UAV imagery as tool to monitor crop growth in West Africa
A case study in a crop fertilization experiment
In the semi-arid tropical zone of Mali, little is known about local growth limiting factors of crops present in the region. A fertilization study with 48 experimental fields in the Sougoumba region was conducted to get more insight in the effect of different fertilizers on the growth of cotton, sorghum, millet, maize and peanuts. The main goal of my internship was to explore the potential of UAV (Unmanned Aerial Vehicle) technology to monitor crop growth through time and to quantify the effect of the different fertilizer treatments. In Mali, my main responsibility was to acquire the UAV image data from all experimental fields. Together with this, most of my activities were aimed at ensuring that the images would be useful for scientific purposes afterwards. This means that much work was directed on the geometric and radiometric correction. Back in the Netherlands, I have payed attention to the quality of the UAV imagery and I examined the effect of the fertilizer application for a selection of the experimental fields.
- The UAV (Sensefly - eBee) has a normal GPS precision (ca. 10m) which means that all pictures need to be geometrically corrected to a precision of ca. 1 cm. In able to do this, one needs reliable, fixed points on the ground, clearly visible from the sky and with known exact coordinates. Since commonly used structures as concrete buildings or wells were missing or scarce in the study area, these points had to be created. In total, 52 GCPs (Ground Control Points) were created during my fieldwork period. Where it was possible, white crosses were painted on bare bed rock. Otherwise, concrete crosses were constructed first, which were (re)painted afterwards (figure 1). These points were later measured exactly using differential GPS.
Figure 1: Two examples of created GCPs. Left:GCP painted on bare bedrock. Right: GCP painted on a constructed concrete cross.
The raw data output of the Canon Powershot S110 NIR camera are digital numbers (DNs). However, one needs to have reflectance factor values for practically all remote sensing purposes. In order to convert the pixels to these reflectance factors, radiometric panels (figure 2) were painted in different shades of grey and subsequently measured with a spectrometer. During a UAV flight the panels were then placed in an open spot. Pre-processing the images, the DN values of the radiometric panel can be correlated with the now known reflectance values, which yields a formula that can be used to radiometrically correct the entire image.
Figure 2: Radiometric panel B
Before analysing the data for fertilization treatment effect, image quality checks are performed. After critical inspection of the images, several issues were encountered concerning the noise and sharpness, which are possibly caused by camera optics and sensor. The sharpness appeared to depend on three different factors: Three different causes of fluctuation in blurriness (or sharpness) are identified:
- Location-imposed sharpness: Pixels located at the periphery of an image are often more blurry than pixels located in the middle;
- Moment-imposed sharpness: Consecutively made pictures in a flight can highly differ in sharpness. Probably this can be due to the roll, pitch and yaw of the UAV;
- Sensor-imposed sharpness: It appeared that the sharpness of the NIR sensor consistently differed from the sharpness of the Red and Green sensor.
The implications of these findings are twofold. First, one should be aware that the DN extraction from the pixels reflecting the radiometric panel is not as precise as it could be, decreasing the precision of the radiometric correction. Secondly, pixel-to-pixel comparison through time will not be reliable, since one will not be able to point out that a difference is caused by on-ground changes or image quality issues.
Potential of UAV technology for crop monitoring
The study shows that application of fertilizers is improving crop growth in the Sougoumba region, but the relative impact of fertilization depends on the local environmental conditions. Results demonstrate that NDVI (Normalized Difference Vegetation Index) values derived from UAV imagery are strongly related to plant height (figure 3). This indicates that pixel values can be translated into crop height, suggesting a high potential for crop growth monitoring in the region. Temporal profiles indicate that differences in NDVI values between crops may be used for automated crop type recognition.
Figure 3: Relationship between the on-ground measured sorghum plant height and the average NDVI pixel values of the experimental plots. Blue line and points = data from 27 August 2014. Orange line and dots = data from 18 September 2014. In the right corner an NDVI picture from the experimental plots of field 3 (18 September 2014) is shown.
Not only quantitative information can be derived from the images. UAV technology can serve as a tool for quality control and for deriving qualitative information as well. For example, figure 4 shows that the presence of ants can be seen from the air as well.
However, pixel-to-pixel comparison is limited because of image quality issues. Use of calibration tarps can be a solution for practical constraints of the radiometric calibration. Improving image quality would increase the reliability of crop yield prediction and allows precise comparison of images acquired at different dates. Further research should include data of all experimental fields to get more insight in the relative effect of the local environment on the fertilization response of the crops. In addition to that, including other crop growth variables (e.g. LAI, chlorophyll, biomass) is likely to improve the linear NDVI-growth model.
Figure 4: Left - NDVI image of a ‘donut-shaped’ feature in the middle of a millet field. Right - example of an ant nest spot where the surrounding plants (sorghum and cotton) have been eaten.
My time in Mali has been a great experience. I have seen how a big multi-disciplinary and international project was organized in a challenging environmental context. Dealing with unexpected problems (malfunctioning data input devices, complications with the UAV, limited financial and material resources etc..) often required finding fast and/or creative solutions. In general, managing the fieldwork in Sougoumba was a highly demanding task, but very rewarding as well. Besides, though this summary is written in the first person, most of the work was executed in close cooperation with Guillaume Chomé. Flying eBee, painting GCPs, carrying out administrative matters and the daily communication with the field team were efforts we shared during my stay. Executing these tasks was only possible given a good collaboration and communication with each other, which I can only consider successful.
Next to the project aspect of my stay I also enjoyed the contact with the people in Koutiala and Sougoumba. During the many miles we walked through the semi-natural habitats and agricultural fields we frequently exchanged information in Bambara with farmers about how we and the families were doing as a way of greeting each other. Another time we weren’t allowed to go away from a farm without taking a bag of recently harvested peanuts with us. Our driver Massaman, who took us everywhere, was always eager to teach us important Bambara phrases. In the evening we went to the restaurant ‘Maman, j’ai faim’ where the ‘aloko’ is unparalleled. Tea is everywhere, and you are always invited to drink along with the people. These and other experiences made my internship a unique experience and are tempting me to come back in the future.