Statistical downscaling

Last updated Oct 28, 2022 Edit Source

See:

• AI/Computer Vision/Super-resolution
• AI4ES/EO
• AI/Computer Vision/Image-to-image translation
• AI4ES/Bias correction, adjustment

• https://en.wikipedia.org/wiki/Downscaling
• Downscaling is any procedure to infer high-resolution information from low-resolution variables. This technique is based on dynamical or statistical approaches commonly used in several disciplines, especially meteorology, climatology and remote sensing. The term downscaling usually refers to an increase in spatial resolution, but it is often also used for temporal resolution.
• Statistical downscaling or what climate can I expect in my own backyard? Statistical DS learns a functional mapping between low and high-resolution climate models from observed data (computationally efficient and scalable across multi-model ensembles) vs dynamical DS, where all the local processes are encoded, such as convective precipitation and vegetation schemes, with subgrid parameters and GCM boundary conditions for HR projections (high computational costs)
• “downscaling” is a climate modeling term while “downsampling” comes from signal processing. Confusingly, “downscaling” is actually equivalent to “upsampling”, both referring to “increasing resolution”
• CarbonPlan - Open data and tools for multiple methods of global climate downscaling

• #TALK What is bias correction/adjustment and statistical downscaling?
• #TALK Different methods for bias adjustment and downscaling
• #TALK Statistical Downscaling (South Central Climate Adaptation Science Center)
• #TALK Webinar: The Ins and Outs of Downscaling: Simple to Complex Techniques Explained Simply (2015)
• Santander group
  • https://meteo.unican.es/trac/estcena/wiki/Downscaling/MetodosDownscaling
  • #TALK Statistical downscaling with deep learning: A contribution to CORDEX-CORE
  • CORDEX
  • https://www.meteo.unican.es/files/posters/2019_Bano-Medina_CORDEX.pdf
  • https://github.com/SantanderMetGroup/DeepDownscaling/blob/master/2020_Bano_GMD_FULL.ipynb
  • https://github.com/SantanderMetGroup/DeepDownscaling/blob/master/2020_Bano_GMD.ipynb
  • https://github.com/SantanderMetGroup/downscaleR.keras/blob/master/R/prepareData.keras.R

• #CODE DL4DS - Deep Learning for empirical DownScaling
  • Python package with state-of-the-art and novel deep learning algorithms for empirical/statistical downscaling of gridded data
• #CODE PyESD
  • Python Package for Empirical Statistical Downscaling. This repository contains all scripts of the pyESD package which is under development. The purpose of the package is to downscale climate variables like precipitation and temperature from large-scale reanalysis datasets (eg. ERA5) to point scale
• #CODE Equidistant quantile matching (EDCDFm)
  • Bias correction and SD in python and R
• #CODE Scikit-downscale
  • https://scikit-downscale.readthedocs.io/en/latest/roadmap.html#models
  • https://github.com/earthcube2020/ec20_hamman_etal
  • https://figshare.com/articles/Scikit-downscale_an_open_source_Python_package_for_scalable_climate_downscaling/12506648/1
• #CODE Bias Correction Spatial Disaggregation
• #CODE XCLIM - Climate indices computations
  • https://xclim.readthedocs.io/en/stable/sdba_api.html
• #CODE DownscaleR package
• #CODE https://github.com/krishaizl/AnalogMethod/

• #BOOK Large-scale machine learning in the earth sciences (Srivastava, CRC 2017)
  • Chapter 4: Statistical Downscaling in Climate with State-of-the-Art Scalable Machine Learning
• #BOOK Statistical Downscaling and Bias Correction for Climate Research (Maraun 2018, CAMBRIDGE)

• #PAPER A comparison of statistical downscaling methods suited for wildfire applications (Abatzoglou 2012)
  • model presented: Bias corrected Spatial Downscaling (BCSD) and the Multivariate Adapted Constructed Analogs (MACA)
  • MACA
  • MACA is a statistical method for downscaling Global Climate Models (GCMs) from their native coarse resolution to a higher spatial resolution that captures reflects observed patterns of daily near-surface meteorology and simulated changes in GCMs experiments
