Forest characterization using traditional forestry methods requires intricate and thorough field measurements across numerous plots. The level of detail and resolution acquired simply has not been reproduced using remote techniques.


Forest characterization using traditional forestry methods requires intricate and thorough field measurements across numerous plots. The level of detail and resolution acquired simply has not been reproduced using remote techniques. Measuring forest plots is expensive and labor-intensive and introduces subjectivity. And is not practical in a tactical environment. Advances in remote technologies (e.g. LiDAR, photogrammetry) have enabled a more affordable and practical method for predicting stand-level parameters. These advances allow for the potential adaptation to military reconnaissance use for mobility and sensor applications. In this regard, we need to develop methods for remote estimation of forest density, a critical requirement for Army mission planning operations in inaccessible areas using a combination of ground-truth and satellite data. Our overall goal is to develop and model relationships between on-the-ground metrics and those extracted from aerial photos in order to construct models that will predict density of a forest given the appropriate metrics obtained from satellite/aerial data.

Use case example: Why it’s important

Addressing this challenge will enable the remote estimation of forest density thus reducing the cost, labor, and time needed to acquire field data. Through the estimation of forest density, we hope to remotely provide on-the-ground insights (or estimation) on mobility in a given unreachable area.

Problem Sponsors


Frequently Asked Questions

How timely is the data?

It depends. We have ground truthed data from NJ that serves to help us validate the forestry/vegetation models. Collecting ground-truth data is time intensive, depending on seasonality, for example. Modeling allows us to capture the density in seconds instead of months/years?


Where have you experienced a lack of data? What is the challenge in accessing data?

Currently we have satellite data for the locations we’re working in. However, because of the resolution of that data, we’re not able to reliably extract useful information.


Have you tried LIDAR and what has been the experience so far?

We have not yet tried LIDAR as an option.


Is it possible to get sample data from forests?

Yes, this information can be provided.


What is the ideal format in which you would like to receive mobility insights derived from remote sensing data? Also, who are the end-users? (e.g., GEOINT analysts or warfighters?)

Long-term, the information is for the warfighter and we want them to have immediate access whether they are in the field or otherwise. Short-term, we want to be able to visualize what density means, like in a heat map and assess the accuracy of our model.


Regarding forest/woodland density and penetration, would forestry companies not be a good resource for determining energy or equipment requirements for penetration of that environment?

We have a forester on the team who is providing them with information regarding what they may need to build a good model.


Do you have regional priorities?

Not right now. Right now, the priority is figuring out how to create a model that accurately represents mobility through a forest.


How much fluctuation is there in forest density? Would the region need to be evaluated on a per mission basis / in real-time or can you rely on “static” maps that are updated once every 1-2 months? Is the penetrability a function of fundamental area density or seasonal change?

Penetrability is a function of both. In the regions that we know, trees and vegetation do not change very fast. Information from one year to another will likely be similar, so you could likely work with static maps that are regularly updated. Seasonal factors will need to be incorporated into modeling, however, as will vegetation type and the types of trees that are there, as different trees have different structures.


Is there interest in differentiating between types of tree species present?

Yes. This is desirable.


Have they considered overlaying DTM, DSMs and bare earth models?

No, they have not yet taken this into account, but it could be useful!


Along with forest density, how important is elevation point clouds, ground conditions, weather, other datasets?

Very important when considering mobility. All of these factors play a role in the analysis and warfighter decisions. These need to be incorporated into the modelling approach. As they are in the beginning of the process, they haven’t figured out how to incorporate these features into their model.


Does the model attempt to calculate each individual vehicles ability penetrate difficult terrain or do you try to support coordinating multiple vehicles to navigate the terrain together using their combined capabilities to overcome obstacles?

At this point in the model, if you give the distance from a tree to the nearest neighbor in the plot (½ hectare), the model will tell you the density of the area. It also gives standard deviation. This has only been tested in one area, and it’s an area that’s flat. Elevation has not been taken into account.


I know archaeologists use satellite imagery to virtually deforest dense jungle areas such as Cambodia and Mexico in reconstructing ancient communities. Do you have access to this same data or technology?

If that’s something that can be done, it would benefit us drastically, but to date, we have not had access to a similar technology.


How quickly are you able to train your model? Have you considered using MI like SageMaker to rapidly adjusting data or parameters to your model?

Right now, we use optimization approach/machine learning (stochastic algorithm) to test the model and minimize the error between the model and the data.


Is this primarily for vehicle accessibility? Have you been able to get data on routes taken successfully?

We are primarily looking at vehicle accessibility, but we do not yet have data on successful routes taken.