Introduction

Over the past century, stand density and fuel loading have increased in forests and rangelands throughout the United States, leading to a general decline in ecosystem health. The weakened condition of the forests and rangelands, coupled with drought stress in the Western United States, now place nearly 200 million acres of Federal forest and rangeland in the contiguous United States at risk of epidemic insect and disease outbreaks and catastrophic wildfires (USDA Forest Service and BLM 2004). In many areas, devastating wildfires and unprecedented insect and disease outbreaks have already occurred.

To address the threat and impact of fire, insects, and diseases to the Nation’s forests and rangelands, the United States Federal Government launched the Healthy Forests Initiative (HFI) in 2002 and enacted the Healthy Forests Restoration Act (HFRA) in 2003. Primary goals of the HFI and HFRA are to facilitate, expedite, and provide national guidance on hazardous-fuel reduction and ecosystem restoration. In areas where insects and diseases have already caused extensive tree mortality, the HFRA calls for accelerated information gathering on the impact of these mortality agents (USDA Forest Service and BLM 2004). However, collecting information on the extent and severity of mortality in a timely and cost-effective manner can be very difficult because of the vast acreages that are affected, coupled with the lack of routine monitoring in some areas. In some cases, the extent of mortality is so great that even traditional assessment methods such as aerial sketch-mapping become impractical in terms of time and cost. An efficient and cost-effective alternative assessment method is needed.

One possible alternative is to sample rather than map mortality. Unlike aerial sketch-mapping, which produces a map, but no quantitative measure of mortality, a sample provides a quantitative measure of mortality, but no map. Not all sample designs are efficient or cost-effective, but some have the potential to be less costly and time consuming than traditional methods of assessing mortality. A multistage sample, for example, is one type of sample design that attempts to optimize sampling efficiency. Sampling efficiency is increased by constraining final sampling units to fall within only certain subsections or regions of a study area rather than distributing them across the entire area. Assessing groups or clusters of sample locations is generally easier and less time consuming than assessing widely dispersed sample locations.

Diagram of a multistage
sampling design.

In a multistage sample design, a study area is initially partitioned into coarse subunits, called primary sampling units (PSUs). A subset of PSUs is selected and subdivided into secondary sampling units (SSUs). This process of selecting and further subdividing sampling units may be repeated as often as necessary, but a three-stage design is common (Figure on the right). In the final stage of the sample, a subset of the previous stage’s sampling units are selected and sampled or assessed. From these samples, the variable of interest can be estimated for the entire study area (Ciesla 2000, Schreuder and others 2004).

The dimensions and layout of sampling units can be arbitrary; however, it is generally advantageous to use strata correlated with the variable of interest as sampling units, particularly in the early stages of the design. Sampling within wisely chosen strata allows a multistage sample to take advantage of known sources of variability in the population and can greatly increase the precision of the estimate. When assessing tree mortality, possible strata might include cover type, stand density, elevation, slope, aspect, soil type, proximity to water, and others.

Multistage sample designs have been used previously to inventory timber, estimate tree mortality and volume loss caused by insects in conifer forests and estimate tree mortality caused by diseases (Ciesla 2000, Langley 1971, Munson and others 1985, White and others 1983). Ciesla (2000) reviewed an assortment of aerial photography-based multistage forest inventories. Typically, these inventories used vegetation type or tree mortality or both as initial-stage strata. Mortality strata were obtained by aerial sketch-mapping or from complete coverage aerial photography. Subsequent stages in these surveys involved collecting aerial photography over plots within the PSU strata. In most of the surveys, the third stage consisted of further subdividing the SSUs into tertiary sampling units (TSUs). A subset of the TSUs was then ground sampled.

Although samples are less costly and time consuming than censuses, multistage samples based on aerial photography and manual photo interpretation can still be expensive and time consuming. Replacing traditional aerial photography in multistage samples with digital aerial and satellite imagery can potentially reduce survey time and related costs. The cost of digital imagery continues to decrease, and economical, yet powerful computers and image processing software can now automate some of the tasks traditionally done by hand. In addition, some field assessments can be greatly reduced by analyzing very high-resolution imagery acquired over the final sampling units.

To aid forest managers in assessing the severity and extent of widespread tree mortality, the USDA Forest Service Remote Sensing Steering Committee sponsored a pilot study to develop a cost-effective, rapid, and statistically rigorous multistage sample design incorporating digital remotely sensed imagery to quantify tree mortality across large geographic areas. The method was evaluated in a piñon pine/juniper woodland.