We present a framework for modeling multi-scale processes, and study its performance in the context of stream-flow forecasting in hydrology. Specifically, we propose a novel hierarchical recurrent neural architecture that factorizes the system dynamics at multiple temporal scales and captures their interactions. This framework consists of an inverse and a forward model. The inverse model is used to empirically resolve the system's temporal modes from data (physical model simulations, observed data, or a combination of them from the past), and these states are then used in the forward model to predict streamflow. Experiments on several catchments from the National Weather Service North Central River Forecast Center show that FHNN outperforms standard baselines, including physics-based models and transformer-based approaches. The model demonstrates particular effectiveness in catchments with low runoff ratios and colder climates. We further validate FHNN on the CAMELS (Catchment Attributes and MEteorology for Large-sample Studies), which is a widely used continental-scale hydrology benchmark dataset, confirming consistent performance improvements for 1–7 day streamflow forecasts across diverse hydrological conditions. Additionally, we show that FHNN can maintain accuracy even with limited training data through effective pre-training strategies and training global models.

The discovery of environmental knowledge depends on labeled task-specific data, but is often constrained by the high cost of data collection. Existing machine learning approaches usually struggle to generalize in data-sparse or atypical conditions. To this end, we propose an Augmentation-Adaptive Self-Supervised Learning (A2SL) framework, which retrieves relevant observational samples to enhance modeling of the target ecosys-tem. Specifically, we introduce a multi-level pairwise learning loss to train a scenario encoder that captures varying degrees of similarity among scenarios. These learned similarities drive a retrieval mechanism that supplements a target scenario with relevant data from different locations or time periods. Further-more, to better handle variable scenarios, particularly under atypical or extreme conditions where traditional models struggle, we design an augmentation-adaptive mechanism that selectively enhances these scenarios through targeted data augmentation. Using freshwater ecosystems as a case study, we evaluate A2SL in modeling water temperature and dissolved oxygen dynamics in real-world lakes. Experimental results show that A2SL signif-icantly improves predictive accuracy and enhances robustness in data-scarce and atypical scenarios. Although this study focuses on freshwater ecosystems, the A2SL framework offers a broadly applicable solution in various scientific domains. Code-https://github.com/shiyuanlsy/A2sl

We introduce TAM-RL (Task Aware Modulation using Representation Learning), a novel multimodal meta-learning framework for few-shot learning in heterogeneous systems, designed for science and engineering problems where entities share a common underlying forward model but exhibit heterogeneity due to entity-specific characteristics. TAM-RL leverages an amortized training process with a modulation network and a base network to learn task-specific modulation parameters, enabling efficient adaptation to new tasks with limited data. We evaluate TAM-RL on two real-world environmental datasets: Gross Primary Product (GPP) prediction and streamflow forecasting, demonstrating significant improvements over existing meta-learning methods. On the FLUXNET dataset, TAM-RL improves RMSE by 18.9% over MMAML with just one month of few-shot data, while for streamflow prediction, it achieves an 8.21% improvement with one year of data. Synthetic data experiments further validate TAM-RL’s superior performance in heterogeneous task distributions, outperforming the baselines in the most heterogeneous setting. Notably, TAM-RL offers substantial computational efficiency, with at least 3x faster training times compared to gradient-based meta-learning approaches while being much simpler to train due to reduced complexity. Ablation studies highlight the importance of pretraining and adaptation mechanisms in TAM-RL’s performance.

Many environmental systems (e.g., hydrology basins) can be modeled as entity whose response (e.g., streamflow) depends on drivers (e.g., weather) conditioned on their characteristics (e.g., soil properties). We introduce Entity-aware Conditional Variational Inference (EA-CVI), a novel probabilistic inverse modeling approach, to deduce entity characteristics from observed driver-response data. EA-CVI infers probabilistic latent representations that can accurately predict response for diverse entities, particularly in out-of-sample few-shot settings. EA-CVI's latent embeddings encapsulate diverse entity characteristics within compact, low-dimensional representations. EA-CVI proficiently identifies dominant modes of variation in responses and offers the opportunity to infer a physical interpretation of the underlying attributes that shape these responses. EA-CVI can also generate new data samples by sampling from the learned distribution, making it useful in zero-shot scenarios. EA-CVI addresses the need for uncertainty estimation, particularly during extreme events, rendering it essential for data-driven decision-making in real-world applications. Extensive evaluations on a renowned hydrology benchmark dataset, CAMELS-GB, validate EA-CVI's abilities.

