Improved Prediction Model Research Articles

Oxygen availability regulates the quality of soil dissolved organic matter by mediating microbial metabolism and iron oxidation

Dissolved organic matter (DOM) plays a vital role in biogeochemical processes and in determining the responses of soil organic matter (SOM) to global change. Although the quantity of soil DOM has been inventoried across diverse spatio-temporal scales, the underlying mechanisms accounting for variability in DOM dynamics remain unclear especially in upland ecosystems. Here, a gradient of SOM storage across 12 croplands in northeast China was used to understand links between DOM dynamics, microbial metabolism, and abiotic conditions. We assessed the composition, biodegradability, and key biodegradable components of DOM. In addition, SOM and mineral-associated organic matter (MAOM) composition, soil enzyme activities, oxygen availability, soil texture, and iron (Fe), Fe-bound organic matter, and nutrient concentrations were quantified to clarify the drivers of DOM quality (composition and biodegradability). The proportion of biodegradable DOM increased exponentially with decreasing initial DOM concentration due to larger fractions of depolymerized DOM that was rich in small-molecular phenols and proteinaceous components. Unexpectedly, the composition of DOM was decoupled from that of SOM or MAOM, but significantly related to enzymatic properties. These results indicate that microbial metabolism exhibited a dominant role in DOM generation. As DOM concentration declined, increased soil oxygen availability regulated DOM composition and enhanced its biodegradability mainly through mediating microbial metabolism and Fe oxidation. The oxygen-induced oxidation of Fe(II) to Fe(III) removed complex DOM compounds with large molecular weight. Moreover, increased oxygen availability stimulated oxidase-catalyzed depolymerization of aromatic substances, and promoted production of protein-like DOM components due to lower enzymatic C/N acquisition ratio. As global changes in temperature and moisture will have large impacts on soil oxygen availability, the role of oxygen in regulating DOM dynamics highlights the importance of integrating soil oxygen supply with microbial metabolism and Fe redox status to improve model predictions of soil carbon under climate change.

Analytical prediction of the formation factor for isotropic mono-sized unconsolidated porous media

Analytical expressions are derived for the formation factor as a function of porosity based on the Representative Unit Cell model for unconsolidated media by applying volume averaging to Ohm’s law. Three arrays are considered, namely an over-staggered, fully staggered and regular array. Numerical simulations are performed to compute the formation factor for two-dimensional versions of the arrays by solving the Laplace equation. The absolute relative percentage difference between the numerical data and corresponding model predictions obtained for the over-staggered array is less than 20% and that for both the fully staggered and regular arrays are less than 10%. Two weighted isotropic models are introduced and a percolation threshold porosity incorporated. The models have been validated against experimental data and models from the literature. The improvement in model prediction with the percolation threshold porosity included is between 100 and 200% at the average porosity of the data set considered.

Towards Precision Dosing of Clozapine in Schizophrenia: External Evaluation of Population Pharmacokinetic Models and Bayesian Forecasting.

Therapeutic drug monitoring and treatment optimization of clozapine are recommended, owing to its narrow therapeutic range and pharmacokinetic (PK) variability. This study aims to assess the clinical applicability of published population PK models by testing their predictive performance in an external data set and to determine the effectiveness of Bayesian forecasting (BF) for clozapine treatment optimization. Available models of clozapine were identified, and their predictive performance was determined using an external data set (53 patients, 151 samples). The median prediction error (PE) and median absolute PE were used to assess bias and inaccuracy. The potential factors influencing model predictability were also investigated. The final concentration was reestimated for all patients using covariates or previously observed concentrations. The 7 included models presented limited predictive performance. Only 1 model met the acceptability criteria (median PE ≤ ±20% and median absolute PE ≤30%). There was no difference between the data used for building the models (therapeutic drug monitoring or PK study) or the number of compartments in the models. A tendency for higher inaccuracy at low concentrations during treatment initiation was observed. Heterogeneities were observed in the predictive performances between the subpopulations, especially in terms of smoking status and sex. For the models included, BF significantly improved their predictive performance. Our study showed that upon external evaluation, clozapine models provide limited predictive performance, especially in subpopulations such as nonsmokers. From the perspective of model-informed prediction dosing, model predictability should be improved using updating or metamodeling methods. Moreover, BF substantially improved model predictability and could be used for clozapine treatment optimization.

