High-resolution Land Cover Map Research Articles

Global high categorical resolution land cover mapping via weak supervision

Global high categorical resolution land cover mapping via weak supervision

The trace of the trees: a historically situated geospatial assessment of deforestation in Bosawas biosphere reserve

ABSTRACT Deforestation is often framed as one of Nicaragua’s most pressing ecological challenges. Historically, the country has addressed the issue by relying on fence-and-control approaches to conservation. This is the case of Bosawas, the largest biosphere reserve in Central America. Despite advancements in remote sensing data, few spatially explicit attempts have been made to assess how this type of conservation scheme interacts with the country’s deforestation trends. This research combines two independently generated high-resolution land cover maps and agrarian census data to assess the extent of deforestation that has taken place in Bosawas from 1990 to 2022. Our findings indicate that the establishment of the biosphere reserve has proven ineffective in preventing deforestation and evidence that its current trend is closely related to the encroachment of capitalistic agrarian-based modes of production. This encroachment is rooted in a long history of state-led interventions, which has led to the gradual dissolution of local customary social orders, allowing the penetration of new ones that privilege individualistic and profit-oriented forms of land tenure and resources exploitation. Hence, conservation strategies that neglect the historical and political-economic roots of deforestation in Bosawas risk aggravating territorial conflicts and facilitating land dispossession modalities affecting mostly local inhabitants.

Deep learning classification of winter wheat from Sentinel optical-radar image time series in smallholder farming areas

As crop yield stagnation, climate change, and the rising demand for agricultural products pose increasing challenges, mapping crop systems is becoming more and more important. Winter wheat is one of the major cereal crops cultivated in China, ranking as the third largest crop in terms of production and harvested area. Accurately mapping winter wheat is necessary for implementing effective farm management practices. While many studies have successfully produced high spatiotemporal resolution land cover maps, relatively few map products of crop types are available in China. The growing archive of satellite image time series provides enormous opportunities to map crops more closely. This research presents a two-step method to map winter wheat based on Sentinel-1 and Sentinel-2 time-series data from Shandong Province using the deep learning approaches. The winter crops were firstly mapped using time-series optical vegetation indices employing the deep learning methods. Then winter wheat was extracted from the winter crops mask by coupling optical and synthetic aperture radar time-series images. The results indicated that the precision of mapping winter wheat using Temporal Convolution Neural Networks (TempCNN) achieved the highest precision in mapping winter wheat, with an overall accuracy of 93.7 %, a kappa coefficient of 0.907, and an F1-score of 0.989. This was followed sequentially by the Residual 1D convolutional neural networks (ResNet), the Multi-Layer Perceptron (MLP), and the Lightweight Temporal Self-Attention Encoder (L-TAE). The Temporal Attention Encoder (TAE) model demonstrated the lowest precision among the compared models. The results agree well with independent county-level official census winter wheat area data (R2 = 0.936). The proposed framework can also be applied in other regions to generate maps of different crops, so future work can extend the proposed model to other agricultural regions, where an increased number of crop types and natural vegetation types can be included and tested.

Big Data-driven MLOps workflow for annual high-resolution land cover classification models

Developing an annual and global high-resolution land cover map is one of the most ambitious tasks in remote sensing, with increasing importance due to the continual rise in validated data and satellite imagery. The success of land cover classification models largely hinges on the data quality, coupled with the application of Big Data techniques and distributed computing. This is essential for efficiently processing the extensive volume of available satellite data. However, maintaining the lifecycle of several annual Machine Learning models presents a complex challenge. The rise of Machine Learning Operations offers an opportunity to automate the maintenance of these models, a feature particularly beneficial in systems that require generating new models each year alongside the continuous integration of validated data. This article details the development of an end-to-end MLOps workflow, meticulously integrating land cover classification models that employ Big Data strategies for processing large-scale, high-resolution spatial data. The workflow is designed within a Kubernetes environment, achieving on-demand auto-scaling, distributed computing, and load balancing. This integration demonstrates the practicality and efficiency of managing and deploying models that treat satellite imagery in an automated, scalable framework, thus marking a significant advancement in remote sensing and MLOps.

