Tuesday, June 27, 2017

2017 Q2 All-Hands - Seattle Edition

That's a wrap!  We just finished up our All-Hands meeting in Seattle, WA.  Because we're a totally remote company, every few months we rent a house and all get together to work on new development projects and get some in-person time with each other.

Our team of seven (and a half!) plus our two summer interns
getting a view of the Space Needle and Mount Rainier

This past week, we made improvements to our biometrics codebase and got the new version of Plot Hound prepped for public launch.  We also scheduled out the rest of our 2017 - lots of exciting developments in carbon assessment, open-source forestry tools, and growth modeling coming over the rest of 2017.

Mission success - we escaped the room at Locurio!

We had a great time at All-Hands and our team is growing all the time.  Interested in helping us build the future of forestry?  Check our jobs page for details.

Thursday, June 22, 2017

Optimal Cruising for Your Forest Type

The following post originally appeared in the June 2017 issue of the Forestry Source.  It was written by SilviaTerra's lead biometrician, Dr. Nan Pond.

We're continuing the discussion in the Society of American Foresters's LinkedIn group - looking forward to hearing your questions and comments!

In the March edition of the Forestry Source, we discussed the importance of evaluating inventory decisions in the context of Cost + Loss. How much does a cruise cost to install, and how much do you lose by making imperfect decisions using the information from that cruise?

Taking this a step further, here’s a look at how this works for two different forest types - an upland hardwood forest and an unthinned loblolly pine plantation. In this example, we’ve simulated forest conditions. The upland hardwood forest simulation covers a 65 year old forest with about 150 ft2 of basal area per acre and 150 trees per acre.  The loblolly pine plantation simulation covers a 13-year-old plantation, with a starting TPA of 400 and BA per acre of 160 ft2.

This analysis follows the process outlined in that previous article - identifying a forest condition of interest, finding a similar FIA plot, simulating a forest and then sampling from it, and then growing the forest and the samples forward and comparing management options. Each forest type was tested using 10 different FIA plots and 100 simulations derived from each plot - a total of 1000 simulations each.

The cruising methods compared were the same for both simulations - BAF 10, BAF 20, and 1/10th acre fixed radius plots, installed at 1 plot per 5 acres and 1 plot per 10 acres - a total of 6 different cruise methods. In both forest conditions, the fixed radius plots were the ‘winning’ methodology.  Let’s note now that these results are very context-specific, and depend on the valuation and markets we chose, the discount rate, and the tested management options.

In the upland hardwood example, there was a tie - 4 of 10 FIA plots examined showed that 1/10th acre plots at 1 plot per 10 acres had the lowest cost+loss while another 4 showed the lowest cost+loss as 1/10th acre plots at 1 plot per 5 acres.  For the loblolly plantation, the optimal cruise methodology was 1/10th acre fixed radius plots at 1 plot per 10 acres.


These results may be surprising - they were to us. My coworker even commented that he’d been “cruising plantations wrong for years!”  The common line of thinking is that variable radius plots are faster, and easier. Because of this they’re far more likely to be chosen by cruisers and cruise managers. It’s cheaper to install a variable radius plot, and even more so to use a 20-factor prism instead of a 10 BAF.

In a cost+loss analysis, we’re able to see the real tradeoffs that are made. The trick here is that often, cruising decisions are made looking solely at the cost part of the equation - not the loss. In our simulations, the BAF plots were absolutely the cheapest approach - they take less time to install and involve measuring fewer trees.  However, the loss side of the equation came into play and tipped the scales each time.  In the loblolly plantation simulations, the mean loss from management decisions based on the BAF 20 cruises was on average twice as much as the loss from BAF 10 or the 1/10th acre plots.  Similarly, the BAF 20 cruises had a mean loss of 3 times greater than cruises using  1/10th acre plots.  The BAF 10 cruises showed twice the loss.   

The takeaway from this shouldn’t be “always use 1/10th acre fixed radius plots” - the key is really to think about the tradeoffs being made when choosing a cruising method. The sampling method used is a meaningful and influential decision. A cruise that doesn’t fully represent the conditions in the stand can lead to costly management mistakes.  

The same methodology can be used to evaluate other inventory approaches.  Is it worth incorporating satellite imagery or LiDAR into your cruising process?  Using the cost+loss approach, you can determine whether the increased precision of your cruise (and the resulting improvement in management) outweighs the cost of the additional data inputs. We’ll cover that in an upcoming article.

Monday, June 19, 2017

Our next "Biometrics Bits" article is now available in the Forestry Source!

