Evaluating NICFI Imagery

Introduction

Norway’s International Climate and Forests Initiative (NICFI) is Norwegian government program seeking to prevent tropical deforestation and improve quality of life surrounding tropical forests. As a part of their initiative, NICFI uses Planet imagery to produce base maps of tropical regions, including tropical Africa. These maps are relatively high-resolution, cover a broad spatial extent, are mosaicked and corrected to make them suitable for remote sensing analysis, and can be made available in Google Earth Engine.

In this post, I investigate NICFI imagery in the study site in Cameroon to determine whether NICFI’s base maps would be useful for future analysis in the region. Specifically, I analyze the technical specifications of the imagery in the context of the 4 domains of remote sensing (spectral, spatial, temporal, and radiometric) and examine actual NICFI images of the region before providing a recommendation as to whether or not to use NICFI basemaps.

The Spectral Domain

First let’s examine the spectral domain. The technical specifications of NICFI Tropical Africa base map are listed here in the Earth Engine Data Catalog. According to this document, NICFI imagery includes red, green, blue, and near-infrared bands. These are sufficient spectral properties from which to create true color and some false color composites of the study site. Figure 1 displays side-by-side true color and false color images, where the false color composite was generated with NIR in the red channel, red in the green channel, and green in the blue channel. Notice how vegetation, which has characteristically large NIR values, really pops out as red. This is one example of how the addition of the NIR band can help us identify land cover.

Figure 1. True color image (left) and false color image (right) of study site using NICFI basemap.

While red, green, blue, and NIR band are sufficient for many purposes, more popular image sources like Landsat 8 and Sentinel-2 include additional bands like shortwave infrared, ultra blue, red edge, and thermal bands. The more bands you have, the easier it is to identify different features in your landscape. The spectral properties of the NICFI basemaps are adequate for some tasks, but may fall short when other bands are necessary to distinguish between visually-similar features.

The Spatial Domain

NICFI imagery performs well in the spatial domain. With a pixel resolution of 4.77 meters, almost 40 NICFI pixels fit within a single Landsat pixel! This incredible pixel resolution makes it easy to identify tiny features on a map. For example, figure 2 illustrates that with NICFI imagery we can identify individual buildings, crop fields, gaps between fields, and roads. Likewise, figure 3 shows that NICFI imagery allows us to see rivers, individual shrubs, and even irrigation channels. While a pixel resolution of 4.77 meters represents a substantial improvement from more common image sources like Landsat 8 (30 m) and Sentinel-2 (10 m), it is important to note that there are other image sources with even better pixel resolutions, down to mere centimeters.

Figure 2. NICFI image showing settlement and agriculture in Cameroon.
Figure 3. NICFI image showing agriculture, settlement, and a river in Cameroon.

‎‎‎ .

Figure 4. An outline of my study site and the first image in the NICFI image collection.

In addition to a fine pixel resolution, NICFI basemaps also have a broad spatial extent. Basemaps are also available that cover the entirety of tropical Asia, tropical Africa, and the tropical Americas. This broad spatial extent allows for both geographically extensive analyses and small scale studies in a wide variety of locations.

However, as figure 4 illustrates, the basemaps frequently have gaps in crucial spots. The very first basemap in the NICFI collection has a gap in part of and surrounding my study site. Thus, while the NICFI basemaps have a broad spatial extent, we often need to merge multiple of them in order to create a smooth picture of our area of interest.

The Temporal Domain

There are two main aspects of the temporal domain: frequency and depth.

In regards to frequency of imaging, NICFI only produces basemaps on a biannual basis and a monthly basis. They produce their basemaps by cloud masking and stitching together many images taken daily throughout the 6-month or 1-month period by Planet Labs. This makes it is impossible to get a snapshot of any particular location on a given day. For this reason, NICFI basemaps are not useful for monitoring rapid changes & short-term events and are less useful than other image sources for conducting analyses that are dependent on the time of the year. This property is illustrated by the side-by-side comparison between the NICFI basemap and Google satellite basemap of the same forest in Cameroon in figure 5. As you can see, this 6-month NICFI basemap is nowhere near as green as the Google satellite basemap, probably because it was produced by averaging out a number of images taken over a 6 month period.

Figure 5. Side-by-side look at the NICFI basemap (left) and Google satellite basemap (right) of a forest in Cameroon.

In regards to temporal depth, the Addendum to Planet Basemaps Product Specifications explains that NICFI began providing their tropical forest basemaps in 2015. This temporal depth is sufficient for some analyses but insufficient for other analyses. Whether or not this is sufficient for an analysis depends on how far back an analysis looks.

The Radiometric Domain

The radiometric resolution of NICFI imagery is 16-bit. In comparison, Sentinel-2, Landsat 8, and some other common satellites are 12-bit. With 12 bits, a computer can store 4,096 different possible levels of a measurement, but with 16 bits, a computer can store 65,536 levels of that measurement. In this manner, NICFI’s high radiometric resolution allows us to distinguish between many more grades of reflectance – 61,440 to be exact. This is a major advantage of the NICFI basemaps.

Recommendations

There are both strengths and weaknesses of NICFI basemaps. As is the case with other image sources, NICFI data is suitable some analyses but not for others. In my opinion, the greatest shortcoming of this dataset is the fact that every image is a mosaick of 1 or 6 months of individual images. Generally speaking, because of its incredible spatial and radiometric resolution, I would recommend this dataset for most analyses in tropical Africa. However, if you’re hoping to examine a location on a particular day, if you’re hoping to use information that changes seasonally, if you need more than 4 bands, or if NICFI has a gap over your study site I would recommend looking for other imagery.