You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

317 lines
11 KiB

#!/usr/bin/env julia
using DataFrames
using Dates
using CSV
using EzXML
using Images
using Colors
#using ImageDraw
using CairoMakie
using Printf
using TimeZones
using PlotUtils
#using Profile
struct StationInfo
id::String
name::String
num::Int64
lat::Float32
lon::Float32
end
function get_station_info(station_id)
stations = readxml("activestations.xml")
station_node = findfirst("/stations/station[@id=\"$station_id\"]", stations)
#<station id="dukn7" lat="36.184" lon="-75.746" elev="7.7"
# name=" 8651370 - Duck Pier, NC " owner="NOS" pgm="NOS/CO-OPS" type="fixed"
# met="y" currents="n" waterquality="n" dart="n"/>
name = strip(station_node["name"])
num, name = split(name, " - ")
lat, lon = parse.(Float32, [station_node["lat"], station_node["lon"]])
return StationInfo(station_id, name, parse(Int, num), lat, lon)
end
function get_station_lat_lon(station_id)
stations = readxml("activestations.xml")
station_node = findfirst("/stations/station[@id=\"$station_id\"]", stations)
return parse.(Float32, [station_node["lat"], station_node["lon"]])
end
tryparsem(T, str) = something(tryparse(T, str), missing)
function get_tile_xy_fraction(lat, lon, zoom)
n = 2^zoom
lat_rad = lat * pi / 180
xtile = (lon + 180) / 360 * n
ytile = (1 - log(tan(lat_rad) + 1 / cos(lat_rad)) / pi) / 2 * n
return xtile, ytile
end
function get_tile_xy(lat, lon, zoom)
xf, yf = get_tile_xy_fraction(lat, lon, zoom)
return (Int(floor(xf)), Int(floor(yf)))
end
function get_tile(x, y, zoom)
url = "https://tile.openstreetmap.org/$zoom/$x/$y.png"
tmp_path = download(url)
tile = load(tmp_path)
return tile
end
function get_tile_by_lat_lon(lat, lon, zoom)
xtile, ytile = get_tile_xy(lat, lon, zoom)
return get_tile(xtile, ytile, zoom)
end
function get_water_levels(station_number, start_date, end_date;
timezone=tz"America/New_York", prediction=false)
start_s = Dates.format(start_date, "yyyymmdd")
end_s = Dates.format(end_date, "yyyymmdd")
if prediction
prod = "predictions"
else
prod = "water_level"
end
url = "https://api.tidesandcurrents.noaa.gov/api/prod/datagetter?product=$(prod)&application=NOS.COOPS.TAC.WL&begin_date=$(start_s)&end_date=$(end_s)&datum=MLLW&station=$(station_number)&time_zone=lst_ldt&units=english&format=csv"
if prediction
url *= "&interval=hilo"
end
tmp_path = download(url)
df = open(tmp_path) do file
CSV.read(file, DataFrame)
end
dt_local = (dt_s) -> DateTime.(dt_s, "yyyy-mm-dd HH:MM")
transform!(df,
"Date Time" => dt_local => "TIME",
" Water Level" => "WLEVEL")
return df[(start_date .<= df.TIME .<= end_date), :]
end
function get_meters_per_pixel(lat, zoom)
px_per_tile = 256
return 156543.03 * cos(lat) / (2^zoom)
end
function get_map(lat, lon, zoom)
px_per_tile = 256
crop_size = 2 * px_per_tile
meters_per_pixel = get_meters_per_pixel(lat, zoom)
xf, yf = get_tile_xy_fraction(lat, lon, zoom)
x0, y0 = (Int(floor(xf)), Int(floor(yf)))
xpoint = 1 + px_per_tile + Int(floor((xf - x0) * px_per_tile))
ypoint = 1 + px_per_tile + Int(floor((yf - y0) * px_per_tile))