  • https://github.com/EricKeenan/cmip6-downscalling/blob/master/notebooks/implement_MACA.ipynb
• #PAPER On the Application of Principal Component Analysis for Accurate Statistical-dynamical Downscaling of Wind Fields (Chavez-Arroyo 2013)
• #PAPER Statistical Downscaling in Climatology (Schoof 2013)
  • https://core.ac.uk/download/pdf/60580596.pdf
  • Downscaling is a term that has been used to describe the range of methods that are used to infer regional‐scale or local‐scale climate information from coarsely resolved climate models
  • This article provides an overview of statistical downscaling with a focus on assumptions, common predictors and predictands, and methodological approaches ranging from interpolation and scaling to regression‐based methods, weather pattern‐based techniques, and stochastic weather generators
• #PAPER VALUE : A framework to validate downscaling approaches for climate change studies (Maraun 2014)
  • VALUE is an open European network to validate and compare downscaling methods for climate change research
  • Systematic validation framework to enable the assessment and comparison of both dynamical and statistical downscaling methods
• #PAPER Spatial downscaling of precipitation using adaptable random forests (He 2016)
• #PAPER Statistical downscaling of precipitation using long short-term memory recurrent neural networks (Misra 2017) ^72531e
  • https://cse.iitkgp.ac.in/~pabitra/paper/taac17.pdf
  • Precipitation use case
  • Proposed a new statistical downscaling model based on Recurrent Neural Network with LSTMs which captures the spatio-temporal dependencies in local rainfall
• #PAPER Intercomparison of ML Methods for Statistical Downscaling: The Case of Daily and Extreme Precipitation (Vandal 2017)
  • Compared four fundamental statistical methods, Bias Correction Spatial Disaggregation (BCSD), Ordinary Least Squares, Elastic-Net, and Support Vector Machine, with three more advanced machine learning methods, Multi-task Sparse Structure Learning (MSSL), BCSD coupled with MSSL, and AI/Deep learning/CNNs to downscale daily precipitation in the Northeast United States
  • Metrics to evaluate of each method’s ability to capture daily anomalies, large scale climate shifts, and extremes are analyzed
  • Found that linear methods, led by BCSD, consistently outperform non-linear approaches. The direct application of state-of-the-art machine learning methods to statistical downscaling does not provide improvements over simpler, longstanding approaches (!!!???)
• #PAPER DeepSD: Generating High Resolution Climate Change Projections through Single Image Super-Resolution (Vandal 2017) ^c8c676
  • #CODE https://github.com/tjvandal/deepsd
  • #CODE https://github.com/tjvandal/srcnn-tensorflow
  • https://www.kdd.org/kdd2017/papers/view/deepsd-generating-high-resolution-climate-change-projections-through-single
  • #THESIS/PHD [Statistical downscaling of global climate models with image super-resolution and uncertainty quantification (Vandal 2018)]( https://www.semanticscholar.org/paper/Statistical-downscaling-of-global-climate-models-Vandal/a5dbe8d2af5b4f49b6f5ee89e5822d3d30c653a4 ^592e4e)
    • https://repository.library.northeastern.edu/files/neu:cj82st065/fulltext.pdf
  • #TALK Super-Resolution and Deep Learning for Climate Downscaling
  • Precipitation use case. PRISM dataset at a 4km daily spatial resolution which aggregates station observations to a grid with physical and topographical information
  • Upscaled the precipitation data to1/8º (12.5km) as our high-resolution observations. Then upscaled further to 1º corresponding to a low-resolution precipitation
  • SD is the problem of mapping a low-resolution climate variable to a high-resolution projection
  • Using the analogy between climate datasets and images, we can relate statistical downscaling to image super-resolution, where one aims to learn a mapping from low- to high-resolution image pairs
  • Local scale projections can be obtained using statistical downscaling, a technique which uses historical climate observations to learn a low-resolution to high-resolution mapping
