Downscaling Land Surface Temperature from MODIS Dataset with Random Forest Approach over Alpine Vegetated Areas

Reconstruction of all-weather land surface temperature based on a combined physical and data-driven model

At present, the remote sensing (RS) thermal infrared (TIR) images that are commonly used to obtain land surface temperature (LST) are contaminated by clouds and thus cannot obtain spatiotemporal integrity of LST. To solve this problem, this study combined a physical model with strong interpretability with a data-driven model with high data adaptability. First, the physical model (Weather Research and Forecast (WRF) model) was used to generate LST source data. Then, combined with multisource RS data, a data-driven method (random forest (RF)) was used to improve the accuracy of the LST, and a model framework for a data-driven auxiliary physical model was formed. Finally, all-weather MODIS-like data with a spatial resolution of 1 km were generated. Beijing, China, was used as the study area. The results showed that in cases of more clouds and fewer clouds, the reconstructed all-weather LST had a high spatial continuity and could restore the spatial distribution details of the LST well. The mean absolute error (MAE), root mean square error (RMSE), and correlation coefficient (ρ) in the case of more (fewer) clouds were ranked as follows: MAE < 1 K (< 2 K), RMSE < 2 K (< 2 K), and ρ > 0.9. The errors obeyed an approximately normal distribution. The total MAE, RMSE, and ρ were 0.80 K, 1.09 K, and 0.94 K, respectively. Generally, the LST reconstructed in this paper had a high accuracy, and the model could provide all-weather MODIS-like LST to compensate for the disadvantages of satellite TIR images (i.e., contamination by clouds and inability to obtain complete LST values).

Assessment of the dynamics of urban surface temperatures and air pollution related to COVID-19 in a densely populated City environment in East Java

The COVID-19 pandemic that has hit the whole world has caused losses in various aspects. Several countries have implemented lockdowns to curb the spread of the SARS-CoV-2 virus that caused death. However, for developing countries such as Indonesia, it is not suitable for lockdown because it considers the economic recession. Instead, the Large-scale Social Restrictions (LSSR) regulation is applied, the same as the partial lockdown. Thus, it is hypothesized that implementing LSSR that limits anthropogenic activities can reduce heat emissions and air pollution. Utilization of remote sensing data such as Terra-MODIS LST and Sentinel-5P images to investigate short-term trends (i.e., comparison between baseline year and COVID-19 year) in surface temperature, Surface Urban Heat Islands Intensity (SUHII), and air pollution such as NO₂, CO, and O₃ in Malang City and Surabaya City, East Java Province. Spatial downscaling of LST using the Random Forest Regression technique was also carried out to transform the spatial resolution of the Terra-MODIS LST image to make it feasible on a city scale. Raster re-gridding was also implemented to refine the Sentinel-5P spatial resolution. The accuracy of LST spatial downscaling results is quite satisfactory in both cities. Surface temperatures in both cities slightly decreased (below 1 °C) during LSSR was applied (P < 0.05). SUHII in both cities experienced a slight increase in both cities during LSSR. NO₂ gas was reduced significantly (P < 0.05) in Malang City (∼38%) and Surabaya City (∼28%) during LSSR phase due to reduced vehicle traffic and restrictions on anthropogenic activities. However, CO and O₃ gases did not indicate anomaly during LSSR. Moreover, this study provides insight into the correlation between SUHII change and the distribution of air pollution in both cities during the pandemic year. Air temperature and wind speed are also added as meteorological factors to examine their effect on air pollution. The proposed models of spatial downscaling LST and re-gridding satellite-based air pollution can help decision-makers control local air quality in the long and short term in the future. In addition, this model can also be applied to other ecological research, especially the input variables for ecological spatial modeling.

