Homologación coberturas (CORINE) a Habitat (IUCN)
Luis H. Romero
28/1/2021
Summary
The R programming language and environment can be really useful for geographic data analysis and modeling due to its capacity of reading, creating, manipulating and analyzing of spatial data with general raster data manipulation functions (Hijmas, 2020; Hijmas, 2021), making static and interactive maps, applying geocomputation to solve real-world problems, etc. (Lovelace, Nowosad & Muenchow, 2021). We can handle several research objectives related with remote sensing, spatial ecology, land use and land cover issues, agriculture, urban expansion, climate change, biological invasions, etc., that involves multitemporal information either in raster or shape data. In this case, we want to get habitat information from land cover data of the continental part of Colombia, as an external input data to improve species distributions models (SDM). The land cover product used here is CORINE Land Cover adjusted to Colombia, and the output result is a habitat map as defined in the International Union for Conservation of Nature (IUCN) habitat classification scheme (Jung et al., 2020). The habitat map had to be reclassified to set links between CORINE land covers and IUCN habitats.
Introduction
The patterns that shape the spatial and temporal distribution of the species is one of the most fundamental pieces of ecological information (Beale & Lennon, 2012). Species distributions models (SDMs) explore the relationship between geographical occurrences of species and corresponding environmental variables (Naimi & Araujo, 2016) and essentially are based in spatial (and some in temporal) interpolation and extrapolation for the present or into the future. The accuracy of each model will depend on how can handle the different processes that a species have to occur within an ecological community like dispersal, environmental tolerance, species interactions, biogeographical history, etc.(Li et al., 2020).
Species distribution modeling has grown rapidly in the last years and with these the software and tools available to display this analysis (specially MaxEnt standalone applications and R programming packages like sdm, zoon R, phyr, etc), and select the most reliable variables for efficient calibrations (kuenm, ENMTools, ENMeval, MaxentVariableSelection, etc) (Golding et al., 2017; Cobos et al., 2019a; 2019b). Each model with its combination of parameters can identify uncertainty in different levels to see where future improvements can be made to diminish overprediction or subprediction on species distribution. One of the ways to identify sources of uncertainty is through acknowledging the species habitat.
According to the definition of Kearney (2006), habitat is a description of a physical place, at a particular scale of space and time, where an organism either actually or potentially lives. So we have an approximation of the habitat if we know the land covers within a particular area of interest, and this combined with sufficient ecological information of a specie could be an interesting input to evaluate a candidate model. This work flow is the second of two documents indicating the process to homologate land covers with habitats. This document will show the homologation between shape land cover data from CORINE Land Cover Colombia and habitat classification scheme of International Union for Conservation of Nature (IUCN).
Package overview
The functions used here are from different packages to read, manipulate and write geographical data in shape and raster format. This libraries and functions are as follows:
- library(sf): refers to a formal standard that describes objects in the real world as a computer representation
- read_sf() read shapefiles with a format easily to read for fasterize() function
- library(dplyr) grammar of data manipulation
- mutate() create, modifiy and delete columns, necessary for the reclassification of the shapefile
- library(fasterize) fast polygon to raster conversion, improve version of rasterize() function
- library(raster)
- raster() load the raster document with land cover data.
- reclassify() the land cover data to make the homologation.
- projectRaster() change the projection of the input raster into another raster of interest.
- resample() transfers values between non matching Raster objects in terms of origin and resolution.
- overlay() create a new raster based on two or more raster objects (Similar to raster calculator in SIG software).
- rasterToPoints() Raster to point conversion.
- rasterFromXYZ() Create a Raster object with x,y and z values, being x and y the spatial coordinates.
- writeRaster() write an entire Raster object to a file.