half_size = Int(floor(crop_size / 2))
ystart = ypoint - half_size
yend = ypoint + half_size
xstart = xpoint - half_size
xend = xpoint + half_size
xpoint -= xstart
ypoint -= ystart
cache_path = "cache/$(lat)_$(lon)_$zoom.png"
if isfile(cache_path)
map = Images.load(cache_path)
return map, xpoint, ypoint
end
map = Array{RGB{N0f8}, 2}(undef, (px_per_tile * 3, px_per_tile * 3))
for i in -1:1
x = x0 + i
ioffset = 1 + (i + 1) * px_per_tile
for j in -1:1
y = y0 + j
joffset = 1 + (j + 1) * px_per_tile
img = get_tile(x, y, zoom)
map[joffset:joffset+px_per_tile-1, ioffset:ioffset+px_per_tile-1] = img
end
end
map = map[ystart:yend, xstart:xend]
#draw!(map, Ellipse(CirclePointRadius(xpoint, ypoint, 6)),
# RGB{N0f8}(0.5, 0.0, 0.5))
Images.save(cache_path, map)
return map, xpoint, ypoint
end
function get_realtime_data_file(station)
station = uppercase(station)
url = "http://www.ndbc.noaa.gov/data/realtime2/$station.txt"
return download(url)
end
function get_realtime_dataframe(station_name; timezone=tz"America/New_York")
station_fname = get_realtime_data_file(station_name)
lines = open(station_fname) do file
readlines(file)
end
dt_array = (y, mon, d, hr, min) -> (astimezone.(ZonedDateTime.(
DateTime.(y, mon, d, hr, min), tz"UTC"), timezone))
deg_c_to_f = passmissing(x -> x * 9.0 / 5.0 + 32.0)
speed_ms_to_mph = passmissing(x -> 0.44704 * x)
lines[1] = lstrip(lines[1], '#')
data = join([lines[1:1]; lines[3:end]], "\n")
df = CSV.read(IOBuffer(data), DataFrame; header=1, delim=" ",
ignorerepeated=true)
select!(df,
[:YY, :MM, :DD, :hh, :mm] => dt_array => :TIME,
:WDIR => x -> tryparsem.(Int, x),
:WSPD => x -> speed_ms_to_mph.(tryparsem.(Float32, x)),
:GST => x -> speed_ms_to_mph.(tryparsem.(Float32, x)),
:PRES => x -> tryparsem.(Float32, x),
:ATMP => x -> deg_c_to_f.(tryparsem.(Float32, x)),
:WTMP => x -> deg_c_to_f.(tryparsem.(Float32, x));
renamecols=false)
return df
end
function plot_station_data(station_name, station_map, station_x, station_y,
df, i)
fig = Figure(resolution=size(station_map), figure_padding=0)
ax = CairoMakie.Axis(fig[1, 1])
hidespines!(ax)
hidedecorations!(ax)
plot_station_data_to_axis(ax, station_name, station_map,
station_x, station_y, df, i)
return fig
end
function plot_station_data_to_axis(ax, station_name, station_map, station_x, station_y,
df, i, old_plots=())
obs_time = df.TIME[i]
obs_time_s = Dates.format(obs_time, "e, u d H:MM")
if length(old_plots) == 0
plot_img = image!(ax, rotr90(station_map))
else
plot_img = old_plots[1]
delete!(ax, old_plots[2])
delete!(ax, old_plots[3])
end
ax.title = "$station_name $obs_time_s"
wind_rad = passmissing(deg2rad)(90 - df.WDIR[i])
wind_mph = df.WSPD[i]
if ismissing(wind_mph) || ismissing(wind_rad)
wind_mph = 0.0
wind_rad = 0
end
atmp = df.ATMP[i]
wtmp = df.WTMP[i]
if ismissing(atmp)
atmp = 0.0
end
if ismissing(wtmp)
wtmp = 0.0
end
txt = @sprintf "Air: %.1f F, Water: %.1f F, Wind: %.1f mph" atmp wtmp wind_mph
wind_vec = 50 * Vec2f(cos(wind_rad), sin(wind_rad)) * wind_mph / 10
station_y += 50 # Why is this off by 50???
