Climate and weather data such as precipitation derived from Global Climate Models (GCMs) and satellite observations are essential for the global and local hydrological assessment. However, most climatic popular precipitation products (with spatial resolutions coarser than 10km) are too coarse for local impact studies and require “downscaling” to obtain higher resolutions. Traditional precipitation downscaling methods such as statistical and dynamic downscaling require an input of additional meteorological variables, and very few are applicable for downscaling hourly precipitation for higher spatial resolution. Based on dynamic dictionary learning, we propose a new downscaling method, PreciPatch, to address this challenge by producing spatially distributed higher resolution precipitation fields with only precipitation input from GCMs at hourly temporal resolution and a large geographical extent. Using aggregated Integrated Multi-satellitE Retrievals for GPM (IMERG) data, an experiment was conducted to evaluate the performance of PreciPatch, in comparison with bicubic interpolation using RainFARM—a stochastic downscaling method, and DeepSD—a Super-Resolution Convolutional Neural Network (SRCNN) based downscaling method. PreciPatch demonstrates better performance than other methods for downscaling short-duration precipitation events (used historical data from 2014 to 2017 as the training set to estimate high-resolution hourly events in 2018).
more »
« less
Downscaling numerical weather models with conditional generative adversarial networks.
Abstract—Numerical simulation of weather is resolution-constrained due to the high computational cost of integrating the coupled PDEs that govern atmospheric motion. For example, the most highly-resolved numerical weather prediction models are limited to approximately 3 km. However many weather and climate impacts occur over much finer scales, especially in urban areas and regions with high topographic complexity like mountains or coastal regions. Thus several statistical methods have been developed in the climate community to downscale numerical model output to finer resolutions. This is conceptually similar to image super-resolution (SR) [1] and in this work we report the results of applying SR methods to the downscaling problem. In particular we test the extent to which a SR method based on a Generative Adversarial Network (GAN) can recover a grid of wind speed from an artificially downsampled version, compared against a standard bicubic upsampling approach and another machine learning based approach, SR-CNN [1]. We use ESRGAN ([2]) to learn to downscale wind speeds by a factor of 4 from a coarse grid. We find that we can recover spatial details with higher fidelity than bicubic upsampling or SR-CNN. The bicubic and SR-CNN methods perform better than ESRGAN on coarse metrics such as MSE. However, the high frequency power spectrum is captured remarkably well by the ESRGAN, virtually identical to the real data, while bicubic and SR-CNN fidelity drops significantly at high frequency. This indicates that SR is considerably better at matching the higher-order statistics of the dataset, consistent with the observation that the generated images are of superior visual quality compared with SR-CNN.
more »
« less
- Award ID(s):
- 1843103
- NSF-PAR ID:
- 10137368
- Date Published:
- Journal Name:
- CLI info
- ISSN:
- 1623-2666
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
more » « less
Abstract Cities need climate information to develop resilient infrastructure and for adaptation decisions. The information desired is at the order of magnitudes finer scales relative to what is typically available from climate analysis and future projections. Urban downscaling refers to developing such climate information at the city (order of 1 – 10 km) and neighborhood (order of 0.1 – 1 km) resolutions from coarser climate products. Developing these higher resolution (finer grid spacing) data needed for assessments typically covering multiyear climatology of past data and future projections is complex and computationally expensive for traditional physics-based dynamical models. In this study, we develop and adopt a novel approach for urban downscaling by generating a general-purpose operator using deep learning. This ‘DownScaleBench’ tool can aid the process of downscaling to any location. The DownScaleBench has been generalized for both in situ (ground- based) and satellite or reanalysis gridded data. The algorithm employs an iterative super-resolution convolutional neural network (Iterative SRCNN) over the city. We apply this for the development of a high-resolution gridded precipitation product (300 m) from a relatively coarse (10 km) satellite-based product (JAXA GsMAP). The high-resolution gridded precipitation datasets is compared against insitu observations for past heavy rain events over Austin, Texas, and shows marked improvement relative to the coarser datasets relative to cubic interpolation as a baseline. The creation of this Downscaling Bench has implications for generating high-resolution gridded urban meteorological datasets and aiding the planning process for climate-ready cities.