Our next "Biometrics Bits" article is out in this month's edition of the Forestry Source!  This month, Dr. Nan Pond wrote about "Optimizing Cruising for Your Forest Type."  This article was a followup to our previous article "What is the 'Right' Amount of Information to Collect?" and in it Nan walks us through how that theory could be applied to a Southern loblolly pine plantation.

We've already gotten a bunch of emails about the article - questions about other forest types and sampling methods.  We love thinking about forest inventory - send us your questions at biometricsbits@silviaterra.com and let's talk!

Friday, June 2, 2017

What is the "right" amount of inventory information to collect?

The following post originally appeared in the Forestry Source.  It was written by SilviaTerra's lead biometrician, Dr. Nan Pond.

A BAF 10 plot every 5 acres, or a 20th acre fixed radius plot every 2.5?  A total height on every other plot, or every plot? What kind of inventory should you pay for, and how much does it matter to your bottom line?
Foresters across America have developed rules of thumb that guide how much inventory information they collect.  Because inventories drive management decisions, these informal guidelines play an important but underappreciated role in forestry.  But are those inventory habits actually optimal?  Is there an opportunity to do better?

We all agree that having inventory information is important for making forest management decisions.  Otherwise we wouldn't cruise at all!  But we also know that collecting information has a cost - a cost that must be justified by improvements in management outcomes.  This is why we measure sample plots rather than conducting a census of every tree - the value of the additional information does not pay back the cost of collecting it.


But there is a cost to having imperfect information.  If we had a perfect information - a complete census - about our forest, we could feed that data to an optimal harvest scheduler and it would give us the best possible management plan.  When we have imperfect information, our harvest scheduler will make some "mistakes" that cause us to harvest some areas too early and others too late.  The more imperfect our information, the more mistakes we'll have in our management plan.

Like most things in life, finding the optimal level of inventory information to collect is all about tradeoffs.  The key question is: "at what point does spending an additional dollar collecting inventory information no longer prevent more than a dollar of management mistakes?"

This turns out to be a challenging question to answer.  While it's
straightforward to know how much we are spending on inventory for a given level of precision (the red and green lines), how do we quantify the cost of the "mistakes" in our suboptimal harvest schedule (the black line)?  There are all sorts of variables that come into play, including forest type, age, product prices, growth predictions, etc.  Until now, no one has been able to come up with a clear way to answer this important question across the US.  In a 2008 paper called "The Value of Timber Inventory Information", Borders et al. discuss the tradeoffs of cost and imperfect information (“cost+loss”) in loblolly pine plantations and natural stands. This paper gives a great economic analysis, but that doesn’t help you decide how many plots, and what type or size of plots to install in your forest!
However, it turns out that we can use the excellent USFS FIA dataset and some clever coding to get to the bottom of this.  The beauty of the FIA dataset is that it has spatially-explicit stem records (as shown in the diagram on the left).  For any region in the US, we can use a collection of nearby FIA plots as a sort of "seed" for simulating a "virtual forest"

We now have perfect information for our virtual forest and can feed that information to a harvest scheduler to develop the optimal management plan.   In this example, the harvest scheduler tells us we'll make $100K in profit.


Now here's the cool part - we can simulate cruising our virtual forest.  For example, we could lay out 50 1/10 acre plots and see which trees are included in each.  Then we use those plots to work up a cruise summary just like for a regular cruise.

We run the cruise summary through a harvest scheduler and it returns a management plan.  But unlike our optimal management plan from before, this management plan is based off of imperfect information.  That means we are almost certainly thinning some stands too early and others too late.

And we can find out!  We apply the harvest schedule to our virtual forest and in this example, we end up with $90K in profit - $10K less than the perfect information optimum. Again, that means that we lost $10K because the information we bought was imperfect. We can do better, with better planning.


If we try again with 70 plots, we might find that we pay $500 more for those 20 extra plots, but end up with $92K in profit because we avoided a couple of management mistakes.  The extra plots were worth it.

Using this method, we finally have a quantitative approach for making decisions about our inventory design.  It's possible to simulate a range of sampling methodologies and see which is the most appropriate for our different forest types and management practices.  This method is also easily extensible to include other aspects of our inventory or planning processes.

By directly relating our sampling methodology to our profitability, this approach lets us identify opportunities to improve our forest management while adding to the bottom line.  There's much more to say about the results of this approach in different forest types - and recommendations we’ll make - stay tuned for more biometrics analysis in next month's issue!

Works Cited
B. E. Borders, W. M. Harrison, M.L. Clutter, B. D. Shiver and R.A. Souter. 2008. The Value of Timber Inventory Information. Can. J. For. Res. 38: 2287-2294.