toolLULCDash.R

rm(list=ls(all=TRUE))
# Shiny app for exploring health facility distributions

library(shiny)
library(shinydashboard)
library(ggplot2)

source('tool functions.R')
tmp=read.csv('data/glm table.csv',as.is=T)
coef1=tmp[,'Estimate']
names(coef1)=tmp$X

# Define UI 
ui <- dashboardPage(
  
  dashboardHeader(title = "Optimal layout of roads", titleWidth = 300),
  
  dashboardSidebar(
    radioButtons(inputId = "landscape",
                 label = "Landscape",
                 choices = c(1, 2),
                 selected = 1),
    textInput("road.cost", label="Road cost (per km)", value=1),
    textInput("pa.cost", label="Cost of deforesting PA (per 1 km2 pixel)", value='1'),
    textInput("forest.cost", label="Cost of deforesting forested UA (per 1 km2 pixel)", value='1'),
    sliderInput("protect", label="Protective effect of PA in reducing deforestation (%)", 
                min=0, max=100, value=30, step=0.5, ticks=T),
    conditionalPanel(
        condition = "input.tipo == 'user-defined'",
        textInput("x", label="x coordinates", value='20,20',placeholder='0-100'),
        textInput("y", label="y coordinates", value='10,90',placeholder='0-100'),
        div("Use commas to specify more than one coordinate (e.g. 3.75, 5.25)", 
            class="form-group shiny-input-container")
    ),
    radioButtons(inputId = "tipo", 
                 label = "Road layout:", 
                 choices = c('straight line', 'user-defined','optimized'),
                 selected='straight line')
    
  ),
    
    dashboardBody(
      p("Roads are a major driver of tropical deforestation and determining the best
        route to connect two places with a road is not trivial given that proximity
        to the road and to the population center often result in higher probability of
        deforestation."),
      p("In this tool, users can explore the implications of different road layouts
        on road cost and deforestation of protected areas (PA) and unprotected areas
        (UA). An optimal road layout depends not only on road cost but also on how
        forests in these areas are valued."),
      p("Disclaimer: The statistical model that underlies this tool was fitted to 
        deforestation data, provided by the Brazilian deforestation monitoring program
        PRODES, from the BR-364 road segment between the cities of Feijo and Manoel
        Urbano (Acre). As such, the outcomes of the model and the calculations involved
        in this app may not be applicable to other localities. Also, we assume that there
        is a causal relationship between deforestation and these distance variables and
        that the estimated relationships do not change for different road layouts."),
      p("Abbreviations: PA = Protected forested Area; UA = Unprotected forested Area"),
      fluidRow(
        column(width = 4,
               infoBoxOutput("rdLength", width = NULL),
               infoBoxOutput("paDef", width = NULL),
               infoBoxOutput("forDef", width = NULL),
               infoBoxOutput("costs", width = NULL)
        ),
        column(
          width = 8,
          box(
            title = "Proposed route",
            width = NULL,
            status = "primary",
            solidHeader = T,
            plotOutput("LULC", height = "600px")
          )
        )
      )
    )
)

# Input list for debug
# input=list();
# input$road.cost='1'
# input$pa.cost='2'
# input$forest.cost='1'
# input$tipo='optimized'
# input$protect='1'
# input$landscape=3

# Define server logic 
server <- function(input, output) {
  # Load data based on landscape
  landscapeList <- reactive({
    l <- as.numeric(input$landscape)
    optimFile <- paste('data/optimized', l, '.csv', sep="")
    gridFile <- paste('data/grid', l, '.csv', sep="")
    optim1 <- read.csv(optimFile,as.is=T)
    grid1 <- read.csv(gridFile,as.is=T)
    
    startEnd <- read.csv('data/startEnd.csv')
    start <- startEnd[startEnd$layout == l, c("startX", "startY")]
    start <- as.numeric(start)
    end <- startEnd[startEnd$layout == l, c("endX", "endY")]
    end <- as.numeric(end)
    
    #get distance to urban centers
    uc=read.csv('data/uc.csv')
    uc=uc[uc$landscape==l,]
    dist=numeric()
    for (i in 1:nrow(uc)){
      x2=(grid1$x-uc$x[i])^2
      y2=(grid1$y-uc$y[i])^2
      dist=cbind(dist,sqrt(x2+y2))
    }
    grid1$dist_uc=apply(dist,1,min)
    
    #points
    # pt <- data.frame(x=c(start[1], end[1]), y=c(start[2], end[2]),
    #                  type=c("End Point"))
    # pt <- rbind(pt, uc)
    
    L <- list(optim1, grid1, start, end, uc)
  })
  
  outList <- reactive({
    optim1 <- landscapeList()[[1]]
    grid1 <- landscapeList()[[2]]
    start <- landscapeList()[[3]]
    end <- landscapeList()[[4]]
    uc <- landscapeList()[[5]]
    