Accurate and timely crop mapping is essential for yield estimation, insurance claims, and conservation efforts. Over the years, many successful machine learning models for crop mapping have been developed that use just the multi-spectral imagery from satellites to predict crop type over the area of interest. However, these traditional methods do not account for the physical processes that govern crop growth. At a high level, crop growth can be envisioned as physical parameters, such as weather and soil type, acting upon the plant leading to crop growth which can be observed via satellites. In this paper, we propose Weather-based Spatio-Temporal segmentation network with ATTention (WSTATT), a deep learning model that leverages this understanding of crop growth by formulating it as an inverse model that combines weather (Daymet) and satellite imagery (Sentinel-2) to generate accurate crop maps. We show that our approach provides significant improvements over existing algorithms that solely rely on spectral imagery by comparing segmentation maps and F1 classification scores. Furthermore, effective use of attention in WSTATT architecture enables detection of crop types earlier in the season (up to 5 months in advance), which is very useful for improving food supply projections. We finally discuss the impact of weather by correlating our results with crop phenology to show that WSTATT is able to capture physical properties of crop growth.

Fairness related to locations (i.e., “where”) is critical for the use of machine learning in a variety of societal domains involving spatial datasets (e.g., agriculture, disaster response, urban planning). Spatial biases incurred by learning, if left unattended, may cause or exacerbate unfair distribution of resources, social division, spatial disparity, etc. The goal of this work is to develop statistically-robust formulations and model-agnostic learning strategies to understand and promote spatial fairness. The problem is challenging as locations are often from continuous spaces with no well-defined categories (e.g., gender), and statistical conclusions from spatial data are fragile to changes in spatial partitionings and scales. Existing studies in fairness-driven learning have generated valuable insights related to non-spatial factors including race, gender, education level, etc., but research to mitigate location-related biases still remains in its infancy, leaving the main challenges unaddressed. To bridge the gap, we first propose a robust space-as-distribution (SPAD) representation of spatial fairness to reduce statistical sensitivity related to partitionings and scales in continuous space. Furthermore, we propose a new SPAD-based stochastic strategy to efficiently optimize over an extensive distribution of fairness criteria, and a bi-level training framework to enforce fairness via adaptive adjustment of priorities among locations. Finally, we extend this framework with a similarity-based training strategy to improve the computational efficiency. Experiments conducted on two real-world problems, crop monitoring in the US and palm oil plantation mapping in Indonesia, show that SPAD can effectively reduce sensitivity in fairness evaluation and the stochastic bi-level training framework can greatly improve the fairness. Controlled experiments also show that similarity-based acceleration can greatly reduce the training time while keeping the prediction performance and fairness results at the same level.

Accurate long-term predictions are the foundations for many machine learning applications and decision-making processes. However, building accurate long-term prediction models remains challenging due to the limitations of existing temporal models like recurrent neural networks (RNNs), as they capture only the statistical connections in the training data and may fail to learn the underlying dynamics of the target system. To tackle this challenge, we propose a novel machine learning model based on Koopman operator theory, which we call Koopman Invertible Autoencoders (KIA), that captures the inherent characteristic of the system by modeling both forward and backward dynamics in the infinite-dimensional Hilbert space. This enables us to efficiently learn low-dimensional representations, resulting in more accurate predictions of long-term system behavior. Moreover, our method’s invertibility design enforces reversibility and consistency in both forward and inverse operations. We illustrate the utility of KIA on pendulum and climate datasets, demonstrating 300% improvements in long-term prediction capability for pendulum while maintaining robustness against noise. Additionally, our method demonstrates the ability to better comprehend the intricate dynamics of the climate system when compared to existing Koopman-based methods.

Creating separable representations via representation learning and clustering is critical in analyzing large unstructured datasets with only a few labels. Separable representations can lead to supervised models with better classification capabilities and additionally aid in generating new labeled samples. Most unsupervised and semisupervised methods to analyze large datasets do not leverage the existing small amounts of labels to get better representations. In this paper, we propose a spatiotemporal clustering paradigm that uses spatial and temporal features combined with a constrained loss to produce separable representations. We show the working of this method on the newly published dataset ReaLSAT, a dataset of surface water dynamics for over 680,000 lakes across the world, making it an essential dataset in terms of ecology and sustainability. Using this large un- labelled dataset, we first show how a spatiotemporal representation is better compared to just spatial or temporal representation. We then show how we can learn even better representations using a constrained loss with few labels. We conclude by showing how our method, using few labels, can pick out new labeled samples from the unlabeled data, which can be used to augment supervised methods leading to better classification.

