My name is Jerome Klaas Vanclay, and I live at 29 King George VI Memorial Drive, Lismore NSW 2480, Australia. I am Professor of Sustainable Forestry at Southern Cross University (SCU) in Lismore. I oversee the forestry programme at SCU and teach the fourth-year subjects of Forest Land Use and Management, Natural Resource Policy, and Extension and Advisory Services. I am also an associate scientist with CIFOR, the Center for International Forestry Research, and lead CIFOR's research on FLORES, the Forest Land-Oriented Resource Envisioning System, a decision support system to help evaluate forest policy options. I am also active in IUFRO, the International Union of Forest Research Organizations, and currently coordinate the IUFRO Subject Group on Tropical Silviculture.
My basic training is in forestry. I studied forestry at the Australian National University, qualified as B..Sc. (For.) with first class honours in 1978, and was awarded the Schlich medal for the best student. I was subsequently awarded a Russell Grimwade scholarship to further my studies at the University of Oxford. I completed my M.Sc. in Forestry and Land Use Management in 1983, and was awarded the Commonwealth Forestry prize for my achievements. I complemented my forestry studies with a graduate Diploma in Computer Science (in 1980) and a Bachelor of Arts in statistics (in 1984) at the University of Queensland, and went on to complete a higher doctorate in forestry (D.Sc.For.) from the University of Queensland in 1992. The title of my doctoral thesis was Modelling the Growth and Yield of Tropical Forests. Part of this thesis formed the basis of my book Modelling Forest Growth and Yield: Applications to tropical moist forests, published by CAB International (Centre for Agriculture and Biosciences, Wallingford, UK) in 1994.
Prior to my appointment as Professor of Sustainable Forestry at Southern Cross University, I was Principal Scientist at the Center for International Forestry Research (CIFOR) during 1995-99. CIFOR is one of 16 centres supported by the CGIAR (Consultative Group for International Agricultural Research) to develop the knowledge and technology to alleviate poverty and conserve biodiversity throughout the tropics. In addition to my research activities, I was involved in training CIFOR staff and partners throughout the tropics in forest growth modelling and yield prediction. In 1996, the World Bank engaged me to train their technical officers in yield prediction and equip them to review allowable cut calculations for Bank projects. I also reviewed the South East Queensland Regional Forest Agreement in 1998 at the request of the Queensland Government's Department of Natural Resources.
During 1991-95, I was Professor of Tropical Forestry at the Royal Veterinary and Agricultural University in Copenhagen, Denmark. My research included indicators of sustainable timber production, silviculture of flood plain forests, growth modelling and yield prediction. I established an M.Sc course in tropical forestry, and supervised eight M.Sc and five Ph.D students through to the completion of their studies. I was an advisor to Demarara Timbers Limited, Mabura Hill, Guyana, and coordinated the IUFRO Working Party on Forest Inventory on Successive Occasions. I also completed consultancies for the Food and Agriculture Organisation of the United Nations, for the International Tropical Timber Organisation, and for the Overseas Development Administration of the British Government. The European Union engaged me as their forestry expert to the EU/FAO Expert Consultation on Crop Yield Forecasting Methods at Villefranche-sur-Mer (France) in 1994. I was a keynote speaker at the University of Oxford's International Symposium on Wise Management of Tropical Forests in 1992, and at the World Resources Institute's Workshop on Sustainability in Natural Tropical Forest Management in Washington DC in 1991.
Prior to 1991, I worked with the Queensland Forest Service (QFS) in various capacities, ultimately as Senior Principal Scientist. I was responsible for yield prediction, remote sensing, resource information systems, and technical support for land use planning within the QFS. I was a member of the Northern Rainforest Management Authority of the Wet Tropics World Heritage Area. I served as secretary (1988-90) and chair (1990-91) of the Australian Forestry Council's Research Working Group on Forest Mensuration and Management.
