Knock Farm case study
Remote sensing in agriculture
Remote sensing is becoming an increasingly important tool in precision agriculture as satellite and drone imagery provide on non-intrusive methods to rapidly screen large areas. The aerial view helps with detecting differences that might not be evident from ground level observation.
Here we are looking at a short case study on how satellite data and drone images could be applied in farm management at Knock Farm located in Northern Scotland (Fig. 1).
Figure 1. Location of Knock Farm in Scotland
Knock Farm is an organic farm that keeps sheep and cattle. The farm grows crops for animal feed to support their lifestock. There are pasture and rough grazing fields within the farm area as well as some grassland, wetland and forest plantation areas. The farm also hosts horses and supply rocks for rockery projects.
Landcover type at Knock Farm were classified (Fig. 2) using fast unsupervised cluster analysis (left) and supervised classification (right) where user trains the classification by selecting representative area for each class. The clustered classification could be further refined by reducing number of classes. Currently three classes included rough grazing type of landcover. Supervised classification could be improved by introducing a rough grazing category. Fast cluster classification is promising as a screening tool.
Figure 2. Satellite image based landcover classifications for Knock Farm (outlined in red).
Figure 3. You can see here which fields got ploughed at Knock Farm this spring (dark grey) and how the crops grew on the them again (colour changing through light grey and light red into dark red. Some of the fields then got ploughed again in November.
False colour images are effective in highlighting changes in vegetation cover (Fig. 3). Plant health and moisture indices can also be visualised online using freely available sentinel-2 data (Fig. 4). Ready-made layers are available for example of NDVI and moisture index data. Swipe below to see how the fields included in the 2017 drone survey (yellow box) are doing currently.
Figure 4. Plant health (Normalised Difference Vegetation Index (NVDI, green image) and Normalised Difference Moisture Index (NDMI, blue image) for the Knock Farm area. The area included in drone survey is highlighted by the yellow box.
Plant health index looks low but no reason to worry as can be seen in Fig. 3 time-lapse the fields have recently been ploughed. NDMI values approaching -1 also indicate bare ground.
Figure 5. Drones collect high resolution imagery allowing visualisation of small details.
Drone imagery allow use to drill into fine detail (Fig. 5). For example an erosion feature caused by cattle trampling can be identified.
Figure 6. DSM of the drone survey area created in Pix4D.
The drone images can also be used to created detailed elevation models of the area. These can be Digital surface models (DSM) that include all surface features (Fig. 6) or digital terrain models (DTM) that show the shape of the ground (trees and buildings and other surface features removed).
Figure 7. Example of volume analysis capability in Pix4D. The red sections above the green base line produce the cut volume and any sections lower than the baseline produce the fill volume.
3D models generated from the drone images can be analysed further to derive volumes of the visible features. Here a cluster of hedge trees and a section of stone fencing were analysed for illustration purposes but this could be your crop volumes.
Figure 8. Illustration of relative plant health index (VARI) derived for the fields in RGB camera imagery captured using a drone. The VARI index successfully identified the erosion feature illustrated in Fig. 5 above)
Plant health index ( Visible Atmospherically Resistant Index VARI ) can be derived from the regular Red Green Blue (RGB) photographs (Fig. 8). This is based on small increase in reflectance in green band relative to red and blue that occurs in healthy plants but not in stressed vegetation. VARI index allows analysis of relative differences in plant health across an area photographed under same conditions.
Recommended remote sensing management regime
Adopt a tiered approach that combine different resolution outputs: satellite imagery from open sources and bespoke drone surveys:
- Carry out regular reviews of satellite data.
- Use rapid cluster based unsupervised landuse classification to screen for mismatches (e.g. a crop type clustering with a different crop) as an alert on potential changes in plant growth, plant health or encroachment by weeds with different reflectance.
- Complement satellite image review with targeted drone surveys in areas where issues (pests, nutrient deficiency, moisture stress) have been suspected based on satellite image screening.
- Use Visible Atmospherically Resistant Index (VARI) calculated from RGB camera output to assess relative plant health across an area photographed under the same conditions.
- Dynamically review the data collection approach as new technologies emerge and become more affordable. An example future application is drone based LIDAR analysis of crop canopies.
