fmfoerster / canhemonr Goto Github PK

---
title: "PWN workflow and functions overview"
author: "Pieter Beck"
date: "Monday, August 11, 2014"
output:
  html_document:
    keep_md: yes
    toc: yes
---
# To do

* Does the fact that orthos have more pixels per polygon, bias the model fits in decline_trees_RS_analysis?
* How will you handle differing pixel and sample size, when you incorporate Landsat?
* When comparing index means, then you should do bandwise means of objects first, before calculating non-linear indices !

# CanHeMonR package
An overview of the tasks that can be performed using the CanHeMonR package.

## Rough crown delineation - watershed_crown_delineation.r
The start of a crown detection algorithm - not very robust yet
#### Challenge
Starting from a multispectral image, you want to roughly identify all the trees as polygons of their crowns

#### Input
* A multi-band raster image
* Spectral bands and/or spectral indices, and respective cut-off values, that allow you to mask out the background
* Criteria for a watershed algorithm

#### Output
* A rough SpatialPolygons object of potential crowns

```{r, eval=FALSE}

require(CanHeMonR)
 watershed_tree_detection(
   image_fname = 'E:/FISE/forest/CanopyHealthMonitoring/PWN/flights_final/150727_mca/150727_mca.bsq',
   #extent = raster::extent(c(747400, 747420, 4463900, 4463920)),
   extent = raster::extent(c(747400, 747600, 4463800, 4464000)),
   index_name = 'NDVI',
   bg_mask = list('NDVI',0.40,800, 1200),
   plott = T,
   neighbour_radius = 1,
   watershed_tolerance = 0.008,
   rough_crowns_shp_fname = 'C:/Users/pieterbeck/Documents/temp/test_treedetection_pols.shp',
   max_npix = 50000,
   nWorkers = 4,
   parallel = T
   )
```

## Crown detection - polygons_to_seeds.r
#### Input
* A rough SpatialPolygons object of potential crowns
* Optionally, a multi-band raster image

#### Output
* A point shapefile denoting (suspected) tree crowns
* The seed can either be put at the polygon's centroid, or at the pixel of highest NDVI


```{r, eval=FALSE}

require(CanHeMonR)
polygons_to_seeds(
   rough_crowns_shp_fname = 'C:/Users/pieterbeck/Documents/temp/test_treedetection_pols.shp',
   seed_placement = 'maxval',
   image_fname = 'E:/FISE/forest/CanopyHealthMonitoring/PWN/flights_final/150727_mca/150727_mca.bsq',
   index_name = 'NDVI',
   nWorkers = 4,
   parallel = T,
   outp_shp_fname = 'C:/Users/pieterbeck/Documents/temp/test_treedetection_seeds.shp'
   )
```

## Filtering suspected crown that are too small to be a tree - remove_small_pols.r

## Filtering crown seeds that don't have a tree-like shadow - shadow_based_seed_filter.r

* if Ganguly's the tree non-tree mask is good, than the shadow-seaker might not be necessary...

```{r, eval=FALSE}
shadow_based_seed_filter(
  seed_shp_fname = 'C:/Users/pieterbeck/Documents/temp/test_treedetection_seeds.shp',
  image_fname 'E:/FISE/forest/CanopyHealthMonitoring/PWN/flights_final/150727_mca/150727_mca.bsq',
  shadow_mask = list(800, 801),
  x_shift = -4,
  y_shift = 4,
  shadow_diameter = 5,
  outp_seed_shp_fname = 'C:/Users/pieterbeck/Documents/temp/test_treedetection_seeds_shadow_masked.shp' )
```

### To Do
 * How does this perform in dense forest?
 * Very crappy trees hardly produce a shadow...

## Warping seed points for crown delineation - tps2dseeds.r
#### Challenge
New imagery doesn't always align perfectly with existing (reference) imagery, for which crown seeds (as points) are available. 
Rather than warp the new imagery to match the reference, this code attempts to warp the available seeds for crown delineation in the new imagery.
#### Input
* A .shp file with many points (seeds) placed in crowns of a reference image
* A .shp file with some points (seeds) placed in crowns of a new image
* A two-column table that gives for points in the new seed file, their respective seed number (FID in Arcgis) in the reference seed file