My contributions to the forestry profession have been recognised in several ways, most notably through the presentation of the Queen's Award for Forestry at the Commonwealth Forestry Conference in 1997, only the fifth time that this award was made. I am regularly called upon to review papers submitted to scientific journals, and have served as referee to some 20 forestry and ecological journals. In addition, I serve on the editorial board of the premier international journals Forest Ecology and Management (published by Elsevier Science B.V.) and Journal of Forest Planning (published by the Japanese Society of Forestry). I have also reviewed grant applications for the New Zealand Foundation for Research Science and Technology. I remain active in IUFRO, primarily as coordinator of IUFRO 1.07 Tropical Silviculture, and was recently elected deputy coordinator of IUFRO Division 1 (Silviculture). I was awarded an IUFRO Certificate of Appreciation in 1993.
In this submission, I want to explain how growth models work, how they can contribute to forest management, and to offer some comment on the insights that we may draw from such models. It is necessary to explain some technical details about these models, because these details are critical. Just as one shouldn't judge a book by its cover (or a car by its paint job), one should not judge a model by what you see on the screen; it is critical to understand how it works and what assumptions are being made. I'll try to do this with a minimum of jargon and unnecessary detail. I begin by describing the stand table, the usual way that foresters describe a patch of forest.
There are lots of ways that we can represent a forest. We can represent a big area with an aerial photograph or a map, and a smaller area with a conventional photograph or a diagram (Figure 1). If we want to store (or to manipulate) details of the forest on a computer, it is convenient to use numbers rather than diagrams. This is easily, and usually, done with a stand table, a summary of the numbers of trees in each of several size classes within a stand, a patch of forest with similar characteristics. Thus the stand table is directly analagous to a diagram or photograph (Figure 1), but is easier to deal with on a computer. However, like the photograph, it remains an abstract (and incomplete) representation of a small part of the forest.
It is possible to compute the average stand table for a large area of forest by averaging the stand tables for several individual stands. This is a useful way to summarize data for a forest estate, but the average so determined need not represent any individual stand.
Foresters are interested to know how forests change over time, and what a given forest stand might look like at some time in the future, and we use a growth model to help us foresee the growth and other natural changes in the forest. The main processes with which we are concerned are growth (increase in size, usually expressed as diameter increment), mortality (the death of trees), and recruitment (growth of seedlings previously too small to consider, across the threshold for inclusion in the stand table). Let me use a fountain as an analogy to explain how we use a growth model to update a stand table and project it into the future.
Let's represent the stand table from Figure 1 as an artificial cascade, with the volume of water in each pool representing the number of trees in a particular size class. (Figure 2). The flow of water between each pool represents growth, and any spillage of water represents mortality. Recruitment can be represented as the addition of fresh tap water into the topmost pool.
It's comparatively easy to measure the growth in the diameter of individual trees. We can do this by measuring the diameter of selected trees at known intervals, or by drilling a small hole in the tree and examining the distance between the growth rings. Obviously, not all trees grow at the same rate; the actual rate for any tree may depend on many factors, but the average growth rate within a size class can be determined and is a useful indicator for forest management. Mortality varies greatly over time (e.g., because of drought), and tends to be clumped (e.g., because of wind damage), so is much more difficult to determine reliably. However, we can make reasonable inferences about mortality and recruitment.
Returning to our analogy, if we know how much water is in each pond, and how much water is flowing between each pond, we have a pretty good idea of the behaviour of the system. Obviously, the water level in any given pool depends on the spillage (mortality), as well as the in- and out-flow (growth), so if we monitor the water level over a period of time, we can infer spillage (mortality). However, given that the system has been in existence for some time, if we are prepared to accept that it is in a near-stable state, we could assume that the water level in each pool remains constant, and infer mortality as the difference between the in- and out-flow. This is a reasonable assumption for all pools, except the first pool, where we also need to know that rate at which new water is added (recruitment).
All we know for sure is the outflow, but we can see that the inflow from the tap and leakage from the pool (recruitment and mortality respectively) must be related, or the water level will change. We can choose a wide range of values for inflow and leakage, without affecting the rest of the system, provided that the inflow from the tap exceeds the outflow to the next pool. We're mainly interested in the last few pools û that's where the big trees are that provide the hollows for the birds and the timber for the harvest. So it suffices to choose a relatively large number for recruitment, and set the mortality at a level that maintains the current number of trees in the class. That's fairly realistic; most natural forests have (at various times, if not always) lots of seedlings on the forest floor, most of which do not survive to grow into big trees.