    #If input is empty or non numeric, make them zero
    road.cost=check.input(input$road.cost) #per length of road
    pa.cost=check.input(input$pa.cost) #per area of deforested pa
    forest.cost=check.input(input$forest.cost) #per area of deforested land
    protect=check.input(input$protect)/100 #% of deforestation probability
    
    str.coords <- user.coords <- data.frame(x=c(start[1],end[1]),y=c(start[2],end[2]))
    user.grid <- grid1
    
    if (input$tipo=='user-defined'){
      #create user coordinates
      x=as.numeric(unlist(strsplit(input$x,split=',')))
      y=as.numeric(unlist(strsplit(input$y,split=',')))
      if (length(x)==length(y)){
        user.coords=data.frame(x=c(start[1],x,end[1]),y=c(start[2],y,end[2]))
      }
    }
    if (input$tipo=='optimized'){
      cost=pa.cost*(1-protect)*optim1$d.pa +
        forest.cost*optim1$d.ua +
        road.cost*optim1$l.road
      ind=which(cost==min(cost))[1]
      x=unlist(optim1[ind,c('x1','x2','x3')])
      y=unlist(optim1[ind,c('y1','y2','y3')])
      user.coords=data.frame(x=c(start[1],x,end[1]),y=c(start[2],y,end[2]))
    }
    
    #get nearest distance
    user.grid$dist_road=get.dist(user.coords, user.grid)
    
    #predict deforestation
    tmp=with(user.grid, exp(coef1['(Intercept)']+
                              coef1['dist_road']*dist_road+
                              coef1['dist_uc']*dist_uc+
                              coef1['dist_road:dist_uc']*dist_road*dist_uc))
    user.grid$prob=def.prob(user.grid, coef1)
    
    #get length of road
    user.length <- get.length(user.coords)
    
    #calculate expected cost
    ecost=get.cost(user.grid, road.cost, pa.cost, forest.cost, protect, user.length)
    
    #change name from "Forest" to "UA"
    cond=user.grid$tipo=='Forest'
    user.grid$tipo[cond]='UA'
    
    #get prob.cor (this helps displaying deforestation probability)
    cond=user.grid$tipo%in%c('UA','PA')
    user.grid2=user.grid[cond,]
    cond=user.grid2$tipo=='PA'
    user.grid2$prob[cond]=user.grid2$prob[cond]*(1-protect)
    prob.thresh=0.1
    cond=user.grid2$prob>prob.thresh
    user.grid2=user.grid2[cond,]
    
    #relabel LC type for plotting
    user.grid$tipo <- factor(user.grid$tipo, levels = c("PA", "UA", "Pasture"),
                             labels = c("Protected Forested Area (PA)", 
                                        "Unprotected Forested Area (UA)",
                                        "Pasture"))
    
    #plot results
    res=ggplot() +
      geom_tile(data = user.grid, alpha = 0.8,aes(x = x, y = y,fill = tipo)) +
      geom_path(data = user.coords, aes(x = x, y = y), show.legend = F, lwd = 1.5) +
      scale_fill_manual(values=c('darkgreen','green','darkseagreen1'),name='Land use type') +
      geom_point(data = user.coords, aes(x = x,y = y), size = 3, show.legend=F) +
      geom_point(data = user.grid2, aes(x = x, y = y, colour = prob), size = 1, alpha = 0.5) +
      scale_colour_continuous(low = "#ffcccc", high = "#ff0000",
                              name = 'Probability\nof deforestation') +
      geom_point(data = uc, aes(x = x, y = y, pch = type), size = 5, colour="blue") +
      scale_shape_manual(values=10, name='') +
      coord_fixed() + theme_bw(base_size = 14)
    
    #calculate straight line cost
    str.grid <- grid1
    str.grid$dist_road <- get.dist(str.coords, str.grid)
    str.grid$prob <- def.prob(str.grid, coef1)
    str.length <- get.length(str.coords)
    
    #sideplots
    rd.len.inc <- (user.length - str.length) / str.length * 100
    d.pa.perc <- ecost$d.pa.prop * 100
    d.ua.perc <- ecost$d.ua.prop * 100
    
    L <- list(main.plot=res, rd.len.inc=rd.len.inc, ecost=ecost)
    L
  }
    
  )
  
  output$LULC <- renderPlot(outList()$main.plot)
  output$rdLength <- renderInfoBox({
    infoBox(
      "Road Length Increased",
      value = paste(round(outList()$rd.len.inc, 1), "%"),
      subtitle = "Proposed vs. straight-line route",
      icon = icon("road"),
      color = "black"
    )
  })
  
  output$paDef <- renderInfoBox({
    infoBox(
      "Deforestation - PA",
      value = paste(round(outList()$ecost$d.pa.prop*100, 1), "%"),
      subtitle = "of original area",
      icon = icon("tree"),
      color = "olive"
    )
  })
  
  output$forDef <- renderInfoBox({
    infoBox(
      "Deforestation - UA",
      value = paste(round(outList()$ecost$d.ua.prop*100, 1), "%"),
      subtitle = "of original area",
      icon = icon("tree"),
      color = "lime"
    )
  })
  
  output$costs <- renderInfoBox({
    ecost = outList()$ecost
    road.cost = round(ecost$road.cost)
    pa.cost = round(ecost$pa.cost)
    forest.cost = round(ecost$forest.cost)
    total.cost = round(road.cost+pa.cost+forest.cost)
    disp.value = paste(total.cost, "=", road.cost, "+", pa.cost, "+", forest.cost)
    infoBox(
      "Costs",
      value = disp.value,
      subtitle = "Total = Road + PA + UA",
      icon = icon("usd"),
      color = "orange"
    )
  })
}

# Run the application 
shinyApp(ui = ui, server = server)