological models. It is known that there are many uncertainties with hydrological models which lead to inaccuracy and unreliability of the predictions. The uncertainty associated with climate change has been described as irreducible and persistent, and downscaling GCM outputs using downscaling methods also lead to considerable uncertainties. The purpose of this study is to propose a method to quantify the propagation effects of uncertainties from statistical downscaling to hydrological modeling. A case study has been provided in this study to demonstrate the feasibility of the proposed method. Statistical downscaling model (SDSM) was applied to downscale H3A2a (A2 emission scenario in Hadley Centre Coupled Model 3) outputs, and the downscaled results were used as inputs to a distributed hydrological model - the soil and water assessment tool (SWAT). The surface runoff prediction has been made for 2016 ~ 2020 by using downscaled precipitation and temperature. The uncertainty associated with statistical downscaling has been quantified through the evaluation of surface runoff simulation from the application of the hydrological modeling study.

Statistical downscaling of precipitation using long short-term memory recurrent neural networks

Hydrological impacts of global climate change on regional scale are generally assessed by downscaling large-scale climatic variables, simulated by General Circulation Models (GCMs), to regional, small-scale hydrometeorological variables like precipitation, temperature, etc. In this study, we propose a new statistical downscaling model based on Recurrent Neural Network with Long Short-Term Memory which captures the spatio-temporal dependencies in local rainfall. The previous studies have used several other methods such as linear regression, quantile regression, kernel regression, beta regression, and artificial neural networks. Deep neural networks and recurrent neural networks have been shown to be highly promising in modeling complex and highly non-linear relationships between input and output variables in different domains and hence we investigated their performance in the task of statistical downscaling. We have tested this model on two datasets—one on precipitation in Mahanadi basin in India and the second on precipitation in Campbell River basin in Canada. Our autoencoder coupled long short-term memory recurrent neural network model performs the best compared to other existing methods on both the datasets with respect to temporal cross-correlation, mean squared error, and capturing the extremes.

Statistical Downscaling of Precipitation and Temperature Using Long Ashton Research Station Weather Generator in Zambia: A Case of Mount Makulu Agriculture Research Station

The Long Ashton Research Station Weather Generator (LARS-WG) is a stochastic weather generator used for the simulation of weather data at a single site under both current and future climate conditions using General Circulation Models (GCM). It was calibrated using the baseline (1981-2010) and evaluated to determine its suitability in generating synthetic weather data for 2020 and 2055 according to the projections of HadCM3 and BCCR-BCM2 GCMs under SRB1 and SRA1B scenarios at Mount Makulu (Latitude: 15.550°S, Longitude: 28.250°E, Elevation: 1213 meter), Zambia. Three weather parameters—precipitation, minimum and maximum temperature were simulated using LARS-WG v5.5 for observed station and AgMERRA reanalysis data for Mount Makulu. Monthly means and variances of observed and generated daily precipitation, maximum temperature and minimum temperature were used to evaluate the suitability of LARS-WG. Other climatic conditions such as wet and dry spells, seasonal frost and heat spells distributions were also used to assess the performance of the model. The results showed that these variables were modeled with good accuracy and LARS-WG could be used with high confidence to reproduce the current and future climate scenarios. Mount Makulu did not experience any seasonal frost. The average temperatures for the baseline (Observed station data: 1981-2010 and AgMERRA reanalysis: 1981-2010) were 21.33°C and 22.21°C, respectively. Using the observed station data, the average temperature under SRB1 (2020), SRA1B (2020), SRB1 (2055), SRA1B (2055) would be 21.90°C, 21.94°C, 22.83°C and 23.18°C, respectively. Under the AgMERRA reanalysis, the average temperatures would be 22.75°C (SRB1: 2020), 22.80°C (SRA1B: 2020), 23.69°C (SRB1: 2055) and 24.05°C (SRA1B: 2055). The HadCM3 and BCM2 GCMs ensemble mean showed that the number of days with precipitation would increase while the mean precipitation amount in 2020s and 2050s under SRA1B would reduce by 6.19% to 6.65%. Precipitation would increase under SRB1 (Observed), SRA1B, and SRB1 (AgMERRA) from 0.31% to 5.2% in 2020s and 2055s, respectively.

Quantifying the sources of uncertainty in upper air climate variables

AbstractFuture estimates of precipitation and streamflow are of utmost interest in hydrological climate change impact assessments. Just as important as the estimate itself, is the variance around the ensemble mean of the projections, this variance being defined as uncertainty in the context of this study. This uncertainty in the hydrological variables of interest is affected by uncertainty in upper air climate variables which are used in statistical downscaling of precipitation or streamflow. Here the extent of uncertainty in upper air climate variables has been assessed for a selection of commonly used atmospheric variables for downscaling, namely, geopotential height and its difference in the north‐south direction, specific humidity, and eastward and northward wind speeds. Generally, in statistical downscaling, no consideration is usually given to the uncertainty of different individual variables, which can result in biases in future climate simulations. The approach of quantifying uncertainty presented here has the potential to enable modelers to better formulate downscaling approaches, leading to more accurate characterization of future precipitation and its associated uncertainty. Based on the spread of multiple‐model outputs, an uncertainty measure called square root of error variance has been used to quantify the contribution of different sources of uncertainty (i.e., models, scenarios, and ensembles) in monthly future climate projections in the 21st century at the 500 hPa and 850 hPa pressure levels. It has been shown that the different climate variables and levels of the atmosphere have distinct patterns in terms of their total future uncertainty and the contributions from the three sources. Scenario and model uncertainties in general contribute reasonably evenly to total uncertainty, with smaller contributions from the initial condition ensembles.