#### Output
A .shp file of the reference seed points warped using the available point-pairs. 
The X and Y coordinates of the original points in reference_seeds_shp are saved as the fields **primalX** and **primalY** in the attribute table of the output
The output is intended to grow crowns in the new image.
```{r, eval=FALSE}
workdir <- 'H:/FISE/forest/CanopyHealthMonitoring/PWN/flights_final/shp+stats/'
target_ref_ID_pairs <- read.csv(file.path(workdir,'150727/150727_mca/link_seeds_to_141120_rgb_all_trees.csv'))
target_ref_ID_pairs <- target_ref_ID_pairs[complete.cases(target_ref_ID_pairs),]

reference_seeds_shp <- 'H:/FISE/forest/CanopyHealthMonitoring/PWN/flights_final/shp+stats/141120/141120_rgb/141120_rgb_all_trees.shp'
target_seeds_shp <- 'H:/FISE/forest/CanopyHealthMonitoring/PWN/flights_final/shp+stats/150727/150727_mca/150727_mca_some_trees.shp'
outp_shp_fname <- 'H:/FISE/forest/CanopyHealthMonitoring/PWN/flights_final/shp+stats/150727/150727_mca/150727_mca_all_trees_by_tps.shp'
require(CanHeMonR)
tps2d_seeds(reference_seeds_shp, target_seeds_shp, target_ref_ID_pairs, outp_shp_fname)
```

## Crown delineation from seed points - region_growing_wrapper.r

#### Input

* A very high resolution image file
* A shape file of seeds (points) placed (manually) at the center of crowns

#### Output

* A polygon shape file of crowns grown around the seeds. 
* The polygons have as attributes primalX and primalY which contain the original coordinates of their seed. These are either passed from the attributes of the
seeds, or, if not present in the seeds' arguments, generated from the seed coordinates.

```{r, eval=FALSE}

#flight <- '150128_cir'
flight <- '141120_rgb_mosaic'
#flight <- '150128_hyper'
#flight <- '150727_rgb'
#flight <- '150727_mca'

#raster image to run crown growing on
r_file <- switch(flight,
                 '150128_cir' = 'H:/FISE/forest/CanopyHealthMonitoring/PWN/flights_final/150128_cir/150128_cir.bsq',
                 '141120_rgb_mosaic' ='H:/FISE/forest/CanopyHealthMonitoring/PWN/flights_final/141120_rgb_mosaic/141120_RGB_mosaic.bsq',
                 '150128_hyper' = 'H:/FISE/forest/CanopyHealthMonitoring/PWN/flights_final/150128_hyperspectral/150128_hyperspectral.bsq',
                 '150727_rgb' = 'H:/FISE/forest/CanopyHealthMonitoring/PWN/flights_final/150727_rgb/150727_rgb.bsq',
                 '150727_mca' = 'H:/FISE/forest/CanopyHealthMonitoring/PWN/flights_final/150727_mca/150727_mca.bsq')

#shapefile of seeds
seed_shpfile <- switch(flight,
                       '150128_cir' = 'H:/FISE/forest/CanopyHealthMonitoring/PWN/flights_final/shp+stats/150128/150128_cir/150128_cir.shp',
                       '141120_rgb_mosaic' = 'H:/FISE/forest/CanopyHealthMonitoring/PWN/flights_final/shp+stats/141120/141120_rgb/141120_rgb_all_trees.shp',
                       '150128_hyper' = 'H:/FISE/forest/CanopyHealthMonitoring/PWN/flights_final/shp+stats/150128/150128_hyper/150128_hyper.shp',
                       '150727_rgb' = 'H:/FISE/forest/CanopyHealthMonitoring/PWN/flights_final/shp+stats/141120/141120_rgb/141120_rgb_all_trees.shp',
                       '150727_mca' = 'H:/FISE/forest/CanopyHealthMonitoring/PWN/flights_final/shp+stats/150727/150727_mca/150727_mca_all_trees_by_tps.shp')

#the probability [in %] cut off in the chi square distribution that determines the cut-off in Mahalanobis distance
#higher prob.cut values produce larger crowns
prob_cut <- 99
#as long as the region grows all the way to the max.crown.radius, prob.cut will be stepped down by prob.step.down and the regino grown again
prob_step_down <- 2
#output shapefile to write crown polygons to
outp_crown_shp_filename <- switch(flight,
                                  '150128_cir' = '',
                                  '141120_rgb_mosaic' = 'E:/beckpie/temp/crown_141120_rgb_all_trees',
                                  '150128_hyper' = '',
                                  '150727_rgb' = 'E:/beckpie/temp/crown_150727_rgb_all_trees',
                                  '150727_mca' = 'E:/beckpie/temp/crown_150727_mca_all_trees')

require(CanHeMonR)
grow_crowns(r_file = r_file,
                        seed_shpfile = seed_shpfile,
                        avoid_recalculation = T,
                        prob_cut = 99,
                        prob_step_down = 2,
                        seed_nrs = NA,#3460:1e6,
                        outp_crown_shp_filename,
                        max_crown_RADIUS_in_m = 4,
                        min_crown_RADIUS_in_m = 0.9,
                        plott = F, RGBseq = c(1,2,3),
                        parallel = T, nWorkers = 22)


