Downscale Sentinel-3 LST Products¶

In this example, a downscaler is trained with previously downloaded Sentinel-3 data which is then used to downscale LST. The results are subsequently plotted.

Import Packages¶

In [1]:

from pathlib import Path

import cartopy.crs as ccrs
import matplotlib.pyplot as plt
from sklearn.linear_model import LinearRegression

from s3lst_ds.downscaling.downscale.downscale import downscale
from s3lst_ds.downscaling.downscale.downscale_config import DownscaleConfig

from pathlib import Path import cartopy.crs as ccrs import matplotlib.pyplot as plt from sklearn.linear_model import LinearRegression from s3lst_ds.downscaling.downscale.downscale import downscale from s3lst_ds.downscaling.downscale.downscale_config import DownscaleConfig

Downscale Products¶

Configure¶

With the configuration below, one intends to:

• use the Sentine-3 products found at directory ./data;
• masking out the data outside of the Lisbon metropolitan area (represented by a WKT string);
• using 5 processors in the wrangling of the data (reprojecting, combining, masking and transforming);
• using a multi-timestamp downscaling Linear Regression model;
• with "FVC", "NDWI", "season" as predictors;
• standardising the numerical predictors ("FVC", "NDWI");
• dummy-encoding the categorical predictors ("season");
• considering timestamp-specific standardization of the spatio-temporal variables ("FVC", "NDWI" and "LST");
• using 5 processors in prediction and scoring;
• training with all timestamps and predicting (downscaling) for timestamps "20210716T104226" and "20210717T111751";
• scoring the model;
• with metrics $R^2$ (r2), Root Mean Square Error (rmse), Mean Absolute Error (mae) and Mean Bias Error (mbe);
• considering residual correction of the fine predictions;
• outputting gridded downscaled products;
• not returning the data wrangler and trained downscaler;
• logging the whole process into the terminal.

In [2]:

downscale_config = DownscaleConfig(
    data_wrangler_path_sentinel3=Path("./data"),
    data_wrangler_aoi="POLYGON ((-9.53 38.57, -9.05 38.57, -9.05 38.92, -9.53 38.92, -9.53 38.57))",
    data_wrangler_max_workers=5,
    downscaler_architecture="multi",
    downscaler_base_model=LinearRegression(),
    downscaler_X=[
        "FVC",
        "NDWI",
        "season",
    ],
    downscaler_masks=["aoi"],
    downscaler_scale="standardize",
    downscaler_encode="dummy",
    downscaler_transform="standardize",
    downscaler_max_workers=5,
    timestamps_infer=["20210716T104226", "20210717T111751"],
    timestamps_fit=None,
    score=True,
    scorers=["r2", "rmse", "rmse_delta", "mae", "mbe"],
    correct=True,
    gridded=True,
    out_data_wrangler=False,
    out_downscaler=False,
    log_mode="console",
)

downscale_config = DownscaleConfig( data_wrangler_path_sentinel3=Path("./data"), data_wrangler_aoi="POLYGON ((-9.53 38.57, -9.05 38.57, -9.05 38.92, -9.53 38.92, -9.53 38.57))", data_wrangler_max_workers=5, downscaler_architecture="multi", downscaler_base_model=LinearRegression(), downscaler_X=[ "FVC", "NDWI", "season", ], downscaler_masks=["aoi"], downscaler_scale="standardize", downscaler_encode="dummy", downscaler_transform="standardize", downscaler_max_workers=5, timestamps_infer=["20210716T104226", "20210717T111751"], timestamps_fit=None, score=True, scorers=["r2", "rmse", "rmse_delta", "mae", "mbe"], correct=True, gridded=True, out_data_wrangler=False, out_downscaler=False, log_mode="console", )

Downscale¶

To train the downscaler and downscale with it, the function below is used.