  • Used static high-resolution topography (elevation from Global 30 Arc-Second Elevation DataSet (GTOPO30) provided by the USGS) data in conjunction with other low-resolution climate variables
  • DeepSD is a generalized stacked super resolution convolutional neural network (SRCNN) framework for statistical downscaling of climate variables
  • each SRCNN increases the resolution by a factor of s and is trained independently (associated input/output pairs)
  • DeepSD augments SRCNN with multi-scale input channels to maximize predictability in statistical downscaling
  • Future work: more variables such as temperature, wind, humidity at different pressure levels, downscaling multiple climate variables (temp, prlr), uncertainty
• #PAPER Coupling dynamical and statistical downscaling for high-resolution rainfall forecasting: case study of the Red River Delta, Vietnam (Anh 2018)
  • hybrid dynamical-statistical downscaling approach is an effort to combine the ability of dynamical downscaling to resolve fine-scale climate changes with the low computational cost of statistical downscaling
  • dynamical downscaling was performed with an RCM driven by the reanalysis to produce nested 30- and 6-km resolution simulations. Subsequently, the 6-km simulation was compared to rain gauge data to examine the ability of the RCM to reproduce known climate conditions. Then, in the statistical downscaling step, the ANN was trained to predict rainfall in the 6-km domain based on weather predictors in the 30-km simulation
  • a Simple multilayer perceptron ANN is used
• #PAPER Comparison of Different Machine Learning Approaches for Monthly Satellite-Based Soil Moisture Downscaling over Northeast China (Liu 2018)
• #PAPER Spatial Statistical Downscaling for Constructing High-Resolution Nature Runs in Global Observing System Simulation Experiments (Ma 2018)
  • Proposed a principled statistical downscaling framework to construct high-resolution Nature Runs (for Observing system simulation experiments - OSSEs) via conditional simulation from coarse-resolution numerical model output
  • Demonstrated these techniques by downscaling a coarse-resolution physical NR at a native resolution of 1∘ latitude×1.25∘ longitude of global surface CO2 concentrations to 655,362 equal-area hexagons
• #PAPER The impact of soil moisture on precipitation downscaling in the Euro-Mediterranean area (Hertig 2018)
• #PAPER DeepDownscale: a deep learning strategy for high-resolution weather forecast (Rodrigues 2018)
  • https://ieeexplore.ieee.org/document/8588749
  • DNN to learn a high-resolution representation from low-resolution predictions using weather forecast as a practical use case
  • Supervised learning approach, since obtaining labeled data can be done automatically
  • The input is a set of n low-resolution forecasts from n different weather models.
  • Each input is interpolated so that they have the same horizontal dimension
  • The input volume is fed into a series of convolutions which are initialized with an approximate delta function up to the last but one layer
  • The architecture is configurable and supports n−1 identity convolutions
  • Results show significant improvement when compared with standard practices and the strategy is still lightweight enough to run on modest computer systems
• #THESIS/MSC Exploring Machine Learning Models for Wind Speed Prediction (Prieto 2018)
• #PAPER Climatologies at high resolution for the earth’s land surface areas (Karger 2017)
• #PAPER Downscaling rainfall using deep learning long short‐term memory and feed-forward neural network (Tran Anh 2019)
  • Precipitation use case and climate projections
  • LSTMs and feedforward neural network methods, for precipitation downscaling for the Vietnamese Mekong Delta
  • Model performances were assessed for 2036–2065 period, using original climate projections from five climate models under the Coupled Model Intercomparison Project Phase 5, for two Representative Concentration Pathway scenarios (RCP 4.5 and RCP 8.5)
  • The results exhibited that there were good correlations between the modelled and observed values of the testing and validating periods at two long‐term meteorological stations