Generating 10-Meter Resolution Land Use and Land Cover Products Using Historical Landsat Archive Based on Super Resolution Guided Semantic Segmentation Network

Generating high-resolution land cover maps using relatively lower-resolution remote sensing images is of great importance for subtle analysis. However, the domain gap between real lower-resolution and synthetic images has not been permanently resolved. Furthermore, super-resolution information is not fully exploited in semantic segmentation models. By solving the aforementioned issues, a deeply fused super resolution guided semantic segmentation network using 30 m Landsat images is proposed. A large-scale dataset comprising 10 m Sentinel-2, 30 m Landsat-8 images, and 10 m European Space Agency (ESA) Land Cover Product is introduced, facilitating model training and evaluation across diverse real-world scenarios. The proposed Deeply Fused Super Resolution Guided Semantic Segmentation Network (DFSRSSN) combines a Super Resolution Module (SRResNet) and a Semantic Segmentation Module (CRFFNet). SRResNet enhances spatial resolution, while CRFFNet leverages super-resolution information for finer-grained land cover classification. Experimental results demonstrate the superior performance of the proposed method in five different testing datasets, achieving 68.17–83.29% and 39.55–75.92% for overall accuracy and kappa, respectively. When compared to ResUnet with up-sampling block, increases of 2.16–34.27% and 8.32–43.97% were observed for overall accuracy and kappa, respectively. Moreover, we proposed a relative drop rate of accuracy metrics to evaluate the transferability. The model exhibits improved spatial transferability, demonstrating its effectiveness in generating accurate land cover maps for different cities. Multi-temporal analysis reveals the potential of the proposed method for studying land cover and land use changes over time. In addition, a comparison of the state-of-the-art full semantic segmentation models indicates that spatial details are fully exploited and presented in semantic segmentation results by the proposed method.

Impervious cover change as an indicator of environmental equity

Pattern-focused environmental equity research has been underpinned by high-resolution remotely sensed data to uncover spatial relationships between environmental amenities (e.g., urban tree cover) and socio-economic status (SES). A constraint imposed by reliance on high-resolution data is the inability to examine temporal patterns, primarily because of the cost of data production and the nascent state of high-resolution land cover mapping. The lack of temporal monitoring is a clear gap in pattern-focused environmental equity research. We examined temporal (2001–2019) relationships between a disamenity, impervious cover (IC), and demographic attributes for the entirety of the conterminous United States. Our main finding was 2001–2019 increases in IC were more pervasive in minority communities but these communities were not necessarily poor, and only rarely poorly educated or non-English speaking. We supported our use of IC from moderate resolution data by comparing it to high-resolution data for 24 cities within the conterminous United States. Mean Absolute Deviation (MAD) was 4.8% overall, ranging from 2.2% to 11.3% across the 24 locations. Differences in classification objectives contributed to differences in %IC estimates between the two sources.

Bridging Domains and Resolutions: Deep Learning-Based Land Cover Mapping without Matched Labels

High-resolution land cover mapping is crucial in various disciplines but is often hindered by the lack of accurately matched labels. Our study introduces an innovative deep learning methodology for effective land cover mapping, independent of matched labels. The approach comprises three main components: (1) An advanced fully convolutional neural network, augmented with super-resolution features, to refine labels; (2) The application of an instance-batch normalization network (IBN), leveraging these enhanced labels from the source domain, to generate 2-m resolution land cover maps for test sites in the target domain; (3) Noise assessment tests to evaluate the impact of varying noise levels on the model’s mapping accuracy using external labels. The model achieved an overall accuracy of 83.40% in the target domain using endogenous super-resolution labels. In contrast, employing exogenous, high-precision labels from the National Land Cover Database in the source domain led to a notable accuracy increase of 2.55%, reaching 85.48%. This improvement highlights the model’s enhanced generalizability and performance during domain shifts, attributed significantly to the IBN layer. Our findings reveal that, despite the absence of native high-precision labels, the utilization of high-quality external labels can substantially benefit the development of precise land cover mapping, underscoring their potential in scenarios with unmatched labels.

