In agroimpacts/arcticdynamics: Arctic Cloud Change Dynamics

setwd(dir = "vignettes/slides")
rmarkdown::render(input = "Presentation.Rmd", output_file = "Presentation.html")
# This is to make the html output appear in the folder where you want it.
knitr::include_graphics(here("images/hcdc2021_09.png"))

knitr::opts_chunk$set(echo = FALSE)

library(here) # set up working directory
here::here() # set working directory for images

Introduction

• Our main questions:
• What influences different altitude cloud cover in the Pacific Arctic? Does this relationship change seasonally?
• How has the Pacific Arctic Region Cloud Cover changed over time and what are their drivers?
• How has the energy budget changed over time?

Data:

• NARR Reanalysis: (1979-2020)
• Cloud cover at three altitude levels (low, medium, and high)
• Wind speed at 10 m
• Air temperature
• Evaporation
• Geopotential height
• Relative humidity
• NASA MODIS-Aqua chlorophyll concentration (2002-2020)
• DoD SSMI/S sea ice concentration (1979-2020)
• NASA CERES Satellite Data (2000-2015)

Original NetCDF Data

knitr::include_graphics(here("images/prelim_subplots.png"))

Methods:
Preprocessing

• Monthly composites were downloaded (external/data/Download_Datasets.R)
• Variables were converted to raster bricks, projected, cropped to the region of interest, and monthly averages were saved to a dataframe and exported to a csv to be combined into one master dataframe (R/Opening_NCDF.R/NARR.R, R/Prepare_Rasters_for_Model.R, R/XGB_Variable_Merge.R)

Figure 1

knitr::include_graphics(here("images/Inset_Study_Site.png"))

Methods:
Extracting Values from DBO 3

• Since the files, were so large, we subset the data to DBO 3, and extracted the values from each variable from within it, then merged all the data to one data frame. (R/Narr_Variable_Merge.R)
• The analysis of each variable's monthly trends were then plotted

Result:

knitr::include_graphics(here("images/ALL_Trends.png"))

Correlation Trend

knitr::include_graphics(here("images/Sea__Ice_vs_Clouds.png"))

Method:
Mann Kendall Trends

• For understanding the spatial trends for the different variables we did the Mann Kendall Test -This was done using the raster.kendall function through the SpatialEco package.
• We also found the p value, z value, confidence and tau values.

Mann Kendall Results

knitr::include_graphics(here("images/kendall_test_results.png"))

Methods:
Gradient Boosting Regression to find the drivers of low cloud cover concentration in the Chukchi Sea.

• Gradient Boosting is a machine learning technique used for regression and classification. Similar to random forest, it uses an ensemble of decision trees to learn how independent variables predict the dependent variable. Unlike random forest, trees are built upon hrough boosting, so trees are interactively improved instead of averaged out by many individual trees. This package extracts information from several raster datasets and builds a gradient boosting model to determine the most important variables that influence low cloud cover concentration over the Chukchi Sea.

• For running the gradient boosting model (XGBoost) the data were partitioned into training and testing, trained a model, and then used accuracy assessment to find the optimum number of trees used to tune the final XGBoost model.

XGBoost Prediction Results

knitr::include_graphics(here("images/XGB_LCC_vs_predicted_testset.png"))

XGBoost Variable Importance Results

knitr::include_graphics(here("images/XGB_VariableImportance.png"))

Future Work:

• Due to the limitation in time there were a few steps we were not able to do.
• This included the detrending of the data from any external factors that would be impacted by it. For instances, with such a large time line, it would be impacted by the Arctic Oscillation or the Pacific Decadal Oscillation
• Improving Accuracy of the Model

agroimpacts/arcticdynamics documentation built on Jan. 1, 2022, 9:19 p.m.