Application of remote sensing technologies to identify impacts of nutritional deficiencies on forests

Abstract

The application of fertiliser is an important silvicultural practice that can induce significant increases in tree growth. However, as responses to fertilisation vary widely, the accurate identification of infertile stands is critical to ensure that this treatment is applied in a cost-effective manner. Through synthesising literature from a range of disciplines the aim of this review is to examine the utility of remotely sensed data for diagnosing nutritional deficiencies in forests.Key symptoms associated with nutrient deficiencies include reductions in leaf area and lower foliar concentrations of important elements and pigments such as chlorophyll, Nitrogen (N), and Phosphorus (P). Methods that have been used for diagnosing nutrient deficiencies include soil and foliage sampling and knowledge of the underlying parent material or soil type. Water availability and forest stand variables that include age, basal area, stand density, and Site Index have been found to influence forest responsiveness to fertiliser application. Although simple models have been developed from these variables their predictive utility is often limited. More recently, diagnostic tools have been developed based on variables that are more physiologically aligned with growth processes influenced by fertilisation, such as leaf area index (LAI), and forest canopy level foliage properties.Remotely sensed data provides a means of scaling predictions of nutrient limitations to the landscape level and has most potential in predicting important stand attributes such as LAI and foliage nutrition. Models of LAI have been developed using both spectral data and LiDAR. However, background effects and increased scattering from non-vegetation components limit the precision of reflectance-based models and these models are most suitable at lower LAI (≲4). In contrast, LiDAR-based models have been found to have a higher precision than reflectance-based models and do not saturate at high LAI. Empirical and semi-physical models developed from discrete LiDAR have been successfully used to predict LAI in a range of forest types and waveform LiDAR shows considerable promise for direct measurement of gap fraction.Canopy level foliage properties have been most commonly predicted using hyperspectral imagery. Considerable research shows that in situ hyperspectral imagery can be used to develop precise models of foliage chlorophyll, N, and P content, but relationships from this data source are weaker for other elements. The most commonly used platform for scaling these predictions to the canopy level has been a fixed wing aircraft although satellite imagery has also been widely used for this purpose. Of the methods used to develop models, machine learning produces the most precise models, followed by partial least squares which is in turn more precise than regression.Although much research has examined how remotely sensed data can be used to predict foliage properties and LAI, very few studies have used these relationships to develop a framework for accurately diagnosing nutrient deficiencies within forests. Further research should focus on developing decision support tools, based on remotely sensed data, for diagnosing nutrient deficiencies. Such an approach is likely to provide managers with a cost-effective means of improving forest productivity.