nocollier / vulnerableecosystems Goto Github PK

I have a section in the README dedicated to providing instructions for obtaining, installing, running the cluster code. But I am not familiar with any of those details. If documented elsewhere, a link would suffice.

A short discussion among @drastogi4, Moet and @minxu74 on Oct. 14, 2020, we discussed that:

• the heat map generated by Moet contains more than 75 indices describing both mean and extreme performances of CMIP6 models in precipitation, surface mean, max, and min temperature. Moet stated that It is important to learn both mean and extremes characteristics of CMIP6 models. He also plan to add other climatic indices including PDO, ENSO and AMO etc.

• Min stated that in order to be consistent for the clusterings of carbon variables in future, we may constrain our model selection to the models with carbon variables outputs. Also we will include the E3SM model despite of its rank in the heat map. Min suggested to add the results of the CMIP6 model mean in the heat map.

• Moet is not sure whether or not global analysis was done before subsetting data over the CONUS for the heat map. He will look at his scripts. He also will set up accounts for Deeksha and Min in his local data server in the office to access the model data and play with his CDO scripts.

• Deeksha has a code to calculate heat waves. She is also exploring a tool to calculated spatial extent of extremes.

• Min has some NCO scripts to calculate the ENSO and AMO indices and python codes to generate drought-related indices,

• We all agreed that the heat map is easy to generate for either the globe, US or other regions in which people are interested, so it can be used in the next big idea presentation. Other indices including climatic, drought indices, and other extreme indices can be applied later or after the initial reduction and selection of indices and models.

• We also briefly talked about creating spatial maps of biases (global or US) in addition to the heat maps. But this will mean one map for each model and index. So, we need to find a way to better present these biases spatially.

-Some of the variables are correlated. So, at some point we need to explore dimensionality reduction techniques (PCA etc) to care of the biases generating from these correlated variables.

• We also agree to upload our scripts to this GitHub for tracking changes and sharing with team members.

@drastogi4, Moet Please add anything that I missed and let us use this issue to track our progress. What do you think?

Thank Moet for organizing the discussion.

I have been playing around with the Dash/Plotly plotting libraries. Plotly is what I have used in our previous LDRD and in RUBISCO to make interactive visualizations. The dash framework is from the same company and is built to allow you to create custom interactive webpages with abstract elements which could be headings, text boxes, buttons, or even plotly plot objects. They also provide a fairly comprehensive callback feature which allows you to update anything on your webpage, when other elements are changed/clicked/mouseovered/selected.

Bottom line: this technology allows us to have multiple scatter plots (say, isoplanes of a cluster parameter space) and then link selections across each plot. This is what the demo script visualization/parameter_space.py does (it also uses ILAMB to do some processing and would require the model data to run). I have also have clickable/selectable global maps working in this framework in the context of some work for RUBISCO/ILAMB.

A few questions stand out in my mind:

1. Is this a good technology to use? I am comfortable saying that it is rich enough to do what we need in terms of features.
2. Are there better alternatives? I know @jitendra-kumar showed a representativeness plot on a webpage but I forget what he used, maybe leaflet?
3. Can we make this work in production? The framework serves up webpages as Flask apps. So you write your code entirely in python, and then once running you open a port on your localhost to see the webpage. I am unclear on what is needed on the server side.