The optical bands of an image tell you what the ground looks like. Three other kinds of data tell you about the ground in other ways: how hot it is, how much light it gives off at night, and how wet the soil is. They are used for heat, drought, crops and the growth of cities. They are also easy to use wrongly, because each one is an indirect measurement, with a unit and a time of day that you need to know.
Land surface temperature
Everything above absolute zero gives off thermal infrared radiation, and the amount depends on its temperature. A thermal sensor measures this radiation in a band near 10 to 12 micrometres, where the atmosphere is fairly clear. From it, software works out the land surface temperature, LST: the temperature of the skin of the ground, which means the soil, the roofs, the leaves of a canopy, as the satellite sees them from above.
Three things make this harder than it sounds. The radiation depends on emissivity, how well a surface gives off heat, and it differs between water, soil and concrete, so the processing estimates it. The atmosphere adds and absorbs radiation, so a correction is applied, often using two neighbouring thermal bands. And the sensor sees only what is not covered by cloud.
LST is not air temperature. On a sunny afternoon, a dry field or a roof can be 10 to 20 degrees hotter than the air a metre above it, and a wet field or a canopy can be close to it. At night they come closer. LST is the right quantity for a heat island, for crop stress and for energy balance, and the wrong one for what a thermometer in the shade reads.
The time of day matters. MODIS flies on two satellites. Terra crosses the equator at about 10:30 in the morning and 22:30 at night. Aqua crosses at about 13:30 in the afternoon and 01:30 at night. The afternoon pass of Aqua sees the heat of the day, so the numbers are higher than those of Terra's morning pass. Never mix the two passes in one series without saying so.
Resolution varies. MODIS LST has a pixel of 1 km, and gives a value every day. Landsat has a thermal band at 100 m, which is resampled to 30 m, and it visits every 16 days per satellite. A park in a city is lost in a 1 km pixel and visible in Landsat. The guide on soil moisture and land surface temperature reads MODIS for Ludhiana and shows afternoon district means of 36 to 46 degrees in May.
Units and no-data
MODIS stores LST in kelvin multiplied by 50, as an integer, so you multiply by 0.02 and subtract 273.15 to get degrees Celsius, and the value 0 means no data. Reading the raw number as degrees, or taking a mean that includes the zeros, are the two classic mistakes.
Night lights
A night lights image shows the light that the Earth gives off after dark, seen from space. The sensor is the day-night band of VIIRS, on the Suomi NPP and NOAA satellites. It is sensitive enough to see the lights of a village, and it passes at about 01:30 local time. The Earth Observation Group at the Colorado School of Mines turns the nightly images into monthly and yearly composites, removing clouds, moonlit nights, fires and gas flares. The original grid is about 500 m (15 arc seconds). The value of a pixel is radiance, in nanowatts per square centimetre per steradian.
What the numbers are good for: following the growth of a city from year to year, finding electrified and unelectrified places, and seeing the effect of a power cut or a war. Brightness tracks population and economic activity closely in many places, and for this reason it is used as a stand-in for them in studies.
What they are not: a count of lamps, people or income. A bright pixel can be a factory, a highway interchange, an airport or a field where crop residue is burning. A dark pixel can be a place that is poor, or one with lamps that point down and spread little light, or one where the lights are weak. Over time, the change from sodium lamps to LEDs can change the numbers without changing the number of lamps, since the sensor sees blue light less well than the old orange light. In the night lights guide, the brightest cell of Ludhiana district is in the city, and 39 percent of the light of the district falls within 10 km of the centre, an area that is 8 percent of the district. That is a measure of concentration, and it says nothing about income.
Soil moisture
Soil moisture is the volumetric water content: the share of the soil volume that is water, in cubic metres of water per cubic metre of soil, written m3/m3. A value of 0.20 means a fifth of the volume is water. Sandy soils hold less than clays, so the same value can be wet in one soil and dry in another. Two more numbers are used in the field. The field capacity is the water a soil holds after it has drained, and the wilting point is the level below which plants cannot take water out.
There are three ways to get soil moisture.
- In-situ sensors are probes in the soil. They are accurate at a point and few.
- Satellite microwave sensors, such as NASA SMAP, which uses a long wavelength, the L band near 1.4 GHz, measure the top few centimetres of soil. Wet soil changes how the ground emits and reflects microwaves. The cell is large, tens of kilometres, and a vegetation canopy and rough ground reduce the quality.
- Models, such as ERA5-Land, calculate the water in the soil, layer by layer, from rain, evaporation and soil properties. ERA5-Land has four layers: 0 to 7 cm, 7 to 28 cm, 28 to 100 cm and 100 to 289 cm. A cell is about 9 km, and the record goes back decades for any place, with no gaps from cloud.
The top layer reacts within hours to rain and dries in days. The deeper layers move slowly, and they matter more for crops with deep roots. The soil moisture guide shows the Ludhiana series for 2023, with the top layer wettest in July and driest in November.
A model of soil moisture does not know about irrigation, which in Punjab can make a field far wetter than the model says. This is the most important limit for the region, and the reason to treat model soil moisture as a regional background and not as a measurement of one field.
Over-reading, a checklist
- Calling LST air temperature.
- Mixing the morning and afternoon passes of MODIS.
- Averaging a thermal image that still holds the no-data value of 0.
- Reading night light brightness as wealth.
- Comparing two years of night lights from different products or versions.
- Using a 9 km soil moisture value for one field.
- Forgetting that all three miss what lies under cloud (LST) or depend on the model (soil moisture).