Chlorophyll-a prediction with Deep Learning¶

Install packages¶

Uncomment the following cell to install the required packages.

In [ ]:

# %pip install "hypercoast[all]"

# %pip install "hypercoast[all]"

Import libraries¶

In [ ]:

import torch
import hypercoast
from hypercoast.chla import *

import torch import hypercoast from hypercoast.chla import *

Download sample data¶

In [ ]:

chla_data_url = "https://github.com/opengeos/datasets/releases/download/hypercoast/chla_test_data.zip"
pace_data_url = "https://github.com/opengeos/datasets/releases/download/hypercoast/PACE_OCI.20241024T182127.L2.OC_AOP.V2_0.NRT.nc"

chla_data_url = "https://github.com/opengeos/datasets/releases/download/hypercoast/chla_test_data.zip" pace_data_url = "https://github.com/opengeos/datasets/releases/download/hypercoast/PACE_OCI.20241024T182127.L2.OC_AOP.V2_0.NRT.nc"

In [ ]:

hypercoast.download_file(chla_data_url, quiet=True)

hypercoast.download_file(chla_data_url, quiet=True)

In [ ]:

pace_filepath = hypercoast.download_file(pace_data_url, quiet=True)

pace_filepath = hypercoast.download_file(pace_data_url, quiet=True)

Train the model¶

In [ ]:

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
# Load the training dataset
train_real_dl, test_real_dl, input_dim, output_dim = load_real_data(
    "data/Chl_RC_PACE.csv", "data/Rrs_RC_PACE.csv"
)
# Load the validation dataset
test_real_Sep, _, _ = load_real_test(
    "data/Chl_RC_PACE_Sep.csv", "data/Rrs_RC_PACE_Sep.csv"
)
# Model output path.
save_dir = "model/VAE_Chla_PACE"
os.makedirs(save_dir, exist_ok=True)

# Create the VAE model and optimizer
model = VAE(input_dim, output_dim).to(device)
opt = torch.optim.Adam(model.parameters(), lr=0.001, weight_decay=1e-3)

best_model_path = "model/vae_trans_model_best_Chl_PACE.pth"
train(model, train_real_dl, epochs=400, best_model_path=best_model_path)
# Load the optimal model
model.load_state_dict(torch.load(best_model_path, map_location=device))

predictions, actuals = evaluate(model, test_real_dl)

predictions_Sep, actuals_Sep = evaluate(model, test_real_Sep)

save_to_csv(predictions, os.path.join(save_dir, "predictions.csv"))
save_to_csv(actuals, os.path.join(save_dir, "actuals.csv"))

device = torch.device("cuda" if torch.cuda.is_available() else "cpu") # Load the training dataset train_real_dl, test_real_dl, input_dim, output_dim = load_real_data( "data/Chl_RC_PACE.csv", "data/Rrs_RC_PACE.csv" ) # Load the validation dataset test_real_Sep, _, _ = load_real_test( "data/Chl_RC_PACE_Sep.csv", "data/Rrs_RC_PACE_Sep.csv" ) # Model output path. save_dir = "model/VAE_Chla_PACE" os.makedirs(save_dir, exist_ok=True) # Create the VAE model and optimizer model = VAE(input_dim, output_dim).to(device) opt = torch.optim.Adam(model.parameters(), lr=0.001, weight_decay=1e-3) best_model_path = "model/vae_trans_model_best_Chl_PACE.pth" train(model, train_real_dl, epochs=400, best_model_path=best_model_path) # Load the optimal model model.load_state_dict(torch.load(best_model_path, map_location=device)) predictions, actuals = evaluate(model, test_real_dl) predictions_Sep, actuals_Sep = evaluate(model, test_real_Sep) save_to_csv(predictions, os.path.join(save_dir, "predictions.csv")) save_to_csv(actuals, os.path.join(save_dir, "actuals.csv"))

In [ ]:

plot_results(predictions, actuals, save_dir, mode="test")

plot_results(predictions, actuals, save_dir, mode="test")

In [ ]:

plot_results(predictions_Sep, actuals_Sep, save_dir, mode="Sep")

plot_results(predictions_Sep, actuals_Sep, save_dir, mode="Sep")

Predict chlorophyll-a concentration¶

In [ ]:

chla_data_file = pace_filepath.replace(".nc", ".npy")

chla_data_file = pace_filepath.replace(".nc", ".npy")

In [ ]:

chla_predict(pace_filepath, best_model_path, chla_data_file, device)

chla_predict(pace_filepath, best_model_path, chla_data_file, device)

Visualize the results¶

In [ ]:

rgb_image_tif_file = "data/snapshot-2024-08-10T00_00_00Z.tif"
output_file = "20241024-2.png"
title = "PACE Chla Prediction"
figsize = (12, 8)
cmap = "jet"

rgb_image_tif_file = "data/snapshot-2024-08-10T00_00_00Z.tif" output_file = "20241024-2.png" title = "PACE Chla Prediction" figsize = (12, 8) cmap = "jet"

In [ ]:

chla_viz(rgb_image_tif_file, chla_data_file, output_file, title, figsize, cmap)

chla_viz(rgb_image_tif_file, chla_data_file, output_file, title, figsize, cmap)