Steps to progressively work towards a LSTM/ML analysis

[bpbond]

• [x] Let's focus on "C" (control plot) and 2022 for the growing season only. That's going to be these files

TMP_C_20220601-20220630_L1_v1-0.csv ~TMP_C_20220701-20220731_L1_v1-0.csv~ ~TMP_C_20220801-20220831_L1_v1-0.csv~

• [x] Make sure these files are in your repo; R reads them in
• [x] Reading the documentation file (TMP_2022_L1_v1-0_metadata.txt) we see that the weather data for this site are in the "GCW" files

GCW_W_20220601-20220630_L1_v1-0.csv ~GCW_W_20220701-20220731_L1_v1-0.csv~ ~GCW_W_20220801-20220831_L1_v1-0.csv~

• [x] Then you can bind (rbind or dplyr::bind_rows) the data into a single data frame -> data_15min

• [x] Filter (subset) the data for the variables that we're interested in: wx_par_den15 (photosynthetically active radiation), wx_tempavg15 (air temperature), wx_windspeed15 (wind speed), plus the thing we're trying to predict: sapflow_2.5cm (sapflow velocity)

• [x] Compute daily DAYTIME averages for all variables, something like this (you'll probably want to use time-handling functions in the lubridate package)

  data_15min %>% 
  mutate(daytime = ...) %>%
  filter(daytime) %>%
  group_by(day, research_name) %>% 
  summarize(value = mean(value)) ->
  data_daily

• [x] Reshape these data (I suggest tidyr::pivot_wider() function) to put them in a form suitable for model-fitting:

[bpbond]

Once we have the daily data in wide form (as above, with one row per day in June 2022, and one column per variable)...

LET'S BUILD A SIMPLE MODEL

Linear regression model would be something like this:

mod <- lm(sap flow_2.5cm ~ wx_par_den15 + wx_tempavg15 + wx_windspeed15, data = data_daily_wide)

[bpbond]

P.S. a presentation I put together on linear regression is here - https://rpubs.com/bpbond/830637 - may be useful