  • Were analyzed extreme indices of precipitation, including the annual maximum wet day frequency (Prcp), 95th percentile of precipitation (P95p), maximum 5‐day consecutive rain (R5d), total number of wet days (Ptot), wet day precipitation (SDII) and annual maximum dry day frequency (Pcdd) to evaluate changes in extreme precipitation events
• #PAPER Harmonized evaluation of daily precipitation downscaled using SDSM and WRF+WRFDA models over the Iberian Peninsula (Gonzalez-Roji 2019)
• #PAPER Configuration and Intercomparison of Deep Learning Neural Models for Statistical Downscaling (Baño-Medina 2019)
  • Deep Neural Networks for Statistical Downscaling of Climate Change Projections (Baño-Medina 2018)
  • Precipitation use case
  • The network includes a first block of three convolutional layers with 50, 25 and 10 (3×3×no. inputs) kernels, respectively, followed by two fully connected (dense) layers with 50 neurons each.The output is modeled through a mixed binomial–lognormal distribution, and the corresponding parameters are estimated by the network,obtaining precipitation as a final product, either deterministically (the expected value) or stochastically (generating a random value from the predicted distribution)
• #PAPER Geographically Weighted Machine Learning and Downscaling for High-Resolution Spatiotemporal Estimations of Wind Speed (Li 2019)
• #PAPER Optimized Artificial Neural Networks-Based Methods for Statistical Downscaling of Gridded Precipitation Data (Hossein Salimi 2019)
• #TALK Downscaling of Low-Resolution Wind Fields using Neural Networks (Kern 2019)
• #THESIS/MSC Development of Multi-Model Ensembles for Climate Downscaling in Ontario, Canada (Li 2019)
• #PAPER Prediction of Long-Term Near-Surface Temperature Based on NA-CORDEX Output (Li 2019)
• #PAPER Time-Dependent Downscaling of PM2.5 Predictions from CAMS Air Quality Models to Urban Monitoring Sites in Budapest (Varga-Balogh 2020)
  • 24-h PM2.5 forecasts obtained from seven individual models of the Copernicus Atmosphere Monitoring Service (CAMS) were downscaled by using hourly measurements at six urban monitoring sites in Budapest for the heating season of 2018–2019
• #PAPER Performance of statistical and machine learning ensembles for daily temperature downscaling (Li 2020)
• #TALK Towards Operational Downscaling of Low Resolution Wind Fields using Neural Networks (Kern 2020)
• #PAPER MeshfreeFlowNet: A Physics-Constrained Deep Continuous Space-Time Super-Resolution Framework (Jiang 2020)
• #PAPER Stochastic Super-Resolution for Downscaling Time-Evolving Atmospheric Fields with GANs (Leinonen 2020) ^20ca2a
  • #CODE https://github.com/jleinonen/downscaling-rnn-gan
  • Precipitation use case
  • Introduced a recurrent, stochastic super-resolution GAN that can generate ensembles of time-evolving high-resolution atmospheric fields for an input consisting of a low-resolution sequence of images of the same field
  • Used two datasets: radar-measured precipitation from Switzerland and cloud optical thickness derived from the Geostationary Earth Observing Satellite 16 (GOES-16)
  • The statistical properties of the generated ensemble are analyzed using rank statistics, a method adapted from ensemble weather forecasting. These analyses indicate that the GAN produces close to the correct amount of variability in its outputs
  • Model:
    • Uses residual blocks
    • Uses ConvGRU layers to learn the temporal evolution of the fields
    • As the GAN generator is fully convolutional, it can be applied after training to input images larger than the images used to train it
  • Compared to Lanczos interpolation, a recurrent CNN (RCNN) trained to optimize RMSE, and the Rain-fall Filtered Autoregressive Model (RainFARM) algorithm. The RCNN uses the same architecture as our GAN generator, except with the noise input disabled
• #PAPER Adversarial super-resolution of climatological wind and solar data (Stengel, 2020) ^7a7640
  • https://www.zotero.org/groups/2448424/ai4es/items/MC8HEHUK/file
  • #CODE https://github.com/NREL/PhIRE
  • Introduced an adversarial deep learning approach to super resolve wind velocity and solar irradiance outputs from global climate models to scales sufficient for renewable energy resource assessment