-
Downscaling coarse global and regional climate models allows researchers to access weather and climate data at finer temporal and spatial resolution, but there remains a need to compare these models with empirical data sources to assess model accuracy. Here, we validate a widely used software for generating North American downscaled climate data, ClimateNA, with a novel empirical data source, 20th century weather journals kept by Admiralty Island, Alaska homesteader, Allen Hasselborg. Using Hasselborg’s journals, we calculated monthly precipitation and monthly mean of the maximum daily air temperature across the years 1926 to 1954 and compared these to ClimateNA data generated from the Hasselborg homestead location and adjacent areas. To demonstrate the utility and potential implications of this validation for other disciplines such as hydrology, we used an established regression equation to generate time series of 95% low duration flow estimates for the month of August using mean annual precipitation from ClimateNA predictions and Hasselborg data. Across 279 months, we found strong correlation between modeled and observed measurements of monthly precipitation ( ρ = 0.74) and monthly mean of the maximum daily air temperature ( ρ = 0.98). Monthly precipitation residuals (calculated as ClimateNA data - Hasselborg data) generally demonstrated heteroscedasticity around zero, but a negative trend in residual values starting during the last decade of observations may have been due to a shift to the cold-phase Pacific Decadal Oscillation. Air temperature residuals demonstrated a consistent but small positive bias, with ClimateNA tending to overestimate air temperature relative to Hasselborg’s journals. The degree of correlation between weather patterns observed at the Hasselborg homestead site and ClimateNA data extracted from spatial grid cells across the region varied by wet and dry climate years. Monthly precipitation from both data sources tended to be more similar across a larger area during wet years (mean ρ across grid cells = 0.73) compared to dry years (mean ρ across grid cells = 0.65). The time series of annual 95% low duration flow estimates for the month of August generated using ClimateNA and Hasselborg data were moderately correlated ( ρ = 0.55). Our analysis supports previous research in other regions which also found ClimateNA to be a robust source for past climate data estimates.more » « less
-
Soil moisture is an important parameter that regulates multiple ecosystem processes and provides important information for environmental management and policy decision-making. Spaceborne sensors provide soil moisture information over large areas, but information is commonly available at coarse resolution with spatial and temporal gaps. Here, we present a modular spatial inference framework to downscale satellite-derived soil moisture using terrain parameters and test the performance of two modeling methods (Kernel-Weighted K-Nearest Neighbormore » « less
and Random Forest ). We generate monthly and weekly gap-free spatial predictions on soil moisture at 1 km using data from the European Space Agency Climate Change Initiative (ESA-CCI; version 6.1) over two regions in the conterminous United States. RF was the method that performed better in cross-validation when comparing with the reference ESA-CCI data, but KKNN showed a slightly higher agreement with ground-truth information as part of independent validation. We postulate that more heterogeneous landscapes (i.e., high topographic variation) may be more challenging for downscaling and predicting soil moisture; therefore, moisture networks should increase monitoring efforts across these complex landscapes. Future opportunities for development of modular cyberinfrastructure tools for downscaling satellite-derived soil moisture are discussed. -
Abstract A dry-air intrusion induced by the tropopause folding split the deep cloud into two layers resulting in a shallow orographic cloud with a supercooled liquid cloud top at around −15°C and an ice cloud above it on 19 January 2017 during the Seeded and Natural Orographic Wintertime Clouds: The Idaho Experiment (SNOWIE). The airborne AgI seeding of this case was simulated by the WRF Weather Modification (WRF-WxMod) Model with different configurations. Simulations at different grid spacing, driven by different reanalysis data, using different model physics were conducted to explore the ability of WRF-WxMod to capture the properties of natural and seeded clouds. The detailed model–observation comparisons show that the simulation driven by ERA5 data, using Thompson–Eidhammer microphysics with 30% of the CCN climatology, best captured the observed cloud structure and supercooled liquid water properties. The ability of the model to correctly capture the wind field was critical for successful simulation of the seeding plume locations. The seeding plume features and ice number concentrations within them from the large-eddy simulations (LES) are in better agreement with observations than non-LES runs mostly due to weaker AgI dispersion associated with the finer grid spacing. Seeding effects on precipitation amount and impacted areas from LES seeding simulations agreed well with radar-derived values. This study shows that WRF-WxMod is able to simulate and quantify observed features of natural and seeded clouds given that critical observations are available to validate the model. Observation-constrained seeding ensemble simulations are proposed to quantify the AgI seeding impacts on wintertime orographic clouds. Significance Statement Recent observational work has demonstrated that the impact of airborne glaciogenic seeding of orographic supercooled liquid clouds is detectable and can be quantified in terms of the extra ground precipitation. This study aims, for the first time, to simulate this seeding impact for one well-observed case. The stakes are high: if the model performs well in this case, then seasonal simulations can be conducted with appropriate configurations after validations against observations, to determine the impact of a seeding program on the seasonal mountain snowpack and runoff, with more fidelity than ever. High–resolution weather simulations inherently carry uncertainty. Within the envelope of this uncertainty, the model compares very well to the field observations.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- The DOI is not currently available.