Yield estimation of Lycium barbarum L. based on the WOFOST model

Previous research about crop growth simulation and yield prediction mainly focused on annual crops with very few such studies relating to perennial fruit tree crops. In this study, field experiments were conducted in different growth periods of Lycium barbarum L. in Gansu province, China to determine and improve the initial WOFOST model parameters with combination of the time-series leaf area index (LAI) obtained from Sentinel-2 satellite and predict yields for L. barbarum shrubs at a field scale for summer fruit (SF) and autumn fruit (AF). The results showed that the performance of initial parameters established by previous research and partial measured data was relatively poor, with a relative error (RE) of 20.95% for annual yield. The simulation after model parameter calibration significantly improved model prediction accuracy. The predicted SF was 2588 kg/ha and AF was 601 kg/ha with a RE of -5.51%. This study provided a strategy to improve modelled yield for fruit tree crops through utilization of remote sensed and field measured data in model parameter calibration, which on the other hand provides useful information for regional development planning and crops marketing.

Urban built form shapes avian richness in green spaces

Urban biodiversity plays an important role in ecological processes and ecosystem services within cities, making conservation a priority in many municipal sustainability plans. Urban green spaces (UGS) have been a key strategy for conservation by providing habitat for wildlife, including avian communities. While the ecological attributes necessary to enhance the habitability of UGS for avian communities are relatively well known, an understanding of how variation in the surrounding urban matrix influences avian richness outcomes in these spaces, is less understood. As new urban areas continue to develop and UGS become increasingly important habitat areas, urban designers and planners will need a better understanding of the ways in which urban built form patterns at the matrix-level influence avian biodiversity outcomes at the site-level in UGS. To that end, this study investigates the influence of 4 urban built form matrix-level variables , capturing three-dimensional (3D) configuration patterns, and 6 UGS site-level variables on total avian richness and avian richness by foraging guild using generalized linear model methods in 22 UGS. This analysis was conducted using high resolution land cover data, LiDAR data, and twenty years of bird occurrence data from the eBird community science program in well-surveyed UGS in Los Angeles, California. The size of the UGS was the best predictor of richness patterns across all groups, confirming previous findings. However, several urban form metrics, when included with UGS size, improved model prediction for Carnivores, Insectivores, and Omnivore foraging guilds, indicating that urban built form does influence some avian groups within UGS. The results of this study suggest that the most important factor for avian richness in UGS is the size of the green space, but that urban built form plays a role too, especially when considered from a 3D perspective.

Development and Validation of Machine Learning Models to Predict Readmission After Colorectal Surgery

BackgroundReadmission after colorectal surgery is common and often implies complications for patients and costs for hospitals. Previous works have created predictive models using logistic regression for this outcome but have shown limited accuracy. Machine learning has shown promise in improving predictions by identifying non-linear patterns in data. We sought to create a more accurate predictive model for readmission after colorectal surgery using machine learning. MethodsPatients who underwent colorectal surgery were identified in the National Quality Improvement Program (NSQIP) database including years 2012–2019 and split into training, validation, and test sets. The primary outcome was readmission within 30 days of surgery. Three types of machine learning models were created, including random forest (RF), gradient boosting (XGB), and neural network (NN). A logistic regression (LR) model was also created for comparison. Model performance was evaluated using area under the receiver operating characteristic curve (AUROC). ResultsThe dataset included 213,827 patients after application of exclusion criteria. A total of 23,083 (10.8%) of patients experienced readmission. NN obtained an AUROC of 0.751 (95% CI 0.743–0.759), compared with 0.684 (95% CI 0.676–0.693) for LR. RF and XGB performed similarly with AUROCs of 0.749 (95% CI 0.741–0.757) and 0.745 (95% CI 0.737–0.753) respectively. Ileus, index admission length of stay, organ-space surgical site infection present at time of surgery, and ostomy placement were identified as the most contributory variables. ConclusionsMachine learning approaches outperformed traditional statistical methods in the prediction of readmission after colorectal surgery. After external validation, this improved prediction model could be used to target interventions to reduce readmission rate.

