remote-sensing-analysis

Skill: Remote Sensing Analysis

Purpose

This skill equips any agent with expert technical knowledge in complete remote sensing, including:

• Optical image processing (multispectral, hyperspectral, panchromatic)
• SAR processing (radiometric calibration, speckle filtering, InSAR, polarimetric decomposition)
• Spectral index calculation and band math
• Radiometric and atmospheric correction
• Change detection (LandTrendr, CCDC, BFAST, neural networks)
• Time series: smoothing, gap-filling, phenological analysis
• ML/DL classification (Random Forest, U-Net, SegFormer, OBIA)
• Object detection in satellite/drone imagery
• Pansharpening and sensor fusion
• Photogrammetry and point cloud processing (LiDAR/SfM)
• LULC product analysis (MapBiomas, CGLS, ESRI, GLAD, ESA WorldCover)
• Spatial statistics and inter-product agreement metrics
• Accuracy assessment (Olofsson 2014) and spatial statistics
• Scientific literature search in indexed journals

1. Raster Image Processing

1.1 Reading, Reprojecting and Aligning

import rasterio
import numpy as np
from rasterio.warp import calculate_default_transform, reproject, Resampling
from pathlib import Path

def read_raster_safe(path: Path, band: int = 1) -> tuple[np.ndarray, dict]:
    """Safe raster reading with nodata handling."""
    with rasterio.open(path) as src:
        data = src.read(band)
        profile = src.profile.copy()
        nodata = src.nodata
        if nodata is not None:
            data = np.where(data == nodata, np.nan, data.astype(float))
    return data, profile

def reproject_match(src_path: Path, ref_path: Path,
                    resampling: str = 'nearest') -> tuple[np.ndarray, dict]:
    """
    Reproject raster to the CRS and grid of a reference raster.
    IMPORTANT: classifications → use 'nearest' (not 'bilinear').
    Reference: Foody (2002) RSE 80(1):185
    """
    from rasterio.enums import Resampling as RS
    method = getattr(RS, resampling)
    with rasterio.open(ref_path) as ref:
        ref_crs, ref_transform = ref.crs, ref.transform
        ref_height, ref_width = ref.height, ref.width
    with rasterio.open(src_path) as src:
        transform, width, height = calculate_default_transform(
            src.crs, ref_crs, ref_width, ref_height,
            left=ref.bounds.left, bottom=ref.bounds.bottom,
            right=ref.bounds.right, top=ref.bounds.top)
        kwargs = src.meta.copy()
        kwargs.update({'crs': ref_crs, 'transform': transform,
                       'width': ref_width, 'height': ref_height})
        data = np.empty((src.count, ref_height, ref_width), dtype=src.dtypes[0])
        reproject(source=rasterio.band(src, list(range(1, src.count + 1))),
                  destination=data, src_transform=src.transform,
                  src_crs=src.crs, dst_transform=transform,
                  dst_crs=ref_crs, resampling=method)
    return data, kwargs