Our fountain example has been useful for explanation, but is hard to computerize. For the computer, it's necessary to represent all this as numbers. So we express the stand table (the depth of water in each of the pools) as a vector (a table with one column), and the dynamic aspects (the in- and out-flows, the leakage, and the input from the tap) as a matrix (a table with the same numbers of rows and columns). This representation is very convenient for computer analysis, and allows some special mathematical gymnastics that offer some insights into the behaviour of the forest. However, although the matrix representation may look sophisticated, it is the exact mathematical equivalent of our fountain analogy, and retains all the limitations of this simplistic view of forest dynamics.
To learn more about forest dynamics, we collate growth data from individual trees in the forest, and it doesn't take long to accumulate a huge pile of data, so we need a concise summary. The matrix model is the best summary û it squeezes a lot of data into a very compact table û but it is not necessarily the most informative summary. It is very simple and convenient to make forecasts using a matrix model, but it is not the only way, and is not necessarily the best way to make these forecasts. Let's begin by looking at the strengths of the matrix approach.
The idea of using a matrix to model changes in populations has been around for at least 60 years, and PH Leslie is credited for formalizing in 1945 this approach for age-based models of animal populations. This approach is a good one, and it is worth considering its strengths before we go on to models for plant populations.
Suppose that we have an animal that lives exactly 3 years. This means our matrix will have 3 rows and 3 columns, but over half of the cells in matrix are redundant and remain empty (or zero). We need to provide estimates of fecundity (the number of offspring born to each individual) in the top row, and estimates for the proportion of individuals surviving and progressing to the next age class (on the lower diagonal).
Let's assume that our animals typically give birth to one offspring after they turn one year old, and to twins after they turn two. Then the top row of our matrix should contain 0, 1, 2 (no offspring, one offspring and twins in the 1st, 2nd and 3rd years of life respectively). And assume that each year, half of the animals die, irrespective of age, and that they all die on their third birthday. Then the first entry in the second row, and the 2nd entry on the 3rd row should contain a half (only half the infants survive to become 1-year olds, and half of those go on to become 2-year olds). This model is logical, and consistent with our understanding of the dynamics of the animal:
I'd like to illustrate two things about of this model, and I'll do this by demonstrating with a simple MS-Excel spreadsheet. Firstly, unless we choose mortality and fecundity carefully, we create a situation where our animals either die out, or where they multiply at an ever-increasing rate (exponentially). Secondly, no matter what numbers we start with, the proportion of animals in each class soon stabilises (in this case, at 4:2:1). This is because of two important assumptions inherent in this modelling approach: (1) that survival and fecundity depend only upon the class an animal is in (and not on the numbers of animals in other classes or in total), and (2) that estimates of survival and fecundity do not change over time. These assumptions are convenient, but not always realistic.
The Leslie model offers an insight into the age structure of a population, but we are often interested in other aspects û specifically, in the case of trees, we often are interested in size and vigour rather than age. The model is easily modified by adding some additional entries - this was first done by Michael Usher who published a matrix model for Scots pine forest in 1966. Let's convert the model above from an age-based model, to a size-based model for small, medium and big trees.
The principal change to be made is that the diagonal needs to have some numbers. In the age-based model, no trees remain the same age in successive years, but in the size-based model, some trees will remain within the same size class. So, lets assume that in any year, half the trees (0.5) remain within a class, thirty percent (0.3) grow into the next class, and twenty percent die. Obviously, the big trees cannot grow into a bigger class, so the bottom right corner represents survival. The bottom left corner remains zero, because trees cannot grow from small to big trees in one time step.
The numbers representing fecundity also need to be re-examined. As it stands, every medium size tree will create another small tree, and every large tree will create another two trees. That's not realistic û while big trees may set more seed, the main effect of big trees in many forests will be to suppress rather than to foster regeneration. There are two solutions that can be used alone or in conjunction. One way is simply to add a realistic amount of regeneration to the top left cell each year. Another is to use negative numbers in the top row of the matrix to try to account for suppression. Both of these "fixes" spoil some of the attractive mathematical properties of the matrix approach, including the possibility that the model may predict a negative number of trees in one or more size classes. Nonetheless, if care is taken in applying the model and interpreting the results, these models can offer insights helpful for forest management.