  • Sort of transfer learning, training on wind and solar data to downscale the CCSM GCM
  • Data:
    • Coarse GCM: NCAR CCSM at ~1deg from 2020 to 2039. To be downscaled (not used in training)
    • Coarsened HR (100m height) wind data, 2km or 0.02 deg resolution, 4-hourly from 2007 to 2013
    • Coarsened HR solar irradiance data, 0.04 deg resolution, 1-hourly from 6AM to 6PM from 2007 to 2013
    • Pairs of LR and HR data. Patches 1000 km2, each patch coarsened down to 10km or 0.1 deg resolution, and to 100km or 1 deg resolution. Similar thing for the solar data
    • 42k and 50k samples (square patches)
  • Two step model, LR -> MR and MR-> HR (e.g. 10x10 to 100x100 to 500x500 pixels)
  • Based on the SRGAN model
  • Demonstrate up to a 50x resolution enhancement of wind and solar data
  • https://renews.biz/61517/nrel-enhances-wind-velocity-data
  • https://www.machinelearningtarragona.com/2020/08/como-generar-rapidamente-informacion.html
• #PAPER An intercomparison of a large ensemble of statistical downscaling methods over Europe: Results from the VALUE perfect predictor cross‐validation experiment (Gutierrez 2020)
  • https://www.zotero.org/groups/2448424/ai4es/items/NB58VG6D/file
  • http://www.value-cost.eu/data
• #PAPER ClimAlign: Unsupervised statistical downscaling of climate variables via normalizing flows (Groenke 2020)
  • Unsupervised downscaling. Downscaling as domain alignment
  • Glow normalizing flow (Kingma 2018), Alignflow (Grover et al 2020).
  • Unpaired data samples, IID from each domain (fine, coarse resolutions)
  • Learn invertible transforms f, from simple prior to informative prior
• #PAPER CliGAN: A Structurally Sensitive Convolutional Neural Network Model for Statistical Downscaling of Precipitation from Multi-Model Ensembles (Chaudhuri 2020)
• #PAPER Deep-learning based down-scaling of summer monsoon rainfall data over Indian region (Kumar 2020)
  • Employed three deep learning-based algorithms derived from the super-resolution convolutional neural network (SRCNN) methods, to precipitation data, in particular, IMD and TRMM data to produce 4x-times high-resolution downscaled rainfall data during the summer monsoon season
  • Among the three algorithms, namely SRCNN, stacked SRCNN, and DeepSD, employed here, the best spatial distribution of rainfall amplitude and minimum root-mean-square error is produced by DeepSD based downscaling
  • Found that spatial discontinuity in amplitude and intensity rainfall patterns is the main obstacle in the downscaling of precipitation
• #PAPER Enabling Smart Dynamical Downscaling of Extreme Precipitation Events With Machine Learning (Shi 2020)
  • https://www.researchgate.net/publication/346390363
• #PAPER Statistical Downscaling of Temperature Distributions from the Synoptic Scale to the Mesoscale Using Deep Convolutional Neural Networks (Sekiyama 2020)
• #PAPER Statistical downscaling of daily temperature and precipitation over China using deep learning neural models: Localization and comparison with other methods (Sun 2020)
  • CNN compared to GLM and two quantile mapping based techniques including bias correction and spatial disaggregation (BCSD) and bias correction and climate imprint (BCCI)
• #PAPER Sequence-based statistical downscaling and its application to hydrologic simulations based on machine learning and big data (Wang 2020)
  • Time series characteristics of meteorological data was considered in statistical downscaling
  • RNN and ANN had different feasibilities in areas, while RNN could improve the performance
  • RNN-RandExtreme method could improve the accuracy of predicting extreme precipitation
• #PAPER PreciPatch: A Dictionary-based Precipitation Downscaling Method (Xu 2020)
• #PAPER Hyper-local, efficient extreme heat projection and analysis using machine learning to augment a hybrid dynamical-statistical downscaling technique (Madaus 2020)
• #PAPER Climate Downscaling Using YNet: A Deep Convolutional Network with Skip Connections and Fusion (Liu 2020)
  • CNN with skip connections and fusion capabilities to perform downscaling for climate variables, on multiple GCMs directly rather than on reanalysis data
