@misc{SafonovaGhazaryanStilleretal., author = {Safonova, Anastasiia and Ghazaryan, Gohar and Stiller, Stefan and Main-Knorn, Magdalena and Nendel, Claas and Ryo, Masahiro}, title = {Ten deep learning techniques to address small data problems with remote sensing}, series = {International Journal of Applied Earth Observation and Geoinformation}, volume = {125}, journal = {International Journal of Applied Earth Observation and Geoinformation}, publisher = {Elsevier BV}, issn = {1569-8432}, doi = {10.1016/j.jag.2023.103569}, pages = {17}, abstract = {Researchers and engineers have increasingly used Deep Learning (DL) for a variety of Remote Sensing (RS) tasks. However, data from local observations or via ground truth is often quite limited for training DL models, especially when these models represent key socio-environmental problems, such as the monitoring of extreme, destructive climate events, biodiversity, and sudden changes in ecosystem states. Such cases, also known as small data problems, pose significant methodological challenges. This review summarises these challenges in the RS domain and the possibility of using emerging DL techniques to overcome them. We show that the small data problem is a common challenge across disciplines and scales that results in poor model generalisability and transferability. We then introduce an overview of ten promising DL techniques: transfer learning, self-supervised learning, semi-supervised learning, few-shot learning, zero-shot learning, active learning, weakly supervised learning, multitask learning, process-aware learning, and ensemble learning; we also include a validation technique known as spatial k-fold cross validation. Our particular contribution was to develop a flowchart that helps DL users select which technique to use given by answering a few questions. We hope that our review article facilitate DL applications to tackle societally important environmental problems with limited reference data.}, language = {en} } @misc{StillerGrahmannGhazaryanetal., author = {Stiller, Stefan and Grahmann, Kathrin and Ghazaryan, Gohar and Ryo, Masahiro}, title = {Improving spatial transferability of deep learning models for small-field crop yield prediction}, series = {ISPRS Open Journal of Photogrammetry and Remote Sensing}, volume = {12}, journal = {ISPRS Open Journal of Photogrammetry and Remote Sensing}, publisher = {Elsevier BV}, issn = {2667-3932}, doi = {10.1016/j.ophoto.2024.100064}, pages = {11}, abstract = {Predicting crop yield using deep learning (DL) and remote sensing is a promising technique in agriculture. In smallholder agriculture (<2 ha), where 84\% of the farms operate globally, it is crucial to build a model that can be useful across several fields (high spatial transferability). However, enhancing spatial model transferability in a small-scale setting faces significant challenges, including spatial autocorrelation, heterogeneity and scale dependence of spatial dynamics, as well as the need to address limited data points. This study aimed to test the hypothesis that spatial cross validation (SCV) is a more suitable model validation practice than random cross validation (RCV) to enhance model transferability for spatial prediction in a small-scale farming setting. We compared the performances of DL models that predict crop yield for several settings including three crop types and two DL architectures based on RCV with and without overlapping samples and SCV. Notably, we conducted model performance tests on external, equally sized fields instead of the field used for training. We used high resolution RGB imagery taken with a drone as input. Our results show that the models using SCV outperformed those using RCV when the models were tested on external fields (on average r = 0.37 for SCV, r = 0.18 for RCV with overlap and r = 0.07 without), even though the models using SCV showed a substantially lower performance for cross validation (CV) than those using RCV (r with SCV and RCV w/o overlap = 0.73 and 0.98/0.73, respectively). The results suggest that RCV leads to over-optimism by overfitting the spatial structure and remembering image-specific information (so called memorization). Our study offers the first empirical evidence in agriculture that SCV is preferable to RCV in small field settings for making DL models more transferable.}, language = {en} } @misc{RazaRyoGhazaryanetal., author = {Raza, Ahsan and Ryo, Masahiro and Ghazaryan, Gohar and Baatz, Roland and Main-Knorn, Magdalena and Inforsato, Leonardo and Nendel, Claas}, title = {Predicting regional-scale groundwater levels at high spatial resolution using spatial random forest models}, series = {International journal of applied earth observation and geoinformation}, volume = {144}, journal = {International journal of applied earth observation and geoinformation}, publisher = {Elsevier BV}, address = {Amsterdam}, issn = {1569-8432}, doi = {10.1016/j.jag.2025.104918}, pages = {1 -- 19}, abstract = {Groundwater is an important resource for sustainable crop growth and agricultural productivity, and groundwater level (GWL) fluctuations directly influence agroecosystems. However, GWL data for most regions is unavailable because of spatial scarcity and discontinuous groundwater observations. In recent studies, machine learning techniques have proven to be a useful tool for producing GWL estimates, but these techniques may not necessarily take into account distances and observations from the nearest points to unmonitored sites as covariates. In this study, we explored the potential of these additional spatial covariates by implementing Random Forest for Spatial Interpolation (RFSI) to generate high-resolution (1 km²) GWL estimates on a monthly scale. To evaluate the effectiveness of the RFSI model, we performed a thorough comparison of the RFSI model with a conventional Random Forest (RF) model, a Random Forest for spatial data (RFsp) and a Support Vector Machine (SVM). The framework was implemented using GWL observation data for the period 2001 to 2022 in Brandenburg, Germany. The RFSI model exhibited the highest predictive accuracy for GWL predictions at testing points, achieving an R² value of 0.86, surpassing the performance of RF, RFsp and SVM with R² values of 0.75, 0.72 and 0.67, respectively. Incorporating nearby-well information substantially improved 1 km GWL predictions: RFSI achieved the lowest RMSE (3.77 m), as compared to RF (4.85 m), RFsp (5.04 m), and SVM (5.59 m). Notably, among the models tested, RF, RFsp, and SVM showed notable inaccuracies, substantially underestimating groundwater levels (GWL) in deeper wells (>10 m) and overestimating levels in shallow wells (1-10 m). In contrast, RFSI performed better, with smaller overestimations (up to ~1.1 m) and less severe underestimations (up to ~-3.9 m). Furthermore, the RFSI model demonstrated significantly higher computational efficiency compared to RFsp and SVM, particularly for large training datasets and high-resolution prediction mapping.}, language = {en} }