In response to Timberlands plans and model for the Maruia Working Circle, Landcare constructed and published a model for Beech forests. Unfortunately, the description of model published by Murray Efford in the Journal of the Royal Society of New Zealand (Vol. 29, No. 2, June 1999) is not entirely consistent with the version made available at http://www.landcare.cri.nz/information_services/media/1999/beechm.exe, and absence of the source code with the latter precludes close scrutiny. The description and model are consistent in terms of initial composition and growth rates, but differ in mortality and recruitment. The Journal description tabulates positive recruitment coefficients (cf. fecundity), but it is apparent that the model simply adds 57 new recruits each year. The model acknowledges the possibility that harvesting may stimulate growth rates by reducing competition (by providing a slider for growth compensation), but does not acknowledge that the same processes may stimulate recruitment and reduce mortality (except indirectly, since faster growth means that trees grow more quickly out of the high-mortality smaller size classes).
The Landcare model is similar to the Timberlands matrix model, except that an adjustment is made to account for the distribution of stems within a size class. The Timberlands results can effectively be reproduced with the Landcare model, so the principal issue is not about the model, but rather the assumptions made in using the model. The principal question relates to the extent that mortality-prone trees can be recognised: if they can be reliably and consistently recognised, the harvest can be subsumed within the model; if not, the harvest will largely be additional to mortality.
Before we examine the implications of using the model, it is prudent to consider the possible consequences of any errors or bias in the model. What happens if the model is wrong? Fortunately, the consequences of model errors are small, provided that there is effective monitoring of the system. If the model is wrong, the anticipated timber supply may not be there in the long term, but the consequences to the forest will be small. The system depends on the ability of field staff to find trees that are about to die, and to harvest with minimal impact, these trees and any neighbouring trees that would be destroyed by the natural fall of dying trees. That should make the system fairly robust, provided that the harvest doesn't increase the overall mortality rate. If the forest is harvested too rapidly, mortality-prone trees will become scarce, and the harvest will fall, so the system should be self-regulating. There are just two caveats that require monitoring: post-harvest mortality needs to be monitored for possible change, and the selection of trees that are about to die should be monitored to ensure that standards do not change without good reason.
There are several things that could be wrong in the model. The estimates for growth, mortality or recruitment may be inappropriate for the stand in question. In practice, because of natural variation in the environment, it is more likely than not that the model will be "wrong" for any given stand, so we should not expect infallible results. However, the model should be unbiased, and should provide reasonable results on average. With some models, we can judge model performance easily by comparing model predictions against actual harvest outturn over an extended period. However, with the Timberlands silvicultural system, any such discrepancy is likely to reflect imprecision in reconnaissance inventory rather than in model performance. It is possible to devise construct experiments to elucidate model weaknesses, but it is unproductive to demonstrate model failure under circumstances beyond the intended scope of the model. The implicit assumption that transition probabilities do not change over time and do not depend on stand condition means that matrix models work best over a limited range of site and stand conditions. It is appropriate to establish the range of sites and stands over which satisfactory performance can be expected, but irrelevant to criticise poor performance beyond this range.
Since temperate trees form annual rings, it is relatively easy to get reasonable estimates of growth, but since not all trees grow at the same rate, the challenge is to establish the general growth trend for trees in the range of situations of interest. The actual growth of a tree may depend on the site, its proximity and size relative to neighbours, and other factors. Some models attempt to take many of these factors into account, but they can only be partially accommodated in the kind of models considered by Timberlands and Landcare, by creating local variants of the basic model.
Obviously, if growth rates used in the models are wrong, predictions of future tree sizes (and hance of future harvests) will also be wrong. However, the nature of the forest, and of timber harvests in the short to medium term depends substantially on the existing forest, and not on short-term growth, so any inaccuracies in growth estimates will have little effect on the forest. Timberlands has established a series of permanent plots that will allow estimates to be refined as data become available.
