Skip to content

Files

Latest commit

41e0095 · Feb 3, 2020

History

History

Analysis

README.md

Analysis Tools

⚠️ This README is incomplete ⚠️

These tools can be used to aid in writing further analysis tools and transparency in model configuration. They will not be available in release archives

Overview

As part of developing this tool, the CALPUFF model was forced with NAM and ECMWF data, the results showed large discrepancies. An experiment was set up (IMO) to compare identical CALPUFF runs over a 1 month period with known observations, and evaluate if the one meteological dataset is inferior. An overview of the results will be hosted in the wiki.

Summary of meteological data:

NAM

  • Resolution:
  • Spatial -
  • Temporal -
  • available to non academic: yes

ECMWF

  • Resolution:
    • Spatial -
    • Temporal -
  • available to non-academic: no

Description of Analysis

  1. Extract Observational data
  2. Fill in missing values with ML alogrithm , but keep raw incomplete data set
  3. Extract model data, nearest point and surrounding points (mean, min max) accept that the model accuracy is not such that the exact station point will be well defined.
  4. Collate data sets and perform statistical analysis
  5. Plot obervations vs model data

Other possible steps:

  • calculate: daily/hourly/month max, min, mean (for area around station), see if it's comparable to data
  • calculate other useful statistics e.g. RMS and std to see variability and error
  • Normalise the data sets to see if there's any peak correleation.

Dec 2019

  • Normalise data,?
  • Composite days?
  • Scatter plot peak location data?
  • Rough estimates of PBL heights between NAM and EMCWF? Afternoon (only) estimates from HOROWORTZ 1964 would be possible vertical profile of theta is 1.5 that of the surface level (), gribs contain the required vars
  • Compiled metseries tool to generate rose plot!

Normalised Plots:

Stats.py plots normalised time series (scaled by max/min), the results are plotted for ElPanama and Pacaya.

Note that ElPanama is only to March 25th as Timeseries incomplete for Observation

  • The models still perform poorly although ECMWF looks marginally better
  • Statistics are in Stats_Normalised.html

Acknowledgements

Sara Barsotti