GELE Photogrammetry & Cartography — Remote Sensing and GISCheat Sheet

Formula

Pixels across swath = Swath width ÷ Spatial resolution

Meaning

Swath width (m), Spatial resolution (m/pixel) → number of pixels across-track

Watch Out

Units must match (both in metres). Don't mix km with m. Result is approximate (may round up if swath doesn't divide evenly).

When To Use

Calculate image dimensions; estimate data volume; sensor comparison

Formula

Ground pixel size (m) = IFOV (rad) × Altitude (m)

Meaning

IFOV = Instantaneous Field of View (sensor angular resolution); Altitude = satellite/aircraft height above ground

Watch Out

IFOV must be in radians, not degrees. Small IFOV (high-resolution sensors) require higher altitude or finer engineering.

When To Use

Determine spatial resolution from sensor specifications; inverse problem: find altitude for required resolution

Formula

Radiometric resolution (bits) → Quantization levels = 2^n

Meaning

n = bit depth; 8-bit = 256 levels, 12-bit = 4096 levels, 16-bit = 65536 levels

Watch Out

Confuse radiometric (tonal depth) with spectral (number of bands) or spatial (pixel size) resolution. Higher bits ≠ always better (depends on noise floor).

When To Use

Compare sensor radiometric depth; assess dynamic range and shadow detail capture

Formula

Spectral reflectance at pixel = Sum(radiance in band) ÷ (solar irradiance × cos(zenith angle))

Meaning

Used in radiometric correction; zenith = angle of sun from vertical; accounts for atmospheric effects and topographic slope

Watch Out

Zenith angle changes throughout day and year—critical for multi-temporal analysis in Philippines (±12° latitude). Omitting correction causes false reflectance trends.

When To Use

Normalize imagery for comparison across seasons/times; remove illumination artifacts

Value

3 × 10⁸ m/s

Symbol

c

Quantity

Speed of light

Value

185 km

Symbol

W_L8

Quantity

Landsat-8 swath width

Value

30 m

Symbol

GSD_L8

Quantity

Landsat-8 spatial resolution (multispectral)

Value

290 km

Symbol

W_S2

Quantity

Sentinel-2A swath width

Value

10 days

Symbol

T_S2

Quantity

Sentinel-2 revisit (single satellite)

Value

250 m (Band 1–2), 500 m (B3–7), 1 km (B8–36)

Symbol

GSD_MODIS

Quantity

MODIS spatial resolution

Value

1–2 days (global coverage)

Symbol

T_MODIS

Quantity

MODIS temporal resolution

Value

0.61 m

Symbol

GSD_QB

Quantity

QuickBird (commercial) panchromatic resolution

Value

1000–2000 m AGL

Symbol

h_LiDAR

Quantity

Typical aircraft LiDAR altitude

Value

2–10 points/m² (varies by mission)

Symbol

ρ_LiDAR

Quantity

LiDAR typical point density

Term

Spatial resolution

Example

Landsat-8: 30 m; Sentinel-2: 10 m; QuickBird: 0.61 m (panchromatic).

Definition

Ground distance represented by one pixel (e.g., 30 m, 10 m, 1 m).

Term

Spectral resolution

Example

Landsat-8: 11 bands; Sentinel-2: 13 bands; hyperspectral: 100–200+ bands.

Definition

Number of spectral bands and their wavelength width; finer = narrower bands = more detail.

Term

Radiometric resolution

Example

MODIS: 12-bit; Landsat-8: 12-bit (processed to 16-bit); standard optical: 8-bit.

Definition

Bit depth; tonal sensitivity (e.g., 8-bit = 256 gray levels; 12-bit = 4096 levels).

Term

Temporal resolution

Example

Landsat-8: 16 days; Sentinel-2: 5 days (dual constellation); MODIS: 1–2 days.

Definition

Revisit time (how often a sensor images the same location).

Term

Passive remote sensing

Example

Landsat, Sentinel-2, MODIS, camera—requires daylight or thermal contrast.