  • Model: encoder-decoder-like architecture with residual learning through skip connections and fusion layers to enable the incorporation of topological and climatological data as auxiliary data
  • the model uses the full blown GCM simulations instead of training on observational/reanalysis data
  • post-upsampling using resize convolution, branch with auxiliary inputs concatenated in the end of the network, main branch encoder-decoder
  • compare the MSE of different methods for the three climate variables: monthly mean precipitation (ppt), monthly maximum temperature (tmax) and monthly minimum temperature (tmin) using three downscaling factors: 2, 4 and 8
• #PAPER Downscaling and Projection of Multi-CMIP5 Precipitation Using Machine Learning Methods in the Upper Han River Basin (Xu 2020)
• #PAPER Downscaling fire weather extremes from historical and projected climate models (Jain 2020)
• #PAPER Statistical downscaling with the downscaleR package (v3.1.0): contribution to the VALUE intercomparison experiment (Bedia 2020) ^downscaleR
  • #code DownscaleR
• #PAPER Understanding Deep Learning Decisions in Statistical Downscaling Models (Baño-Medina 2020)
• #PAPER RainNet v1.0: a convolutional neural network for radar-based precipitation nowcasting (Ayzel 2020)
  • #CODE https://github.com/hydrogo/rainnet
• #PAPER RainNet: A Large-Scale Dataset for Spatial Precipitation Downscaling (Chen 2020)
  • https://neuralchen.github.io/RainNet/
  • REAL (non-simulated) Large-Scale Spatial Precipitation Downscaling Dataset, RainNet, which contains 62,424 pairs of low-resolution and high-resolution precipitation maps for 17 years (2002 - 2018)
  • These data are collected from satellites, radars and gauge stations, which can reveal the multi-source characteristics of meteorological data
  • Models use L1 and perceptual loss (pre-trained VGG19 network)
  • Eight metrics specifically considering the physical property of the data set are raised, while fourteen models are evaluated on the proposed dataset
• #PAPER A comparative study of convolutional neural network models for wind field downscaling (Hohlein 2020)
  • CNNs for downscaling of short-range forecasts of near-surface winds on extended spatial domains
  • DeepRU, a novel U-Net-based CNN architecture, which is able to infer situation-dependent wind structures that cannot be reconstructed by other models
  • compare: Localized multi-linear regression model (LinearEnsemble), linear shallow CNN (no activation), deepSD (vandal, pre-upsampling), FSRCNN (post-upsampling, smaller kernels 3x3 and 1x1), EnhanceNet (residual connections) and deepRU (UNET-style network)
• #PAPER Downscaling and Evaluation of Seasonal Climate Data for the European Power Sector (Ostermoller 2021)
  • The global model, GCFS2.0, has limited skill of forecasting surface temperature and precipitation over Europe, which also applies for other seasonal prediction systems
  • An improvement of hindcast skill can in some cases be achieved by using multi-model ensembles of seasonal forecasts, depending on season and region
  • Applied the empirical-statistical downscaling method EPISODES, a perfect prognosis method based on large-scale predictors for the variables of interest
  • The predictors of temperature and precipitation were determined by cross-validation to produce low-biased output for the considered region
  • No relevant increase or decrease of hindcast skill in terms of ACC due to downscaling could be detected for the two case study regions
  • The skill remains about the same, but the seasonal data is now available at a higher spatial resolution (approx. 6 km)
  • ANAREG, analogs (by grid point, in a 5x5 window, 20 days in the past)
• #PAPER A perfect prognosis downscaling methodology for seasonal prediction of local-scale wind speeds (Ramon 2021)
  • SD with a perfect prognosis approach to produce seasonal predictions of near-surface wind speeds at the local scale
  • Hybrid predictions combine a dynamical prediction of the four main Euro-Atlantic Teleconnections (EATC) and a multilinear statistical regression (fitted with observations and EATC)