Enhancing water balance assessment in urban areas through high-resolution land cover mapping: Case study of Debrecen, Hungary

Land cover (LC) mapping in urban areas is one of the core applications in remote sensing (RS), and it plays an important role in modern urban planning and management. In order to support up-to-date simplified water balance calculation, LC modelling needs to be developed for precise and high-resolution operations to monitor rapidly expanding urban areas. This study provides a spatial decision support tool for urban water balance calculations considering hydrological parameters at pixel scale. The study is aimed to create a high-resolution LC map for more precise water balance calculations to identify and link the most important parameters, such as crop coefficient, for estimating crop evapotranspiration, runoff coefficients, and infiltration rate. Meteorological data from 2016 to 2019 were involved in evapotranspiration estimation. The study site is Debrecen, a city in northeast Hungary with a population of about 200,000. By integrating Landsat 8 imagery, Google Earth (GE), and Corine Land Cover (CLC), a LC map of 30 m resolution was created with 81.2 % overall accuracy (OA) and a coefficient of kappa greater than 0.78. For the classification of Landsat 8, seven classes were assigned (forests, sealed surfaces, areas with crop cover, grassland, semi-sealed surfaces, bare ground and surface water bodies). Classification results were validated by 101 ground truth samples using other satellite images and aerial photographs. Water balance parameters (i.e. surface runoff, infiltration and evapotranspiration) were then defined and calculated at a pixel scale and for larger areas. The technique developed in this study can also be utilized in other urban areas.

Resolving heterogeneous fluxes from tundra halves the growing season carbon budget

Abstract. Landscapes are often assumed to be homogeneous when interpreting eddy covariance fluxes, which can lead to biases when gap-filling and scaling up observations to determine regional carbon budgets. Tundra ecosystems are heterogeneous at multiple scales. Plant functional types, soil moisture, thaw depth, and microtopography, for example, vary across the landscape and influence net ecosystem exchange (NEE) of carbon dioxide (CO2) and methane (CH4) fluxes. With warming temperatures, Arctic ecosystems are changing from a net sink to a net source of carbon to the atmosphere in some locations, but the Arctic's carbon balance remains highly uncertain. In this study we report results from growing season NEE and CH4 fluxes from an eddy covariance tower in the Yukon–Kuskokwim Delta in Alaska. We used footprint models and Bayesian Markov chain Monte Carlo (MCMC) methods to unmix eddy covariance observations into constituent land-cover fluxes based on high-resolution land-cover maps of the region. We compared three types of footprint models and used two land-cover maps with varying complexity to determine the effects of these choices on derived ecosystem fluxes. We used artificially created gaps of withheld observations to compare gap-filling performance using our derived land-cover-specific fluxes and traditional gap-filling methods that assume homogeneous landscapes. We also compared resulting regional carbon budgets when scaling up observations using heterogeneous and homogeneous approaches. Traditional gap-filling methods performed worse at predicting artificially withheld gaps in NEE than those that accounted for heterogeneous landscapes, while there were only slight differences between footprint models and land-cover maps. We identified and quantified hot spots of carbon fluxes in the landscape (e.g., late growing season emissions from wetlands and small ponds). We resolved distinct seasonality in tundra growing season NEE fluxes. Scaling while assuming a homogeneous landscape overestimated the growing season CO2 sink by a factor of 2 and underestimated CH4 emissions by a factor of 2 when compared to scaling with any method that accounts for landscape heterogeneity. We show how Bayesian MCMC, analytical footprint models, and high-resolution land-cover maps can be leveraged to derive detailed land-cover carbon fluxes from eddy covariance time series. These results demonstrate the importance of landscape heterogeneity when scaling carbon emissions across the Arctic.

Time-first approach for land cover mapping using big Earth observation data time-series in a data cube – a case study from the Lake Geneva region (Switzerland)

ABSTRACT Accurate, consistent, and high-resolution Land Use & Cover (LUC) information is fundamental for effectively monitoring landscape dynamics and better apprehending drivers, pressures, state, and impacts on land systems. Nevertheless, the availability of such national products with high thematic accuracy is still limited and consequently researchers and policymakers are constrained to work with data that do not necessarily reflect on-the-ground realities impending to correctly capture details of landscape features as well as limiting the identification and quantification of drivers and rate of change. Hereafter, we took advantage of the Switzerland’s official LUC statistical sampling survey and dense time-series of Sentinel-2 data, combining them with Machine and Deep Learning methods to produce an accurate and high spatial resolution land cover map over the Lake Geneva region. Findings suggest that time-first approach is a valuable alternative to space-first approaches, accounting for the intra-annual variability of classes, hence improving classification performances. Results demonstrate that Deep Learning methods outperform more traditional Machine Learning ones such as Random Forest, providing more accurate predictions with lower uncertainty. The produced land cover map has a high accuracy, an improved spatial resolution, while at the same time preserving the statistical significance (i.e. class proportion) of the official national dataset. This work paves the way towards the objective to produce a yearly high resolution land cover map of Switzerland and potentially implement a continuous land change monitoring capability. However further work is required to properly address challenges such as the need for increased temporal resolution for LUC information or the quality of training samples.

