Wetland Mapping with GeoAI¶

Install packages¶

In [ ]:

# %pip install geoai-py

# %pip install geoai-py

Import libraries¶

In [ ]:

import geoai
import leafmap

import geoai import leafmap

Create an interactive map¶

In [ ]:

m = geoai.Map(center=[47.229011, -99.878662], zoom=13)
m.add_basemap("Esri.WorldImagery")
m

m = geoai.Map(center=[47.229011, -99.878662], zoom=13) m.add_basemap("Esri.WorldImagery") m

Draw an area of interest¶

In [ ]:

if m.user_roi is not None:
    bbox = m.user_roi_bounds()
else:
    bbox = [-99.9057, 47.2143, -99.8686, 47.2419]

if m.user_roi is not None: bbox = m.user_roi_bounds() else: bbox = [-99.9057, 47.2143, -99.8686, 47.2419]

Get the year of NWI data¶

In [ ]:

nwi_year = leafmap.get_nwi_year(bbox=bbox, return_geometry=False)[0]
nwi_year

nwi_year = leafmap.get_nwi_year(bbox=bbox, return_geometry=False)[0] nwi_year

In [ ]:

nwi_year_geom = leafmap.get_nwi_year(bbox=bbox)
nwi_year_geom

nwi_year_geom = leafmap.get_nwi_year(bbox=bbox) nwi_year_geom

In [ ]:

m.add_gdf(nwi_year_geom, layer_name="NWI Year")

m.add_gdf(nwi_year_geom, layer_name="NWI Year")

Search for NAIP images for the corresponding NWI year¶

In [ ]:

items = geoai.pc_stac_search(
    collection="naip",
    bbox=bbox,
    time_range=f"{nwi_year}-01-01/{nwi_year}-12-30",
)
items

items = geoai.pc_stac_search( collection="naip", bbox=bbox, time_range=f"{nwi_year}-01-01/{nwi_year}-12-30", ) items

Visualize NAIP images¶

In [ ]:

geoai.view_pc_items(items=items)

geoai.view_pc_items(items=items)

Download NAIP images¶

In [ ]:

images = geoai.pc_stac_download(items, output_dir="naip", assets=["image"])
images

images = geoai.pc_stac_download(items, output_dir="naip", assets=["image"]) images

In [ ]:

first_image = list(images.values())[0]["image"]
first_image

first_image = list(images.values())[0]["image"] first_image

In [ ]:

second_image = list(images.values())[1]["image"]
second_image

second_image = list(images.values())[1]["image"] second_image

Download NWI data¶

In [ ]:

image_bbox = leafmap.image_bbox(first_image, to_crs="EPSG:4326")
image_bbox

image_bbox = leafmap.image_bbox(first_image, to_crs="EPSG:4326") image_bbox

In [ ]:

nwi_gdf = leafmap.get_nwi(
    geometry=image_bbox, clip=True, add_class=True, output="wetlands.geojson"
)
nwi_gdf

nwi_gdf = leafmap.get_nwi( geometry=image_bbox, clip=True, add_class=True, output="wetlands.geojson" ) nwi_gdf

In [ ]:

wetlands_classes = nwi_gdf["WETLAND_TY"].unique()
wetlands_classes

wetlands_classes = nwi_gdf["WETLAND_TY"].unique() wetlands_classes

In [ ]:

num_classes = len(wetlands_classes) + 1
num_classes

num_classes = len(wetlands_classes) + 1 num_classes

Visualize NWI data¶

In [ ]:

m.remove_layer(m.layers[-1])
m.add_raster(first_image, layer_name="NAIP")
m.add_nwi(nwi_gdf, layer_name="NWI")
m

m.remove_layer(m.layers[-1]) m.add_raster(first_image, layer_name="NAIP") m.add_nwi(nwi_gdf, layer_name="NWI") m

Select training and test images¶

In [ ]:

train_raster_path = first_image
test_raster_path = second_image
train_vector_path = "wetlands.geojson"

train_raster_path = first_image test_raster_path = second_image train_vector_path = "wetlands.geojson"

In [ ]:

geoai.view_vector_interactive(
    train_vector_path, column="WETLAND_TY", tiles=train_raster_path
)

geoai.view_vector_interactive( train_vector_path, column="WETLAND_TY", tiles=train_raster_path )

