RELIABLE VINEYARD ANALYTICS: TURNING DATA into INFORMATION

Welcome to the fourth installment of our series on science and technology solutions for the viticulture industry!

The general theme of this week's post revolves around the fact that there is a very significant difference between information and data - though these terms are often used interchangeably. Data are simply raw numbers, while information is the end-result of research, processing, and testing. While many organizations in the marketplace collect and deliver data as a final product, at Hawk Aerial, we're in the business of delivering information to our clients.

As vineyard vigor and disease mapping are based on evolving science and technology, the creation of maps that confer accurate, useful information depends on:

  • deployment of highly specialized equipment;
  • skilled application of data capture techniques;
  • a history of perpetual investment in sound scientific research and experimentation.

This post will be a brief treatment of how analytical mapping must be executed in order to serve as an effective tool for the vineyard manager. 

There are nearly 300,000 species of flowering plants on our planet, all of which reflect slightly different ranges of solar energy wavelengths, depending on their genus, species, variety, and health status. Thus, vine foliage reflects very slightly different portions of the light spectrum than that of the potato plant, oak leaves, or blueberry bushes. The application of a general vegetation index to fields planted in any crop is a common practice in today's marketplace - yet this approach is far too broad to be of any real use to the vineyard manager. 

Accurate and usable wavelength reflectance data is best captured from a drone at no more than several hundred feet above ground. Capturing this data accurately from an airplane requires extremely sophisticated equipment, complex scheduling, and is reliant on favorable weather conditions.
Data captured from drones features higher ground resolution, is virtually free of distortion created by the diffractive effect of water vapor in the atmosphere, and features precise locational repeatability of image capture in subsequent missions. It also relates only to objects in the area of interest (read: vines), lessening the interference created by the abundance of color present in large field-of-view images. The benefits conferred by these factors include:

  • Vigor or disease data for each individual vine;
  • Much less need for corrective processing, which can degrade the raw data and thus decrease the reliability of the final product;
  • Eliminates potential effects of varying position and altitude during image capture while comparing different vigor or disease datasets over time. 


The reflectance data that is fundamental to accurate vigor and disease mapping in grapevines resides in very narrow, specific ranges, known as bands. The bands corresponding to normal vine health shift at different phenological stages, and even between varietals, in infinitesimal but very significant gradations.
Our research partners have isolated the various iterations of significant wavelength bands through the painstaking use of ultra-accurate hyperspectral imaging, coupled with exhaustive and rigorous research in California vineyards, over the past 15 years.

 

Once these highly specific ranges of wavelength reflectance data are collected and preprocessed, the process of algorithmic referencing and verification begins. This is made possible by use of our proprietary vitis vinifera health databases - storehouses of verified, ground-truthed vine health information. We've been compiling these deep databases through a decade and a half of research to date. We are continually joining new data to them as our research forges forward - driving their growth, and in turn, the evolution of our best-in-class service, which is getting smarter and more accurate every day.  

So there you have it - the other guys give you data. We give you information. 

METHODS of AERIAL PHOTOGRAPHY in vineyard mapping

 

Hi! And welcome to the third installment in our four-part series unpacking the science and methods behind application of remote sensing techniques enabling important analytics to vineyard management!

This week, we'll zoom out from the specifics of camera technology to a discussion of the methods of image capture used to generate analytical vine-vigor and disease maps.

Aerial imaging has been applied to agriculture since at least the early 1960s, when spy-planes were used to estimate crop production behind the Iron Curtain during the Cold War.

The areas of crop research, camera technology, imaging techniques, global positioning technology, and aeronautics have all undergone a quantum leap in the past five decades - bringing a whole new set of tools to the agricultural field. 

The most effective use of cameras to produce imagery of cultivated fields is to mount them to aircraft. Images captured from directly overhead at altitude (known as nadir photography) provide a real-time, map-like view of fields.
Images captured by this method also offer a much greater level of detail than satellite imagery - and their clarity suffers much less from distortion due to atmospheric effects and cloud cover.

With the advent of using airplanes to capture images of fields in crops, came the discipline of photogrammetry - the science of joining adjacent and overlapping photographs together to render mosaics, enabling the generation of expansive maps.