Fine-resolution land cover mapping over large and mountainous areas for Lāna‘i, Hawaii using posterior probabilities, and expert knowledge

ABSTRACT The task of accurately mapping species-specific vegetation cover in remote and topographically complex regions like those found in Hawaiʻi presents unique challenges. This study leverages a machine learning approach to accurately classify vegetation into fine species-specific classes across the island of Lāna‘i, Hawaii, offering a novel methodology for tackling such challenges. Utilizing high-resolution WordView-2 satellite imagery, a neural network classifier and a custom lidar-based geometric correction, we introduced two new approaches to refine our high-resolution land cover classifications. This included the implementation of prior-based adjustments to class posterior probabilities to enhance land cover classification accuracy. Moreover, we developed mixed hierarchical classification maps that use class posterior probabilities to identify, at the pixel level, the finest land cover class that meets a user-defined confidence threshold. The resulting high-resolution land cover map for Lāna‘i captures the rich diversity and distribution of native and invasive plant species with high overall accuracy, generally exceeding 95%, based on independent ground control data. The capacity to produce wall-to-wall species-level vegetation maps provides a new window into monitoring vegetation dynamics on Lāna‘i and similarly remote and topographically complex regions, and contributes to our broader understanding of ecosystem responses to invasive species, climatic changes, and land management practices such as erosion and sediment control planning. Our approach offers a blueprint for similar efforts in other complex and remote ecosystems.

High-Resolution Land-Cover Mapping Based on a Cross-Resolution Deep Learning Framework and Available Low-Resolution Labels

High-Resolution Land-Cover Mapping Based on a Cross-Resolution Deep Learning Framework and Available Low-Resolution Labels

Knowledge evolution learning: A cost-free weakly supervised semantic segmentation framework for high-resolution land cover classification

Despite the success of deep learning in land cover classification, high-resolution (HR) land cover mapping remains challenging due to the time-consuming and labor-intensive process of collecting training samples. Many global land cover products (LCP) can reflect the low-level commonality (LLC) knowledge of land covers, such as basic shape and underlying semantic information. Therefore, we expect to use LCP as weakly supervised information to guide the semantic segmentation of HR images. We regard high-level specialty (HLS) knowledge as HR information unavailable in the LCP. We believe LLC knowledge can gradually evolve into HLS knowledge through self-active learning. Hence, we design a knowledge evolution learning strategy from LLC to HLS knowledge and correspondingly devise a knowledge evolution weakly supervised learning framework (KE-WESUP) based on LCP. KE-WESUP mainly includes three tasks: (1) Abstraction of LLC knowledge. KE-WESUP first adopts a training method based on superpixel to alleviate the inconsistency between LCP and HR images and directly learns the LLC knowledge from LCP according to the feature-fitting capacity of convolutional neural networks. (2) Automatic exploration of HLS knowledge. We propose a dynamic label optimization strategy to obtain a small number of point labels with high confidence and encourage the model to automatically mine HLS knowledge through the knowledge exploration mechanism, which prompts the model to adapt to complexHR scenes. (3) Dynamic interaction of LLC and HLS knowledge. We adopt the consistency regularization method to achieve further optimization and verification of LLC and HLS knowledge. To verify the effectiveness of KE-WESUP, we conduct experiments on USDA National Agriculture Imagery Program (1 m) and GaoFen-2 (1 m) data using WolrdCover (10 m) as labels. The results show that KE-WESUP achieves outstanding results in both experiments, which has significant advantages over existing weakly supervised methods. Therefore, the proposed method has great potential in utilizing the prior information of LCP and is expected to become a new paradigm for large-scale HR land cover classification.