```

request up to nWorkers*170 trees for a 9 days period of processing

## Inspect crown delineation efficiency - evaluate_crown_growing_efficiency.r

#### Challenge

After running crown delineation, it is valuable to inspect how computational time was spent, and whether the parameters of the delineation
algorithm could be improved. When the prob_cut setting in region_growing_wrapper is too high, or the maximum crown diameter too small, then
the algorithm will grow the crown too large (ie up to the maximum diameter), reduce the cut-off for the similarity criterion (mahalanobis distance) and start growing the crown all over... until it doesn't grow up to the maximum diameter anymore. This can eat up a disproportionate amount of computational time. **evaluate_crown_growing_efficiency** allows you to inspect where computational time was spent.

#### Input

* A polygon shapefile as produced by region_growing_wrapper

#### Output

Graphs of distributions that show on which crowns most computational time is spent

```{r, eval=FALSE}
workdir <- 'H:/FISE/forest/CanopyHealthMonitoring/PWN/flights_final'
  workdir <- 'C:/Users/pieterbeck/Documents/temp/PWN/flights_final'
  
  #to be run for each of the flights
  flight <- '141120_rgb_mosaic'
  flight <- '150727_rgb' #700. 546, 435
  flight <- '150727_mca'
  
  outp_crown_shp_filename <- switch(flight,
                                    '150727_mca' = file.path(workdir,'shp+stats/150727/150727_mca/crown_150727_mca_all_trees_99.shp'),
                                    '141120_rgb_mosaic' = file.path(workdir,'shp+stats/141120/141120_rgb/crown_141120_rgb_all_trees'),
                                    '150727_rgb' = file.path(workdir,'shp+stats/150727/150727_rgb/crown_150727_rgb_all_trees')
  )
  evaluate_crown_growing_efficiency(outp_crown_shp_filename)
```

## Warping crowns using segmentation results

#### Challenge

Sometimes image segmentation is able to detect crowns well, yet when the image doesn't overlay well on the existing database of crowns (with the identifying attributes
primalX and primalY) then it is hard to link the segmented crowns with the database.
**shift_polygons_to_points** uses an existing polygon .shp of crowns and warps these to the nearest centroids of crowns estimated using segmentation.
The thin plate spline reprojection transformation can be calibrated using the centroids of the polygons to be warped, or the centroids of a different polygon data set. This is useful in the case the centroids to be warped have undergone crown trimming. In that cases their
untrimmed ancestor crowns can be used to calibrate the tps model.
If the exisiting polygon .shp has attributes like primalX and primalY they can be maintained in the output


#### Output
A crown polygon .shp for the segmented image, with primalX and primalY attributes.

```{r, eval=FALSE}
workdir <- 'H:/FISE/forest/CanopyHealthMonitoring/PWN/flights_final/'
Polygon_shp <- file.path(workdir,'/shp+stats/150727/150727_mca/crown_150727_mca_all_trees_99.shp')

wd <- 'H:/FISE/forest/CanopyHealthMonitoring/PWN/flights_final/shp+stats/150727/150727_flir'
point_shp <- file.path(wd,'indices_centroide2.shp')
outp_fname <- file.path(wd,'crown_150727_flir_all_trees_99_tps.shp')
extent_to_process <- raster::extent(c(747287,748000,4463535,4464600))
shift_polygons_to_points(polygon_shp, point_shp, outp_fname, extent_to_process = NA, attributes_to_retain = c('primalX' ,'primalY'))
```

## Mask the shadow sections of a crown - assign_crowncells_quantile / assign_crowncells_quantile_parallel

#### Challenge

Portions of crowns can be quite heavily shaded, comprising the mean spectral indices estimated for the crown. Furthermore, when crowns delineated in 
multispectral imagery, are used to estimated crown temperature in thermal imagery, it is particularly useful to exclude the cooler, shadow-section of the crown.
This code allows one to calculate for every pixel in a crown its quantile, ie rank, based on a single spectral band or index.
This value (e.g. NIR reflectance rank in the crown) can then be used as a mask while extracting thermal metrics from crowns.