Rapid advancement in inverse modeling methods have brought into light their susceptibility to imperfect data. This has made it imperative to obtain more explainable and trustworthy estimates from these models. In hydrology, basin characteristics can be noisy or missing, impacting streamflow prediction. We propose a probabilistic inverse model framework that can reconstruct robust hydrology basin characteristics from dynamic input weather driver and streamflow response data. We address two aspects of building more explainable inverse models, uncertainty estimation (uncertainty due to imperfect data and imperfect model) and robustness. This can help improve the trust of water managers, handling of noisy data and reduce costs. We also propose an uncertainty based loss regularization that offers removal of 17% of temporal artifacts in reconstructions, 36% reduction in uncertainty and 4% higher coverage rate for basin characteristics. The forward model performance (streamflow estimation) is also improved by 6% using these uncertainty learning based reconstructions.

In many environmental applications, recurrent neural networks (RNNs) are often used to model physical variables with long temporal dependencies. However, due to minibatch training, temporal relationships between training segments within the batch (intra-batch) as well as between batches (inter-batch) are not considered, which can lead to limited performance. Stateful RNNs aim to address this issue by passing hidden states between batches. Since Stateful RNNs ignore intra-batch temporal dependency, there exists a trade-off between training stability and capturing temporal dependency. In this paper, we provide a quantitative comparison of different Stateful RNN modeling strategies, and propose two strategies to enforce both intra- and inter-batch temporal dependency. First, we extend Stateful RNNs by defining a batch as a temporally ordered set of training segments, which enables intra-batch sharing of temporal information. While this approach significantly improves the performance, it leads to much larger training times due to highly sequential training. To address this issue, we further propose a new strategy which augments a training segment with an initial value of the target variable from the timestep right before the starting of the training segment. In other words, we provide an initial value of the target variable as additional input so that the network can focus on learning changes relative to that initial value. By using this strategy, samples can be passed in any order (mini-batch training) which significantly reduces the training time while maintaining the performance. In demonstrating the utility of our approach in hydrological modeling, we observe that the most significant gains in predictive accuracy occur when these methods are applied to state variables whose values change more slowly, such as soil water and snowpack, rather than continuously moving flux variables such as streamflow.

Collecting large annotated datasets in Remote Sensing is often expensive and thus can become a significant obstacle for training advanced machine learning models. Standard techniques for addressing this issue, based on the underlying idea of pre-training the Deep Neural Networks (DNN) on freely available large datasets, cannot be used for Remote Sensing due to the unavailability of such large-scale labeled datasets and the heterogeneity of data sources caused by the varying spatial and spectral resolution of different sensors. Self-supervised learning is an alternative approach that learns feature representation from unlabeled images without human annotations. In this paper, we introduce a new method for land cover mapping by using a clustering-based pretext task for self-supervised learning. We demonstrate the method's effectiveness in two societally relevant applications from the aspect of segmentation performance, discriminative feature representation learning, and the underlying cluster structure. We also show the effectiveness of the active sampling using the clusters obtained from our method in improving the mapping accuracy given a limited budget for annotating. Finally, a real-world application of the developed framework in identifying intra-class categories of well-managed and poorly-managed plantations demonstrates its utility in a problem of considerable societal importance.

As the adoption of machine learning continues to thrive, fairness of the algorithms has become a key factor determining their long-term success and sustainability. Among them, location-based fairness - or spatial fairness - is critical for a variety of essential societal applications that commonly rely on spatial data, including agriculture, disaster response, urban planning, etc. Spatial biases incurred by learning, if left unattended, may cause or exacerbate unfair distribution of resources, spatial disparity, social division, etc. However, very limited understanding has been developed on location-based fairness and bias in machine learning. Compared to traditional fairness-preserving techniques, the spatial consideration introduces two major layers of complication: (1) Space is continuous with no well-defined categories (e.g., categories by race or gender); and (2) Categorizations given by space-partitionings are known to be subject to high statistical sensitivity (e.g., gerrymandering). Under these challenges, we formally explore and demonstrate the fragility of learning methods in the spatial fairness-bias sphere. Specifically, we present a set of techniques that can maneuver the training process towards various targeted fairness-bias outcomes, while maintaining the same level of overall prediction performance (i.e., for "free"). Extensive experiments are carried out on two real-world problems: crop monitoring in the US and palm oil plantation mapping in Indonesia. The results demonstrate the effectiveness of the manipulation algorithms and the importance of explicitly regulating location-based fairness using a diverse set of criteria.