Statistical downscaling of precipitation using local regression and high accuracy surface modeling method

Downscaling precipitation is required in local scale climate impact studies. In this paper, a statistical downscaling scheme was presented with a combination of geographically weighted regression (GWR) model and a recently developed method, high accuracy surface modeling method (HASM). This proposed method was compared with another downscaling method using the Coupled Model Intercomparison Project Phase 5 (CMIP5) database and ground-based data from 732 stations across China for the period 1976–2005. The residual which was produced by GWR was modified by comparing different interpolators including HASM, Kriging, inverse distance weighted method (IDW), and Spline. The spatial downscaling from 1° to 1-km grids for period 1976–2005 and future scenarios was achieved by using the proposed downscaling method. The prediction accuracy was assessed at two separate validation sites throughout China and Jiangxi Province on both annual and seasonal scales, with the root mean square error (RMSE), mean relative error (MRE), and mean absolute error (MAE). The results indicate that the developed model in this study outperforms the method that builds transfer function using the gauge values. There is a large improvement in the results when using a residual correction with meteorological station observations. In comparison with other three classical interpolators, HASM shows better performance in modifying the residual produced by local regression method. The success of the developed technique lies in the effective use of the datasets and the modification process of the residual by using HASM. The results from the future climate scenarios show that precipitation exhibits overall increasing trend from T1 (2011–2040) to T2 (2041–2070) and T2 to T3 (2071–2100) in RCP2.6, RCP4.5, and RCP8.5 emission scenarios. The most significant increase occurs in RCP8.5 from T2 to T3, while the lowest increase is found in RCP2.6 from T2 to T3, increased by 47.11 and 2.12 mm, respectively.

Predictor Selection Associated With Statistical Downscaling of Precipitation over Zambia

Predictor Selection Associated With Statistical Downscaling of Precipitation over Zambia

Statistical downscaling of precipitation and temperature using sparse Bayesian learning, multiple linear regression and genetic programming frameworks

This study attempted to investigate two approaches to downscale temperature and four approaches to downscale precipitation. The first approach was an implementation of multiple linear regression (MLR) in the form of backward stepwise regression. The second approach applied canonical correlation analysis (CCA) with a sparse Bayesian learning (SBL) approach called relevance vector machine (RVM). For precipitation downscaling, two additional approaches which combined genetic programming (GP) as the predictor processing method with sparse Bayesian learning (SBLGP) and multiple linear regression (MLRGP) were also presented. The results showed that SBL outperformed MLR in downscaling temperature. For all stations, temperature was better downscaled than precipitation. For precipitation downscaling, the SBLGP approach outperformed all other approaches. MLRGP, on the other hand, did not bring about much improvement in the results and was in many cases outperformed by MLR.

Evaluación de modelos climáticos globales del CMIP5 sobre el noroeste de América del Sur

Objective: The purpose of this study was to validate and compare the performance of six Global Climate Models (GCMs) simulations from the 5th Coupled Model Intercomparison Project (CMIP5) in the area of northern South America. Methodology: The validation study is carried out for sea level pressure (SLP) and sea surface temperature (SST), because of these are two of the most important global variables in describing the climate of Colombia. So, both variables are susceptible be used as predictors in statistical downscaling of precipitation in Colombia. To this end we compare the mean and variance fields of SLP and SST in different models of the CMIP5 with those obtained from the NCEP reanalysis data for the period 1950-2005. Furthermore, we have compared the main modes of variability derived from a Princi- pal Component Analysis (PCA). Results: The results show how the models reproduce reasonably well the mean fields of SLP and SST, although some models, such as CCSM4, tend to show more zonally SLP patterns, strengthening subtropical highs. Conclusions: For the PCA, all models reproduce the main variability modes associated with ENSO and tropical Atlantic reasonably well, while many of them tend to overestimate the variance associated with the first mode.

Statistical downscaling of precipitation on a spatially dependent network using a regional climate model

We present a detailed downscaling simulation methodology for generating precipitation events, conditioned on external climate covariates, on a network of meteorological stations. These events can be input to hydrological models. To simulate an event on a future day t, the method uses the K-nearest neighbor algorithm to identify close neighbors from the historical record, and resamples a past event with resampling probabilities determined from external covariates. This preserves the spatial dependence of precipitation on the network, and other important distributional features of precipitation. Large numbers of arbitrary tuning parameters and model assumptions are reduced through the use of a multivariate Gaussian model relating climate covariates to historical precipitation. The approach is demonstrated by simulating daily precipitation, maximum temperature, and minimum temperature on a network of 93 locations in North Carolina, all conditioned on climate model output. The downscaling is based on a regional climate model (RCM) embedded within a global NCEP reanalysis model. The method is demonstrated using precipitation in North Carolina with the Canadian climate RCM as an NCEP driven RCM.