Establishment of a feed intake prediction model based on eating time, ruminating time and dietary composition

The individual feed intake of dairy cows is an indicator of oestrus and health status and can be used as an abnormality warning in health monitoring. However, existing methods for monitoring individual feed intake are limited by the feeding environment and equipment and are difficult to apply in commercial production. In this experiment, a feed intake monitoring system for dairy cows was used to determine the daily feed intake of 10 dairy cows. The dairy cows were fed 8 diets with different concentration-to-forage ratios according to different phases or time intervals. The adaptation period of the cows to each diet was 7 days, and the trial period was 7 days. Smart collars were used to monitor the daily eating time and ruminating time of the dairy cows, which were used along with dietary components as input parameters to establish five feed intake prediction models: a linear regression model (LRM), an artificial neural network model, a support vector machine model, a K-nearest neighbour model and a chi-square automatic interaction detector model. The research results showed that R2 was 0.73 for the LRM and ≥0.82 for the other four models, which indicated that the relationship between feed intake and the predictors may not be a simple linear relationship. The five models had root mean square errors (RMSEs) ≤ 1.5 kg/d and standard deviations (SDs) ≤ 0.87 kg/d, and in external verification, the five models achieved R2 values ≥ 0.80, RMSEs ≤ 0.75 kg/d, and SDs ≤ 0.90 kg/d. This study shows that adding dietary ingredient data to collar-collected data can improve prediction model accuracy.

Uncertainty quantification for ecological models with random parameters.

There is often considerable uncertainty in parameters in ecological models. This uncertainty can be incorporated into models by treating parameters as random variables with distributions, rather than fixed quantities. Recent advances in uncertainty quantification methods, such as polynomial chaos approaches, allow for the analysis of models with random parameters. We introduce these methods with a motivating case study of sea ice algal blooms in heterogeneous environments. We compare Monte Carlo methods with polynomial chaos techniques to help understand the dynamics of an algal bloom model with random parameters. Modelling key parameters in the algal bloom model as random variables changes the timing, intensity and overall productivity of the modelled bloom. The computational efficiency of polynomial chaos methods provides a promising avenue for the broader inclusion of parametric uncertainty in ecological models, leading to improved model predictions and synthesis between models and data.

RSU-Aided Optimal Member Replacement Scheme with Improved Mobility Prediction for Vehicular Clouds in VANETs

The technique of vehicular clouds is considered an attractive approach in VANETs, because it provides a requester vehicle the ability to use resources of neighborhood vehicles (called cloud member vehicles) to construct a vehicular cloud to use next-generation vehicular applications during driving. Generally, member vehicles can move along different routes from the route of the requester vehicle in intersections and, as a result, leave the vehicular cloud. Then, the leaving member vehicle should be replaced by new member vehicles at intersections to reconstruct the vehicular cloud. However, identifying optimal replacement vehicles among many vehicles at intersections is a very difficult task involving minimizing the waste of resources of vehicles due to their irregular mobility. Thus, we propose an optimal member replacement scheme that finds optimal replacement vehicles through the improved mobility prediction of vehicles by borrowing the computational ability of RSUs on intersections. The proposed scheme first makes an improved mobility prediction model by combining both the trajectory prediction of vehicles using the Markov model and the location prediction of vehicles using the Gaussian distribution. Through the improved mobility prediction model, the proposed scheme then determines the leaving member vehicles and calculates their own leaving time. Next, the proposed scheme addresses the problem to find optimal replacement vehicles to minimize the waste resource and solves it through an integer linear programming. For the performance evaluation of the proposed scheme, we implement it in an NS-3 simulator, which includes the Manhattan mobility model, to reflect the mobility of vehicles on roads. Simulation results conducted in various environments verify that the proposed scheme achieves better performance than the existing scheme.

Experimental Study on Small-Strain Shear Modulus of Unsaturated Silty-Fine Sand

The small-strain stiffness of soil is significant in the accurate prediction of the deformation caused by interactions between foundation soil and structures. Considering the whole range of small strain (10−6~10−3), a bending element-resonant column (BE-RC) combined test system was developed to conduct continuous tests on the shear modulus of unsaturated soil. Under the dehydration path, it was used to investigate the small-strain shear modulus of unsaturated silty-fine sand in Hangzhou Bay, China. The results show that the shear modulus under different net stresses and matrix suctions appeared to non-linearly decay with the increase in strain until stable values were reached at a large strain. At the beginning from the saturated state, the Gmax value increased slowly with decreasing saturation and reached its maximum value at the optimum saturation (Sr)opt; then, it rapidly decayed to the level in the saturated, once the saturation degree decreased to a level lower than (Sr)opt. Additionally, an improved prediction model was proposed for the Gmax of unsaturated sand, considering different saturations. Based on the mesoscopic evolution of internal pore water morphology and the variation in intergranular stress caused by capillary action, the variation in the Gmax could be divided into three segments of saturation: the boundary effect stage, the transition stage and the unsaturated residual stage.

