The interpretation of remote sensing : a feasibility study

This thesis describes research done to ascertain the possibilities and limitations of the use of remote sensing observations for agriculture. The topic is defined in Chapter 1. In Chapter 2 the possible applicability of certain existing models for this study is examined. Three models are developed further in Chapter 3. Two of them describe the relation between properties of a crop and its reflective behaviour; the third is a model for soil reflection. Chapter 4 presents calculations on hypothetical crops; in Chapter 5 some calculations based on actual crop data are described.In the early years of the use of remote sensing, the agricultural applications of this technique were almost exclusively qualitative. However, the quantitative applications have been steadily increasing. These applications always require measured electromagnetic radiative energy to be translated into parameters of agricultural interest, such as coverage, leaf area index, biomass, development stage or health conditions of a crop, and in most situations the qualitative information which crop is being grown must also be determined from the observed radiation.The problem is defined in Chapter 1. The conclusion is that the optical behaviour of a crop is determined by a number of more or less independent parameters. This number generally exceeds the number of data measured. Consequently, in many situations it is very difficult to reliably estimate the quantities that cause this behaviour, if these estimations are only based on the observations. Two additional complicating factors are that the relations between the optical and the agricultural properties of a crop are surely not unequivocal, and that the reflection of a crop also depends on the spatial distribution of the incident radiation. This justifies the conclusion that it is useful to study the relations between the agricultural properties of a crop and the upward radiation, measured under different irradiance conditions.In Chapter 2, it is investigated if, and under which conditions, existing models could be used as reflection models. To determine the requirements of reflection models, an existing analytical model (Kubelka & Munk, 1931) is used to calculate the ratio of reflected radiation to absorbed radiation, and a new analytical model is used to ascertain the spatial distribution of the reflected radiation. From the calculations with the Kubelka-Munk model it is concluded that for the calculation of the absorption of incident radiation by a crop the modelling of the spatial distribution of the intercepted and remitted radiation can be strongly simplified, without compromising the quality of the absorption calculations. This justifies that in the models used by plant physiologists for calculating the absorption of photosynthetically active radiation, the spatial distribution of this secondary radiation is strongly simplified. Calculations with the distribution model show that both the crop properties and the spatial distribution of the incoming radiation may strongly influence the reflected radiation in one direction.Four existing models are examined to ascertain their applicability in the situation under consideration. As might be expected, the two absorption models (De Wit, 1965; Goudriaan, 1977) lack the level of detail that is required for the modelling of the reflected radiation. Of the two models that have primarily been developed as reflection models (Suits, 1972; Chen, 1984), the first is a very theoretical model, so if it is applied to a real crop it only permits qualitative statements. The second model should, in principle, be applicable, but its intensive use of computer resources means that a prohibitive number of calculations is required. The conclusion is that it makes sense to develop a new reflection model, incorporating several aspects of the existing models both in its theoretical basis and in its implementation.Chapter 3 is devoted to the three models that are developed in this study: the TURTLE, HARE and SOIL models. TURTLE and HARE describe a crop and SOIL describes a, non-flat soil. Both TURTLE and HARE are based on the description of a crop as 8, stack of thin crop layers. For this description, 46 directions are defined in a semispace. These directions, which are used as representatives of all possible directions, are distributed in such way that each represents an equal solid angle (0.14 sr). The mutual angle between adjacent directions is 0.42 rad. Because each model direction represents all directions within a fairly regular pentagonal or hexagonal conic sector of the space, the angle between a representative and a represented direction never exceeds 0.24 rad. These model directions are used both to represent radiation patterns and to define leaf angle distributions. In the latter case, the directions are used as vectors perpendicular to possible leaf orientations.The optical properties of a single crop layer are derived from the optical properties of the crop components and of the leaf angle distribution of this layer. The optical properties of a crop layer are described in the form of a set of four 46*46 matrices: one for the upperside reflection of the layer, one for the underside reflection, one for the upward transmission and one for the downward transmission through the layer. In the models as described, upperside and underside are assumed to be identical, so only two matrices have to be computed: one for the layer reflection and one for the layer transmission. From these matrices, and in combination with another 46*46 matrix for the soil reflection, one matrix is calculated. This matrix describes the optical behaviour of the total crop. The TURTLE and HARE models differ in the aspect that the TURTLE model allows