A Global 250-m Downscaled NDVI Product from 1982 to 2018

Satellite-based normalized difference vegetation index (NDVI) time series data are useful for monitoring the changes in vegetation ecosystems in the context of global climate change. However, most of the current NDVI products cannot effectively reconcile high spatial resolution and continuous observations in time. Here, to produce a global-scale, long-term, and high-resolution NDVI database, we developed a simple and new data downscaling approach. The downscaling algorithm considers the pixel-wise ratios of the coefficient of variation (CV) between the coarse- and fine-resolution NDVI data and relative changes in the NDVI against a baseline period. The algorithm successfully created a worldwide monthly NDVI database with 250 m resolution from 1982 to 2018 by translating the fine spatial information from MODIS (Moderate-resolution Imaging Spectroradiometer) data and the long-term temporal information from AVHRR (Advanced Very High Resolution Radiometer) data. We employed the evaluation indices of root mean square error (RMSE), mean absolute error (MAE), and Pearson’s correlation coefficient (Pearson’s R) to assess the accuracy of the downscaled data against the MODIS NDVI. Both the RMSE and MAE values at the regional and global scales are typically between 0 and 0.2, whereas the Pearson’s R values are mostly above 0.7, which implies that the downscaled NDVI product is similar to the MODIS NDVI product. We then used the downscaled data to monitor the NDVI changes in different plant types and places with significant vegetation heterogeneity, as well as to investigate global vegetation trends over the last four decades. The Google Earth Engine platform was used for all the data downscaling processes, and here we provide a code for users to easily acquire data corresponding to any part of the world. The downscaled global-scale NDVI time series has high potential for the monitoring of the long-term temporal and spatial dynamics of terrestrial ecosystems under changing environments.

Step-By-Step Downscaling of Land Surface Temperature Considering Urban Spatial Morphological Parameters

Land surface temperature (LST) is one of the most important parameters in urban thermal environmental studies. Compared to natural surfaces, the surface of urban areas is more complex, and the spatial variability of LST is higher. Therefore, it is important to obtain a high-spatial-resolution LST for urban thermal environmental research. At present, downscaling studies are mostly performed from a low spatial resolution directly to another high resolution, which often results in lower accuracy with a larger scale span. First, a step-by-step random forest downscaling LST model (SSRFD) is proposed in this study. In our work, the 900-m resolution Sentinel-3 LST was sequentially downscaled to 450 m, 150 m and 30 m by SSRFD. Then, urban spatial morphological parameters were introduced into SSRFD, abbreviated as SSRFD-M, to compensate for the deficiency of remote-sensing indices as driving factors in urban downscaling LST. The results showed that the RMSE value of the SSRFD results was reduced from 2.6 °C to 1.66 °C compared to the direct random forest downscaling model (DRFD); the RMSE value of the SSRFD-M results in built-up areas, such as Gulou and Qinhuai District, was reduced by approximately 0.5 °C. We also found that the underestimation of LST caused by considering only remote-sensing indices in places such as flowerbeds and streets was improved in the SSRFD-M results.

Improving LST Downscaling Quality on Regional and Field-Scale by Parameterizing the DisTrad Method

Many scientists have been investigating Land Surface Temperature (LST) because of its relevance in water management science due to its direct influence on the hydrological water cycle. This effect stems from being one of the most significant variables influencing evapotranspiration. One of the most important reasons for the evapotranspiration retrieved from MODIS data’s limited suitability for scheduling and planning irrigation schemes is the lack of spatial resolution. As a result, high-resolution LST is required for estimating evapotranspiration. The goal of this study is to improve the resolution of the available LST data, to improve evapotranspiration (ETa) estimation using statistical downscaling with Normalized Difference Vegetation Index (NDVI) as a predictor. The DisTrad (Disaggregation of Radiometric Surface Temperature) method was used for the LST downscaling procedure, which is based on aggregating the NDVI map to the LST map resolution and then calculating the coefficient of variation of the native NDVI map within the aggregated pixel and classifying the aggregated map into three classes: NDVI < 0.2 for the bare soil, 0.2 ≤ NDVI ≤ 0.5 for the partial vegetation, and NDVI > 0.5 for the full vegetation. DisTrad uses 25% of the pixels with the lowest coefficient of variation from each class to calculate the regression coefficients. In this work, adjustments to the DisTrad method were implemented to enhance downscaling LST and to examine the impacts of that alteration on the evapotranspiration estimation. The linear regression model was tested as an alternative to the original second-order polynomial. In using 10% of the pixels instead of the originally proposed 25% with the lowest coefficient of variation values, it is assumed that a group of pixels with a lower coefficient of variation represents a more homogeneous area, thus it gives more accurate values. The downscaled LST map retrieval was validated using Landsat 8 thermal maps (100 m). Applying the modified DisTrad approach to disaggregate Landsat LST to 30 m (NDVI resolution) yielded an R2 of 0.72 for the 10%, 0.74 for the 25% and 0.61 for the second-order polynomial lowest coefficient of variation compared to native LST Landsat, which means that 10% can be used as an alternative. Applying the downscaled LST map to estimate ETa yielded R2 0.84 in both cases, compared to ETa yielded from the native Landsat LST. These results prove that using the robust linear regression provided better results than using polynomial regression. With the downscaled Land Surface Temperature data, it was possible to create detailed ETa maps of the small agricultural fields in the test area.