In [3]:

data_lst_ds = downscale(downscale_config)["y_fine_pred"]  # type: ignore

data_lst_ds = downscale(downscale_config)["y_fine_pred"] # type: ignore

INFO     Downscaling Sentinel-3 LST products                                            

INFO     The data will now be wrangled.                                                 
INFO     Getting a SingleDataWrangler instance for each timestamp and performing        
         wrangling of the respective data...                                            
         100% ━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 11/11  [ 0:00:01 < 0:00:00 , 53 timestamp/s ]:--:-- , ? timestamp/s ]
INFO     Data wrangled.                                                                 

INFO     The downscaler will now be trained.                                            
⠋        Training downscaler with the coarse data...
INFO     Downscaler trained.                                                            

INFO     LST will now be downscaled.                                                    
INFO     Predicting raw target...                                                       
         100% ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 2/2  [ 0:00:00 < 0:00:00 , 37 timestamp/s ]
INFO     LST downscaled.                                                                

INFO     The downscaler will now be scored with respect to the coarse training data     
         using Sentinel-3 as ground truth.                                              
INFO     Predicting raw target and scoring...                                           
         100% ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 11/11  [ 0:00:00 < 0:00:00 , ? timestamp/s ]
INFO     Downscaler scored with respect to coarse training data using Sentinel-3 as     
         ground truth.                                                                  

INFO     The downscaler will now be scored with respect to the coarse inference data    
         using Sentinel-3 as ground truth.                                              
INFO     Predicting raw target and scoring...                                           
         100% ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 2/2  [ 0:00:00 < 0:00:00 , ? timestamp/s ]
INFO     Downscaler scored with respect to coarse inference data using Sentinel-3 as    
         ground truth.                                                                  

INFO     The downscaler attained the following metrics:                                 
                                                                                        
                                  Metrics                                               
         ┏━━━━━━━┳━━━━━━━━┳━━━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━┓                     
         ┃ Batch ┃ Grid   ┃ Ground truth ┃ Metric     ┃     Value ┃                     
         ┡━━━━━━━╇━━━━━━━━╇━━━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━┩                     
         │ train │ coarse │ sentinel     │ r2         │   0.63545 │                     
         │       │        │              │ rmse       │    3.5807 │                     
         │       │        │              │ rmse_delta │   0.76731 │                     
         │       │        │              │ mae        │    2.6391 │                     
         │       │        │              │ mbe        │ -0.067016 │                     
         │ infer │ coarse │ sentinel     │ r2         │   0.39128 │                     
         │       │        │              │ rmse       │    3.4761 │                     
         │       │        │              │ rmse_delta │   0.79423 │                     
         │       │        │              │ mae        │    2.4232 │                     
         │       │        │              │ mbe        │    0.2788 │                     
         └───────┴────────┴──────────────┴────────────┴───────────┘                     
                                                                                        

Plot Downscaled LST Products¶

In [4]:

# Define the coordinate reference system to use in the plot
crs = ccrs.PlateCarree()


# Initialise figure and axes using a given CRS
fig, axs = plt.subplots(
    nrows=1,
    ncols=len(data_lst_ds),
    figsize=(12, 5),
    subplot_kw={"projection": crs},
)

fig.suptitle(
    "Downscaled LST Data",
    fontsize=16,
    y=1.05,
)

for ax, timestamp in zip(axs, data_lst_ds):
    img = data_lst_ds[timestamp].plot(  # type: ignore
        ax=ax,
        cmap="Spectral_r",
        alpha=0.8,
        cbar_kwargs={
            "orientation": "horizontal",
            "pad": 0.05,
            "label": None,
        },
        vmin=290,
        vmax=320,
        # "neither" to make the colorbar to exactly end at vmin and vmax
        extend="neither",
        # Project the data to the given CRS if it is not already in that CRS
        transform=crs,
    )  # type: ignore

    ax.set_title(f"Timestamp {timestamp.strftime('%Y-%m-%d %H:%M:%S')}", fontsize=14)  # type: ignore

plt.show()

# Define the coordinate reference system to use in the plot crs = ccrs.PlateCarree() # Initialise figure and axes using a given CRS fig, axs = plt.subplots( nrows=1, ncols=len(data_lst_ds), figsize=(12, 5), subplot_kw={"projection": crs}, ) fig.suptitle( "Downscaled LST Data", fontsize=16, y=1.05, ) for ax, timestamp in zip(axs, data_lst_ds): img = data_lst_ds[timestamp].plot( # type: ignore ax=ax, cmap="Spectral_r", alpha=0.8, cbar_kwargs={ "orientation": "horizontal", "pad": 0.05, "label": None, }, vmin=290, vmax=320, # "neither" to make the colorbar to exactly end at vmin and vmax extend="neither", # Project the data to the given CRS if it is not already in that CRS transform=crs, ) # type: ignore ax.set_title(f"Timestamp {timestamp.strftime('%Y-%m-%d %H:%M:%S')}", fontsize=14) # type: ignore plt.show()