 In the mosaicing process, images taken at different locations can be stitched together to form a larger image by matching features appearing in different frames.

In the mosaicing process, images taken at different locations can be stitched together to form a larger image by matching features appearing in different frames.

If you've ever taken a group picture, you are familiar with backing away from your subjects in order to increase the field of view, or FOV, allowing you to capture them all in the image. You've also noticed the further away you snap your photo, the less detail of your subjects your photograph captures.

The same concept applies to aerial photography; the higher up the camera, the more ground you capture, and the less detailed the photo. Thus, an airplane flying higher will capture a field or vineyard faster and in fewer pictures, but less details of the plants can be discerned.

 Airplanes capture larger areas, faster, and with lower detail. Drones capture smaller areas, at greater detail, and operate in a slow and controlled fashion.

Airplanes capture larger areas, faster, and with lower detail. Drones capture smaller areas, at greater detail, and operate in a slow and controlled fashion.

As such, airplanes are a good choice for capturing very large areas and generating a coarse, but expansive dataset.

Inversely, remotely piloted drones, AKA unmanned aerial vehicles, which typically fly at 1/10 or lower of the altitude of manned aircraft, are a better choice for capturing smaller areas in crisp, high-resolution data. Drone-acquired imagery captured at an altitude of 350 feet can have over ten times the detail and resolution of that captured from manned aircraft flying at 3500 feet, even when the latter is using incredible, advanced 100-megapixel cameras. 

Although the quality of an aerial map can be judged on many parameters, the benchmark is ground-sampling distance, or GSD. This metric refers to the linear size of the area on the ground that is represented by one pixel - remember from last week's post that a pixel is the elemental component of a digital image - a square containing only one color value.
Thus, an image with a lower GSD features higher resolution, and vice-versa.

By way of example, if the GSD is three feet, that means all colors captured from a 9 square-foot area on the ground are represented by one pixel - a single square of one color, averaged from all the colors captured in that area.
If the GSD is six inches, then that same 9-square foot area is represented by 36 pixels - thus capturing much more detailed color information - and thus a much richer data input for analysis.

 Two images of the same area in a vineyard. The image on top is was taken at low altitude using a 24-megapixel camera. It features low ground-sampling distance. The lower image was taken at ten times the altitude using a state of the art 100-megapixel camera, and features high ground-sampling distance.. For analysis of relatively small areas, the image at the top yields much more accurate information, for obvious reasons. 

Two images of the same area in a vineyard. The image on top is was taken at low altitude using a 24-megapixel camera. It features low ground-sampling distance. The lower image was taken at ten times the altitude using a state of the art 100-megapixel camera, and features high ground-sampling distance.. For analysis of relatively small areas, the image at the top yields much more accurate information, for obvious reasons. 

Imagery from manned flights are usually delivered on a set schedule - resulting from the aircraft performing regular imaging flyovers on a schedule that suits the imaging organization.
Mapping using drone-acquired imagery, on the other hand, can be arranged at any time suiting the end customer's needs - on very short notice, with a rapid turnaround time. 

Most drones performing precision agricultural imaging are of the rotorcraft variety - featuring four, six, or eight propellers providing lift - think of a miniature, multi-propeller helicopter.

 A rotorcraft, or multi-copter type drone, in controlled flight.

A rotorcraft, or multi-copter type drone, in controlled flight.

This form factor provides extreme agility, slow, controlled flight (usually around 10 miles an hour) and the ability to hover. Thanks to extremely precise global positioning electronics, drones can create maps from images captured at exactly the same locations in your vineyard week after week, month after month, and year after year. This results in maps that are very consistent over time, recording spectral data from plant foliage under nearly identical conditions.
Manned aircraft, on the other hand, fly very high overhead at speeds usually in excess of a hundred miles an hour, making repeating image capture at precise locations challenging, and introducing considerable uncertainty related to speed, distance from the subject, and angle of spectral data reflected from plant foliage. 

 A drone captures images within a very tight coordinate range in a repeatable fashion.

A drone captures images within a very tight coordinate range in a repeatable fashion.

