Seonyoung Park

Relationship between land cover patterns and surface temperature in urban areas

The relationship between land cover patterns and surface temperature was examined using random forest as well as simple linear regression for two urban sites in Denver, Colorado, USA. Among four land cover types of buildings, trees, grass, and roads and parking lots, only trees and roads and parking lots show significant spatial metrics affecting surface temperature using both the methods. For trees, total class area seems the most important factor affecting surface temperature (R2 = 0.47; percentage of increased mean standard error when mean patch area is excluded %IncMSE = 5.35 for Site B in July), followed by aggregation metrics measuring physical connectedness (R2 for patch cohesion index = 0.42) and patch isolation (%IncMSE for mean Euclidean nearest neighbor distance = 6.01 for Site A in July). For roads and parking lots, the existence of dominantly large patches is the most important factor (R2 for range in patch area = 0.40, for largest patch index = 0.40, for Site B in July), followed by total class area (R2 = 0.39 for Site B in July). Despite some limitations, the findings of this study provide useful information for alleviating urban heat stress especially during summer and reducing adverse impacts of urban drought.

Downscaling of MODIS One kilometer evapotranspiration using Landsat-8 data and machine learning approaches

This study presented a MODIS 8-day 1 km evapotranspiration (ET) downscaling method based on Landsat 8 data (30 m) and machine learning approaches. Eleven indicators including albedo, land surface temperature (LST), and vegetation indices (VIs) derived from Landsat 8 data were first upscaled to 1 km resolution. Machine learning algorithms including Support Vector Regression (SVR), Cubist, and Random Forest (RF) were used to model the relationship between the Landsat indicators and MODIS 8-day 1 km ET. The models were then used to predict 30 m ET based on Landsat 8 indicators. A total of thirty-two pairs of Landsat 8 images/MODIS ET data were evaluated at four study sites including two in United States and two in South Korea. Among the three models, RF produced the lowest error, with relative Root Mean Square Error (rRMSE) less than 20%. Vegetation greenness related indicators such as Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVI), Soil Adjusted Vegetation Index (SAVI), and vegetation moisture related indicators such as Normalized Difference Infrared Index—Landsat 8 OLI band 7 (NDIIb7) and Normalized Difference Water Index (NDWI) were the five most important features used in RF model. Temperature-based indicators were less important than vegetation greenness and moisture-related indicators because LST could have considerable variation during each 8-day period. The predicted Landsat downscaled ET had good overall agreement with MODIS ET (average rRMSE = 22%) and showed a similar temporal trend as MODIS ET. Compared to the MODIS ET product, the downscaled product demonstrated more spatial details, and had better agreement with in situ ET observations (R2 = 0.56). However, we found that the accuracy of MODIS ET was the main control factor of the accuracy of the downscaled product. Improved coarse-resolution ET estimation would result in better finer-resolution estimation. This study proved the potential of using machine learning approaches for ET downscaling considering their effectiveness and ease of implementation. Future research includes development of the spatial-temporal fusion models of Landsat data and MODIS ET in order to increase temporal resolution of downscaled ET.

Detection of tropical overshooting cloud tops using himawari-8 imagery

Abstract: Overshooting convective cloud Top (OT)-accompanied clouds can cause severe weather conditions, such as lightning, strong winds, and heavy rainfall. The distribution and behavior of OTs can affect regional and global climate systems. In this paper, we propose a new approach for OT detection by using machine learning methods with multiple infrared images and their derived features. Himawari-8 satellite images were used as the main input data, and binary detection (OT or nonOT) with class probability was the output of the machine learning models. Three machine learning techniques—random forest (RF), extremely randomized trees (ERT), and logistic regression (LR)—were used to develop OT classification models to distinguish OT from non-OT. The hindcast validation over the Southeast Asia and West Pacific regions showed that RF performed best, resulting in a mean probabilities of detection (POD) of 77.06% and a mean false alarm ratio (FAR) of 36.13%. Brightness temperature at 11.2 μm (Tb11) and its standard deviation (STD) in a 3 × 3 window size were identified as the most contributing variables for discriminating OT and nonOT classes. The proposed machine learning-based OT detection algorithms produced promising results comparable to or even better than the existing approaches, which are the infrared window (IRW)-texture and water vapor (WV) minus IRW brightness temperature difference (BTD) methods.

AMSR2 soil moisture downscaling using multisensor products through machine learning approach

Soil moisture is important to understand the interaction between the land and the atmosphere, and has an influence on hydrological and agricultural processes such as drought and crop yield. In-situ measurements at stations have been used to monitor soil moisture. However, data measured in the field are point-based and difficult to represent spatial distribution of soil moisture. Remote sensing techniques using microwave sensors provide spatially continuous soil moisture. The spatial resolution of remotely sensed soil moisture based on typical passive microwave sensors is coarse (e.g., tens of kilometers), which is inadequate for local or regional scale studies. In this study, AMSR2 soil moisture was downscaled to 1km using MODIS products that are closely related to soil moisture through statistical ordinary least squares (OLS) and random forest (RF) machine learning approaches. RF (r2=0.96, rmse=0.06) outperformed OLS (r2=0.47, rmse=0.16) in modeling soil moisture possibly because RF is much flexible through randomization and adopts an ensemble approach. Both approaches identified T·V (i.e., multiplication between land surface temperature and normalized difference vegetation index) and evapotranspiration. AMSR2 soil moisture produced from the VUA-NASA algorithm appeared overestimated at high elevation areas because the characteristics of ground data for validation and correction used in the algorithm were different from those in our study area. In future study, AMSR2 soil moisture based on the JAXA algorithm will be evaluated with additional input variables including land cover, elevation and precipitation.

Classification and Mapping of Paddy Rice by Combining Landsat and SAR Time Series Data

Rice is an important food resource, and the demand for rice has increased as population has expanded. Therefore, accurate paddy rice classification and monitoring are necessary to identify and forecast rice production. Satellite data have been often used to produce paddy rice maps with more frequent update cycle (e.g., every year) than field surveys. Many satellite data, including both optical and SAR sensor data (e.g., Landsat, MODIS, and ALOS PALSAR), have been employed to classify paddy rice. In the present study, time series data from Landsat, RADARSAT-1, and ALOS PALSAR satellite sensors were synergistically used to classify paddy rice through machine learning approaches over two different climate regions (sites A and B). Six schemes considering the composition of various combinations of input data by sensor and collection date were evaluated. Scheme 6 that fused optical and SAR sensor time series data at the decision level yielded the highest accuracy (98.67% for site A and 93.87% for site B). Performance of paddy rice classification was better in site A than site B, which consists of heterogeneous land cover and has low data availability due to a high cloud cover rate. This study also proposed Paddy Rice Mapping Index (PMI) considering spectral and phenological characteristics of paddy rice. PMI represented well the spatial distribution of paddy rice in both regions. Google Earth Engine was adopted to produce paddy rice maps over larger areas using the proposed PMI-based approach.