New Caledonia¶

Downloading data in parallel¶

We can use geefcc to download forest cover change for large countries, for example New-Caledonia. The country will be divided into several tiles which are processed in parallel. If your computer has n cores, n-1 cores will be used in parallel.

import os
from pathlib import Path
import time

import ee
import geefcc
import pandas as pd
from tabulate import tabulate

We initialize Google Earth Engine.

# Initialize GEE
ee.Initialize(project="deforisk",
              opt_url=("https://earthengine-highvolume."
                       "googleapis.com"))

We can compute the number of cores used for the computation.

ncpu = os.cpu_count() - 1
ncpu

Using TMF product¶

Downloading data¶

We download the forest cover change data from GEE for New Caledonia for years 2001, 2010 and 2020, using a tile size of one degree. We use the TMF product.

years_tmf = [2001, 2010, 2020]
out_dir_tmf = Path("out_tmf")
ofile_tmf = out_dir_tmf / "forest_tmf.tif"

start_time = time.time()
if not ofile_tmf.is_file():
    geefcc.get_fcc_loss(
        aoi=(163.5, -23, 168.15, -19.51),
        buff=0.0,
        years=years_tmf,
        source="tmf",
        tile_size=1.0,
        crop_to_aoi=True,
        output_file=ofile_tmf,
        parallel=True,
        ncpu=ncpu,
    )
end_time = time.time()

We estimate the computation time to download 20 1-degree tiles using several cores.

elapsed_time = (end_time - start_time) / 60
print('Execution time:', round(elapsed_time, 2), 'minutes')

Execution time: 1.2 minutes

Transform multiband fcc raster in one band raster¶

We transform the data to have only one band describing the forest cover change with 0 for non-forest, 1 for deforestation on the period 2001–2009, 2 for deforestation on the period 2010–2019, and 3 for the remaining forest in 2020. To do so, we just sum the values of the raster bands.

fcc_file_tmf = out_dir_tmf / "fcc_tmf.tif"
if not fcc_file_tmf.is_file():
    geefcc.sum_raster_bands(
        input_file=ofile_tmf,
        output_file=fcc_file_tmf,
        verbose=False,
    )

Plot the forest cover change map¶

We plot the forest cover change map. The raster is automatically resampled to a coarser resolution before plotting as it exceeds the default pixel threshold. Country borders and the tile grid are overlaid directly from their vector files.

geefcc.plot_fcc_loss(
    input_file=fcc_file_tmf,
    years=years_tmf,
    output_file="fcc_tmf.png",
    title="Forest cover change 2001-2010-2020, TMF",
    dpi=200,
    borders=Path("data") / "borders_NCL.gpkg",
    grid=out_dir_tmf / "grid.gpkg",
    xlim=(163, 169),
    ylim=(-23.25, -18.75),
)

Lines in black represent country borders. One degree tiles in grey cover the whole buffer and were used to download the data in parallel.

Reproject in EPSG:3163 for area computation¶

res_df_tmf = geefcc.stat_fcc_loss(
    input_file=fcc_file_tmf,
    years=years_tmf,
    epsg=32758,
    output_file="fcc_statistics_tmf.csv"
)
tabulate(res_df_tmf, headers=res_df_tmf.columns, tablefmt="orgtbl", showindex=False)

category	label	count	area_ha
0	non-forest in 2001	197249917	17752493
1	deforestation 2001-2010	292122	26291
2	deforestation 2010-2020	257516	23176
3	forest in 2020	9379101	844119

Using GFC product and tree cover >= 80%¶

Downloading data¶

We download the forest cover change data from GEE for New Caledonia for years 2001, 2010 and 2020, using a tile size of one degree. We use the GFC product and a tree cover percentage >= 80 to define the forest.

years_gfc = [2001, 2010, 2020]
out_dir_gfc80 = Path("out_gfc80")
ofile_gfc80 = out_dir_gfc80 / "forest_gfc80.tif"

start_time = time.time()
if not ofile_gfc80.is_file():
    geefcc.get_fcc_loss(
        aoi=(163.5, -23, 168.15, -19.51),
        buff=0.0,
        years=years_gfc,
        source="gfc",
        perc=80,
        tile_size=1.0,
        crop_to_aoi=True,
        output_file=ofile_gfc80,
        parallel=True,
        ncpu=ncpu,
    )
end_time = time.time()

We estimate the computation time to download 20 1-degree tiles using several cores.

elapsed_time = (end_time - start_time) / 60
print('Execution time:', round(elapsed_time, 2), 'minutes')

Execution time: 1.0 minutes

Transform multiband fcc raster in one band raster¶

fcc_file_gfc80 = out_dir_gfc80 / "fcc_gfc80.tif"
if not fcc_file_gfc80.is_file():
    geefcc.sum_raster_bands(
        input_file=ofile_gfc80,
        output_file=fcc_file_gfc80,
        verbose=False,
    )

Plot the forest cover change map¶

geefcc.plot_fcc_loss(
    input_file=fcc_file_gfc80,
    years=years_gfc,
    output_file="fcc_gfc80.png",
    title="Forest cover change 2001-2010-2020, GFC 80%",
    dpi=200,
    borders=Path("data") / "borders_NCL.gpkg",
    grid=out_dir_gfc80 / "grid.gpkg",
    xlim=(163, 169),
    ylim=(-23.25, -18.75),
)

Lines in black represent country borders. One degree tiles in grey cover the whole buffer and were used to download the data in parallel.