Since timber harvests depend on anticipated mortality, any bias in expected growth rates will only be manifested in the very long term, when log sizes may be larger or smaller than anticipated.
Mortality is more difficult to quantify, partly because it tends to be clustered in time and space, and partly because there is no unambiguous record of past mortality - unlike the neat growth rings present in every tree. Thus mortality estimates in both the Timberlands and Landcare simulations have been obtained by assuming that the present beech stands are in equilibrium, and if left undisturbed, would tend to maintain their present size structure. In the absence of data to the contrary, that is a reasonable assumption, but it remains an untested assumption.
Any bias in mortality estimates will have an immediate and direct effect on the timber harvest, and our expectations about the structure of the future forest will not materialize. However, since Timberlands harvests only dying trees, the consequences for the forest are negligible û the forest will continue to develop in a natural way, but not in the way we anticipated. The consequences for Timberlands are more significant, since the harvest depends not on the predicted mortality, but on actual mortality trends, so any bias will quickly become evident in the number of trees available for harvest.
The real concern is whether harvesting will influence natural mortality rates. While it is impossible to rule this out completely, it seems unlikely, as very little unnatural disturbance occurs. The main differences between harvesting and natural death is that the tree is felled a few years before its anticipated natural death, and that the useful part of the bole of the tree is removed by helicopter.
In conventional logging operations, it is not uncommon that machines may damage some trees and disturb the soil, but the Timberlands approach causes very little disturbance indeed. It is possible that trees may grow asymmetrically to fill gaps created by harvesting, and that this may predispose them to leaning, and subsequently falling, in that direction, but any such effect must be natural, since Timberlands attempts to emulate natural tree fall.
The estimates of recruitment are possibly the most contentious in the model. Growth rates are easily checked by measuring growth rings, and mortality estimates are based on a reasonable assumption based on maintaining the status quo, but the assumption of 57 recruits per hectare each year is based on slim evidence indeed. However, it seems a reasonable estimate since it maintains the status quo in simulations studies, and seems to be a conservative estimate for the abundant seedlings usually present in the field.
Any bias in this estimate is likely to be inconsequential, at least in the short to medium term. Regeneration and recruitment usually depends on light reaching the forest floor, and on gaps in the canopy, so in the absence of major disturbances, is likely to be self-compensating. Any shortfall in recruitment should become evident through routine monitoring long before it presents an ecological problem.
Parameters such as growth, mortality and recruitment, are relatively to check and explore using sensitivity analyses. Other assumptions can be of greater significance and can be harder to test. Perhaps the most important issue is the structure of the model itself. Is a stand table based modelling approach appropriate for the ecological system and the management issues under consideration?
Few detailed comparisons of different modelling approaches have been reported in the scientific literature. In my work with north Queensland rainforests, I have used both stand table and tree list based models, and found the overall inferences about total harvest volume to be comparable. However, the stand table based approach does not provide the same amount of detail as alternative approaches, and is not really suitable for examining inter-tree competition and growth responses following harvesting. Thus care must be taken when attempting to draw inferences from stand table based models. However, the model should be suitable to give an overall impression of the future structure of beech forests.
The Timberlands and Landcare models help focus attention on the dynamics of the beech forests, but in some respects these models are irrelevant to the practical issues of harvesting dying trees from these forests. The harvesting system relies on field recognition and harvesting of trees that are about to die, and not on computer predictions of mortality. If the system succeeds in its purest form, the forest will continue to develop in a natural way. The only difference between a Timberlands forest and an entirely natural forest will be that fallen logs do not remain on the forest floor, but are lifted out of the forest and put good use. The role of the computer model is to help foresters gain a better understanding of forest dynamics, and to help Timberlands anticipate the medium to long-term timber supply.
There are two critical practical issues implicit in the Timberlands system: the ability of Timberlands staff to recognise trees that are about to die, and to harvest these trees without causing additional mortality. Both issues can be satisfied, and can be tested through routine monitoring.
Most foresters can confidently appraise the vigour of a tree, and give a reasonable assessment of the likely longevity of a tree exhibiting symptoms of senescence, decay or other defects. This skill can be honed through training and tested through routine monitoring.