#### Input
* A raster file to draw values from, and that will serve as the spatial template for the output
* A polygon shapefile depicting crowns that overlays r
* A filename for the raster output (only for non-parallel)
* An optional filename from .shp output of trimmed crown polygons
* The quantile values that are used to determine the trimmed crowns

#### Output
A raster file, with the same spatial attributes as r and values between 0 and 100 for cells in polygons, and NA outside polygons. (only for non-parallel)
An optional shp file of trimmed polygons.

```{r, eval=FALSE}
flight <- '150727_mca'
workdir <- 'H:/FISE/forest/CanopyHealthMonitoring/PWN/flights_final/'

r_layer <- switch(flight,
                 '150727_mca' = 
                  raster::raster(file.path(workdir, '/150727_mca/150727_mca.bsq'),band=5))
crown_shp <- switch(flight,
                  '150727_mca' = file.path(workdir,'/shp+stats/150727/150727_mca/crown_150727_mca_all_trees_99.shp'))
outp_tif <- switch(flight,
                   '150727_mca' = 'E:/beckpie/temp/crown_150727_mca_quantile_800.tif')

temp_outp <-'E:/beckpie/temp/trimmed.tif'
outpname_shp <- 'E:/beckpie/temp/trimmed.tif'
outpname_r <-  'E:/beckpie/temp/trimmed.shp'
require(CanHeMonR)

#debug(assign_crowncells_quantile)
test <- assign_crowncells_quantile(r = r_layer, crown_shp = crown_shp, outpname_r = outpname_r, outpname_shp = outpname_shp, plott=F, quantiles_to_keep = 66:100)
```


## Build a database of spectral indices for crowns - crown_indices_from_bigshape.r

This wrapper for spectral_indices_for_crowns.r allows you to build a master SpatialPolygonsDataFrame containing crown-level information 
obtained from different flights, sensors, and tree locations.

#### Input
* A SpatialPolygons(DataFrame) to extract values for
* An Image to extract values from
* There's an option to convert SpatialPolygons to SpatialPoints before extracting pixel values. This can be useful and save lots of time when the polygons are smaller than the cel size of the image. Default is FALSE.
* Names of flights for which to extract data. If 'ALL' then data from all flights will be extracted.
* Names of sensors for which to extract data. If 'ALL' then data from all sensors will be extracted.
* Names of Indices to calculate. If NULL (the default) only individual bands values are returned. This is useful when image_fname is a DEM for example.
* Optionally, the bandnames of the image_fname. If set to NULL (the default), bandnames will be read from image_fname

#### Output

* A SpatialPolygonsDataFrame.


## Calculate spectral indices for crowns - spectral_indices_for_crowns.r

Calculate one or more spectral indices for individual
tree crowns using a provided raster image and write the output to a .shp

#### Input
* Shapefile depicting crowns
* Raster file of the image to calculate the spectral index from.
The bands of the raster should be named following the Quantalab convention X700.000000.Nanometers
* Character vector with the names of a spectral indices implemented as a function in CanHeMonR. e.g. "G_over_B".
* outname Filename for a shapefile of the output

#### Output
A point shapefile with attribute table, with the points placed at the centroid of each crown.
The attribute table of the shp file contains a column for each spectral index.


```{r, eval=FALSE}
require(CanHeMonR)
flight <- '150727_mca'

crown_shp <- switch(flight,
                  '150727_mca' = 'E:/beckpie/temp/crown_150727_mca_all_trees_99.shp')
r_fname <- switch(flight,
                  '150727_mca' = 'H:/FISE/forest/CanopyHealthMonitoring/PWN/flights_final/150727_mca/150727_mca.bsq')

spectral_indices_for_crowns(crown_shp = crown_shp, r_fname = r_fname, index_names = c("G_over_B","BGI2","RDVI","NDVI"), 
                            outname = "E:\\beckpie\\temp\\output_example2.shp")

```

### A library of spectral indices - spectral_indices.r

Implementations of spectral indices that can be called from **spectral_indices_for_crowns**.
The same functions can be applied to data frames where columns represent spectral bands, or raster bricks,
where layers do the same, as long as the columns/layers are named using the convention *X700.000000.Nanometers*

```{r, eval=FALSE}
# Create and NDVI image from a multispectral image
my_ms_data <- raster::brick('H:/FISE/forest/CanopyHealthMonitoring/PWN/flights_final/150727_mca/150727_mca.bsq')

require(CanHeMonR)
my_NDVI <- NDVI(my_ms_data, outp_fname = 'E:/beckpie/temp/crown_150727_mca_NDVI.tif')

# Calculate NDVI from a data frame extracted per-polygon from a multipsectral image
```