Definition

Detects naturally available energy: reflected sunlight (optical, NIR) or thermal emission (thermal IR).

Term

Active remote sensing

Example

SAR (Synthetic Aperture Radar), LiDAR; penetrates clouds, sees through canopy.

Definition

Sensor emits its own energy (radar/microwave or LiDAR) and measures backscatter; all-weather, day/night.

Term

IFOV (Instantaneous Field of View)

Example

Smaller IFOV = finer pixels; typical: 0.005°–0.05° rad depending on satellite.

Definition

Angular footprint of one sensor pixel; defines spatial resolution with altitude.

Term

Spectral signature

Example

Water: low NIR; vegetation: high NIR, low red (NDVI); concrete: moderate all bands.

Definition

Unique pattern of reflectance/emission across spectral bands for a material or land cover.

Formula

Buffer distance = desired buffer radius (m) ÷ pixel size (m) [for raster]

Meaning

Determines number of pixel rings around a feature; scale buffer operation appropriately

Watch Out

Raster buffers create stepped boundaries (discrete cells); vector buffers are smooth curves. Buffer size vs pixel resolution trade-off: fine cells = memory cost but smoother result.

When To Use

Plan buffer analysis in raster GIS; set exclusion zones around roads, utilities, structures

Formula

Map projection distortion ≈ 1 + (distance from standard parallel / radius of Earth)²

Meaning

Approximate formula for metric (not angular) distortion; shows increase with distance from reference (standard) parallel

Watch Out

Distortion is zero on standard parallels; increases away from them. UTM (conformal) preserves angles but not area; equal-area projections preserve area but not angles. Choose based on application (cadastre = conformal; forest area = equal-area).

When To Use

Assess which projection suits area of interest; ensure accuracy requirements met; UTM zones every 6°

Formula

Overlay analysis: output_value(x,y) = f(raster1(x,y), raster2(x,y), ...)

Meaning

Combine multiple raster layers cell-by-cell using logical/arithmetic operators

Watch Out

All input rasters must have identical extent, projection, and cell size. Misalignment causes systematic error. No-data handling critical—propagates through calculations.

When To Use

Multi-criteria suitability analysis, land-use change detection, hazard mapping

Formula

Spatial interpolation error ≈ variance(observed) × (1 – R² from cross-validation)

Meaning

R² from cross-validation (0–1) indicates interpolation reliability; higher R² = less error

Watch Out

Cross-validation R² can overfit if training points cluster. Interpolation cannot reliably extrapolate beyond data range. Always validate at withheld test points.

When To Use

Validate kriging, IDW, or spline interpolation for elevation, temperature, pollutants

Value

6° longitude

Symbol

w_UTM

Quantity

UTM zone width

Value

500 km (500,000 m)

Symbol

FE

Quantity

UTM false easting

Value

0 m at equator

Symbol

FN_N

Quantity

UTM false northing (N. Hemisphere)

Value

10 million m (10,000,000 m)

Symbol

FN_S

Quantity

UTM false northing (S. Hemisphere)

Value

6,378,137 m

Symbol

a_GRS80

Quantity

GRS80 ellipsoid semi-major axis (a)

Value

6,356,752.314 m

Symbol

b_GRS80

Quantity

GRS80 ellipsoid semi-minor axis (b)

Value

1/298.257222101

Symbol

f

Quantity

GRS80 flattening (f)

Value

Zones 50–54 (approximately 120°–130°E)

Symbol

PH_zones

Quantity

Philippine archipelago UTM zones

Value

0.1–1 m (aerial), 0.5–2.5 m (satellite)

Symbol

GSD_ortho

Quantity

Typical ortho-photo spatial resolution

Term

Vector data model

Example

Cadastral parcels (polygons), road centerlines (lines), boreholes (points).

Definition

Represents discrete features as points (node coords), lines (vertex sequences), or polygons (closed rings); precise boundaries; attribute table.

Term

Raster data model

Example

Satellite imagery, DEM, rainfall grid, temperature surface.