CORINE Land Cover Colombia
As well as ESA’s products are good references of coverage at a global level, in the case of Colombia there is the National Legend of Land Cover, CORINE Land Cover methodology adapted for Colombia scale 1: 100,000 which consolidates the characterization of the natural and anthropized covers present in the Colombian territory and has been an input for the study of various national processes such as ecosystem maps, land use conflict, watershed and territory management, monitoring of forest deforestation, forest inventories, etc. (IDEAM, 2010). This coverage map can be downloaded for free from the IDEAM website http://www.ideam.gov.co/web/ecosistemas/coberturas-nacionales. The same homologation exercise between covers and habitats will be carried out as was presented in the first document with the ESA product, only this time it will be based on Shapefile data.
corine <- read_sf("CbTr_2010_2012.shp")Shapefile preparation (Optional)
Variables from the Shapefile that were added in other routines will be removed. This step does not need to be replicated.
colnames(corine)## [1] "OBJECTID_1" "OBJECTID" "APOYO" "CAMBIO" "CONFIABILI"
## [6] "INSUMO" "CODIGO" "LEYENDA3N" "Area_ha" "Cod"
## [11] "Categoria" "Cod_Finer" "Cod_ESA" "Cod_Coper" "geometry"eliminar <- c("OBJECTID_1", "APOYO", "CAMBIO", "CONFIABILI", "INSUMO", "Cod_ESA", "Cod_Coper", "Corine_hab")
corine_2 <- corine[, !(names(corine) %in% eliminar)]Homologation
Here we present the proposal of homologation between CORINE Land Cover and IUCN Habitat classification. For the most of the habitat classes we used the third level of Corine´s legend, but some important habitats are best differentiated with fourth levels or higher land cover codes. That is why in the next code we used in some cases the “LEYENDA3N” variable and in others the “CODE” variable of the input shapefile.
corine_2 <- corine_2 %>%
mutate(
habitat = case_when(
LEYENDA3N %in% "1.1.1. Tejido urbano continuo" ~ 1404,
LEYENDA3N %in% "1.1.2. Tejido urbano discontinuo" ~ 1404,
LEYENDA3N %in% "1.2.1. Zonas industriales o comerciales" ~ 1404,
LEYENDA3N %in% "1.2.2. Red vial, ferroviaria y terrenos asociados" ~ 1404,
LEYENDA3N %in% "1.2.3. Zonas portuarias" ~ 1404,
LEYENDA3N %in% "1.2.4. Aeropuertos" ~ 1404,
LEYENDA3N %in% "1.2.5. Obras hidraulicas" ~ 1404,
LEYENDA3N %in% "1.3.1. Zonas de extraccion minera" ~ 1404,
LEYENDA3N %in% "1.3.2. Zona de disposicion de residuos" ~ 1404,
LEYENDA3N %in% "1.4.1. Zonas verdes urbanas" ~ 1404,
LEYENDA3N %in% "1.4.2. Instalaciones recreativas" ~ 1404,
LEYENDA3N %in% "2.1.1. Otros cultivos transitorios" ~ 1401,
LEYENDA3N %in% "2.1.2. Cereales" ~ 1401,
LEYENDA3N %in% "2.1.3. Oleaginosas y leguminosas" ~ 1401,
LEYENDA3N %in% "2.1.4. Hortalizas" ~ 1401,
LEYENDA3N %in% "2.1.5. Tuberculos" ~ 1401,
LEYENDA3N %in% "2.2.1. Cultivos permanentes herbaceos" ~ 1401,
LEYENDA3N %in% "2.2.2. Cultivos permanentes arbustivos" ~ 1401,
LEYENDA3N %in% "2.2.3. Cultivos permanentes arboreos" ~ 1401,
LEYENDA3N %in% "2.2.4. Cultivos agroforestales" ~ 1401,
LEYENDA3N %in% "2.2.5. Cultivos confinados" ~ 1401,
