Posts

Training Data for GeoAI: Labeling Imagery Without Fooling Yourself

A model can only learn what its labels teach it. If the labels are wrong, uneven or unexplained, the model will still produce a map. That map will just look more trustworthy than it deserves. This post covers the part of GeoAI work that happens before any model runs: building training data. We will use flood and landslide mapping from satellite imagery as examples. The focus is on five decisions that decide whether your results mean anything. A label is the answer you give the model for a piece of imagery, such as "flooded" or "not flooded". Training data is the set of images or pixels paired with those labels. Points, polygons or masks? You can record a label in three common ways. Each has a cost and a limit. Points A point label marks one location with a class. It is fast to collect. You can click through hundreds of locations in QGIS in an afternoon, or reuse field survey points. Points work well for classification of single pixels or small patche...

NDVI and NDWI: What Spectral Indices Actually Show

A satellite image is a grid of numbers. Each pixel stores how much light the ground reflected in each band. A spectral index is a simple formula that combines two or more of those bands into one new number. That number highlights something you care about, such as healthy plants or open water. NDVI and NDWI are the two most common indices. They are easy to compute, but they are also easy to misread. This post explains the idea behind them, how to read the results, and the mistakes that trip up beginners. If you need a refresher on bands and resolution, see our earlier post on satellite imagery basics. The reflectance idea behind indices Reflectance is the fraction of incoming sunlight that a surface reflects back toward the sensor. Different materials reflect different amounts in different wavelengths. This pattern is called a spectral signature . Three simple cases show why indices work: Healthy leaves absorb most red light for photosynthesis. They reflect a lot of near-i...

Whitebox QGIS Practice 32 — Sediment Transport Index of Two Basins: A 0.4 Power on Contributing Area Removed the Main-Channel Skew

Image
In practice 31 I compared two basins with the Stream Power Index (SPI). The mean was higher in the interval basin A−B and the median was higher in basin B. What pushed the A−B mean up was the 1% of cells along the main A channel. So what happens if I look at erosion through a different index built from the same inputs? This time I made the Sediment Transport Index (STI) with Whitebox SedimentTransportIndex and compared the same two basins again. STI is a terrain-based form of the slope-length factor LS in the soil loss equation (USLE/RUSLE). The formula is STI = (n+1) × (SCA / 22.13) n × (sin β / 0.0896) m , and the Whitebox defaults are n = 0.4 and m = 1.3. 22.13 m is the slope length of the standard USLE plot and 0.0896 is the sine of a 9% slope, so the two ratios are dimensionless. The biggest difference from SPI is the 0.4 power on the contributing area. If SCA grows ten thousand times, its part of STI grows only about 40 times. Data I copied the practice 31 input folder into ...

Satellite Imagery Basics: Resolution, Bands and Revisit Time

When someone says "use satellite imagery" for a disaster project, the next question should always be: which satellite, which image, and why. Not all satellite images are the same. Some show fine detail but visit a place rarely. Some cover huge areas every day but show only a blur. Picking the wrong one wastes time or gives you a wrong answer with confidence. This post covers three properties that define any satellite image: spatial resolution, temporal resolution and spectral resolution. Once you understand these, reading a satellite data page or a band list stops being confusing. Spatial resolution: how much detail you get Spatial resolution is the size of one pixel on the ground. A 10 m resolution image means each pixel represents a 10 by 10 meter square. Anything smaller than that square gets blended into a single average value. Coarse resolution (250 m to 1 km): good for watching weather, smoke plumes, or large-scale vegetation patterns. One pixel can cover seve...

Coordinate Systems and EPSG Codes in Korea: Belt TM, UTM-K, and Why 5181 Is Not 5186

Image
The first thing I check on any GIS dataset is its coordinate reference system. The metadata often gives only a number such as "EPSG:5186", and data for the same area in Korea can arrive as 5181, 5186, 5179 or 4326. The numbers alone do not tell you which ones match and where they will disagree. From now on, this blog writes the name and the EPSG code together the first time a coordinate system appears in a post (for example, East Belt (GRS80) EPSG:5187). This post sets out the lookup table behind that rule and the places where it actually goes wrong in practice. The basic difference between geographic and projected coordinate systems is covered in Coordinate Systems in QGIS: EPSG:4326 vs EPSG:5186 . Here I focus on the codes you meet in Korean data and how they differ. What an EPSG code bundles To turn a pair of numbers into one point on the ground you need several things: the ellipsoid that approximates the Earth, how that ellipsoid is tied to the Earth (the geodetic d...

Whitebox QGIS Practice 31 — Stream Power Index of Two Basins: The Main Channel Set the Mean, the Slopes Set the Median

Image
In practice 30 I compared where two basins are likely to stay wet, using the topographic wetness index (TWI). The same inputs answer the opposite question too. Not where water collects, but where the collected water, running down a slope, has the most power to scour its bed. When I want a first pass over erosion-prone ground before thinking about check dams or drains, the index I reach for is the Stream Power Index (SPI). The formula is SPI = SCA × tan β. SCA is the D-infinity specific contributing area from practice 30 (m²/m) and β is the slope. TWI is the log of SCA divided by tan β; SPI is SCA multiplied by it. It grows where a cell receives a lot of water from upslope and is steep as well. This time I took the practice 30 outputs as they were, ran Whitebox StreamPowerIndex , and compared the two basins with Zonal statistics . The comparison flipped depending on which summary value I looked at, and most of this post is about that. Data I copied the practice 30 inputs and output...

Hazard, Exposure and Vulnerability: The Risk Equation Behind Disaster Maps

A flood hazard map shows where water may reach. It does not show who lives there, or how badly they would be hurt. To answer those questions, you need a fuller idea of risk. Disaster risk is usually described with three parts: hazard , exposure and vulnerability . This post explains each one, shows how each becomes a map layer, and walks through a simple overlay that combines them. The risk equation A common way to write the idea is: Risk = Hazard × Exposure × Vulnerability This is a concept, not a physical law. Different agencies and papers define the terms slightly differently, and some use addition or other combinations instead of multiplication. What matters is the logic: if any one part is missing, risk is low. A strong earthquake in an empty desert: high hazard, no exposure, so little risk. A town on a flood plain with a very high levee and a good warning system: high hazard, high exposure, but low vulnerability, so lower risk. Keep this logic in mind. It explai...