Use of High-Resolution Land Cover Maps to Support the Maintenance of the NWI Geospatial Dataset: A Case Study in a Coastal New Orleans Region

The National Wetlands Inventory (NWI) is the most comprehensive wetland geospatial dataset in the United States. However, it can be time-consuming and costly to maintain. This study introduces automated algorithms and methods to support NWI maintenance. Through a wall-to-wall comparison between NWI and Coastal Change Analysis Program (C-CAP) datasets, a pixel-level difference product was generated at 1 m resolution. Building upon this, supplementary attributes describing wetland changes were incorporated into each NWI polygon. Additionally, new water polygons were extracted from C-CAP data, and regional statistics regarding wetland changes were computed for HUC12 watersheds. The 1 m difference product can indicate specific wetland change locations, such as wetland loss to impervious surfaces, the gain of open water bodies from uplands, and the conversion of drier vegetated wetlands to open water. The supplementary attributes can indicate the amount and percentage of wetland loss or water regime change for NWI polygons. Extracted new water polygons can serve as preliminary materials for generating NWI standard-compliant products, expediating NWI maintenance processes while reducing costs. Regional statistics of wetland change can help target watersheds with the most significant changes for maintenance, thereby reducing work areas. The approaches we present hold significant value in supporting NWI maintenance.

The impact of the German timber footprint on potential species loss in supply regions

Previous own assessments have shown that a) Germany has a wood consumption above global average, b) is strongly dependent on imports and c) has a domestic roundwood production that is at the limit of the sustainable harvest potential. Thereby Germany further increases the pressures on global forests which are already stressed by climate-change related impacts and a continuously growing global demand for wood. This leads to negative impacts on the biodiversity in the areas where wood is harvested. This paper aims to show the connection between Germany's timber consumption footprint and the impact on the biodiversity in the regions where the roundwood is sourced. A two-step process is used. In the first step, high-resolution land cover and land use maps are used as a basis for the countryside species-area relationship model, assessing the projected loss of the four taxa amphibians, birds, mammals and reptiles in relation to undisturbed natural ecosystems due to forests occupied for roundwood production. In the second step, roundwood equivalents consumed in Germany in 2015 are traced back to the region of origin using an environmentally-extended input-output analysis and the thereby induced potential species loss is calculated. We show that the highest impact on projected species richness loss caused by roundwood logging is taking place in Oceania (3.34E−03 species/m3), Carribean (1.56E−04 species/m3), and East Asia (1.43E−04 species/m3). German roundwood consumption has the highest projected species loss in the United States (7.4 species), followed by China (7.3 species) and Brazil (4.8 species). From a biodiversity impact perspective, Germany could theoretically reduce its impact by relocating imports to European countries. In view of the planetary boundary of sustainable roundwood consumption, which has already been exceeded, reducing consumption appears to be the only viable long-term option for high-consumption countries such as Germany to reduce negative impacts on global biodiversity.

Map of Land Cover Agreement: Ensambling Existing Datasets for Large-Scale Training Data Provision

Land cover information plays a critical role in supporting sustainable development and informed decision-making. Recent advancements in satellite data accessibility, computing power, and satellite technologies have boosted large-extent high-resolution land cover mapping. However, retrieving a sufficient amount of reliable training data for the production of such land cover maps is typically a demanding task, especially using modern deep learning classification techniques that require larger training sample sizes compared to traditional machine learning methods. In view of the above, this study developed a new benchmark dataset called the Map of Land Cover Agreement (MOLCA). MOLCA was created by integrating multiple existing high-resolution land cover datasets through a consensus-based approach. Covering Sub-Saharan Africa, the Amazon, and Siberia, this dataset encompasses approximately 117 billion 10m pixels across three macro-regions. The MOLCA legend aligns with most of the global high-resolution datasets and consists of nine distinct land cover classes. Noteworthy advantages of MOLCA include a higher number of pixels as well as coverage for typically underrepresented regions in terms of training data availability. With an estimated overall accuracy of 96%, MOLCA holds great potential as a valuable resource for the production of future high-resolution land cover maps.

Data availability of the published paper: Advancing High-Resolution Land Cover Mapping in Colombia: The Importance of a Locally Appropriate Legend.

Data availability of the published paper: Advancing High-Resolution Land Cover Mapping in Colombia: The Importance of a Locally Appropriate Legend.

