Mehdi Gholamnia

This study presents a novel approach for assessing urban accessibility by leveraging Large Language Models (LLMs) to analyze crowdsourced textual data. Using Airbnb review content from Dublin, we developed a pipeline that extracts perceptions of accessibility—including travel time to public transportation and the city center—directly from user-generated reviews. The FLAN-T5 LLM was tested on a manually curated synthetic test dataset, achieving 84% and 86% accuracy in extracting public transport and city center travel times, respectively. Correlation analysis further confirmed strong negative associations between inferred accessibility by LLM and user-provided location ratings ( r = − 0.91 and r = − 0.95), validating the semantic relevance of textual data. This approach highlights how LLMs can transform unstructured review content into structured urban mobility insights, offering a scalable and cost-effective tool for planners. By capturing lived experiences at scale, the method supports more inclusive and data-driven urban policy. The findings contribute to the growing intersection of AI, spatial information science, and participatory urban analytics.

Green roofs are essential for urban greening and climate adaptation, especially in densely populated areas. Analyzing runoff reduction parameters is crucial for effectively designing and implementing these systems. This study enhances traditional assessments using advanced sensors to gather meteorological and hydrological data from four green roof installations at University College Dublin (UCD) in Dublin, Ireland. The comprehensive dataset enabled detailed modeling of runoff hydrograph parameters using rainfall hyetographs, which were subsequently analyzed through sophisticated machine learning algorithms. This research introduces an innovative approach by identifying the optimal combination of variables for modeling key runoff characteristics, including Water Retention Amount (WRA), Total RUnoff Volume (TRUV), Peak Runoff Discharge (PRD), and Peak Flow Reduction (PFR). The findings are compelling, with Support Vector Regression (SVR) achieving R² values ranging from 0.67 to 0.82 and RMSE values ranging from 0.37 to 1.51 millimeters for WRA, TRUV, PRD, and PFR. XGBoost (XGB) demonstrated superior performance, with R² values ranging from 0.77 to 0.84 and RMSE values ranging from 0.28 to 1.26 millimeters for the same parameters. Random Forest Regression (RF) also showed robust results, with R² values ranging from 0.76 to 0.84 and RMSE values ranging from 0.31 to 1.29 millimeters. Overall, the green roof system demonstrated a water retention rate of 55.69% for the studied events. The study identifies Cumulative Rainfall Volume (CRV) and Peak Rainfall Intensity (PRI) as crucial for modeling runoff, highlighting green roofs’ potential as sustainable urban infrastructure and offering key insights for their design and optimization.

Wetlands, essential for Earth’s health, ecological balance, and local economies, require accurate monitoring and assessment for effective conservation. Data-driven models based on remote sensing are highly capable of monitoring the status and classification of wetlands. This study developed a semi-supervised framework for mapping wetland covers in Zrebar, Iran, using Landsat time series data from 1984 to 2022. A pixel purification technique was applied to the temporal candidate images to refine the initial training data (conventional scenario) and generate purified training data (proposed scenario). The Support Vector Machine (SVM) algorithm was utilized to classify the land cover within the wetland, and the accuracy of the two scenarios was evaluated and compared. Over the study period, the analysis of land cover changes within Zrebar Wetland revealed significant spatial and temporal changes in soil and farmland, reed, and water from 1984 to 2022. The omission error rates for the classes soil and farmland, reed, and water were decreased from 0.14, 0.14, and 0.12 for scenario 1 to 0.03, 0.05, and 0.05 for scenario 2, respectively. In addition, the commission error for these classes decreased from 0.13, 0.18, and 0.09 for scenario 1 to 0.04, 0.06, and 0.04 after applying the filtered training data in the scenario 2. Finally, the overall accuracy of the initial training data (scenario 1) and the filtered training data (scenario 2) were 0.86 and 0.94, respectively. These results underscore the effectiveness of the proposed strategy in enhancing the accuracy of land cover classification within the wetland over time, highlighting its potential for future wetland conservation efforts.