 Due to the high speed, high altitude, and the effect of crosswinds, an airplane has much less control over the exact location in which an image is captured.

Due to the high speed, high altitude, and the effect of crosswinds, an airplane has much less control over the exact location in which an image is captured.

When the goal is to create a very detailed map, the best approach is low, slow, and precise - best delivered by a remotely piloted drone fitted with a finely-tuned multispectral camera.

Stay tuned for next week's installment, where we bring together elements of all the posts to date for an unpacking of the unique and cutting-edge science and research that enables the delivery of our vigor and disease maps, which are the only maps on the market delivering reliable, actionable information.  

 

The Power of Multispectral Cameras

Hi! Welcome to the 2nd weekly installment in our four-part vineyard-oriented science and technology series! Last week we talked briefly about the electromagnetic spectrum. This week, we’ll narrow the focus a bit and discuss the use of imaging systems in viticulture, particularly with regard to multispectral cameras.

Infrared radiation, or IR, (discussed in the last email, and in greater detail in last week’s blog post), is generally sensed by us humans as radiant heat. The photosynthetic foliage of plants reflects IR in a direct and positive relationship to plant health and vigor, i.e., more active photosynthesis = more intense infrared reflection. We cannot see infrared with our eyes, but a camera’s sensor can detect it much as you and I detect visible light.

Infrared film was invented about a hundred years ago, but it took until the early 1970s to discover that IR reflectance from plants is a dynamic phenomenon containing actionable information. In the past 20 years, traditional photographic film has largely been replaced by the photodetector - silicon-based imaging sensors ubiquitous in the modern age - you more than likely have one on your person right now, in your smartphone camera. The photodetector makes infrared capture more practical - and opens up a world of possibility for its use in viticulture.

A multispectral camera works much the same way as a regular digital camera, with a few important differences.

In general, a regular digital camera has one sensor with three stacked layers – each sensitive to only red, green, or blue (RGB). This sensor is divided into pixels (a truncation of the words picture and element) which are the elemental unit of a digital image. Each pixel, when exposed to light, records digital numbers - an expression of the intensity of light each color-sensitive detects. Thus, each pixel in an RGB image has three values, for example, (150, 34, 231), corresponding to (red, green, blue).
 

A multispectral camera, however, normally features four (or more) separate sensors, each receptive only to energy within a given wavelength range. Hence, instead of three values, each sensor will only report one value per pixel when an image is captured. A typical setup is as follows: one sensor only picks up light in the red wavelength range, one in the green, one in the blue, and one in the infrared - yielding four separate images. This allows much more precise color segmentation.

During analysis of vegetation health, each of the four images are stacked on top of each other in GIS or photo-processing software. Here, the ratios between reflectance of the four different radiation wavelength ranges are determined, allowing the researcher to draw conclusions.

To be usable in viticulture, multispectral imagery must be captured and processed according to rigorous standards.

Firstly, access to a database of vine foliage reflectance ratios that are typical of photosynthetic vigor, water status, and disease symptoms, is fundamental.
Information gained from any given imaging session must be referenced to a sound dataset in order to produce a reliable vine-health map.

The second thing that is needed is the ability to focus on a very small portion of the spectral range in question, by using filters for each sensor that exclude wavelength ranges not relevant to the analysis.

 Desired light wavelength ranges from foliage reflectance reaching the sensor ca be re installed on the lens to the right. 

Desired light wavelength ranges from foliage reflectance reaching the sensor ca be re installed on the lens to the right. 

To effectively record an accurate picture of your vine health, the camera's filters must be adjustable. The correct settings to capture exactly the light wavelengths required must be informed by data compiled by decades of meticulous research. Without the correct filter settings, a multispectral camera is geared towards general agriculture - an “overview” scope that does not have much relevance to the particularities of grapevines.

In conclusion, multispectral cameras are an incredibly powerful tool, but are not likely to deliver actionable results without:

- rigorous cross-referencing with databases of grapevine reflectance.
- filter settings calibrated to pick up relevant wavelength ranges.

Signing off for this week! Stay tuned - next week's post will unpack the ins and outs of aerial image capture techniques.