Create image chips for training¶

In [ ]:

out_folder = "output"

out_folder = "output"

In [ ]:

tiles = geoai.export_geotiff_tiles(
    in_raster=train_raster_path,
    out_folder=out_folder,
    in_class_data=train_vector_path,
    tile_size=1024,
    stride=256,
    buffer_radius=0,
)

tiles = geoai.export_geotiff_tiles( in_raster=train_raster_path, out_folder=out_folder, in_class_data=train_vector_path, tile_size=1024, stride=256, buffer_radius=0, )

Train a segmentation model¶

In [ ]:

# Train U-Net model
geoai.train_segmentation_model(
    images_dir=f"{out_folder}/images",
    labels_dir=f"{out_folder}/labels",
    output_dir=f"{out_folder}/unet_models",
    architecture="unet",
    encoder_name="resnet34",
    encoder_weights="imagenet",
    num_channels=4,
    num_classes=num_classes,  # background and wetlands classes
    batch_size=8,
    num_epochs=50,
    learning_rate=0.001,
    val_split=0.2,
    verbose=True,
)

# Train U-Net model geoai.train_segmentation_model( images_dir=f"{out_folder}/images", labels_dir=f"{out_folder}/labels", output_dir=f"{out_folder}/unet_models", architecture="unet", encoder_name="resnet34", encoder_weights="imagenet", num_channels=4, num_classes=num_classes, # background and wetlands classes batch_size=8, num_epochs=50, learning_rate=0.001, val_split=0.2, verbose=True, )

Evaluate the model¶

In [ ]:

geoai.plot_performance_metrics(
    history_path=f"{out_folder}/unet_models/training_history.pth",
    figsize=(15, 5),
    verbose=True,
)

geoai.plot_performance_metrics( history_path=f"{out_folder}/unet_models/training_history.pth", figsize=(15, 5), verbose=True, )

Run inference on a single image¶

In [ ]:

# Define paths
masks_path = second_image.replace("naip", "prediction")
model_path = f"{out_folder}/unet_models/best_model.pth"

# Define paths masks_path = second_image.replace("naip", "prediction") model_path = f"{out_folder}/unet_models/best_model.pth"

In [ ]:

geoai.semantic_segmentation(
    input_path=test_raster_path,
    output_path=masks_path,
    model_path=model_path,
    architecture="unet",
    encoder_name="resnet34",
    num_channels=4,
    num_classes=6,
    window_size=1024,
    overlap=256,
    batch_size=4,
)

geoai.semantic_segmentation( input_path=test_raster_path, output_path=masks_path, model_path=model_path, architecture="unet", encoder_name="resnet34", num_channels=4, num_classes=6, window_size=1024, overlap=256, batch_size=4, )

Vectorize the results¶

In [ ]:

output_path = masks_path.replace(".tif", "_mask.geojson")
gdf = geoai.raster_to_vector(
    masks_path, output_path, min_area=300, simplify_tolerance=1
)

output_path = masks_path.replace(".tif", "_mask.geojson") gdf = geoai.raster_to_vector( masks_path, output_path, min_area=300, simplify_tolerance=1 )

In [ ]:

geoai.view_vector_interactive(output_path, tiles=test_raster_path)

geoai.view_vector_interactive(output_path, tiles=test_raster_path)

Download more images¶

In [ ]:

items = geoai.pc_stac_search(
    collection="naip",
    bbox=bbox,
)
items

items = geoai.pc_stac_search( collection="naip", bbox=bbox, ) items

In [ ]:

images = geoai.pc_stac_download(items, output_dir="naip", assets=["image"])
images

images = geoai.pc_stac_download(items, output_dir="naip", assets=["image"]) images

Run inference on multiple images¶

In [ ]:

geoai.semantic_segmentation_batch(
    input_dir="naip",
    output_dir="prediction",
    model_path=model_path,
    architecture="unet",
    encoder_name="resnet34",
    num_channels=4,
    num_classes=6,
    window_size=1024,
    overlap=256,
    batch_size=4,
)

geoai.semantic_segmentation_batch( input_dir="naip", output_dir="prediction", model_path=model_path, architecture="unet", encoder_name="resnet34", num_channels=4, num_classes=6, window_size=1024, overlap=256, batch_size=4, )