The correction of Soil Moisture (SM) estimates in Land Surface Models (LSMs) is considered essential for improving the performance of numerical weather forecasting and hydrologic models used in weather and climate studies. Along with surface screen-level variables, the satellite data, including Brightness Temperature (BT) from passive microwave sensors, and retrieved SM from active, passive, or combined active–passive sensor products have been used as two critical inputs in improvements of the LSM. The present study reviewed the current status in correcting LSM SM estimates, evaluating the results with in situ measurements. Based on findings from previous studies, a detailed analysis of related issues in the assimilation of SM in LSM, including bias correction of satellite data, applied LSMs and in situ observations, input data from various satellite sensors, sources of errors, calibration (both LSM and radiative transfer model), are discussed. Moreover, assimilation approaches are compared, and considerations for assimilation implementation are presented. A quantitative representation of results from the literature review, including ranges and variability of improvements in LSMs due to assimilation, are analyzed for both surface and root zone SM. A direction for future studies is then presented.

The existence of several NDVI products in Qinghai-Tibetan Plateau (QTP) makes it challenging to identify the ideal sensor for vegetation monitoring as an important factor for landslide detection studies. A pixel-based analysis of the NDVI time series was carried out to compare the performances of five NDVI products, including ETM+, OLI, MODIS Series, and AVHRR sensors in QTP. Harmonic analysis of time series and wavelet threshold denoising were used for reconstruction and denoising of the five NDVI datasets. Each sensor performance was assessed based on the behavioral similarity between the original and denoised NDVI time series, considering the preservation of the original shape and time series values by computing correlation coefficient (CC), mean absolute error (MAE), root mean square error (RMSE), and signal to noise ratio (SNR). Results indicated that the OLI slightly outperformed the other sensors in all performance metrics, especially in mosaic natural vegetation, grassland, and cropland, providing 0.973, 0.015, 0.022, and 27.220 in CC, MAE, RMSE, and SNR, respectively. AVHRR showed similar results to OLI, with the best results in the predominant type of land covers (needle-leaved, evergreen, closed to open). The MODIS series performs lower across all vegetation classes than the other sensors, which might be related to the higher number of artifacts observed in the original data. In addition to the satellite sensor comparison, the proposed analysis demonstrated the effectiveness and reliability of the implemented methodology for reconstructing and denoising different NDVI time series, indicating its suitability for long-term trend analysis of different natural land cover classes, vegetation monitoring, and change detection.

With the development of remote sensing algorithms and increased access to satellite data, generating up-to-date, accurate land use/land cover (LULC) maps has become increasingly feasible for evaluating and managing changes in land cover as created by changes to ecosystem and land use. The main objective of our study is to evaluate the performance of Support Vector Machine (SVM), Artificial Neural Network (ANN), Maximum Likelihood Classification (MLC), Minimum Distance (MD), and Mahalanobis (MH) algorithms and compare them in order to generate a LULC map using data from Sentinel 2 and Landsat 8 satellites. Further, we also investigate the effect of a penalty parameter on SVM results. Our study uses different kernel functions and hidden layers for SVM and ANN algorithms, respectively. We generated the training and validation datasets from Google Earth images and GPS data prior to pre-processing satellite data. In the next phase, we classified the images using training data and algorithms. Ultimately, to evaluate outcomes, we used the validation data to generate a confusion matrix of the classified images. Our results showed that with optimal tuning parameters, the SVM classifier yielded the highest overall accuracy (OA) of 94%, performing better for both satellite data compared to other methods. In addition, for our scenes, Sentinel 2 date was slightly more accurate compared to Landsat 8. The parametric algorithms MD and MLC provided the lowest accuracy of 80.85% and 74.68% for the data from Sentinel 2 and Landsat 8. In contrast, our evaluation using the SVM tuning parameters showed that the linear kernel with the penalty parameter 150 for Sentinel 2 and the penalty parameter 200 for Landsat 8 yielded the highest accuracies. Further, ANN classification showed that increasing the hidden layers drastically reduces classification accuracy for both datasets, reducing zero for three hidden layers.