Prediction of response to input drivers by unmonitored entities has been recognized as one of the most important problems in many scientific problems. This problem is challenging due to the non-stationary processes that underlie the dynamics of data observations over space and time. Hence, directly transferring models from well-observed data entities to unmonitored target entity often lead to sub-optimal performance due to the shift in data distribution. This paper proposes a new meta-transfer learning framework that automatically estimates the similarity amongst entities to transfer knowledge from well-observed entities to unmonitored entities. A sequence autoencoder embeds temporal behaviors of time series data and simulations generated by traditional physics-based models. This embedding model is trained in a meta-transfer learning framework under the guidance of source-to-source transferring experiences. We tested this method in streamflow prediction for multiple river segments in the Delaware River Basin, an ecologically diverse region along the eastern coast of the United States. The experimental results demonstrate the superiority of the proposed method in predicting streamflow for unmonitored stream segments compared to a diverse set of baselines. Our method also creates meaningful similarity estimates amongst segments to guide the transfer learning process.

Machine Learning is beginning to provide state-of-the-art performance in a range of environmental applications such as streamflow prediction in a hydrologic basin. However, building accurate broad-scale models for streamflow remains challenging in practice due to the variability in the dominant hydrologic processes, which are best captured by sets of process-related basin characteristics. Existing basin characteristics suffer from noise and uncertainty, among many other things, which adversely impact model performance. To tackle the above challenges, in this paper, we propose a novel Knowledge-guided Self-Supervised Learning (KGSSL) inverse framework to extract system characteristics from driver(input) and response(output) data. This first-of-its-kind framework achieves robust performance even when characteristics are corrupted or missing. We evaluate the KGSSL framework in the context of stream flow modeling using CAMELS (Catchment Attributes and MEteorology for Large-sample Studies) which is a widely used hydrology benchmark dataset. Specifically, KGSSL outperforms baseline by 16% in predicting missing characteristics. Furthermore, in the context of forward modelling, KGSSL inferred characteristics provide a 35% improvement in performance over a standard baseline when the static characteristic are unknown.

Accurate predictions of water temperature are the foundation for many decisions and regulations, with direct impacts on water quality, fishery yields, and power production. Building accurate broad-scale models for lake temperature prediction remains challenging in practice due to the variability in the data distribution across different lake systems monitored by static and time-series data. In this paper, to tackle the above challenges, we propose a novel machine learning based approach for integrating static and time-series data in deep recurrent models, which we call Invertibility-Aware-Long Short-Term Memory(IA-LSTM), and demonstrate its effectiveness in predicting lake temperature. Our proposed method integrates components of the Invertible Network and LSTM to better predict temperature profiles (forward modeling) and infer the static features (i.e., inverse modeling) that can eventually enhance the prediction when static variables are missing. We evaluate our method on predicting the temperature profile of 450 lakes in the Midwestern U.S. and report relative improvement of 4% to capture data heterogeneity and simultaneously outperform baseline predictions by 12% when static features are unavailable.

Mapping and monitoring crops is a key step to-wards sustainable intensification of agriculture and addressing global food security. A dataset like ImageNet that revolutionized computer vision applications can accelerate development of novel crop mapping techniques. Currently, the United States Department of Agriculture (USDA) annually releases the Cropland Data Layer (CDL) which contains crop labels at 30m resolution for the entire United States of America. While CDL is state of the art and is widely used for a number of agricultural applications, it has a number of limitations (e.g., pixelated errors, labels carried over from previous years and errors in classification of minor crops). In this work, we create a new semantic segmentation benchmark dataset, which we call CalCROP21, for the diverse crops in the Central Valley region of California at 10m spatial resolution using a Google Earth Engine based robust image processing pipeline and a novel attention based spatio-temporal semantic segmentation algorithm STATT. STATT uses re-sampled (interpolated) CDL labels for training, but is able to generate a better prediction than CDL by leveraging spatial and temporal patterns in Sentinel2 multi-spectral image series to effectively capture phenologic differences amongst crops and uses attention to reduce the impact of clouds and other atmospheric disturbances. We also present a comprehensive evaluation to show that STATT has significantly better results when compared to the resampled CDL labels. We have released the dataset and the processing pipeline code for generating the benchmark dataset.

