sheilasaia/paper-ndfd-nccoast-study

Analysis associated with ndfd shellcast paper.

paper-ndfd-nccoast-study's Issues

check cocorahs data max

https://www.cocorahs.org/Content.aspx?page=mod_measurements

rerun 2015 and 2016 data grab to fix column names

normals!

check out John's code here: https://github.ncsu.edu/ncsco/corn_web/blob/main/data_functions.php#L409

convert to metric units

create roc analysis plots

• plots without subgroup for season
• plots with subgroup for season

get all historic precip data

want data for all applicable stations, for years = 2015 & 2016

checkout the following for data from the SCO API:

• ASOS - altered surface observation station from nws and faa
• AWOS - automated weather observation station from faa
• BUOY - buoy network from national data buoy center
• CMAN - coastal marine automated network from national data buoy center
• CoCoRaHS - have this!
• COOP - cooperative network from national center of environmental information
• Duke - duke energy sites from duke energy
• ECONET - environmental and climate observing network from nc sco
• NCSU - econet sites on ncsu campus from ncsco
• NOS - national ocean service from national data buoy center
• RAWS - remote automatic weather stations from usfs
• ThreadEx - threaded station extremes from regional noaa climate centers
• TVA - tenseness valley authority sites from tva
• USACE - us army corps sites from usace
• USCRN - us climate reference network by national centers of environmental information
• USGS - usgs gages

Others:

• PRISM
• Cooperative Observer NOAA
• Emergency Mgmt
• NC FIMAN
• SECOORA

remove cmus that are provisional and have no rainfall events for the entire 2015-2016 period

fix metadata distinct_all() issue in SCO API call

see cloudyR script

add type of network (community-based, govt, etc.) to precip data

change historic to observation and ndfd to forecast throughout scripts

Add citation.cff file

Details: https://twitter.com/natfriedman/status/1420122675813441540

repeat analysis with 2016 data

rerun part 4 & 6 of ndfd_convert_raster_to_point_script.R

county based shape files are incorrect on work pc and need to rerun (see part 4 "county based: classify historic precip as urban/non-urban and coast/non-coast")

compare actual depth to ndfd depths

includes loop!

compile all sco historic precip data

(not just cocorahs)

rerun convert scripts with aggregated data and new projection

issue with clouds api data duplicates

fix percent complete on 04 script line 50!

create cmu class map (showing if cmu's are approved or conditionally approved)

calculate error statistics

Hi Sheila,

Could you please calculate the R2 (coefficient of determination) and RMSE for each of the valid periods shown in the plot below?

In the paper, I'm thinking we can have the above plot alongside this one:

Note that, in the figure above, I manually added the R2 and RMSE text boxes in powerpoint. Here's the code I used to create the plot (I pulled code from one of your scripts, but it looks slightly different than the original plot - I must have pulled the code for a different figure).

my_validhrs_colors <- c("#66c2a5", "#fc8d62", "#8da0cb")

ggplot(data = d) +
  geom_point(aes(x = precip_test*2.54, y = precip_pred*2.54, fill = as.factor(valid_period)), shape = 21, alpha = 0.50, size = 3) +
  geom_abline(slope = 1, intercept = 0, lty = 2) +
  facet_wrap(~ as.factor(valid_period)) +
  labs(x = "Forecasted (cm)", y = "Observed (cm)", fill = "Valid Period Hours") +
  theme_classic() +
  scale_fill_manual(values = my_validhrs_colors) +
  theme(axis.text = element_text(size = 16),
        axis.title = element_text(size = 16),
        text = element_text(size = 16),
        panel.grid.major = element_blank(), 
        panel.grid.minor = element_blank(),
        panel.background = element_blank())
ggsave("sim_obs_ml.pdf", width = 12, height = 4, units = "in")

compile all gps data and rerun 2015 data (along with 2016 data)

that is re-run ndfd_convert_raster_to_point_script.R

finalize DF to raster script (including clipping)

lag 48h and 72h ndfd forecast values

Currently, the qpf and pop values for the 48h and 72h valid periods are lined up with the 24h observed rainfall values (e.g., a 48h forecast issued on Jan 1 is being compared to actual precip from Jan 1, but should be compared to actual precip from Jan 2). New date columns should be added that correspond to the different ndfd valid periods, and the data should then be reorganized to ensure the observed precip values are aligned with the appropriate forecast valid periods.

To implement this change, edit the roc_wrangle_data_script.R script. From discussion today, it seems like a good approach would be to add columns for date_48h and date_72h when creating temp_ndfd_data_sel (line 307). The data would still need to be realigned so the precip matches the forecast valid periods (maybe done after the completion of the i/j/k loops?).

rerun get ndfd script for 2015 and 2016 with aggregated functionality