It is inevitable that some deaths cannot be anticipated (e.g., due to lightning strikes on a previously vigorous tree), and that some predictable deaths will be overlooked. Hence it is appropriate that Timberlands does not plan to harvest all the anticipated mortality. Conversely, it is possible that some trees assessed as likely to die could be tougher than anticipated, and if not harvested, might survive in the forest for a few more decades. The incidence and fate of such "sad but tough" trees is likely to be low, and can be monitored.
Most logging operations create a significant amount of disturbance, require the removal of trees to provide access (logging roads and snig tracks), and cause additional mortality through damage to residual trees. The Timberlands system is different: tree fellers arrive on foot, directional felling is used, and the log is lifted out by helicopter. It seems likely that there will be some, but very little additional mortality. Since trees are deliberately felled before their death, they may have bigger and heavier crowns than they would have at the time of natural tree fall, and they may thus leave a larger "footprint" when they are felled. Conversely, it is possible that the deliberate felling for harvesting may direct trees into gaps and away from established young trees, and may thus reduce the mortality of understorey trees. It is impossible to assert that there will be no additional mortality, but it seems that with suitable training, and care and attention to detail, any such additional mortality will be minimal.
Sustainability is generally assessed on the basis of ecological, economic and social criteria. I prefer to comment only on some ecological and social aspects relating to the physical structure of the forest. The Timberlands system attempts to pre-empt and harvest only the natural mortality in the forest. Where it succeeds, the only difference in between Timberlands forests and natural forests should be the number of "stags" (standing dead trees) and the volume and dimensions of dead logs on the forest floor. This difference may have implications for entomology, but is unlikely to detract from the visual appearance, the nutrient dynamics or the growth rates of the forest.
Perhaps the most important question relating to sustainability is a simple one: Will the forest still exist in the future, and what will it look like? For me, the answer is clear. I expect that the forest will look virtually indistinguishable from comparable areas from which timber harvesting has been excluded, with just two exceptions: the provision of roads may make the forest more accessible, and predator control may mean that the Timberlands forests may have more birds.
The Timberlands system relies on two important principles (recognising trees about to die, and harvesting them without causing additional mortality), and it is important to monitor operations to ensure that these conditions are satisfied. Such monitoring systems should provide an early warning system to alert us if something is amiss, and should provide information for management so that operations can be progressively and continually fine-tuned. Fortunately, such monitoring is easy to implement, and has been implemented by Timberlands.
Some trees assessed as about to die are not harvested due to their high level of defect; these are uniquely identified, tagged and their locations recorded, and these trees can help confirm field assessments of both likely death and direction of fall. It is possible that such defective trees are more likely to die, or are easier to recognise and appraise than the bulk of the harvested trees, and the permanent plot system established by Timberlands will provide a more reliable basis for checking prognoses.
The locations of all harvested trees is recorded, and can be plotted onto maps, providing a revealing insight into how quickly the forest is being worked over. At a more local scale, the distribution of harvested stems offers other interesting insights: whether the forest has been methodically searched or whether patches have been missed; and whether any spatial pattern is evident amongst senescent trees. Together, these records should provide a good basis to monitor operations and forewarn of any possible problems. Monitoring procedures should reveal any discrepancies in field appraisals, any significant additional mortality, or any detectable changes in growth rates or other ecological processes, and should allow detection of any such changes before any major impacts are observed.
There has been much rhetoric about sustainable forestry, but few practical examples exist. Worldwide, most natural forests are exploited for short-term gain rather than managed for sustainable multiple use. A few forests have been certified as sustainable, but these are managed as production forests and no longer resemble natural forests. The rainforests of north Queensland were once promoted as a model of sustainable forestry, but these are now managed as National Parks within the World Heritage Area. In South Africa, there is one example of a harvesting scheme similar to Timberlands, but we desperately needs more viable examples of this kind if we are to promote sustainable forestry as a feasible option.
New Zealand may achieve more for forest conservation globally by promoting the sustainable use of these beech forests, than can be achieved locally by adding to the already extensive beech conservation reserves.