## Merge spectral indices calculated from different images - merge_pnt_shp_files_by_attributes.r

The extraction of spectral indices from tree crowns is done independently for different images, using their particular crown data set.
**spectral_indices_for_crowns.r** ensures that during the index extraction process the original id of the crown is maintained (through the 'primal_X'
and 'primal_Y' attributes).
**merge_pnt_shp_files_by_attributes** combines the indices extracted from different images into a single point .shp.

The attributes in the output have a 'i#' prefex referring
to the input it drew from. It also writes away a text file that links the ix numbering in the attribute names of the .shp back to the filenames
of the single-image files. The filename of the text file is outp_shp with '.txt' appended

### Visualize result of merger - pairsplot

**pairsplot** A customized pairsplot function that allows you to only display particular attributes of the merged shapefile.

## Take a statistical sample of the crowns, stratified by multiple spectral indices

For field inspections and validation, not all mapped crowns can be inspected. Therefore a sample of the crown population needs to be drawn
that covers as exhaustively as possible, the range of values of _all_ the spectral indices.

```{r, eval=FALSE}
flight <- '150727_mca'
workdir <- 'H:/FISE/forest/CanopyHealthMonitoring/PWN/flights_final'
inp_shp <- switch(flight,
                  '150727_mca' = file.path(workdir,'shp+stats/150727/150727_mca/crown_150727_mca_all_trees_99_indices.shp'))
shp_outname <-  switch(flight,
                  '150727_mca' = 'E:/beckpie/temp/testsamp.shp') 
require(CanHeMonR)
tt <- samplestratified_shp(shp_fname = inp_shp, shp_outname, plott=T,nsamples=100)

```

## Visually inspect crowns - crown_inspector.r

####
To ensure that crown delineation was succesful, it is useful to plot delineated crowns on a background of the image used to create it,
and along with the seed used to start the delineation.
Furthermore, it is useful to this delineation for the representation of an individual crown in multiple image sources (e.g. rgb and multispectral).

#### Input
* Point shapefiles of seeds
* Matching image filenames used to delineate the crowns
* Polygon shapefile depicting the result of the crown delineation

#### Output
* Plots of the inputs written to a seperate file for each crown in a directory of choice

## PWN_functions.R
Supporting functions used a.o. in **decline_trees_compile_RS.r** and in **decline_trees_RS_analysis.r**

### Avoid Data Frames With Too Many Rows - samp_df
### Clip Spatialpolygons To The Extent Of A Raster - clip_polygons2rasters
### Find A Polygon's Neighbour Polygons - identify_n_neighbours
### Subtract Band And Polygon Specific Median From Digital Numbers - subtract_control_median
### Tile the extent object of a raster object for tiled processing - tile_raster_extent
### Eliminate the effect of tiled processing on delineated polygons - merge_pols_intersected_by_lines
### Convert a polygon to the point where a raster is at its maximum - point_at_highest_value_in_polygon

## PWN_plotting_functions.R
Functions to mask to the PWN buffer zone or to add it to a map. Used in **??**

### Add Outline of the PT PWN Buffer Zone to an Existing Plot - add_PT_buffer_parishes_borders
### Mask a Raster File to Only Retain Data within the PT PWN Buffer Zone - mask_to_PT_buffer_zone
## segment_classification_functions.R
Functions that support the classification of segments into target objects

## fit_function_to_canopy_health.index.r
Fits a simple polynomial of the form y ~ scale*(x^rate)+interc
 to a time series of canopy health index (x) using nls
(called from **decline_trees_RS_analysis**)

### region_growing_functions.r
Executes region growing from windows of single or multi-layer rasters around seed points
(called from **region_growing_wrapper.r**)
calls **remoteness_functions.r**

### remoteness_functions.r
(called from **region_growing_functions.r**)


#### Input
* Polygon .shp with tree crowns
* Raster image to calculate spectral indices from
* Names of spectral indices to calculate
* Name of the output shapefile

#### Output 
Output will be generated as a point shapefile with attribute table, with the points placed at the centroid of each crown.
The attribute table of the shp file contains a column for each spectral index.