LEYENDA3N %in% "2.3.1. Pastos limpios" ~ 1402,
LEYENDA3N %in% "2.3.2. Pastos arbolados" ~ 1402,
LEYENDA3N %in% "2.3.3. Pastos enmalezados" ~ 1402,
LEYENDA3N %in% "2.4.1. Mosaico de cultivos" ~ 1401,
LEYENDA3N %in% "2.4.2. Mosaico de pastos y cultivos" ~ 1401,
LEYENDA3N %in% "2.4.3. Mosaico de cultivos, pastos y espacios naturales" ~ 1401,
LEYENDA3N %in% "2.4.4. Mosaico de pastos con espacios naturales" ~ 1402,
LEYENDA3N %in% "2.4.5. Mosaico de cultivos con espacios naturales" ~ 1401,
CODIGO %in% 311 ~ 105,
CODIGO %in% 3111 ~ 105,
CODIGO %in% 31111 ~ 105,
CODIGO %in% 31112 ~ 107,
CODIGO %in% 311121 ~ 107,
CODIGO %in% 311122 ~ 107,
CODIGO %in% 311123 ~ 107,
CODIGO %in% 3112 ~ 105,
CODIGO %in% 31121 ~ 105,
CODIGO %in% 31122 ~ 107,
CODIGO %in% 312 ~ 105,
CODIGO %in% 3121 ~ 105,
CODIGO %in% 31211 ~ 105,
CODIGO %in% 31212 ~ 107,
CODIGO %in% 3122 ~ 105,
CODIGO %in% 31221 ~ 105,
CODIGO %in% 31222 ~ 107,
LEYENDA3N %in% "3.1.3. Bosque fragmentado" ~ 105,
LEYENDA3N %in% "3.1.4. Bosque de galeria y ripario" ~ 107,
LEYENDA3N %in% "3.1.5. Plantacion forestal" ~ 1401,
CODIGO %in% 321 ~ 405,
CODIGO %in% 3211 ~ 405,
CODIGO %in% 32111 ~ 405,
CODIGO %in% 321111 ~ 405,
CODIGO %in% 321112 ~ 405,
CODIGO %in% 321113 ~ 405,
CODIGO %in% 32112 ~ 406,
CODIGO %in% 321121 ~ 406,
CODIGO %in% 321122 ~ 406,
CODIGO %in% 321123 ~ 406,
CODIGO %in% 3212 ~ 405,
CODIGO %in% 32121 ~ 405,
CODIGO %in% 32122 ~ 405,
LEYENDA3N %in% "3.2.2. Arbustal" ~ 305,
LEYENDA3N %in% "3.2.3. Vegetacion secundaria o en transicion" ~ 1406,
LEYENDA3N %in% "3.3.1. Zonas arenosas naturales" ~ 600,
LEYENDA3N %in% "3.3.2. Afloramientos rocosos" ~ 600,
LEYENDA3N %in% "3.3.3. Tierras desnudas y degradadas" ~ 600,
LEYENDA3N %in% "3.3.4. Zonas quemadas" ~ 600,
LEYENDA3N %in% "3.3.5. Zonas glaciares y nivales" ~ 600,
LEYENDA3N %in% "4.1.1. Zonas Pantanosas" ~ 600,
LEYENDA3N %in% "4.1.2. Turberas" ~ 600,
LEYENDA3N %in% "4.1.3. Vegetacion acuatica sobre cuerpos de agua" ~ 600,
LEYENDA3N %in% "4.2.1. Pantanos costeros" ~ 1300,
LEYENDA3N %in% "4.2.2. Salitral" ~ 1300,
LEYENDA3N %in% "4.2.3. Sedimentos expuestos en bajamar" ~ 1300,
LEYENDA3N %in% "5.1.1. Rios (50 m)" ~ 500,
LEYENDA3N %in% "5.1.2. Lagunas, lagos y cienagas naturales" ~ 500,
LEYENDA3N %in% "5.1.3. Canales" ~ 1500,
LEYENDA3N %in% "5.1.4. Cuerpos de agua artificiales" ~ 1500,
LEYENDA3N %in% "5.2.1. Lagunas costeras" ~ 1300,
LEYENDA3N %in% "5.2.2. Mares y oceanos" ~ 900,
LEYENDA3N %in% "5.2.3. Estanques para acuicultura marina" ~ 1500,
TRUE ~ 1700
)
)Land Cover rasterization with fasterize function
As an input for uncertainty protocols in species distribution models, habitat information from land covers can be more easily manipulated in raster format. Since a considerably large area is being shaped to raster, it is recommended to use the fasterize function, which is an enhancement to the rasterize function, processing the supplied shapefile information about a hundred times faster. In the raster_template attribute of the fasterize function, the habitat map built with the ESA land cover data will be located, or any other raster with the desired extent, dimensions and resolution can be placed.