Gully erosion susceptibility considering spatiotemporal environmental variables: Midwest U.S. region

Study regionThe study was tested in Jefferson County in Illinois, USA, whose land use is a typical representation of row crop cultivation in the Midwestern USA. Study focusThis study aimed to predict the gully erosion susceptibility in agricultural land using remote sensed environmental data (topographic, pedologic, land cover, precipitation, and vegetation development) considering their spatio-temporal variability in a modeling framework based on the maximum entropy model MaxEnt. The methodology thoroughly evaluated each environmental factor contributing to gully erosion prediction and used a set of rules based on accuracy, transferability, and efficiency to evaluate the model performance. New Hydrological Insights for the RegionThis study developed a data-driven modeling framework that can be applied across other regions. The modeling framework indicates that fifteen factors were the most relevant for developing the gully erosion susceptibility map, where 7.4% of the agricultural land in the study area was found at elevated risk of developing gully erosion. Slope, land cover, organic matter, seasonal LAI, and maximum daily precipitation were the most contributing environmental factors to the study area. Furthermore, this study identified the importance of high temporal resolution in varying seasonal factors (i.e., leaf area index and precipitation) to improve model predictability compared to annual temporal discretization.

Optimal Clipping of Structural Subgrid Stress Closures for Large-Eddy Simulation

Structural subgrid stress models for large-eddy simulation often allow for backscatter of energy from unresolved to resolved turbulent scales, but excessive model backscatter can eventually result in numerical instability. A commonly employed strategy to overcome this issue is to set predicted subgrid stresses to zero in regions of model backscatter. This clipping procedure improves the stability of structural models, however, at the cost of reduced correlation between the predicted subgrid stresses and the exact subgrid stresses. In this paper, we propose an alternative strategy that removes model backscatter from model predictions through the solution of a constrained minimization problem. This procedure, which we refer to as optimal clipping, results in a parameter-free mixed model, and it yields predicted subgrid stresses in higher correlation with the exact subgrid stresses as compared with those attained with the traditional clipping procedure. We perform a series of a priori and a posteriori tests to investigate the impact of applying the traditional and optimal clipping procedures to Clark’s gradient subgrid stress model, and we observe that optimal clipping leads to a significant improvement in model predictions as compared to the traditional clipping procedure.

A physics-inspired neural network model for short-term wind power prediction considering wake effects

Accurate short-term wind power prediction plays an essential role in the wind farm control and the dispatch of wind energy into the power system. Incorporating physical factors that have a major impact on wind farm power generation into machine learning algorithms has always been an important way to improve prediction accuracy. Overlooked in the literature, however, is the influence of wind turbines wakes in improving model predictions. In this work, a physics-inspired neural network model for short-term wind power prediction is developed considering wake effects. Different from traditional neural network models, part of the nodes in the proposed model are determined by the analytical wake model, which enhances the statistical prediction model physically. In this way, the model can be well adapted to the wake effects in the wind farm. Verifications in the actual wind farm case illustrate that there is a good agreement with the prediction results and measured data. Compared with traditional models, the wind power prediction performance of the proposed model has improved by more than 20% in terms of RMSE. Based on this work, we recommend that the wake effect should be considered in the short-term wind power prediction model, which is of great benefit to improving its accuracy.

Mathematical modeling and parameter estimation for 1,6-Hexanediol diacrylate photopolymerization with bifunctional initiator

A dynamic model is proposed for photopolymerization of 1,6-hexanediol diacrylate (HDDA) using bifunctional initiator bis-acylphosphine oxide (BAPO). The proposed model accounts for branching, backbiting and cyclization reactions, and for diffusion-dependent reaction rates during photopolymerization. The proposed model contains 40 adjustable kinetic and free-volume parameters. Experimental data available for parameter estimation are vinyl group conversions obtained using a variety of light intensities and exposure times, and monomer conversions for three experiments. Systematic parameter ranking and estimation is used to evaluate the influence of phenomena included in the model on the quality of the fit. Estimation and ranking results indicate that branching, backbiting, and cyclization reactions have important influences on conversion. Reactions involving two large molecules and propagation reactions become diffusion-dependent. Incorporating diffusion-dependent initiator efficiency results in improved model predictions.