Definition

Grid of cells (regular or irregular); each cell holds one value (elevation, reflectance, land class); efficient for continuous surfaces.

Term

Attribute table

Example

Parcel table: ParcelID, Owner, Area_m2, LandUse, DateSurveyed.

Definition

Database records linked to spatial features; each row = one feature, columns = properties.

Term

Projection (map projection)

Example

UTM (conformal, 6° wide zones), Transverse Mercator, Equal-Area Conic.

Definition

Mathematical transformation from 3D ellipsoid (lat/lon) to 2D plane; all projections distort shape, area, distance, or angle.

Term

Datum

Example

WGS84 (global, GNSS), PRS92 (Philippine), NAD83 (North America).

Definition

Reference ellipsoid (shape of Earth) and origin point; defines latitude/longitude; PRS92 (GRS80 ellipsoid) is Philippine standard.

Term

Coordinate reference system (CRS)

Example

EPSG:32651 (UTM Zone 51N, WGS84); EPSG:3857 (Web Mercator).

Definition

Combination of datum + projection; fully defines spatial coordinates; must be identical across overlaid GIS layers.

Term

Topology

Example

Roads must connect at intersections; parcel polygons must not overlap; shared boundaries are stored once.

Definition

Spatial relationships (adjacency, connectivity, containment) explicitly stored; ensures consistency (no slivers, overshoots, undershoots).

Term

Overlay

Example

Intersect parcels with zoning polygons; multiply slope raster by landuse raster (suitability analysis).

Definition

Combine two or more spatial datasets (vector or raster) to create new features/values; fundamental GIS operation.

Term

Buffer

Example

100 m buffer around contamination source; 50 m protection zone around heritage structure.

Definition

Create polygon(s) or zone(s) at fixed distance from feature(s); used for exclusion/inclusion zones.

Term

Interpolation

Example

Predict elevation between survey points; estimate rainfall at ungauged weather station.

Definition

Estimate values at unsampled locations from point observations; methods: IDW (Inverse Distance Weighted), Kriging, Spline, Nearest Neighbour.

Formula

NDVI = (NIR − Red) ÷ (NIR + Red); range: −1 to +1

Meaning

Normalized Difference Vegetation Index; NIR = band (0.7–1.3 µm), Red = band (0.6–0.7 µm); >0.5 = dense vegetation, 0–0.3 = sparse/barren

Watch Out

NDVI alone cannot distinguish vegetation type (grass vs forest). Sensitive to atmospheric haze—preprocess with radiometric/atmospheric correction. Values near 0 = confusion between water and urban; use water index (NDWI) for disambiguation.

When To Use

Quick vegetation health/vigor assessment; crop monitoring; deforestation detection; simple, requires only 2 bands

Formula

NDWI = (NIR − SWIR) ÷ (NIR + SWIR); or (Green − NIR) ÷ (Green + NIR) for water mapping

Meaning

Normalized Difference Water Index; SWIR = short-wave infrared (1.3–3 µm); positive = high water content; >0.3 = open water

Watch Out

NIR vs SWIR formulation differs by author; specify which bands used. Cloud and cloud shadow have NDWI values similar to water; visual inspection or auxiliary data (elevation) needed.

When To Use

Flood mapping, water body delineation, moisture stress in vegetation, wetland classification

Formula

Overall accuracy = (sum of diagonal pixels) ÷ (total pixels in confusion matrix)

Meaning

Confusion matrix rows = reference (ground truth), columns = classification result; diagonal = correct pixels

Watch Out

Overall accuracy ignores class imbalance (e.g., 95 % accuracy if 90 % of area is one class, accidentally classifies everything as that class). Use producer's/user's accuracy, F1-score, or Kappa coefficient instead.

When To Use

Assess classification performance; threshold for acceptance varies (>85 % often target, >90 % for critical applications)

Formula

Kappa coefficient = (Accuracy_observed − Accuracy_expected) ÷ (1 − Accuracy_expected)

Meaning

Accounts for chance agreement; ranges 0 (no better than random) to 1 (perfect); Kappa > 0.6 generally acceptable

Watch Out

Kappa can be paradoxically low if a class is rare (even with high overall accuracy). Use with producer's/user's accuracy, not as sole metric.

