Agricultural drones

Why this matters

A farmer managing a 30-rai field cannot walk all of it often, and usually notices a problem only once leaves have turned yellow, which may be too late. A drone with a multispectral camera sees plant stress in wavelengths the human eye cannot, and a spray drone applies fertiliser or chemicals only where needed. But a beautiful image of the field helps no one unless you understand what its numbers mean and what decision follows.

The whole course follows one hypothetical case: a 300 × 160 m (30 rai) rice field of an agricultural cooperative, monitored with a multispectral drone, followed by zone-based liquid fertiliser application. All numbers are synthetic training data. The Python code for every module can be downloaded from /downloads/uat-365/.

Precision agriculture

The International Society of Precision Agriculture (ISPA) gives a short definition: precision agriculture is a management strategy that takes account of temporal and spatial variability to improve the sustainability of agricultural production. Put simply, no field is the same in every square metre. One corner is saline, another waterlogged, another short of nitrogen. Applying the same amount of fertiliser everywhere wastes it where it is not needed and leaves too little where it is.

The FAO and ITU report (2018) describes many roles for drones in agriculture. This course groups them in two:

• Survey drones carry RGB, multispectral or thermal cameras to map fields and monitor crop health (modules 2, 3 and 5)
• Spray and spreading drones carry a liquid tank or seed hopper to apply fertiliser, crop protection products or seed (module 4)

How plants reflect light

NASA Earth Observatory explains that chlorophyll in leaves strongly absorbs visible light, especially red and blue, while the cell structure of the leaf strongly reflects near-infrared (NIR) light. A healthy plant therefore reflects little red and much NIR. When a plant is stressed, chlorophyll falls and leaf structure changes, so red reflectance rises and NIR falls. The steep rise in reflectance between red and NIR is called the red edge.

A multispectral camera such as the MicaSense RedEdge-P has the five bands shown in Figure 1: blue 475 nm, green 560 nm, red 668 nm, red edge 717 nm and NIR 842 nm. Each band is only 12–57 nm wide, so it measures reflectance in the chosen range more precisely than an ordinary RGB camera.

Drone, satellite or walking the field

ESA’s Sentinel-2 provides red and NIR bands at 10 m resolution and revisits the same place every 5 days with two satellites. The data is free, but often cloud-covered in the rainy season. A drone resolves centimetres and can fly under cloud, but needs a pilot and covers far less area.

FIELD_L, FIELD_W, RAI = 300, 160, 1600      # m, m, m² per rai

for name, gsd in (("drone 2 cm", 0.02), ("drone 4 cm", 0.04), ("Sentinel-2 10 m", 10.0)):
    print(f"{name:<16} {RAI / gsd ** 2:>12,.0f} pixels per rai")

def pixels(length, offset, size=10):
    first = offset // size
    last = (offset + length - 1e-9) // size
    inside = sum(1 for i in range(int(first), int(last) + 1)
                 if i * size >= offset and (i + 1) * size <= offset + length)
    return inside, int(last - first + 1)

nx_in, nx_all = pixels(FIELD_L, 5)
ny_in, ny_all = pixels(FIELD_W, 5)
print(f"field shifted 5 m from the satellite grid: {nx_in * ny_in} pixels fully inside, "
      f"{nx_all * ny_all} touching the field")

drone 2 cm          4,000,000 pixels per rai
drone 4 cm          1,000,000 pixels per rai
Sentinel-2 10 m            16 pixels per rai
field shifted 5 m from the satellite grid: 435 pixels fully inside, 527 touching the field

The three work together. Drone or satellite images show “where something is wrong”; walking into the field shows “why”.

Area units and Thai requirements

The Weights and Measures Act B.E. 2542 (1999) defines 1 rai = 1,600 m², 1 ngan = 400 m² and 1 square wa = 4 m². International spray and fertiliser rates usually use hectares (10,000 m²).

def thai_area(m2):
    rai, rest = divmod(m2, 1600)
    ngan, rest = divmod(rest, 400)
    return int(rai), int(ngan), rest / 4

for length, width in ((300, 160), (250, 137)):
    area = length * width
    rai, ngan, wa = thai_area(area)
    print(f"{length} x {width} m = {area:,} m2 = {rai} rai {ngan} ngan {wa:g} sq wa = {area / 10_000:.3f} ha")

300 x 160 m = 48,000 m2 = 30 rai 0 ngan 0 sq wa = 4.800 ha
250 x 137 m = 34,250 m2 = 21 rai 1 ngan 62.5 sq wa = 3.425 ha

Before using agricultural drones in Thailand, check at least three sets of requirements:

• NBTC: register the drone’s radio equipment
• CAAT: register and obtain permission for the pilot under the rules in force (see UAT 313)
• Department of Agriculture: in December 2024 it launched a standard operating practice for spraying with agricultural unmanned aircraft and ID cards for contracted spray operators who pass its training. Spray operators must check the latest details directly with the Department (module 4)

Module lab

Common mistakes

Summary

• Precision agriculture manages the variability within a field rather than treating it uniformly
• Healthy plants reflect little red and much NIR, stressed plants the reverse; multispectral cameras measure this difference
• Drones, satellites and ground scouting complement each other; choose to fit the question
• 1 rai = 1,600 m², and agricultural drone use must meet NBTC, CAAT and Department of Agriculture requirements

Check your understanding

1. How many rai is a 200 × 120 m field?
2. How do red and NIR reflectance usually change when a plant is stressed?
3. How many pixels are in one rai at 5 cm GSD?
4. What are the main advantage and limitation of Sentinel-2 compared with a drone?
5. How many rai are 5 hectares?