Warm bias of cold sea surface temperatures in the East Sea (Japan Sea)

The East/Japan Sea (ES) is regarded as a natural laboratory for predicting future changes in the global Meridional Overturning Circulation (MOC) under warming climates, as the ES MOC (EMOC) changes rapidly in comparison with the global MOC. Specifically, intermediate and deep-water masses of the ES are formed in its northern reaches via wind-driven subduction of surface water, and convection from the surface to deep layers during the winter. Accordingly, it is important to investigate the variation of winter sea surface temperatures (SSTs) for characterizing and predicting the EMOC; however, global SST products must be corrected and optimized for the ES, as they fail to incorporate the local marginal sea conditions. Here, a warm bias in cold SST was identified for three SST products, such as optimally interpolated sea surface temperatures (OISSTs), microwave SSTs, and operational SST and sea ice analysis products, suggesting the potential usefulness of a correction method incorporating Argo float data. When comparing OISSTs with 5 m temperature estimates from Argo float data during 2000–2020, under the assumption that the mixed layer depth is deeper than 8 m, a nearly normalized histogram of biases was produced, and the robust warm bias (mean = 0.9°C) was detected in the range of relatively cold SSTs (-2°C to 10°C), yet no significant bias in warm SSTs (> 10°C) was found. To minimize the warm bias in cold SSTs, OISSTs were corrected with an inverse 4th-order polynomial fitting method. Subsequently, the mean bias between the corrected SSTs and top depth temperatures of Argo float data was significantly reduced to less than 0.1°C. Moreover, the warm bias of cold SSTs resulted in severe underestimations of the outcropping area colder than 1°C over the northern region, as well as the occurrence period of 1°C to 5°C SSTs in the north-western ES. These results highlight the importance of local bias correction for SST products, and it is expected that the newly suggested correction method will improve model predictions of EMOC change by enhancing SST data quality in the northern ES.

Microbe-iron interactions control lignin decomposition in soil

Lignin decomposition is critically linked to terrestrial carbon (C) cycle due to the enormous C mass of lignin and its importance in controlling overall rates of litter decomposition. Interactions between lignin and iron (Fe) minerals have been increasingly recognized as key mediators of lignin decomposition in experimental studies. However, we still lack a quantitative understanding of how Fe minerals interact with microbes to control lignin decomposition. Here, we leveraged experimental results from an incubation of Fe-rich soil, in which lignin decomposition rates were measured at aerobic conditions after four levels of pre-treated O2 availability, to examine microbe-Fe (MiFe) interactions in lignin decomposition with a MiFe model. We quantified how Fe redox cycling interacted with microbial activities to control lignin decomposition via data-model integration. Our results showed that the MiFe model with time-dependent growth and mortality functions better represented CO2 release from lignin decomposition (R2 ranging from 0.96 to 0.97) than models assuming either first-order or Michaelis-Menten kinetics. Reduction of Fe(III) to Fe(II) after pre-treatments with lower O2 availability stimulated the Fenton reaction to break down macro-molecular lignin into small molecules available to microbes. The small molecules of lignin and necromass bounded with oxidized Fe and were protected from decomposition. After 1-year incubation, the model implied that most of C stabilized with Fe minerals was derived from small molecular lignin C. Our quantitative analysis of microbe-Fe interactions sheds new light on lignin decomposition and preservation and helps improve model prediction of soil C persistence under global change.

Crown dieback and mortality of urban trees linked to heatwaves during extreme drought

Cities have been described as ‘heat islands’ and ‘dry islands’ due to hotter, drier air in urban areas, relative to the surrounding landscape. As climate change intensifies, the health of urban trees will be increasingly impacted. Here, we posed the question: Is it possible to predict urban tree species mortality using (1) species climate envelopes and (2) plant functional traits? To answer these, we tracked patterns of crown dieback and recovery for 23 common urban tree and shrub species in Sydney, Australia during the record-breaking austral 2019–2020 summer. We identified 10 heat-tolerant species including five native and five exotic species, which represent climate-resilient options for urban plantings that are likely to continue to thrive for decades. Thirteen species were considered vulnerable to adverse conditions due to their mortality, poor health leading to tree removal, and/or extensive crown dieback. Crown dieback increased with increasing precipitation of the driest month of species climate of origin, suggesting that species from dry climates may be better suited for urban forests in future climates. We effectively grouped species according to their drought strategy (i.e., tolerance versus avoidance) using a simple trait-based framework that was directly linked with species mortality. The seven most climate-vulnerable species used a drought-avoidance strategy, having low wood density and high turgor loss points along with large, thin leaves with low heat tolerance. Overall, plant functional traits were better than species climate envelopes at explaining crown dieback. Recovery after stress required two mild, wet years for most species, resulting in prolonged loss of cooling benefits as well as economic losses due to replacement of dead/damaged trees. Hotter, longer, and more frequent heatwaves will require selection of more climate-resilient species in urban forests, and our results suggest that future research should focus on plant thermal traits to improve prediction models and species selection.