When To Use

Compare classifiers fairly; account for class frequency bias; standard in remote-sensing validation papers

Formula

User's accuracy (class i) = (correct pixels of class i) ÷ (total pixels classified as i)

Meaning

How reliable is the map for users? For class i, what fraction of mapped pixels are actually that class? (commission error = 1 − UA)

Watch Out

Low user's accuracy means high commission error (false positives); conversely, producer's accuracy measures omission (false negatives).

When To Use

Assess map utility for application; e.g., user's accuracy for 'Forest' tells farmer how likely mapped forest is true forest

Formula

Producer's accuracy (class i) = (correct pixels of class i) ÷ (total reference pixels of class i)

Meaning

How well does classifier detect class i? What fraction of true class-i pixels were correctly identified? (omission error = 1 − PA)

Watch Out

Low producer's accuracy = false negatives (missed real pixels). Inverse of user's accuracy; both needed for full picture.

When To Use

Assess classifier completeness; e.g., did the classifier find all farmland?

Value

> 0.5

Symbol

NDVI_healthy

Quantity

NDVI healthy vegetation threshold

Value

0.2–0.5

Symbol

NDVI_sparse

Quantity

NDVI sparse/stressed vegetation

Value

< 0.2

Symbol

NDVI_barren

Quantity

NDVI barren/water

Value

> 0.3

Symbol

NDWI_water

Quantity

NDWI open water threshold

Value

> 85 %

Symbol

Acc_min

Quantity

Acceptable classification accuracy

Value

> 90 %

Symbol

Acc_excellent

Quantity

Excellent classification accuracy

Value

> 0.6

Symbol

Kappa_min

Quantity

Kappa coefficient threshold (acceptable)

Term

Supervised classification

Example

Collect GCPs for 'Rice', 'Corn', 'Urban', 'Forest'; classifier learns spectral signatures; applies to entire image.

Definition

Train classifier using labelled reference data (training pixels); common methods: Maximum Likelihood, Random Forest, SVM, Neural Net.

Term

Unsupervised classification

Example

K-means with k=5 groups image into 5 spectral clusters; analyst must manually interpret which is vegetation, water, etc.

Definition

Cluster pixels by spectral similarity without prior labels; methods: K-means, ISODATA; produces classes but not directly interpretable.

Term

Training data / Training pixels

Example

Collect 50–200 pixels per class using GPS, ortho-photo, or field visit; ensure geographic spread (not all in one corner).

Definition

Sample of manually verified reference pixels used to teach classifier; must represent all classes and spectral variation.

Term

Confusion matrix

Example

If 100 reference rice pixels, 95 classified correctly, 3 classified corn, 2 classified fallow → row sums to 100, diagonal has 95.

Definition

Rows = reference class, columns = classified class; diagonal = correctly classified, off-diagonal = errors.

Term

Spectral separability

Example

Water (low NIR) vs vegetation (high NIR) = good separability; grass vs wheat = poor separability without time series.

Definition

Degree to which spectral signatures of two classes differ; higher separability = easier classification.

Term

Atmospheric correction

Example

FLAASH, 6S models; converts Landsat DN → TOA radiance → surface reflectance; enables temporal comparison.

Definition

Remove atmospheric effects (scattering, absorption) to derive surface reflectance from top-of-atmosphere (TOA) radiance.

Term

Radiometric normalization

Example

Normalize 2010 and 2020 satellite images to 2015 conditions for change detection.

Definition

Adjust radiance/reflectance to common reference condition (time, illumination, sensor); allows multi-temporal/multi-sensor comparison.

Term

Change detection

Example

NDVI(2010) − NDVI(2020) > −0.2 suggests vegetation loss; likely deforestation if confirmed by post-classification.

Definition

Compare classifications or indices from two or more dates to identify land-cover conversion; methods: direct comparison, post-classification, index differencing.