#water_temp_color = RGB(0, 0, 0.1 + 0.9 * (wtmp - 40) / 40)
#air_temp_color = RGB(0.5 + (atmp - 40) / 90, 0, 0)
plot_tooltip = tooltip!(ax, station_x, station_y, txt;
offset=15, placement=:above)
plot_arrow = arrows!(ax, [Point2f(station_x, station_y)], [wind_vec],
fxaa=true, linewidth=2, align=:center, arrowhead='⟰')
return plot_img, plot_tooltip, plot_arrow
end
function timestamp_range(timestamps, t0::Dates.DateTime)
return Dates.value.(timestamps) .- Dates.value(t0)
end
struct DateTimeTicks
t0::Dates.DateTime
end
# See https://github.com/MakieOrg/Makie.jl/issues/442#issuecomment-1446004881
function Makie.get_ticks(t::DateTimeTicks, any_scale, ::Makie.Automatic, vmin, vmax)
dateticks, dateticklabels = PlotUtils.optimize_datetime_ticks(
Dates.value(Dates.DateTime(Dates.Millisecond(Int64(vmin)) + t.t0)),
Dates.value(Dates.DateTime(Dates.Millisecond(Int64(vmax)) + t.t0)),
)
dtformat = "e H:MM"
labels = [Dates.format(convert(Dates.DateTime, Millisecond(v)), dtformat)
for v in dateticks]
return dateticks .- Dates.value(t.t0), labels
end
function save_animation(station_name, out_path, hours, real_hour_per_animation_second)
info = get_station_info(station_name)
lat, lon = info.lat, info.lon
station_map, station_x, station_y = get_map(lat, lon, 14)
df = get_realtime_dataframe(station_name; timezone=tz"America/New_York")
diff_hours = (df.TIME[1] - df.TIME[2]).value / 1000 / 3600
stride = Int(ceil(real_hour_per_animation_second / diff_hours))
end_index = Int(min(stride * hours + 1, size(df, 1)))
data_indexes = range(1, end_index; step=stride)
df_recent = reverse(@view df[1:end_index, :])
df_ts = select(df_recent,
:TIME => x -> Dates.DateTime.(x),
renamecols=false)
df_tides = get_water_levels(info.num, df_ts.TIME[1], df_ts.TIME[end])
fig_res = (size(station_map, 1), Int(floor(size(station_map, 2) * 1.5)))
fig = Figure(resolution=fig_res, figure_padding=0)
t0 = df_ts.TIME[1]
ax_temp = CairoMakie.Axis(fig[1, 1], xticks=DateTimeTicks(t0), ylabel="°F")
ax_wind = CairoMakie.Axis(fig[1, 1], xticks=DateTimeTicks(t0), ylabel="mph",
yaxisposition=:right, ygridvisible=false,
xgridvisible=false, xticksvisible=false,
xticklabelsvisible=false)
ax_tide= CairoMakie.Axis(fig[2, 1], xticks=DateTimeTicks(t0), ylabel="ft")
ax = CairoMakie.Axis(fig[3, 1])
hidespines!(ax)
hidedecorations!(ax)
rowsize!(fig.layout, 1, Auto(0.3))
rowsize!(fig.layout, 2, Auto(0.2))
xs_time = timestamp_range(df_ts.TIME, t0)
lines!(ax_temp, xs_time, df_recent.WTMP, color=:darkblue)
lines!(ax_temp, xs_time, df_recent.ATMP, color=:brown)
lines!(ax_wind, xs_time, df_recent.WSPD, color=:skyblue)
old_vline = vlines!(ax_temp, xs_time[1], color=:black, linewidth=5)
lines!(ax_tide, xs_time, df_tides.WLEVEL, color=:blue)
old_plots = ()
record(fig, out_path, data_indexes; framerate=1) do i
old_plots = plot_station_data_to_axis(ax, station_name, station_map,
station_x, station_y, df_recent,
i, old_plots)
delete!(ax_temp, old_vline)
old_vline = vlines!(ax_temp, xs_time[i], color=:black, linewidth=5)
end
save(replace(out_path, r"\..*" => ".png"), fig)
end
function main()
if (length(ARGS) < 1)
println("Usage: create_latest_map.jl station_name [out_path]")
exit(1)
end
station_name = ARGS[1]
out_fname = "$station_name.png"
if length(ARGS) > 1
out_fname = ARGS[2]
end
if any(endswith.(out_fname, [".gif", ".webm", ".mp4", ".mkv"]))
save_animation(station_name, out_fname, 48, 1)
else
lat, lon = get_station_lat_lon(station_name)
station_map, station_x, station_y = get_map(lat, lon, 14)
df = get_realtime_dataframe(station_name; timezone=tz"America/New_York")
fig = plot_station_data(station_name, station_map, station_x, station_y,
df, 1)
save(out_fname, fig)
end
end
main()
#Profile.print()