Principal Component Based Sampling for the Continuous Maintenance of Hydraulic Models

The calibration and continuous maintenance of hydraulic models is essential for the optimisation, planning and management of water distribution networks (WDNs). This process requires model fitting against multiple hydraulic states. However, when extended time series of hydraulic data are considered, it is essential to know which hydraulic states to use for model fitting as too many become computationally impractical. This paper presents a novel principal component analysis (PCA) based sampling method, which evaluates the significance of newly observed hydraulic data to be included in the model calibration and maintenance. A framework is presented, which allows for different sized batches of hydraulic data to be utilised. This enables both model calibration on extended time series of data, or model calibration and maintenance with continuous data. An extensive experimental program was conducted to investigate the performance of the proposed framework with different sampling methods compared to other conventional approaches. The results demonstrate that the presented sampling methods can maintain and improve model prediction accuracy both for retrospective and concurrent implementations when used with extended time series data, which capture different hydraulic states. The hydraulic model of the operational WDN used in this study and the acquired hydraulic data are provided as supplementary data to facilitate reproducible research.

Genetic risk score enhances the risk prediction of severe obesity in adult survivors of childhood cancer.

Adult survivors of childhood cancer have high rates of obesity, which, in combination with the cardiotoxic effects of specific cancer therapies, places them at high risk for cardiovascular morbidity. Here we show the contribution of genetic risk scores (GRSs) to increase prediction of those survivors of childhood cancer who are at risk for severe obesity (body mass index ≥40 kg m-2) as an adult. Among 2,548 individuals of European ancestry from the St. Jude Lifetime Cohort Study who were 5-year survivors of childhood cancer, the GRS was found to be associated with 53-fold-higher odds of severe obesity. Addition of GRSs to risk prediction models based on cancer treatment exposures and lifestyle factors significantly improved model prediction (area under the curve increased from 0.68 to 0.75, resulting in the identification of 4.3-times more high-risk survivors), which was independently validated in 6,064 individuals from the Childhood Cancer Survivor Study. Genetic predictors improve identification of patients who could benefit from heightened surveillance and interventions to mitigate the risk of severe obesity and associated cardio-metabolic complications.

Retrospective comparisons of competing demographic models give clarity from "messy" management on a Scottish grouse moor.

Retrospective comparison of predictive models that describe competing hypotheses regarding system function can shed light on regulatory mechanisms within the framework of adaptive resource management. We applied this approach to a 28-year study of red grouse (Lagopus lagopus scotica) in Scotland, with the aims of reducing uncertainty regarding important drivers of grouse population dynamics, and of evaluating the efficacy of using seasonal versus annual model assessments. We developed three sets of models that predicted pre-breeding and post-breeding grouse density, matching the timing of grouse counts on the ground. We updated conditions and management through time in the spirit of a real-time, adaptive management program and used a Bayesian model weight updating process to compare model predictions with empirical grouse densities. The first two model sets involved single annual updates from either pre-breeding or post-breeding counts; the third set was updated twice a year. Each model set comprised seven models representing increasingly complex hypotheses regarding potentially important drivers of grouse: the baseline model included weather and parasite effects on productivity, shooting losses and density-dependent overwinter survival; subsequent models incorporated the effect of habitat gain/loss (HAB), control of non-protected predators (NPP) and predation by protected hen harriers (Circus cyaneus, HH) and buzzards (Buteo buteo, BZ). The weight of evidence was consistent across model sets, settling within 10 years on the harrier (NPP + HH), buzzard (NPP + HH + BZ) and buzzard + habitat (NPP + HH + BZ + HAB) models, and downgrading the baseline + habitat, non-protected predator, and non-protected predator + habitat models. By the end of the study only the buzzard and buzzard + habitat models retained substantial weights, emphasizing the dynamical complexity of the system. Habitat inclusion failed to improve model predictions, implying that over the period of this study habitat quantity was unimportant in determining grouse abundance. Comparing annually and biannually assessed model sets, the main difference was in the baseline model, whose weight increased or remained stable when assessed annually, but collapsed when assessed biannually. Our adaptive modeling approach is suitable for many ecological situations in which a complex interplay of factors makes experimental manipulation difficult.