In this study, the spatiotemporal behavior of vegetation cover in the Kurdistan province of Iran was analyzed for the first time by TIMESAT and Breaks for Additive Season and Trend (BFAST) algorithms. They were applied on Normalized Vegetation Index (NDVI) time series from 2000 to 2016 derived from Moderate Resolution Imaging Spectroradiometer (MODIS) observations. The TIMESAT software package was used to estimate the seasonal parameters of NDVI and their relation to land covers. BFAST was applied for identifying abrupt changes (breakpoints) of NDVI and their magnitudes. The results from TIMESAT and BFAST were first reported separately, and then interpreted together. TMESAT outcomes showed that the lowest and highest amplitudes of NDVI during the whole time period happened in 2008 and 2010. The spatial distribution of the number of breakpoints showed different behaviors in the west and east of the study area, and the breakpoint frequency confirmed the extreme NDVI amplitudes in 2008 and 2010 found by TIMESAT. For the first time in Iran, a correlation analysis between accumulated precipitations and maximum NDVIs (from one to seven months before the NDVI maximum) was conducted. The results showed that precipitation one month before had a higher correlation with the maximum NDVIs in the region. Overall, the results describe the NDVI behavior in terms of greenness, lifetime, abrupt changes for the different land covers, and across the years, suggesting how the northwest and west of the study area can be more susceptible to drought conditions.

The air temperature is an essential variable in many applications related to Earth science. Sporadic spatial distribution of weather stations causes a low spatial resolution of measured air temperatures. This study focused on modeling the air diurnal temperature cycle (DTC) based on the land surface temperature (LST) DTC. The air DTC model parameters were estimated from LST DTC model parameters by a regression analysis. Here, the LST obtained from the INSAT-3D geostationary satellite and the air temperature extracted from weather stations were used within the time frame of 4 March 2015 to 22 May 2017 across Iran. Constant parameters of the air DTC model for each weather station were estimated based on an experimental approach over the time period. Results showed these parameters decrease as elevation increases. The mean absolute error (MAE) and the root mean square error (RMSE) for three hours sampling were calculated. The MAE and RMSE ranges were between [0.1, 4] °C and [0.1, 3.3] °C, respectively. Additionally, 95% of MAEs and RMSEs were less than 2.9 °C and 2.4 °C values, correspondingly. The range of the mean values of MAEs and RMSEs for a three-hour sampling time were [−0.29, 0.6] °C and [2, 2.11] °C. The DTC model results showed a meaningful statistical fitting in both air DTCs modeled from LST and weather station-based DTCs. The variability of mean error and RMSE in different land covers and elevation classes were also investigated. In spite of the complex behavior of the environmental variables in the study area, the model error bar did not show significantly biased estimations for various classes. Therefore, the developed model was less sensitive to variations of land covers and elevation changes. It can be conclude that the coefficients of regression between LST and air DTC could model properly the environmental factors.

In many disciplines, knowledge on the accuracy of Land Surface Temperature (LST) as an input is of great importance. One of the most efficient methods in LST evaluation is cross validation. Well-documented and validated polar satellites with a high spatial resolution can be used as references for validating geostationary LST products. This study attempted to investigate the discrepancies between a Moderate Resolution Imaging Spectro-radiometer (MODIS) and Indian National Satellite (INSAT-3D) LSTs for high temperatures, focusing on six deserts with sand dune land cover in the Middle East from 3 March 2015 to 24 August 2016. Firstly, the variability of LSTs in the deserts of the study area was analyzed by comparing the mean, Standard Deviation (STD), skewness, minimum, and maximum criteria for each observation time. The mean value of the LST observations indicated that the MYD-D observation times are closer to those of diurnal maximum and minimum LSTs. At all times, the LST observations exhibited a negative skewness and the STD indicated higher variability during times of MOD-D. The observed maximum LSTs from MODIS collection 6 showed higher values in comparison with the last versions of LSTs for hot spot regions around the world. After the temporal, spatial, and geometrical matching of LST products, the mean of the MODIS—INSAT LST differences was calculated for the study area. The results demonstrated that discrepancies increased with temperature up to +15.5 K. The slopes of the mean differences were relatively similar for all deserts except for An Nafud, suggesting an effect of View Zenith Angle (VZA). For modeling the discrepancies between two sensors in continuous space, the Diurnal Temperature Cycles (DTC) of both sensors were constructed and compared. The sample DTC models approved the results from discrete LST subtractions and proposed the uncertainties within MODIS DTCs. The authors proposed that the observed LST discrepancies in high temperatures could be the result of inherent differences in LST retrieval algorithms.