Compute statistics¶

res_df_gfc80 = geefcc.stat_fcc_loss(
    input_file=fcc_file_gfc80,
    years=years_gfc,
    epsg=32758,
    output_file="fcc_statistics_gfc80.csv",
)
tabulate(res_df_gfc80, headers=res_df_gfc80.columns, tablefmt="orgtbl", showindex=False)

category	label	count	area_ha
0	non-forest in 2001	199835640	17985208
1	deforestation 2001-2010	41700	3753
2	deforestation 2010-2020	27890	2510
3	forest in 2020	7273426	654608

Using GFC product and tree cover >= 60%¶

Downloading data¶

We download the forest cover change data from GEE for New Caledonia for years 2001, 2010 and 2020, using a tile size of one degree. We use the GFC product and a tree cover percentage >= 60 to define the forest.

out_dir_gfc60 = Path("out_gfc60")
ofile_gfc60 = out_dir_gfc60 / "forest_gfc60.tif"

start_time = time.time()
if not ofile_gfc60.is_file():
    geefcc.get_fcc_loss(
        aoi=(163.5, -23, 168.15, -19.51),
        buff=0.0,
        years=years_gfc,
        source="gfc",
        perc=60,
        tile_size=1.0,
        crop_to_aoi=True,
        output_file=ofile_gfc60,
        parallel=True,
        ncpu=ncpu,
    )
end_time = time.time()

We estimate the computation time to download 20 1-degree tiles using several cores.

elapsed_time = (end_time - start_time) / 60
print('Execution time:', round(elapsed_time, 2), 'minutes')

Execution time: 1.12 minutes

Transform multiband fcc raster in one band raster¶

fcc_file_gfc60 = out_dir_gfc60 / "fcc_gfc60.tif"
if not fcc_file_gfc60.is_file():
    geefcc.sum_raster_bands(
        input_file=ofile_gfc60,
        output_file=fcc_file_gfc60,
        verbose=False,
    )

Plot the forest cover change map¶

geefcc.plot_fcc_loss(
    input_file=fcc_file_gfc60,
    years=years_gfc,
    output_file="fcc_gfc60.png",
    title="Forest cover change 2001-2010-2020, GFC 60%",
    dpi=200,
    borders=Path("data") / "borders_NCL.gpkg",
    grid=out_dir_gfc60 / "grid.gpkg",
    xlim=(163, 169),
    ylim=(-23.25, -18.75),
)

Lines in black represent country borders. One degree tiles in grey cover the whole buffer and were used to download the data in parallel.

Compute statistics¶

res_df_gfc60 = geefcc.stat_fcc_loss(
    input_file=fcc_file_gfc60,
    years=years_gfc,
    epsg=32758,
    output_file="fcc_statistics_gfc60.csv",
)
tabulate(res_df_gfc60, headers=res_df_gfc60.columns, tablefmt="orgtbl", showindex=False)

category	label	count	area_ha
0	non-forest in 2001	197186817	17746814
1	deforestation 2001-2010	74006	6661
2	deforestation 2010-2020	60484	5444
3	forest in 2020	9857349	887161

Summary of the results¶

def areas_from_df(df, product, version, perc=""):
    fc1 = df.loc[df["category"].isin([1, 2, 3]), "area_ha"].sum()
    fc2 = df.loc[df["category"].isin([2, 3]), "area_ha"].sum()
    fc3 = df.loc[df["category"] == 3, "area_ha"].sum()
    d1 = df.loc[df["category"] == 1, "area_ha"].values[0]
    d2 = df.loc[df["category"] == 2, "area_ha"].values[0]
    n_years1 = years_gfc[1] - years_gfc[0]
    n_years2 = years_gfc[2] - years_gfc[1]
    return {"product": product, "version": version, "perc": perc,
            "fc2001": fc1, "fc2010": fc2, "fc2020": fc3,
            "d1": round(d1 / n_years1), "d2": round(d2 / n_years2)}

tmf_areas = areas_from_df(res_df_tmf, "tmf", "v1_2025", "")
gfc80_areas = areas_from_df(res_df_gfc80, "gfc", "v1_13(2025)", 80)
gfc60_areas = areas_from_df(res_df_gfc60, "gfc", "v1_13(2025)", 60)

res_df = pd.DataFrame([tmf_areas, gfc80_areas, gfc60_areas])
res_df.to_csv("comparison_geefcc_nc.csv", index=False)
tabulate(res_df, headers=res_df.columns, tablefmt="orgtbl")

**Comparing forest-cover change products for New Caledonia.** fc: forest cover (in ha), d1: mean annual deforestation (in ha) in the first period 2001–2010, d2: mean annual deforestation (in ha) in the second period 2010–2020, **perc**: tree cover threshold (in %) used to define the forest with GFC.¶
	product	version	perc	fc2001	fc2010	fc2020	d1	d2
0	tmf	v1_2025		893586	867295	844119	2921	2318
1	gfc	v1_13(2025)	80	660871	657118	654608	417	251
2	gfc	v1_13(2025)	60	899266	892605	887161	740	544

Forest cover for TMF and GFC with tree cover >= 60% are similar in 2020 (mean at about 865,000 ha) but the annual deforestation is 4-5 times lower when using the GFC product (e.g. 544 ha/yr for GFC in the period 2010–2020 against 2322 ha/yr for TMF for the same period).

geefcc

Navigation

Related Topics

New Caledonia¶

Downloading data in parallel¶

Using TMF product¶

Downloading data¶

Transform multiband fcc raster in one band raster¶

Plot the forest cover change map¶

Reproject in EPSG:3163 for area computation¶

Using GFC product and tree cover >= 80%¶

Downloading data¶

Transform multiband fcc raster in one band raster¶

Plot the forest cover change map¶

Compute statistics¶

Using GFC product and tree cover >= 60%¶

Downloading data¶

Transform multiband fcc raster in one band raster¶

Plot the forest cover change map¶

Compute statistics¶

Summary of the results¶