Land use zones and land use patterns in the Atlantic Zone of Costa Rica : a pattern recognition approach to land use inventory at the sub-regional scale, using remote sensing and GIS, applying an object-oriented and data-driven strategy

This thesis describes an approach to land use inventory at the sub-regional scale in the Guacimo-Rio Jiménez-Siquirres (GRS) area in the Atlantic Zone of Costa Rica. Therefore, the concept of "land use zones" is introduced. The land use zone (LUZ) plays a central role in the definition of an observational methodology as well for structuring dynamics in land use. Land use is described in terms of the land use pattern (LUP). The LUP denotes the farming systems and land utilization types (LUTs) occurring within a land use zone.This thesis formulates a methodology for the inventory of land use and land use change that is object-oriented and data-driven. "Object-oriented" means that land use is expressed in terms of a collection of objects (land use zones) with specific geometric and thematic characteristics. A classification system is developed so that each class contains land use zones with a characteristic thematic description, geometry, aggregation structure and dynamics. The handling of such complex object information requires that emphasis is put on the definition of a data model.For inventory purposes satellite imagery and aerial photos are used. The use of these materials involves pattern recognition. The "data-driven" approach in this case means that the classes to describe land use are not a-priori but inductive, i.e. they result from the inventory process. The data- driven approach is a strategy to gain insight in the sub-regional land use expressed in the land use patterns. The complex land use inventory process is unravelled into a number of sequentially ordered processing steps, described in the various chapters.This thesis consists of three parts.The first part of this thesis defines LUZs as a tool for the inventory of land use and land use change. From the comparison of aerial photos of 1948-1952 and 1984 we learn that the LUZs, that belong to the agricultural area, have stable boundaries. This implies that the land use zone may serve as a reference area for monitoring land use change.Data on farm size distribution and farming system composition of a number of zones were obtained by means of a farm survey. The data show significant differences in the farming system composition, on the basis of which we define the land use patterns. That these clear differences occur indicates that the LUZ serves as a spatial unit for the land use inventory at the sub-regional level. The differences in LUP relate to differences observed in land cover composition and farm size distributions. This relation indicates that information on LUP may be inferred from composite land cover and farm size characteristics, so that satellite imagery and aerial photos can be used as tools for land use inventory, if the proper rules for interpretation are applied.The correspondence between LUP and composite land cover implies that change in LUP may be inferred from change in land cover composition, under the condition that the geometric characteristics do not change. Change in land cover composition of LUZs between 1986 and 1990 was investigated using satellite imagery. Clear trends in land use change were observed, when the proper interpretation rules are applied. These trends were a decrease in the area for the cultivation of maize and in pasture land, and an increase in the area for banana and macadamia production and reforestation. Besides changes in area of crops, a change in the condition of pastures and banana plantations could be indicated.The second part describes the pattern recognition process. This concerns the identification and classification of the LUZs. First, the stratification of the GRS area into sub-regions is described. The spatial pattern, which is determined by the size, form and arrangement of the agricultural fields, is used as a key to the aerial photo interpretation and as a criterion to identify the LUZs.Once the LUZs are identified, their field size characteristics and the land cover composition are determined. A procedure is described for the per pixel land cover classification. This procedure will guide the image analyst in the complex task of defining a set of training classes with statistical properties suitable for the maximum likelihood classification. Emphasis is on the training phases. The procedure presented here makes use of supervised as well as unsupervised approaches.Special attention has been paid to the definition of LUZ classes. Statistical methods are used to identify and define the different patterns as a key to classification. Field size characteristics of the LUZs were determined. One-way analysis of variance and multiple comparison were used to evaluated differences in mean field size. This resulted in the definition of five classes for mean field size.A hierarchical cluster analysis was performed to evaluate the difference in land cover composition between the LUZs. To derive the relevant groups of corresponding LUZs from the results (represented by a dendrogram) a critical distance is defined. The critical distance denotes the minimum distance at which LUZs (or groups of LUZs) are considered significantly different with respect to their land cover composition. The critical distance reflects accuracy of data on land cover composition, which is determined by the accuracy of the land cover classification and the geometric accuracy of the land cover map and the LUZ map. The resulting composite land cover classes provide information on land use of the LUZs.Land use information is obtained by interpretation of the land cover composition and field size characteristics of the LUZs. This interpretation involves the transformation of data classes into information categories by using mapping rules (also termed decision rules). Mapping rules assign a conditional class label to an object, whereby the condition refers to a particular context. The mapping involves complex decisions. Insight in the complex decision structure is gained by putting the decision rules in hierarchical order. The result is a decision tree for the classification of LUZs in terms of LUPs. The decision tree leads to stepwise classification of LUZs. The decision tree provides a formalized description of the decisions in the classification of the LUZs.In Part Three the land use in the GRS area is evaluated with respect to bio-physical land potentials. The LUZ map and the physiographic soil map were combined. The soil unit boundaries and the land use zone boundaries corresponded to a high degree. But this does not mean that land use is in agreement with the (bio-physical) land potentials. Results show that 18 % of the GRS area is at risk of land degradation, while 51 % of the area is considered to have potential for more intensive use. Expert judgement is used to determine the suitability of the soil types for specific land utilization types (LUTs). However, the exact position of the soil type or the LUT cannot be determined at a sub-regional scale, with units being composite in nature. This introduces a fundamental uncertainty with respect to the statements on land use suitability. The study, therefore, has an exploratory character. The figures denote expectations.In the last chapter the variation in banana yield within one plantation (representing a particular LUZ) is investigated. Soil survey data explained 67 % of the variation. Combining Landsat-TM and soil data did not provide a better estimation of yields. The explained variation remained 67 %.