Formula

Root Mean Square Error (RMSE) = √[ Σ(predicted − observed)² ÷ n ]

Meaning

n = number of samples; measures average magnitude of prediction errors; in same units as variable (e.g., metres for planimetry)

Watch Out

RMSE is sensitive to outliers (one large error inflates it); Median Absolute Error (MAE) more robust. For spatial accuracy, report separately for X, Y, Z; RMSE_XY ≤ 1 pixel typical for ortho-photo.

When To Use

Assess georeferencing quality (GCP residuals), interpolation accuracy, classification uncertainty; compare methods

Formula

Horizontal positional accuracy (CE90) ≈ 1.65 × RMSE_XY

Meaning

Circular error at 90 % confidence; standard for geospatial products per NSSDA (National Standard for Spatial Data Accuracy)

Watch Out

Different standards (NSSDA, ASPRS, ISO) define accuracy differently; specify which. PRS92 cadastral survey must meet ±0.5 m horizontal (CA 141, PD 1529).

When To Use

State map accuracy in geospatial metadata; ensure meets application spec (cadastral: ±0.5 m; topographic: ±5 m; reconnaissance: ±50 m)

Formula

Data integration rule: All layers must share same datum + projection + coordinate system

Meaning

Transformation required if not identical; systematic offset if overlooked

Watch Out

Subtle transformation errors (wrong datum) cause 0.5–2 m shifts (deadly for cadastre). Always verify with known control points.

When To Use

Check EPSG codes before overlay; transform as needed; document source CRS for audit trail

Value

0.5–1 pixel (15–30 m for Landsat)

Symbol

RMSE_GCP

Quantity

Typical GCP horizontal RMSE (good setup)

Value

≤ 12 m (Level-1T product)

Symbol

CE90_L1T

Quantity

Landsat orthorectified CE90

Value

0.5–2 m (0.1–1 m GSD)

Symbol

CE90_ortho

Quantity

High-res ortho-photo CE90 target

Value

± 0.5 m horizontal (relative to adjacent parcels)

Symbol

Acc_cadastral

Quantity

Cadastral survey accuracy (PD 1529, CA 141)

Value

± 0.5–5 m (scale-dependent: ±0.5 mm at map scale)

Symbol

Acc_topo

Quantity

Topographic map accuracy standard

Value

± 16 m (90 % confidence, global average)

Symbol

LE90_SRTM

Quantity

SRTM DEM vertical accuracy

Value

± 17 m (RMS)

Symbol

RMSE_ASTER

Quantity

ASTER GDEM vertical accuracy

Term

Geospatial metadata

Example

Landsat-8 orthorectified image: WGS84/UTM51N, 30 m GSD, L1T processing level, July 2023, CE90 = 12 m.

Definition

Documentation of data source, coordinate system, accuracy, processing history, lineage; essential for reproducibility and compliance (ISO 19115).

Term

Ground control points (GCPs)

Example

Survey intersection point, road junction, building corner identified in image and field-surveyed; typically 6–10 GCPs minimum for 30–50 m² area.

Definition

Measured locations (lat/lon, UTM, or local coords) tied to image features; used to georeference imagery and assess accuracy.

Term

Orthorectification

Example

Landsat raw scene has ~35 m RMS error; orthorectified product has ~12 m CE90 (certified accuracy); suitable for mapping.

Definition

Geometric correction to remove sensor tilt, Earth curvature, and terrain relief; project image to map plane (true orthophoto); requires DEM and GCPs.

Term

Pansharpening

Example

Landsat 8: 30 m multispectral + 15 m panchromatic → 15 m RGB ortho-photo (Brovey or IHS method).

Definition

Fuse low-resolution multispectral image with high-resolution panchromatic band to create high-res colour composite.

Term

Interoperability

Example

GeoTIFF with EPSG code readable by ArcGIS, QGIS, PostGIS without format conversion.

Definition

Ability of different GIS/remote-sensing systems to exchange and use data seamlessly; requires open standards (OGC, ISO 19115/19139).