Solar-Induced Chlorophyll Fluorescence Trends and Mechanisms in Different Ecosystems in Northeastern China

Solar-induced chlorophyll fluorescence (SIF), when used as a proxy for plant photosynthesis, can provide an indication of the photosynthesis rate and has the potential to improve our understanding of carbon exchange mechanisms within an ecosystem. However, the relationships between SIF and vegetation indices (VIs) operating within different ecological contexts and the effect of other environmental factors on SIF remain unclear. This study focused on three ecosystems (cropland, forest, and grassland), with different ecological characteristics, located in Northeast China. These areas provide case studies where numerous relationships can be explored, including the correlations between the Orbiting Carbon Observatory-2 (OCO-2) SIF and MODIS products, meteorological factors, and the differences in the relationships between the three different ecosystems. Some interesting results and conclusions were obtained. First, in different ecosystems, the relationships between SIF and MODIS products show different correlations, whereby the enhanced vegetation index (EVI) has a close relationship with SIF in all the three ecosystems of forest, cropland, and grassland. Second, forest-type ecosystems appear to be sensitive to changes in daily temperature, whereas cropland and grassland areas respond more closely to changes in previous 16-day daily minimum temperature. Compared with forest and cropland areas, grasslands were more sensitive to precipitation (although the R2 value was small). Third, different ecosystems have different mechanisms of photosynthesis. Hence, we suggest that it is better to use SIF in areas exhibiting different ecological characteristics, and different models should be employed while simulating SIF.

MSG-GAN-SD: A Multi-Scale Gradients GAN for Statistical Downscaling of 2-Meter Temperature over the EURO-CORDEX Domain

One of the most important open challenges in climate science is downscaling. It is a procedure that allows making predictions at local scales, starting from climatic field information available at large scale. Recent advances in deep learning provide new insights and modeling solutions to tackle downscaling-related tasks by automatically learning the coarse-to-fine grained resolution mapping. In particular, deep learning models designed for super-resolution problems in computer vision can be exploited because of the similarity between images and climatic fields maps. For this reason, a new architecture tailored for statistical downscaling (SD), named MSG-GAN-SD, has been developed, allowing interpretability and good stability during training, due to multi-scale gradient information. The proposed architecture, based on a Generative Adversarial Network (GAN), was applied to downscale ERA-Interim 2-m temperature fields, from 83.25 to 13.87 km resolution, covering the EURO-CORDEX domain within the 1979–2018 period. The training process involves seasonal and monthly dataset arrangements, in addition to different training strategies, leading to several models. Furthermore, a model selection framework is introduced in order to mathematically select the best models during the training. The selected models were then tested on the 2015–2018 period using several metrics to identify the best training strategy and dataset arrangement, which finally produced several evaluation maps. This work is the first attempt to use the MSG-GAN architecture for statistical downscaling. The achieved results demonstrate that the models trained on seasonal datasets performed better than those trained on monthly datasets. This study presents an accurate and cost-effective solution that is able to perform downscaling of 2 m temperature climatic maps.