raster_template <- raster("Habitats_ESA.tif") # If you want to upload a raster file
#raster_template <- esa_hab #If you already have your raster template
clc_raster <- fasterize(corine_2, raster_template, field = "habitat", fun = "sum")
plot(clc_raster, main= "Homologación Coberturas-Hábitats CORINE")
Add altitude variable
The IUCN has habitats with a very particular geographic delimitation for Colombia, but it is not possible to assign a coverage category to them. These habitats include 1.9. Forest - Subtropical/Tropical moist montane, 3.7 Shrubland - High altitude and 4.7 Grassland - High altitude. Many species use these habitats in Colombia, so they must be differentiated; For this, a DEM of 90 meters will be used that will allow the delimitation of these habitats. To do this correctly, the DEM must be reprojected and have the same extent, resolution and dimensions as the ESA map.
dem_colombia <- raster("dem_colombia3.tif")
dem_proj <- projectRaster(dem_colombia, crs = crs(clc_raster))
dem_al <- resample(dem_proj, clc_raster, method= "ngb")It was proposed to establish a limit of 500 meters to differentiate lowland from montane habitats.
corine_hab <- overlay(dem_al, clc_raster,
fun= function(altura, cover){
ifelse((altura >= 500) & (cover == 105),109,
ifelse((altura >= 500) & (cover == 305),307,
ifelse((altura >= 500) & (cover == 405),407,cover)))
})
plot(corine_hab, main= "Habitats_CORINE_300m") #Visualmente se ve igual al anterior pero es por la disposición de los códigos
Map visualization
This is a representation with colors more suited to the kinds of habitats that can be generated. However, as it requires an additional transformation of the data, it is suggested to work with the map corine_hab.
dfc <- data.frame(rasterToPoints(corine_hab))
raster_testc <- rasterFromXYZ(dfc[1:3])
cut <- sort(unique(corine_hab))
mypal <- c("darkcyan", "darkgreen", "darkorchid", "red", "chartreuse", "aquamarine3", "cyan", "blue3", "darkorange", "black", "darkkhaki", "hotpink", "blanchedalmond", "black")
plot(raster_testc,
main= "Habitats_CORINE_300m",
col= mypal,
breaks= cut)
Create raster file
This habitat map built from ESA covers can be an important input in updating and refining routines for the construction of species distribution models. For this reason, it is important to save a copy in .tif format in the working directory.
#writeRaster(corine_hab, "Habitats_CORINE.tif", overwrite= T)Confusion matrix between land cover products
This routine corresponds to the second one referring to homologation between land covers and habitats. Two homologation proposals have been built from ESA and CORINE products, and one way to compare them is through a confusion matrix. In this way, the degree of similarity that these two products have when studying the habitat of a species will be determined.