Term

Open Geospatial Consortium (OGC)

Example

WMS (Web Map Service): request map tile from server; used in many online map viewers.

Definition

International body setting standards for geospatial web services (WMS, WFS, WCS, GML); ensure data interoperability.

Term

INSPIRE Directive (EU) / PRS92 Standard (PH)

Example

All Philippine government spatial data must reference PRS92 (GRS80 ellipsoid) or WGS84 with documented transformation.

Definition

Regulatory framework for consistent spatial data sharing; EU requires harmonized coordinate systems; Philippines adopted PRS92 (RA 4374, RA 8560).

Values

• Relies on sun (optical, NIR) or target thermal emission
• Sensor emits own energy; measures backscatter

Property

Energy source

Values

• Blocked by clouds; cannot sense through dense canopy
• Penetrates clouds (radar); some canopy penetration (LiDAR limited)

Property

Cloud penetration

Values

• Requires daylight (optical); thermal works night (but lower detail)
• Works day and night; radar all-weather

Property

Day/night operation

Values

• Rich: visible, NIR, SWIR, thermal bands; many free platforms (Landsat, Sentinel-2, MODIS)
• Limited: radar intensity ± phase (SAR); LiDAR monochromatic (intensity/range only)

Property

Spectral information

Values

• Good to excellent (0.6–30 m commercial; 250 m MODIS free)
• Good: SAR 1–100 m; LiDAR 1–10 m (point spacing)

Property

Spatial resolution

Values

• Free (Landsat, Sentinel, MODIS); commercial (QuickBird, Pleiades) expensive
• SAR free (Sentinel-1); LiDAR airborne only: expensive (€1–5/km²)

Property

Cost per image

Values

• Land-use/cover classification, vegetation health (NDVI), water detection, crop monitoring
• DEM/canopy height (LiDAR), flood/inundation (SAR), change detection, biomass (SAR coherence)

Property

Primary applications

Values

• Landsat-8/9, Sentinel-2, MODIS, Spot, QuickBird
• Sentinel-1 (radar), ALOS PALSAR (radar), airborne LiDAR, Icsat-2 (LiDAR space-borne)

Property

Common platforms

Values

• Points (X,Y), lines (sequences of vertices), polygons (closed rings); linked to attribute table
• Grid of regular cells (pixels); each cell has one value (reflectance, elevation, land-class code)

Property

Representation

Values

• Shapefile (.shp + .shx + .dbf), GeoJSON, GeoPackage, PostGIS
• GeoTIFF, HDF5, NetCDF, JP2000, ASCII Grid

Property

File format

Values

• Small file size (1000 parcel polygons ≈ 100 KB); scales with feature count
• File size = pixels × bit-depth (1 km² at 1 m resolution, 8-bit ≈ 1 MB); grows with area, not complexity

Property

Storage efficiency

Values

• High precision; boundaries definable to 0.01 m
• Limited by pixel size (e.g., 30 m Landsat = ±15 m inherent positional uncertainty)

Property

Positional accuracy

Values

• Ideal for discrete objects (buildings, roads, parcels, hydrography with defined banks)
• Ideal for continuous phenomena (elevation, rainfall, temperature, satellite reflectance)

Property

Feature representation

Values

• Geometric intersection; results in new vector features (may create slivers if inputs unaligned)
• Cell-by-cell arithmetic/logic (fast); all inputs must have identical extent, resolution, projection

Property

Overlay operation

Values

• Topology explicitly stored (shared edges, adjacency); prevents overlaps/gaps if validation enforced
• No explicit topology; cell grid is implicit; single cell cannot be split between classes

Property

Topological enforcement

Values

• Rich attributes (100s of columns per feature); supports relational database (SQL queries)
• Single-value per cell (or limited: e.g., 3-band RGB); must use separate raster layers for multiple attributes

Property

Attribute complexity

Values

• Precise geometric operations (buffer, intersect, union); network analysis (shortest path); spatial statistics on attributes
• Fast map algebra (slope from DEM, NDVI from bands); interpolation; statistical summaries (e.g., mean elevation in watershed)