# Workflow

# Summarizing PT-ESP buffer zone

## fty_PTES_border_zones.r
Calculates 25 m FTY maps (services element 2 of high res forest layers) for the official buffer zone along the PT-ES border, and for 20 km and 40 km buffer zones either side of the border. It saves tables of FTY (in km2) within the buffer zones

#### output
"E:/beckpie/PWN/R_code/rdata/fty_PTES_border_zones_tables.rdata"

## fty_PTES_border_zones.Rmd
Summarizes the forest cover in the official and 20  and 40 km buffer zones

# Flight campaign planning
## Areas_to_fly_summer_2015.r
Creates and measures the 40.000 ha area designated for imaging this summer. It then cuts it up into lots and blocks and gives summary statistics for each

## Flight_cost_TravellingSalesman_approach.r
Simulating cost of covering buffer zone from aircraft. Calculate the cost/distance required to cover a progressive percentage of coniferous pixels in the PT buffer zone.

## Handling ortophoto collections

### create_overview_of_rasters.r

#### Challenge
When vizualizing in ArcGis a collection of ortophotos it can be hard to figure out which photo covers a particular location.
_create_overview_of_rasters.r_ creates a polygon shapefile of all the rasters in a directory, with filenames as attributes.
These can be used as labels in vizualisations or in queries.


#### Output
Tables with required flight distance to cover a given % of coniferous cover in a chosen area. Different output files 
are produced for each threshold beyond which a coarse (1 km pixel) is considered 'coniferous'
"E://beckpie//PWN//R_code//rdata//flight_length_simulation_",aggregation.minimum.cover,"perc_aggregation_v2.rdata"

# Feasibility study 2014-2015 

## Spain_field_sites_size_calculations.r
Calculate the exact area of the Granadilla forest site

## Field data summary - Granadilla_forest_field_data_summary.Rmd
Summarizes the field survey data from Granadilla forest

#### Input
"E:/beckpie/PWN/R_code/rdata/fty_PTES_border_zones_tables.rdata"

## Graphs for feasibility study report

### boxplot_indice_hyper.r & boxplot_indice_cir.r
Generates figure for the pilot study report. Plots hyperspectral indices per field-observed canopy decline level.
The hyperspectral index values are those provided in tab format by Pablo.

### field_dat_graphs_for_report.r
Summarizes the field data for the pilot study report


# Processing image segmentation results

### classify_segmentation_supervised.r
rF Classifies a polygon shapefile  into 'EG' and 'NO' based on a training field called 'veg'(with values 'EG' and 'NO') and predictor
variables in the other fields (band statistics for the polygons generated by Definiens).
TO DO: Alternatively and additionally, an NDVI cut-off can be applied to the polygons in test_decline_trees_complie_RS.r

#### Input
polygons that result from image segmentation with polygon level attributes (e.g. mean and sd radiance per band) as attributes

#### Output
Polygon shapefile with only the polygons classified as 'EG' by the rF. These do NOT display properly in Arcgis !?

## Case studies

### PTtest_decline_trees_compile_RS.r
Executes the workflow for the PT test area (summer 2014). This site had insufficient field data

#### executes
decline_trees_compile_RS.r

### PTtest_decline_trees_RS_analsysis.r
Executes the workflow for the PT test area (summer 2014). This site had insufficient field data

#### executes
decline_trees_RS_analysis.r

## gathering RS data over segments

### decline_trees_compile_RS.r
#### Input
shapefile of coniferous forest patches from classify_segmentation.r
names of relevant RS data

* choose the maximum nr of RS pixels per polygon
max.samp.per.polygon <- 1000

* choose the inward buffering to apply to the polygons
inward.buffer <- -10 #in m (the units of the polygon data)

* minimum polygon area after buffering*
min.polarea <- 10*10

* choose the nr of neighbouring polygons to consider when checking for change
n.neighbours <- 8

* RS data
in lists with elements, date, fname (filenames of raster files), sensors, bands

#### Output
* Decline.list
list for each patch/polygon
containing attributes from shapefile (ie declinging trees or  confirmed no declining trees
and RS data
* Decline.cluster
SpatialPolygons of original EG patches

## Detecting change

### decline_trees_RS_analysis.r
#### Input
list created by decline_trees_compile_RS.r

.shp file with chosen attributes for each patch/polygon, based on comparison of RS data for a polygon and its neighbours

# TerraBella processing

## get_TerraBella_scene_date.r

# Sandbox

## crowns/crown_inspector.r

## Project_PT_nuts_web_app.r
prepares the PT nuts for upload in http://geoportal.lneg.pt/geoportal/mapas/index.html