---
title: "Temporally Downscale Meteorology"
output: html_vignette
vignette: >
  %\VignetteIndexEntry{Temporally Downscale Meteorology}
  %\VignetteEngine{knitr::rmarkdown}
---


### Subdaily Training Data

Examples:

* Fluxnet2015 30-minute/hourly
* Ameriflux 30-minute/hourly

### Extract Training Data and Merge All Years Into 1 File

```{r, eval=FALSE}
library(PEcAn.data.atmosphere)
library(PEcAn.DB)

# Download the Fluxnet2015 file
sitename   <- 'Niwot Ridge Forest/LTER NWT1'
lat        <- 40.0329
lon        <- -105.546
start_date <- '2000/01/01'
end_date   <- '2014/12/31'
outfolder  <- "~/Example"
# TODO show where `bety` comes from
con <- db.open(params = bety)
format    <- query.format.vars(input.id = 5000000005, bety = con)
in.prefix <- 'FLX_US-NR1_FLUXNET2015_SUBSET_HH_1998-2014_1-3'
download.Fluxnet2015(sitename, lat, lon, start_date, end_date, outfolder, username = 'pecan', overwrite = FALSE)

# Convert the data to CF metadata standards
met2CF.csv(in.path = file.path(outfolder), in.prefix, outfolder = file.path(outfolder, "CF"), start_date, end_date, format, lat, lon)

# Fill the gaps in the flux tower dataset
metgapfill(lst = -7, in.path = file.path(outfolder, "CF"), in.prefix, outfolder = file.path(outfolder, "gapfill"), start_date, end_date)

# Merge the individual years of the gapfilled NC files together
nc.merge(outfolder = file.path(outfolder, "training_data"), in.path = file.path(outfolder, "gapfill"), in.prefix, start_date, end_date, upscale = "hour")
```

### Generate Linear Regression Models From Training Data

Note: This requires ~ 120 GB of space if using the entire training dataset
```{r, eval=FALSE}
in.prefix      <- "US-NR1"
dat.train.file <- "~/Example/training_data/FLX_US-NR1_FLUXNET2015_SUBSET_HH_1998-2014_1-3_dat.train.nc"
n.beta         <- 10 # Number of betas for the linear regression model to create, we'll choose 10 for time's sake
day.window     <- 5  # Number of days around a specific timestamp to gather statistics from (at the same timestamp)
gen.subdaily.models(outfolder = dir.create(outfolder, "lm_models_base"), dat.train.file, in.prefix, n.beta, day.window)
```

### Coarse Data to Downscale

Examples:

* MACA daily
* CRU-NCEP 6 hourly

### Extract Data We Want To Downscale

```{r, eval=FALSE}
start_date      <- "2020-01-01"
end_date        <- "2020-12-31"
site_id         <- 772
model           <- "BNU-ESM"
scenario        <- "rcp85"
ensemble_member <- "r1i1p1"
download.MACA(outfolder, start_date, end_date, site_id, model, scenario, ensemble_member)

```

### Predict Subdaily Data Using Statistics From Training Data

```{r, eval=FALSE}
in.path        <- "~/Example/MACA_site_0-772"
in.prefix      <- "MACA.BNU-ESM.rcp85.r1i1p1" # this is the data we are going to downscale
lm.models.base <- "~/Example/lm_model_output" # where we stored the lm models
n.ens          <- 3 # number of ensemble members we want to generate
predict_subdaily_met(outfolder = file.path(outfolder, "downscaled_data"), in.path, in.prefix, lm.models.base, n.ens, start_date, end_date)
```