Development in mountainous areas is inevitable in countries with high population densities, but the actual relationship between development and landslides remains uncertain. Clarifying the key current or historical factors resulting in landslides is crucial for hazard prevention and mitigation. This study focused on the Shihmen Reservoir catchment in Taiwan. Two combinations of explanatory variables in five different years (1946, 1971, 2001, 2004, and 2012) collected from a geodatabase and digital archives were used to conduct proximity and discrete logistic regression analyses. The results demonstrate that landslides increased dramatically from 1946 to 2012 in the catchment area. The proximity and overlapping of human development with landslides increased. However, the logistic regression results indicated that variation in susceptibility to landslides was due to natural causes, with the exception of historical deforestation and newly constructed road systems. Therefore, well-recovered historical woodland sites might currently be landslide-prone areas. We suggest that cumulative historical events should be considered as explanatory variables in future landslide prediction analysis.

Tea is an important but vulnerable economic crop in East Asia, highly impacted by climate change. This study attempts to interpret tea land use/land cover (LULC) using very high resolution WorldView-2 imagery of central Taiwan with both pixel and object-based approaches. A total of 80 variables derived from each WorldView-2 band with pan-sharpening, standardization, principal components and gray level co-occurrence matrix (GLCM) texture indices transformation, were set as the input variables. For pixel-based image analysis (PBIA), 34 variables were selected, including seven principal components, 21 GLCM texture indices and six original WorldView-2 bands. Results showed that support vector machine (SVM) had the highest tea crop classification accuracy (OA = 84.70% and KIA = 0.690), followed by random forest (RF), maximum likelihood algorithm (ML), and logistic regression analysis (LR). However, the ML classifier achieved the highest classification accuracy (OA = 96.04% and KIA = 0.887) in object-based image analysis (OBIA) using only six variables. The contribution of this study is to create a new framework for accurately identifying tea crops in a subtropical region with real-time high-resolution WorldView-2 imagery without field survey, which could further aid agriculture land management and a sustainable agricultural product supply.

Precisely estimating the yield of paddy rice is crucial for national food security and development evaluation. Rice yield estimation based on satellite imagery is usually performed with global regression models; however, estimation errors may occur because the spatial variation is not considered. Therefore, this study proposed an approach estimating paddy rice yield based on global and local regression models. In our study area, the overall per-field data might not available because it took lots of time and manpower as well as resources. Therefore, we gathered and accumulated 26 to 63 ground survey sample fields, accounting for about 0.05% of the total cultivated areas, as the training samples for our regression models. To demonstrate whether the spatial autocorrelation or spatial heterogeneity exists and dominates the estimation, global models including the ordinary least squares (OLS), support vector regression (SVR), and the local model geographically weighted regression (GWR) were used to build the yield estimation models. We obtained the representative independent variables, including 4 original bands, 11 vegetation indices, and 32 texture indices, from SPOT-7 multispectral satellite imagery. To determine the optimal variable combination, feature selection based on the Pearson correlation was used for all of the regression models. The case study in Central Taiwan rendered that the error rate was between 0.06% and 13.22%. Through feature selection, the GWR model’s performance was more relatively stable than the OLS model and nonlinear SVR model for yield estimation. Where the GWR model considers the spatial autocorrelation and spatial heterogeneity of the relationships between the yield and the independent variables, the OLS and nonlinear SVR models lack this feature; this led to the rice yield estimation of GWR in this study be more stable than those of the other two models.

When original aerial photographs are combined with deep learning to classify forest vegetation cover, these photographs are often hindered by the interlaced composition of complex backgrounds and vegetation types as well as the influence of different deep learning calculation processes, resulting in unpredictable training and test results. The purpose of this research is to evaluate (1) data preprocessing, (2) the number of classification targets, and (3) convolutional neural network (CNN) approaches combined with deep learning’s effects on high-resolution aerial photographs to identify forest and vegetation types. Data preprocessing is mainly composed of principal component analysis and content simplification (noise elimination). The number of classification targets is divided into 14 types of forest vegetation that are more complex and difficult to distinguish and seven types of forest vegetation that are simpler. We used CNN approaches to compare three CNN architectures: VGG19, ResNet50, and SegNet. This study found that the models had the best execution efficiency and classification accuracy after data preprocessing using principal component analysis. However, an increase in the number of classification targets significantly reduced the classification accuracy. The algorithm analysis showed that VGG19 achieved the best classification accuracy, but SegNet achieved the best performance and overall stability of relative convergence. This proves that data preprocessing helps identify forest and plant categories in aerial photographs with complex backgrounds. If combined with the appropriate CNN algorithm, these architectures will have great potential to replace high-cost on-site forestland surveys. At the end of this study, a user-friendly classification system for practical application is proposed, and its testing showed good results.

This study proposed a novel methodology to classify the shape of gaps using landscape indices and multivariate statistics. Patch-level indices were used to collect the qualified shape and spatial configuration characteristics for canopy gaps in the Lienhuachih Experimental Forest in Taiwan in 1998 and 2002. Non-hierarchical cluster analysis was used to assess the optimal number of gap clusters and canonical discriminant analysis was used to generate the discriminant functions for canopy gap classification. The gaps for the two periods were optimally classified into three categories. In general, gap type 1 had a more complex shape, gap type 2 was more elongated and gap type 3 had the largest gaps that were more regular in shape. The results were evaluated using Wilks’ lambda as satisfactory (p < 0.001). The agreement rate of confusion matrices exceeded 96%. Differences in gap characteristics between the classified gap types that were determined using a one-way ANOVA showed a statistical significance in all patch indices (p = 0.00), except for the Euclidean nearest neighbor distance (ENN) in 2002. Taken together, these results demonstrated the feasibility and applicability of the proposed methodology to classify the shape of a gap.