Property

Analysis capability

Values

• Primary: vector (polygon) parcels with attributes (Owner, Area_m2, LandUse, etc.); defines exact boundaries
• Supplementary: raster orthophoto (background), DEM (height reference), land-use raster (zoning visualization)

Property

Typical use in cadastre/surveys

Values

• 30 m (MS), 15 m (Pan)
• 11 bands (VIS, NIR, SWIR, TIR, Pan)
• 16 days
• Free
• Regional LULC, vegetation, water, change detection; standard for long-term analysis (archive since 1972)

Property

Landsat-8

Values

• 10 m (bands 2,3,4,8), 20 m (others), 60 m (atmospheric)
• 13 bands (MS); higher spectral detail than Landsat
• 5 days (dual sat)
• Free
• High-res LULC, crop/vegetation detail, water boundaries; newer, better atmospheric correction; EU standard

Property

Sentinel-2A/2B

Values

• 250 m (B1–2), 500 m (B3–7), 1 km (B8–36)
• 36 bands (visible to TIR)
• 1–2 days
• Free
• Global/continental scales: rainfall, snow, ocean colour, aerosol; for rapid change detection (fire, floods)

Property

MODIS (Aqua/Terra)

Values

• 0.61 m (Pan), 2.44 m (MS)
• 4 bands (Pan, RGB, NIR) or 1 band (Pan only)
• 1–5 days (tasking)
• $15–30/km²
• Urban mapping, 3D modelling, precise feature extraction; expensive; used for validation/critical projects

Property

QuickBird (Maxar)

Values

• 10 m GRD, 5 m SLC
• 2 polarisations (VV+VH or HH+HV)
• 6 days (single sat), 3 days (dual)
• Free
• All-weather mapping, flood detection, change detection, DEM generation (InSAR); works at night, through clouds

Property

Sentinel-1 (SAR radar)

Values

• 25 m (HH), 50 m (HV)
• L-band radar (HH, HV)
• 46 days (archive); new data limited
• Free (archive)
• Forest/biomass mapping, DEM (archived 2006–2011); L-band penetrates vegetation better than C-band (Sentinel-1)

Property

ALOS PALSAR (SAR radar)

Values

• 0.5–2 m point spacing (1–4 points/m²)
• Monochromatic (intensity); optionally RGB from camera
• On-demand (expensive mission)
• $1–5/km²
• High-precision DTM/DSM, urban canopy, forest structure, utility corridor mapping; cadastral-grade accuracy possible

Property

Airborne LiDAR

Values

• Diagonal sum / Total pixels
• 0–100 %
• Fraction of all pixels correctly classified; ignores class imbalance
• Quick check; but insufficient alone (can be high even if rare class omitted)

Property

Overall Accuracy (OA)

Values

• True Positives / (TP + False Positives) per class
• 0–100 %
• Of mapped pixels in class i, how many are truly class i? (map reliability for users)
• Assess commission error (false positives); critical for applications where false alarms costly (e.g., contamination zones)

Property

User's Accuracy (UA) / Precision

Values

• True Positives / (TP + False Negatives) per class
• 0–100 %
• Of true class-i pixels, how many were found? (detector sensitivity)
• Assess omission error (false negatives); critical if missing real cases costly (e.g., disease/forest area)

Property

Producer's Accuracy (PA) / Recall

Values

• (OA − Expected_OA) / (1 − Expected_OA)
• 0–1 (or 0–100 %)
• 0 = random, 0.6 = acceptable, 1.0 = perfect; accounts for chance agreement
• Compare classifiers; essential for peer-reviewed papers; corrects for class imbalance better than OA

Property

Kappa Coefficient

Values

• 2 × (Precision × Recall) / (Precision + Recall)
• 0–1
• Harmonic mean of UA and PA; balances false positives and false negatives
• Binary classification (e.g., built-up vs non-built); multi-class requires weighted F1

Property

F1 Score