Numerical climate model simulations run at high spatial and temporal resolutions generate massive quantities of data. As our computing capabilities continue to increase, storing all of the data is not sustainable, and thus is it important to develop methods for representing the full datasets by smaller compressed versions. We propose a statistical compression and decompression algorithm based on storing a set of summary statistics as well as a statistical model describing the conditional distribution of the full dataset given the summary statistics. We decompress the data by computing conditional expectations and conditional simulations from the model given the summary statistics. Conditional expectations represent our best estimate of the original data but are subject to oversmoothing in space and time. Conditional simulations introduce realistic small-scale noise so that the decompressed fields are neither too smooth nor too rough compared with the original data. Considerable attention is paid to accurately modeling the original dataset--one year of daily mean temperature data--particularly with regard to the inherent spatial nonstationarity in global fields, and to determining the statistics to be stored, so that the variation in the original data can be closely captured, while allowing for fast decompression and conditional emulation on modest computers.

Fire emissions are a major contributor to atmospheric composition, affecting atmospheric oxidizing capacity and air quality. Transported amounts from Northern Hemisphere boreal fires can reach the pristine Arctic atmosphere as well as impact air quality in populated regions. Carbon monoxide (CO) is a useful trace gas emitted from fires that can be used to link extreme fire events with climate variability. We use our recently developed statistical tool to investigate the climate drivers of satellite measured CO variability in two Northern Hemisphere boreal fire regions: northwest Canada and Siberia. Our focus is on quantifying the ability of climate mode indices for the Pacific, Atlantic, Indian and Arctic Oceans in predicting CO amounts in these regions. Climate mode indices El Niño Southern Oscillation (ENSO), Tropical North Atlantic (TNA), the Dipole Mode Index (DMI) and the Arctic Oscillation (AO) are used to develop statistical models of column CO interannual variability from the Measurements of Pollution In The Troposphere (MOPITT) satellite instrument, for the time period covering 2001-2017. In addition, we assess the ability of fire emission inventories to reproduce CO, including the Fire Inventory from NCAR (FINN), the NASA Quick Fire Emissions Dataset (QFED) and the Copernicus Atmosphere Monitoring Service (CAMS) Global Fire Assimilation System (GFAS). These are implemented in the NCAR Community Atmosphere Model with chemistry (CAM-chem) and subsequently evaluated against MOPITT CO observations. Emission uncertainty contribution to inter-inventory differences are quantified, and the modeled contribution of fires to CO interannual variability is determined.

Baker et al (2015) developed the Community Earth System Model Ensemble Consistency Test (CESM-ECT) to provide a metric for software quality assurance by determining statistical consistency between an ensemble of CESM outputs and new test runs. The test has proved useful for detecting statistical difference caused by compiler bugs and errors in physical modules. However, detection is only the necessary first step in finding the causes of statistical difference. The CESM is a vastly complex model comprised of millions of lines of code which is developed and maintained by a large community of software engineers and scientists. Any root cause analysis is correspondingly challenging. We propose a new capability for CESM-ECT: identifying the sections of code that cause statistical distinguishability. The first step is to discover CESM variables that cause CESM-ECT to classify new runs as statistically distinct, which we achieve via Randomized Logistic Regression. Next we use a tool developed to identify CESM components that define or compute the variables found in the first step. Finally, we employ the application Kernel GENerator (KGEN) created in Kim et al (2016) to detect fine-grained floating point differences. We demonstrate an example of the procedure and advance a plan to automate this process in our future work.