Getting the Data

Tech stack

The data was collected with the following technologies:

• Satellite images: The satellite images were obtained through Sentinelhub which gives access to a number of satellites including Harmonized Landsat Sentinel which was used for this project.

  Satellite images were collected for 3 different time periods (for the 3 months in a season) and 6 bands (blue, green, red, NIR_Narrow, SWIR1, SWIR2) for each region.

• Saving images: Osgeo GDAL was used to save the images to a geotiff file in the proper format that can be analyzed by the computer vision model.

• Geospatial data: The geospatial data of the exact latitude/longitude and address of the selected region was obtained through the Nominatim API.

Code snippets

Getting geospatial data

const fetchCoords = (district: string) => {
    const url = "https://nominatim.openstreetmap.org/search?q=" + district + "&format=json";
    fetch(url)
        .then(response => response.json())
        .then(data => {
            setPlaces(data);
        })
}

const fetchCity = (lat: any, lng: any) => {
    fetch(`https://nominatim.openstreetmap.org/reverse?lat=${lat}&lon=${lng}&zoom=6&format=json`)
        .then(response => response.json())
        .then((data: any) => {
            setName(data["display_name"]);
            if (name !== "Empty") {
                fetchCoords(data["display_name"]);
            }
            setMessage("");
            setError("");
        })
}

Getting satellite images

def get_data(glacier_bbox, time_interval, evalscript):
    """
    glacier_bbox is the bounding coordinates of the region
    time_interval is the time interval for the satellite image
    evalscript is a script telling which band of wavelength (blue, green, infrared, etc) to return
    """

    request = SentinelHubRequest(
        evalscript=evalscript,
        input_data=[
            SentinelHubRequest.input_data(
                data_collection=DataCollection.HARMONIZED_LANDSAT_SENTINEL,
                time_interval=time_interval,
            )
        ],
        responses=[SentinelHubRequest.output_response("default", MimeType.PNG)],
        bbox=glacier_bbox,
        size=glacier_size,
        config=config,
    )

    return request.get_data()[0]

Saving the images

def write_geotiff(filename, arrays, extent):
    def getGeoTransform(extent, nlines, ncols):
        resx = (extent[2] - extent[0]) / ncols
        resy = (extent[3] - extent[1]) / nlines
        return [extent[0], resx, 0, extent[3] , 0, -resy]
    
    arr_type = gdal.GDT_Int16
    driver = gdal.GetDriverByName("GTiff")
    out_ds = driver.Create(filename, arrays[0].shape[1], arrays[0].shape[0], 18, arr_type)
    srs = osr.SpatialReference()
    out_ds.SetProjection(srs.ExportToWkt())
    out_ds.SetGeoTransform(getGeoTransform(extent, arrays[0].shape[0], arrays[0].shape[1]))
    for i, arr in enumerate(arrays, 1):
        band = out_ds.GetRasterBand(i)
        band.WriteArray(arr)
    band.FlushCache()