Residential green space improves cognitive performances in primary schoolchildren independent of traffic-related air pollution exposure

BackgroundCognitive performances of schoolchildren have been adversely associated with both recent and chronic exposure to ambient air pollution at the residence. In addition, growing evidence indicates that exposure to green space is associated with a wide range of health benefits. Therefore, we aimed to investigate if surrounding green space at the residence improves cognitive performance of primary schoolchildren while taking into account air pollution exposure.MethodsCognitive performance tests were administered repeatedly to a total of 307 primary schoolchildren aged 9-12y, living in Flanders, Belgium (2012–2014). These tests covered three cognitive domains: attention (Stroop and Continuous Performance Tests), short-term memory (Digit Span Forward and Backward Tests), and visual information processing speed (Digit-Symbol and Pattern Comparison Tests). Green space exposure was estimated within several radii around their current residence (50 m to 2000 m), using a aerial photo-derived high-resolution (1 m2) land cover map. Furthermore, air pollution exposure to PM2.5 and NO2 during the year before examination was modelled for the child’s residence using a spatial–temporal interpolation method.ResultsAn improvement of the children’s attention was found with more residential green space exposure independent of traffic-related air pollution. For an interquartile range increment (21%) of green space within 100 m of the residence, a significantly lower mean reaction time was observed independent of NO2 for both the sustained-selective (-9.74 ms, 95% CI: -16.6 to -2.9 ms, p = 0.006) and the selective attention outcomes (-65.90 ms, 95% CI: -117.0 to -14.8 ms, p = 0.01). Moreover, green space exposure within a large radius (2000 m) around the residence was significantly associated with a better performance in short-term memory (Digit-Span Forward Test) and a higher visual information processing speed (Pattern Comparison Test), taking into account traffic-related exposure. However, all associations were attenuated after taking into account long-term residential PM2.5 exposure.ConclusionsOur panel study showed that exposure to residential surrounding green space was associated with better cognitive performances at 9–12 years of age, taking into account traffic-related air pollution exposure. These findings support the necessity to build attractive green spaces in the residential environment to promote healthy cognitive development in children.

Random Forest Classification of Multitemporal Landsat 8 Spectral Data and Phenology Metrics for Land Cover Mapping in the Sonoran and Mojave Deserts

Geospatial data and tools evolve as new technologies are developed and landscape change occurs over time. As a result, these data may become outdated and inadequate for supporting critical habitat-related work across the international boundary in the Sonoran and Mojave Deserts Bird Conservation Region (BCR 33) due to the area’s complex vegetation communities and the discontinuity in data availability across the United States (US) and Mexico (MX) border. This research aimed to produce the first 30 m continuous land cover map of BCR 33 by prototyping new methods for desert vegetation classification using the Random Forest (RF) machine learning (ML) method. The developed RF classification model utilized multitemporal Landsat 8 Operational Land Imager spectral and vegetation index data from the period of 2013–2020, and phenology metrics tailored to capture the unique growing seasons of desert vegetation. Our RF model achieved an overall classification F-score of 0.80 and an overall accuracy of 91.68%. Our results portrayed the vegetation cover at a much finer resolution than existing land cover maps from the US and MX portions of the study area, allowing for the separation and identification of smaller habitat pockets, including riparian communities, which are critically important for desert wildlife and are often misclassified or nonexistent in current maps. This early prototyping effort serves as a proof of concept for the ML and data fusion methods that will be used to generate the final high-resolution land cover map of the entire BCR 33 region.

Exploring Global Climate Model Downscaling Based on Tile-Level Output

Abstract Statistical and dynamical modeling techniques are used to downscale global climate model (GCM) outputs to practical resolutions for local- or regional-scale applications. Current techniques do not incorporate the effects of land-use and land-cover changes, although research has shown that such changes can substantially affect climate locally. Here, we explore a new downscaling technique that uses tile-level GCM outputs provided under phase 6 of the Coupled Model Intercomparison Project (CMIP6). The method, land-cover tile downscaling (LTD), spatially locates the tile-level GCM outputs by mapping them to corresponding classes in high-resolution land-cover maps. Furthermore, it applies an elevation-based correction to account for the effect of topography on the local climate. LTD is applied to near-surface temperature outputs from the Community Earth System Model, version 2 (CESM2) and U.K. Earth System Model, version 1 (UKESM1), and surface temperature output from CESM2 and evaluated against observations. In comparison with grid-averaged control data, LTD outputs show an overall bias reduction that is not spatially consistent. Moreover, LTD performs better on air temperature data than on surface temperature and better on areas dominated by primary/secondary land and crops than on urban land. This could arise from simplifications in methods, like land-cover reclassification and simplified lapse rate estimates. However, the difference in response between the two variables and land-cover types implies that biases also stem from model structural features involved in estimating their tile-level outputs. This is supported by the differences between grid average data provided by the models and the same data reconstructed from tile-level outputs. Therefore, a thorough evaluation and quality control of tile-level outputs is recommended.