classes_esa <- sort(unique(raster_template))
classes_corine <- sort(unique(corine_hab))
classes_habitats <- sort(unique(c(classes_esa, classes_corine)))
rf1_corine <- factor(corine_hab[],levels= classes_habitats)
rf2_esa <- factor(raster_template[],levels= classes_habitats)
confusionMatrix(rf1_corine,rf2_esa)## Confusion Matrix and Statistics
##
## Reference
## Prediction 101 105 107 109 305 307 405 406
## 101 0 0 0 0 0 0 0 0
## 105 33 4241130 20644 4826 8771 13 12251 4013
## 107 30 477756 65827 22665 4526 8577 111529 11402
## 109 34 4803 2422 953892 14 7803 47 513
## 305 0 50693 2722 291 18166 71 21783 499
## 307 9 284 237 62124 75 23835 8 149
## 405 31 67990 3321 50 17238 78 561377 13337
## 406 13 26024 19042 34 2437 27 283164 163376
## 407 5 70 179 34066 109 80282 32 2663
## 500 12 28897 9271 3772 2596 618 4828 15039
## 600 24 18426 9242 1178 16557 3429 36898 26368
## 900 0 41 67 0 0 0 2 14
## 1300 92 93 1602 0 90 0 129 614
## 1401 19 150220 11875 333998 14546 80739 73783 4000
## 1402 13 288453 9771 369901 96394 166728 195642 9903
## 1404 13 3072 310 1491 1735 1286 2656 980
## 1406 4 176960 6439 147996 11177 11829 10149 888
## 1500 0 338 254 359 77 36 34 160
## 1700 0 721 234 6783 14 324 0 40
## Reference
## Prediction 407 500 600 900 1300 1401 1402 1404
## 101 0 0 0 0 0 0 0 0
## 105 37 1643 1070 0 0 12364 3974 50
## 107 742 6041 223 0 0 21631 3965 68
## 109 4356 30 208 0 0 10180 2209 58
## 305 6 385 17219 0 0 22214 2700 90
## 307 9364 13 103 0 0 2064 81 55
## 405 20 180 3784 0 0 1192 347 66
## 406 0 2820 320 0 0 15096 1098 156
## 407 83249 41 1490 0 0 3187 283 237
## 500 766 95311 166 0 0 7829 764 243
## 600 6673 10115 27732 0 0 16558 1977 365
## 900 0 127 2 0 0 2 0 0
## 1300 0 2938 2428 0 0 19 0 19
## 1401 20705 1891 1566 0 0 157569 27408 1364
## 1402 32363 3278 406 0 0 523120 196295 1575
## 1404 1385 348 604 0 0 8621 705 19866
## 1406 2200 1212 37 0 0 24725 5991 56
## 1500 21 3340 34 0 0 113 11 46
## 1700 110 0 46 0 0 317 112 4
## Reference
## Prediction 1406 1500 1700
## 101 0 0 0
## 105 20709 0 0
## 107 57830 0 0
## 109 11049 0 0
## 305 25455 0 0
## 307 7894 0 0
## 405 94165 0 0
## 406 31575 0 0
## 407 11049 0 7
## 500 6145 0 0
## 600 20809 0 692
## 900 2 0 0
## 1300 32 0 0
## 1401 108115 0 0
## 1402 255912 0 0
## 1404 4255 0 0
## 1406 21889 0 0
## 1500 117 0 0
## 1700 227 0 0
##
## Overall Statistics
##
## Accuracy : 0.5563
## 95% CI : (0.556, 0.5566)
## No Information Rate : 0.4645
## P-Value [Acc > NIR] : < 2.2e-16
##
## Kappa : 0.4426
##
## Mcnemar's Test P-Value : NA
##
## Statistics by Class:
##
## Class: 101 Class: 105 Class: 107 Class: 109 Class: 305
## Sensitivity 0.000e+00 0.7661 0.402713 0.49083 0.093388
## Specificity 1.000e+00 0.9858 0.938152 0.99562 0.987706
## Pos Pred Value NaN 0.9791 0.083030 0.95617 0.111933
## Neg Pred Value 1.000e+00 0.8293 0.991224 0.90938 0.984998
## Prevalence 2.786e-05 0.4645 0.013716 0.16307 0.016322
## Detection Rate 0.000e+00 0.3559 0.005523 0.08004 0.001524
## Detection Prevalence 0.000e+00 0.3635 0.066524 0.08371 0.013618
## Balanced Accuracy 5.000e-01 0.8760 0.670432 0.74322 0.540547
## Class: 307 Class: 405 Class: 406 Class: 407 Class: 500
## Sensitivity 0.061801 0.42713 0.64332 0.513892 0.734784
## Specificity 0.992850 0.98097 0.96727 0.988627 0.993133
## Pos Pred Value 0.224234 0.73558 0.29967 0.383726 0.540750
## Neg Pred Value 0.969365 0.93250 0.99204 0.993270 0.997070
## Prevalence 0.032361 0.11028 0.02131 0.013593 0.010884
## Detection Rate 0.002000 0.04710 0.01371 0.006985 0.007997
## Detection Prevalence 0.008919 0.06404 0.04575 