The availability of massive earth observing satellite data provides huge opportunities for land use and land cover mapping. However, such mapping effort is challenging due to the existence of various land cover classes, noisy data, and the lack of proper labels. Also, each land cover class typically has its own unique temporal pattern and can be identified only during certain periods. In this article, we introduce a novel architecture that incorporates the UNet structure with Bidirectional LSTM and Attention mechanism to jointly exploit the spatial and temporal nature of satellite data and to better identify the unique temporal patterns of each land cover class. We compare our method with other state-of-the-art methods both quantitatively and qualitatively on two real-world datasets which involve multiple land cover classes. We also visualise the attention weights to study its effectiveness in mitigating noise and in identifying discriminative time periods of different classes. The code and dataset used in this work are made publicly available for reproducibility.

Knowledge discovery from spatial data is essential for many important societal applications including crop monitoring, solar energy estimation, traffic prediction and public health. This paper aims to tackle a key challenge posed by spatial data - the intrinsic spatial heterogeneity commonly embedded in their generation processes - in the context of deep learning. In related work, the early rise of convolutional neural networks showed the promising value of explicit spatial-awareness in deep architectures (i.e., preservation of spatial structure among input cells and the use of local connection). However, the issue of spatial heterogeneity has not been sufficiently explored. While recent developments have tried to incorporate awareness of spatial variability (e.g., SVANN), these methods either rely on manually-defined space partitioning or only support very limited partitions (e.g., two) due to reduction of training data. To address these limitations, we propose a Spatial-Net to simultaneously learn a space-partitioning scheme and a deep network architecture with a Significance-based Grow-and-Collapse (SIG-GAC) framework. SIG-GAC allows collaborative training between partitions and uses an exponential reduction tree to control the network size. Experiments using real-world datasets show that Spatial-Net can automatically learn the pattern underlying heterogeneous spatial process and greatly improve model performance.

Spatial data are ubiquitous, massively collected, and widely used to support critical decision-making in many societal domains, including public health (e.g., COVID-19 pandemic control), agricultural crop monitoring, transportation, etc. While recent advances in machine learning and deep learning offer new promising ways to mine such rich datasets (e.g., satellite imagery, COVID statistics), spatial heterogeneity – an intrinsic characteristic embedded in spatial data - poses a major challenge as data distributions or generative processes often vary across space at different scales, with their spatial extents unknown. Recent studies (e.g., SVANN, spatial ensemble) targeting this difficult problem either require a known space-partitioning as the input, or can only support very limited number of partitions or classes (e.g., two) due to the decrease in training data size and the complexity of analysis. To address these limitations, we propose a model-agnostic framework to automatically transform a deep learning model into a spatial-heterogeneity-aware architecture, where the learning of arbitrary space partitionings is guided by a learning-engaged generalization of multivariate scan statistic and parameters are shared based on spatial relationships. We also propose a spatial moderator to generalize learned space partitionings to new test regions. Experiment results on real-world datasets show that the spatial transformation and moderation framework can effectively capture flexibly-shaped heterogeneous footprints and substantially improve prediction performances.

Many real-world phenomena are observed at multiple resolutions. Predictive models designed to predict these phenomena typically consider different resolutions separately. This approach might be limiting in applications where predictions are desired at fine resolutions but available training data is scarce. In this paper, we propose classification algorithms that leverage supervision from coarser resolutions to help train models on finer resolutions. The different resolutions are modeled as different views of the data in a multi-view framework that exploits the complementarity of features across different views to improve models on both views. Unlike traditional multi-view learning problems, the key challenge in our case is that there is no one-to-one correspondence between instances across different views in our case, which requires explicit modeling of the correspondence of instances across resolutions. We propose to use the features of instances at different resolutions to learn the correspondence between instances across resolutions using attention mechanism. Experiments on the real-world application of mapping urban areas using satellite observations and sentiment classification on text data show the effectiveness of the proposed methods.

In this work, we present a multi-view framework to classify spatio-temporal phenomena at multiple resolutions. This approach utilizes the complementarity of features across different resolutions and improves the corresponding models by enforcing consistency of their predictions on unlabeled data. Unlike traditional multi-view learning problems, the key challenge in our case is that there is a many-to-one correspondence between instances across different resolutions, which needs to be explicitly modeled. Experiments on the real-world application of mapping urban areas using spatial raster data-sets from satellite observations show the benefits of the proposed multi-view framework.