Optimizing yield data collection efforts for forest management planning

by 1965- Gilabert, Horacio

Forest yield estimates are an essential component of linear programming forest planning models since the volume of the forest and the classification of analysis areas or land base are usually derived using this type of data. Forest yield is normally acquired by a forest inventory or using growth and yield models that predict future yields. Since forest inventories and growth and yield models are often subject to multiple errors, when this information is used as input in the planning models the assumption of certainty for LP models is not accomplished. In this framework, a trade-off develops since underlying the acquisition of data is the belief that better data leads to better decisions and at the same time obtaining better data costs money. In regard to yield data from forest inventories and growth and yield models, and their utilization for forest planning, some research questions arise concerning the relationships between the yield information and the decisions that are adopted using uncertain yield data. This work uses a harvest scheduling linear programming model to calculate optimal inventory and growth and yield modeling efforts. The harvest scheduling model maximizes the net present value of harvest plus the value of the forest remaining after the planning horizon. Special formulations of the harvest scheduling models are built (True or Augmented Real Model) using several formulations of other harvest scheduling models (Perturbed Models, where sub-samples of inventory plots and experimental plots were used to build a yield table and an age-site class distribution of forest areas), and a reference formulation (Real Model, where all of the available inventory and experimental plots information is used to build the yield table and the age-site distribution). This Augmented Real Models were used to calculate the value of the Loss variable that is considered to be a measure of monetary losses as a result of the use of imperfect yield information in prescribing optimal harvest policies. Loss was used to estimate the trade-offs between making better decisions – i.e., harvest policies – when better information is available and the cost of obtaining better information. The value of Loss and the costs of acquiring the yield data were used to fit and empirical relationship between the amount of information (number of inventory and experimental plots, and size of the inventory plot) used in the harvest scheduling models and the total costs, i.e., Cost+Loss, in order to derive optimal inventory sampling sizes, optimal inventory plot size and the optimal number of experimental plots. An example was developed with radiata pine information in southern Chile. Results from the minimization of the Cost-plus-Loss functions indicate that when the expected monetary losses are considered, sampling sizes for the inventory are much larger than the current practices in Chile and the number of experimental plots is much smaller than the current system used in the example although the growth models were simpler than the current models used for radiata pine management in chile. The results were inconclusive in regard to the size of the inventory plot. iv The Cost-plus-Loss approach indicate that the number of inventory and experimental plots used for harvest planning purposes should be increased until the marginal value of the information provided by these plots equals the marginal cost of collecting the data. This is an alternative approach to the current statistics-oriented criteria for inventory planning. The analysis of the Loss function indicates that inventory information is more important for harvest scheduling purposes, than experimental plots information. Finally, the potential distribution and value of the objective function of the harvest scheduling model is dependent on the yield information used to develop the yield table and the distribution of forest area in age-site classes. The results indicate that there are statistical significant differences on the mean value between different sampling sizes for the inventory and for the experimental plots. v