0.018204 0.014789
## Balanced Accuracy 0.527325 0.70405 0.80529 0.751260 0.863958
## Class: 600 Class: 900 Class: 1300 Class: 1401 Class: 1402
## Sensitivity 0.482816 NA NA 0.19058 0.79177
## Specificity 0.985725 1.000e+00 0.999324 0.92514 0.83261
## Pos Pred Value 0.140741 NA NA 0.15952 0.09131
## Neg Pred Value 0.997466 NA NA 0.93877 0.99471
## Prevalence 0.004820 0.000e+00 0.000000 0.06938 0.02080
## Detection Rate 0.002327 0.000e+00 0.000000 0.01322 0.01647
## Detection Prevalence 0.016534 2.156e-05 0.000676 0.08288 0.18038
## Balanced Accuracy 0.734270 NA NA 0.55786 0.81219
## Class: 1404 Class: 1406 Class: 1500 Class: 1700
## Sensitivity 0.816926 0.032321 NA 0.000e+00
## Specificity 0.997691 0.964445 0.9995855 9.993e-01
## Pos Pred Value 0.419760 0.051925 NA 0.000e+00
## Neg Pred Value 0.999625 0.942995 NA 9.999e-01
## Prevalence 0.002040 0.056825 0.0000000 5.865e-05
## Detection Rate 0.001667 0.001837 0.0000000 0.000e+00
## Detection Prevalence 0.003971 0.035372 0.0004145 7.495e-04
## Balanced Accuracy 0.907308 0.498383 NA 4.996e-01It is observed that the two land cover products have a similarity of only 55.63%. This difference must be taken into account when choosing a land cover product as an approximation to the habitat of the species, depending on whether what is desired is a better discrimination between habitats (Offered by CORINE), or if monitoring is preferred at different instants of time (Offered by ESA and its annual update).
References
Beale, C. M. & Lennon, J. J. (2012). Incorporating uncertainty in predictive species distribution modelling. Philosophical Transactions of the royal society B, 367, 247-258.
Cobos, M. E., Peterson, A. T., Barve, N. & Osorio-Olvera, L. (2019a). kuenm: an R package for detailed development of ecological niche models using Maxent. Bioinformatics and genomics. PeerJ 7:e6281 https://doi.org/10.7717/peerj.6281
Cobos, M. E., Peterson, A. E., Osorio-Olvera, L. & Jiménez-García, D. (2019b). An exhaustive analysis of heuristic methods for variable selection in ecological niche modeling and species distribution modeling. Ecological Informatics, 53, 1-10.
Golding, N., August, T. A., Lucas, T. C. D., Gavaghan, D. J., van Loon, E. E. & McInerny, G. (2017). The zoon r package for reproducible and shareable species distribution modelling. Methods in ecology and evolution, 9, 260-268.
Hijmas, R. (2020). Raster package v.3.4-5. Available in https://www.rdocumentation.org/packages/raster/versions/3.4-5
Hijmas, R. (2021). The raster package. Available in https://rspatial.org/raster/pkg/RasterPackage.pdf
IDEAM (2010). Leyenda Nacional de Coberturas de la Tierra. Metodología CORINE Land Cover, adaptada para Colombia Escala 1:100.000. Instituto de Hidrología, Metereología y Estudios Ambientales. Bogotá, D.C., 72 p.
Jung, M., Dahal, P. R., Butchart, S. H. M., et al. (2020). A global map of terrestrial habitat types. Nature - Scientific data, 7:256, 1-8
Kearney, M. (2006). Habitat, environment and niche: what are we modelling?. OIKOS, 115, 186-191.
Li, D., Dinnage, R., Nell, L. A., Helmus, M. R. & Ives, A. R. (2020). phyr: An r package for phylogenetic species‐distribution modelling in ecological communities. Methods in ecology and evolution, 11, 1455-1463.
Lovelace, R., Nowosad, J. & Muenchow, J. (2021). Geocomputation with R. The R series. CRC Press Taylor & Francis Group. Available in: https://geocompr.robinlovelace.net/
Naimi, B. & Araújo, M. B. (2016). sdm: a reproducible and extensible R platform for species distribution modelling. Ecography, 39, 368-375.