  • Comparison with hybrid preds. from ERA5 100-metre wind speed (observational reference)
  • SD techniques are either in:
    • perfect prognosis: model is fitted with observational data for both predictors and predictands
    • model output statistics: using data from a GCM
  • Selection of predictors is vital
• #PAPER Augmented Convolutional LSTMs for Generation of High-Resolution Climate Change Projections (Harilal 2021)
  • Pre-upsampling
  • LSTMs for spatio-temporal data
• #PAPER Deep Learning for Daily Precipitation and Temperature Downscaling (Wang 2021)
  • Super Resolution Deep Residual Network (SRDRN) for downscaling daily precipitation and temperature
  • transfer learning, the trained SRDRN model in one region was directly applied to downscale precipitation in another region with a different environment, and the results showed notable improvement compared to classic statistical downscaling method
  • The outstanding performance of the SRDRN approach stemmed from its ability to fully extract spatial features without suffering from degradation and overfitting issues due to the incorporations of residual blocks, batch normalizations, and data augmentations
• #PAPER Fast and accurate learned multiresolution dynamical downscaling for precipitation (Wang 2021)
  • precipitation: 1-year outputs from two RCM simulations using the Weather Research and Forecasting model, both at 50 km resolution and at 12 km resolution
  • supervised MSE and CGAN trainings
  • inception-style modules, CBAM attention modules, transposed convolution post-upsampling
  • validation analysis, extremes
• #PAPER Deconditional Downscaling with Gaussian Processes (Chau 2021)
  • #CODE https://github.com/shahineb/deconditional-downscaling
  • scalable Bayesian solution to the mediated statistical downscaling problem, which handles unmatched multi-resolution data. The proposed approach combines Gaussian Processes with the framework of deconditioning using RKHSs - reproducing kernel Hilbert space
• #PAPER On the suitability of deep convolutional neural networks for continental-wide downscaling of climate change projections (Baño-Medina 2021)
  • CNNs for downscaling future climate change projections using Global Climate Model (GCM) outputs as predictors
• #PAPER Spatio-Temporal Downscaling of Climate Data Using Convolutional and Error-Predicting Neural Networks (Serifi 2021)
• #PAPER Super-resolution and denoising of fluid flow using physics-informed convolutional neural networks without high-resolution labels (Gao 2021)
• #PAPER Convolutional conditional neural processes for local climate downscaling (Vaughan 2021) ^ccnp4ds
  • #CODE https://github.com/annavaughan/convCNPClimate
  • See AI/Deep learning/Neural processes
  • ConvCNPs are a recently developed class of models that allow deep learning techniques to be applied to off-the-grid spatio-temporal data
  • This model has a substantial advantage over existing downscaling methods in that the trained model can be used to generate multisite predictions at an arbitrary set of locations, regardless of the availability of training data
• #POSTER Intercomparison of DL techniques for empirical statistical downscaling over North America (Gonzalez-Abad 2022)
• #PAPER Downscaling Multi-Model Climate Projection Ensembles with Deep Learning (DeepESD): Contribution to CORDEX EUR-44 (Baño-Medina 2022)
  • See AI4ES/Ensembles, multi-models
  • temperature and precipitation over the European EUR-44i (0.5º) domain, based on eight GCMs from the CMIP5
• #PAPER DeepUrbanDownscale: A physics informed deep learning framework for high-resolution urban surface temperature estimation via 3D point clouds (Chen 2022)
  • DeepUrbanDownscale (DUD) for high-resolution urban surface temperature estimation - novel physics informed neural network (PINN) based framework
  • This network, ingesting the high-precision land surface geometry information derived from 3D point clouds and guided by the atmospheric physics related to surface temperature, constructs a physics informed data-driven framework to fit high-resolution temperature distribution