Spatial Downscaling of Land Surface Temperature over Heterogeneous Regions Using Random Forest Regression Considering Spatial Features

Land surface temperature (LST) is one of the crucial parameters in the physical processes of the Earth. Acquiring LST images with high spatial and temporal resolutions is currently difficult because of the technical restriction of satellite thermal infrared sensors. Downscaling LST from coarse to fine spatial resolution is an effective means to alleviate this problem. A spatial random forest downscaling LST method (SRFD) was proposed in this study. Abundant predictor variables—including land surface reflection data, remote sensing spectral indexes, terrain factors, and land cover type data—were considered and applied for feature selection in SRFD. Moreover, the shortcoming of only focusing on information from point-to-point in previous statistics-based downscaling methods was supplemented by adding the spatial feature of LST. SRFD was applied to three different heterogeneous regions and compared with the results from three classical or excellent methods, including thermal image sharpening algorithm, multifactor geographically weighted regression, and random forest downscaling method. Results show that SRFD outperforms other methods in vision and statistics due to the benefits from the supplement of the LST spatial feature. Specifically, compared with RFD, the second-best method, the downscaling results of SRFD are 10% to 24% lower in root-mean-square error, 5% to 20% higher in the coefficient of determination, 11% to 25% lower in mean absolute error, and 4% to 17% higher in structural similarity index measure. Hence, we conclude that SRFD will be a promising LST downscaling method.

Resolution Enhancement of Remotely Sensed Land Surface Temperature: Current Status and Perspectives

Remotely sensed land surface temperature (LST) distribution has played a valuable role in land surface processes studies from local to global scales. However, it is still difficult to acquire concurrently high spatiotemporal resolution LST data due to the trade-off between spatial and temporal resolutions in thermal remote sensing. To address this problem, various methods have been proposed to enhance the resolutions of LST data, and substantial progress in this field has been achieved in recent years. Therefore, this study reviewed the current status of resolution enhancement methods for LST data. First, three groups of enhancement methods—spatial resolution enhancement, temporal resolution enhancement, and simultaneous spatiotemporal resolution enhancement—were comprehensively investigated and analyzed. Then, the quality assessment strategies for LST resolution enhancement methods and their advantages and disadvantages were specifically discussed. Finally, key directions for future studies in this field were suggested, i.e., synergy between process-driven and data-driven methods, cross-comparison among different methods, and improvement in localization strategy.

Spatial Downscaling of Land Surface Temperature Based on a Multi-Factor Geographically Weighted Machine Learning Model

Land surface temperature (LST) is a critical parameter of surface energy fluxes and has become the focus of numerous studies. LST downscaling is an effective technique for supplementing the limitations of the coarse-resolution LST data. However, the relationship between LST and other land surface parameters tends to be nonlinear and spatially nonstationary, due to spatial heterogeneity. Nonlinearity and spatial nonstationarity have not been considered simultaneously in previous studies. To address this issue, we propose a multi-factor geographically weighted machine learning (MFGWML) algorithm. MFGWML utilizes three excellent machine learning (ML) algorithms, namely extreme gradient boosting (XGBoost), multivariate adaptive regression splines (MARS), and Bayesian ridge regression (BRR), as base learners to capture the nonlinear relationships. MFGWML uses geographically weighted regression (GWR), which allows for spatial nonstationarity, to fuse the three base learners’ predictions. This paper downscales the 30 m LST data retrieved from Landsat 8 images to 10 m LST data mainly based on Sentinel-2A images. The results show that MFGWML outperforms two classic algorithms, namely thermal image sharpening (TsHARP) and the high-resolution urban thermal sharpener (HUTS). We conclude that MFGWML combines the advantages of multiple regression, ML, and GWR, to capture the local heterogeneity and obtain reliable and robust downscaled LST data.

Downscaling Aster Land Surface Temperature over Urban Areas with Machine Learning-Based Area-To-Point Regression Kriging

Land surface temperature (LST) is a vital physical parameter of earth surface system. Estimating high-resolution LST precisely is essential to understand heat change processes in urban environments. Existing LST products with coarse spatial resolution retrieved from satellite-based thermal infrared imagery have limited use in the detailed study of surface energy balance, evapotranspiration, and climatic change at the urban spatial scale. Downscaling LST is a practicable approach to obtain high accuracy and high-resolution LST. In this study, a machine learning-based geostatistical downscaling method (RFATPK) is proposed for downscaling LST which integrates the advantages of random forests and area-to-point Kriging methods. The RFATPK was performed to downscale the 90 m Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) LST 10 m over two representative areas in Guangzhou, China. The 10 m multi-type independent variables derived from the Sentinel-2A imagery on 1 November 2017, were incorporated into the RFATPK, which considered the nonlinear relationship between LST and independent variables and the scale effect of the regression residual LST. The downscaled results were further compared with the results obtained from the normalized difference vegetation index (NDVI) based thermal sharpening method (TsHARP). The experimental results showed that the RFATPK produced 10 m LST with higher accuracy than the TsHARP; the TsHARP showed poor performance when downscaling LST in the built-up and water regions because NDVI is a poor indicator for impervious surfaces and water bodies; the RFATPK captured LST difference over different land coverage patterns and produced the spatial details of downscaled LST on heterogeneous regions. More accurate LST data has wide applications in meteorological, hydrological, and ecological research and urban heat island monitoring.