the radiation pattern and the radiation intensity throughout the complete crop to be calculated. To enable this, in the calculations the layers are stacked one by one, in an upward direction, starting with the soil matrix. In the HARE model, the matrices for stacks of identical model layers are calculated by means of a doubling method. This way of combining layer matrices prohibits the calculation of a radiation pattern throughout the crop, but this limitation is not very drastic in remote sensing applications. The advantage of this method is that it reduces the computer time required by a factor of 5 to 10, compared with the TURTLE model. Also, the incoming radiation is described as a, vector that comprises in the 46 previously mentioned directions. Multiplying this vector by the crop reflection matrix yields the radiation remitted by the crop. The TURTLE model allows the radiation regime within the crop to be calculated in a similar way.The SOIL model is developed to investigate if the influence of a non-flat soil to the reflected radiation is so large that it would be unrealistic to model the soil as a flat reflecting surface. The calculations show that only if the coverage of the crop is very low (< 0.25), the spatial distribution of the radiation reflected by the soil is of importance in case individual wavelength bands are considered, but that this influence vanishes if instead of single spectral bands, the ratio between two bands (or a function of such a, ratio) is used as a basis for the interpretation of remote sensing observations.In Chapter 4, a large number of calculations with the HARE model are presented. For these calculations, test values based on values that are found in literature are chosen for the parameters that can be varied in the model. These parameters concern the leaf angle distribution, the leaf reflection and the leaf transmission coefficient, the reflective behaviour of the leaves, the soil reflection coefficient, the spatial distribution of the incoming radiation and the direction of observation. The calculations always concern the relation between the upward radiation and the primary crop property, namely the vertically measured coverage. It is investigated how this relation is influenced by variations in the values of the given parameters. Based on these calculations it can be concluded that these variations can be large, but that it may be expected that the relation between the vertical coverage and the ratio between the reflection in the infrared and the red bands (or a derived function such as the normalized difference between these two, the vegetation index), will be much less sensitive to the mentioned variations. For this reason a second series of calculations is done. In these calculations, the sensitivity of the relation between coverage and vegetation index for the same parameter variations is examined. It appears that this relation is indeed much less sensitive, except for changes in the observation direction. The latter phenomenon is investigated separately. The conclusion of the latter investigation is that the commonly used method to reduce the directional dependency that is based on a quadratic regression, only enhances the quality of the interpretation under special conditions, but that in some cases, this correction yields an even worse result than the result that would have been found if the correction had not been applied at all. It is indicated how, by means of the HARE model, the calculations may be improved.Finally, Chapter 5 discusses the interpretation of remote sensing observations applied to winter wheat and sugarbeet. The aim of the wheat calculation is to investigate which variation in the crop can be determined and at which moments. For the calculations, a, normal developing wheat crop was constructed, based on literature data. Some variants with a higher and a lower LAI and also three variants with strongly yellowing leaves were derived from this crop. The calculation indicates that the vegetation index only gives information about the LAI as long as the crop is green and the LAI does not exceed 3.5. Higher LAI-values cannot be distinguished, and if the crop turns yellow, it cannot be distinguished from a crop with a much lower LAI. If, besides the vegetation index the ratio red/green is also applied, the interpretation possibilities increase somewhat. The red/green ratio decreases until the crop reaches an LAI of 6, higher LAI-values cannot be determined. It is also possible, if repeated observations are carried out, to distinguish yellowing in a late growth phase from a decreasing LAI-value. The use of the vegetation index or the red/green ratio causes a drastic decrease in the influence of factors that are of no agricultural interest, but which were present in the reflection in the individual bands.The aim of the calculations with sugarbeet was to ascertain which conditions are most applicable for the detection of places where the crop droops its leaves. For this purpose the spatial pattern of the reflected radiation by a beet crop in different wavelength bands is calculated. These calculations are done both for a healthy crop and for a dehydrated crop. For both crop types the infrared/red ratio is calculated for all possible observation directions, and then the quotient of these ratios is calculated. It appears that the areas where leaf drooping has occurred can easily be identified, providing that the observation direction is chosen well (e.g. facing the sun, and with an inclination that is approximately the complement of the sun's inclination). These areas can be distinguished, even if the observations are carried out under full cloud cover. The latter conclusion is especially important if the observations are done on a small scale by using a micro-light airplane.