Integrating LiDAR, Multispectral and SAR Data to Estimate and Map Canopy Height in Tropical Forests

Developing accurate methods to map vegetation structure in tropical forests is essential to protect their biodiversity and improve their carbon stock estimation. We integrated LIDAR (Light Detection and Ranging), multispectral and SAR (Synthetic Aperture Radar) data to improve the prediction and mapping of canopy height (CH) at high spatial resolution (30 m) in tropical forests in South America. We modeled and mapped CH estimated from aircraft LiDAR surveys as a ground reference, using annual metrics derived from multispectral and SAR satellite imagery in a dry forest, a moist forest, and a rainforest of tropical South America. We examined the effect of the three forest types, five regression algorithms, and three predictor groups on the modelling and mapping of CH. Our CH models reached errors ranging from 1.2–3.4 m in the dry forest and 5.1–7.4 m in the rainforest and explained variances from 94–60% in the dry forest and 58–12% in the rainforest. Our best models show higher accuracies than previous works in tropical forests. The average accuracy of the five regression algorithms decreased from dry forests (2.6 m +/− 0.7) to moist (5.7 m +/− 0.4) and rainforests (6.6 m +/− 0.7). Random Forest regressions produced the most accurate models in the three forest types (1.2 m +/− 0.05 in the dry, 4.9 m +/− 0.14 in the moist, and 5.5 m +/− 0.3 the rainforest). Model performance varied considerably across the three predictor groups. Our results are useful for CH spatial prediction when GEDI (Global Ecosystem Dynamics Investigation lidar) data become available.

Estimation of Climatologies of Average Monthly Air Temperature over Mongolia Using MODIS Land Surface Temperature (LST) Time Series and Machine Learning Techniques

The objective of this research was to develop a robust statistical model to estimate climatologies (2002–2017) of monthly average near-surface air temperature (Ta) over Mongolia using Moderate Resolution Imaging Spectroradiometer (MODIS) land surface temperature (LST) time series products and terrain parameters. Two regression models were analyzed in this study linking automatic weather station data (Ta) with Earth observation (EO) images: partial least squares (PLS) and random forest (RF). Both models were trained to predict Ta climatologies for each of the twelve months, using up to 17 variables as predictors. The models were applied to the entire land surface of Mongolia, the eighteenth largest but most sparsely populated country in the world. Twelve of the predictor variables were derived from the LST time series products of the Terra MODIS satellite. The LST MOD11A2 (collection 6) products provided thermal information at a spatial resolution of 1 km and with 8-day temporal resolution from 2002 to 2017. Three terrain variables, namely, elevation, slope, and aspect, were extracted using a Shuttle Radar Topography Mission (SRTM) digital elevation model (DEM), and two variables describing the geographical location of weather stations were extracted from vector data. For training, a total of 8544 meteorological data points from 63 automatic weather stations were used covering the same period as MODIS LST products. The PLS regression resulted in a coefficient of determination (R2) between 0.74 and 0.87 and a root-mean-square error (RMSE) from 1.20 °C to 2.19 °C between measured and estimated monthly Ta. The non-linear RF regression yielded even more accurate results with R2 in the range from 0.82 to 0.95 and RMSE from 0.84 °C to 1.93 °C. Using RF, the two best modeled months were July and August and the two worst months were January and February. The four most predictive variables were day/nighttime LST, elevation, and latitude. Using the developed RF models, spatial maps of the monthly average Ta at a spatial resolution of 1 km were generated for Mongolia (~1566 × 106 km2). This spatial dataset might be useful for various environmental applications. The method is transparent and relatively easy to implement.