Regional-scale modelling and predictive uncertainty analysis of cumulative groundwater impacts from coal seam gas and coal mining developments

Abstract

The geological basins in Australia across which the hydrogeological Great Artesian Basin (GAB) exists, hold significant coal and coal seam gas (CSG) resources. Resource development from deep sedimentary basins often involves the risk of impacting groundwater resources. Predictive analysis of potential impacts on water resources are important for the risk analysis of resource development projects. A regional-scale numerical groundwater model was developed to probabilistically assess potential groundwater impacts due to additional coal resource development from a deep sedimentary basin underlying the GAB. The probabilistic simulation considered the plausible variability of the model parameters and accounted for uncertainties. Predictive uncertainty analysis was undertaken using a rejection sampling method after screening the model runs using predefined objective functions to evaluate the performance of the model runs with respect to historical observations. The predictive simulations were undertaken using 2,618 model runs to obtain maximum head drawdown caused by CSG and coal mining developments. The results showed that the water-table drawdown from an individual coal mining development becomes insignificant (maximum difference in drawdown <0.2 m) beyond 5 km. In general, less water-table drawdown was produced near the CSG development site but small amounts of drawdown spread further from it. Separation of surficial aquifers from deeper coal formations, as generally found in the GAB, limits the propagation of CSG-induced drawdown into the aquifers closer to the surface. While this study was specifically done for the Namoi region, similar outcomes could be expected in the broader GAB and other basins where equivalent hydrogeological conditions exist.

Résumé

Les bassins géologiques en Australie, parmi lesquels le Grand Bassin hydrogéologique Artésien (GBA), contiennent d’importantes ressources en charbon et gaz de houille (Coal Seam Gas - CSG). Le développement des ressources des bassins sédimentaires profonds implique souvent le risque d’impacter les ressources d’eaux souterraines. L’analyse prédictive des impacts potentiels sur les ressources en eau est importante pour l’analyse de risque des projets de développement de ressources minières. Un modèle numérique des eaux souterraines à l’échelle régionale a été développé pour évaluer de manière probabiliste les impacts potentiels sur les eaux souterraines dus au développement d’une ressource additionnelle de charbon dans un bassin sédimentaire profond sous-jacent au GBA. La simulation probabiliste a considéré une variabilité plausible des paramètres du modèle et a pris en compte les incertitudes. L’analyse prédictive d’incertitude a été entreprise en utilisant une méthode d’échantillonnage de type rejet après examen des résultats du modèle, en utilisant des fonction-objectif prédéfinies pour évaluer la pertinence des résultats du modèle par rapport aux observations historiques. Les simulations prédictives ont été entreprises en utilisant 2618 exécutions du modèle pour obtenir l’abaissement maximal du niveau principal provoqué par l’exploitation de CSG et de charbon. Les résultats montrent que l’abaissement du niveau piézométrique sous l’effet de l’exploitation d’une extraction individuelle de charbon devient insignifiant (rabattement maximal <0.2 m) au-delà de 5 kilomètres. En général, le rabattement du niveau piézométrique est moindre à proximité du site d’exploitation des CSG mais de faibles rabattements se propagent plus loin. L’individualisation des aquifères superficiels par rapport aux formations houillères plus profondes, ce qui est généralement le cas dans le GBA, limite la propagation des rabattements induits par l’exploitation des CSG dans les couches aquifères plus proches de la surface. Même si cette étude a été spécifiquement réalisée pour la région de Namoi, des résultats semblables pourraient être attendus plus largement dans le GBA et dans d’autres bassins où des conditions hydrogéologiques équivalentes existent.

Resumen

Las cuencas geológicas de Australia, a través de las cuales existe la Great Artesian Basin (GAB), contienen importantes recursos de carbón y gas de veta de carbón (CSG). El desarrollo de recursos de cuencas sedimentarias profundas a menudo implica el riesgo de afectar los recursos de aguas subterráneas. El análisis predictivo de los impactos potenciales sobre los recursos hídricos es importante para el análisis de riesgos de los proyectos de desarrollo de los recursos. Se desarrolló un modelo numérico de aguas subterráneas a escala regional para evaluar probabilísticamente los impactos potenciales debido al desarrollo adicional de recursos de carbón de una cuenca sedimentaria profunda subyacente al GAB. La simulación probabilística consideró la variabilidad plausible de los parámetros del modelo y tomó en cuenta las incertidumbres. El análisis predictivo de la incertidumbre se llevó a cabo utilizando un método de muestreo de rechazo después del filtrado de las corridas del modelo utilizando funciones objetivas predefinidas para evaluar el rendimiento de las corridas del modelo con respecto a las observaciones históricas. Las simulaciones predictivas se realizaron utilizando 2618 corridas del modelo para obtener la máxima reducción de carga causada por los desarrollos de CSG y de la minería de carbón. Los resultados mostraron que la profundización de la capa freática de un desarrollo minero individual de carbón se vuelve insignificante (diferencia máxima de profundización <0.2 m) más allá de 5 km. En general, se produjo una menor reducción del nivel de agua cerca del sitio de desarrollo de CSG, pero pequeñas cantidades de reducción del nivel de agua se extendieron más allá de él. La separación de los acuíferos superficiales de las formaciones de carbón más profundas, como se encuentra generalmente en el GAB, limita la propagación de la reducción inducida por el CSG hacia los acuíferos más cercanos a la superficie. Si bien este estudio se realizó específicamente para la región de Namoi, se pueden esperar resultados similares en la gran extensión del GAB y en otras cuencas donde existen condiciones hidrogeológicas equivalentes.

摘要

澳大利亚的地质盆地横穿水文地质自流大盆地(GAB), 蕴藏着丰富的煤矿和煤层气(CSG)资源。深层沉积盆地的资源开发往往会对地下水资源造成威胁。对水资源的潜在影响的预测分析对于资源开发项目的风险分析具有重要意义。建立了一个区域尺度的地下水数值模型, 以概率评估GAB盆地下伏深层沉积盆地的其他煤炭资源开采对地下水的潜在影响。概率模拟考虑了模型参数的似然可变性及不确定性。筛选模型运行后采用拒绝抽样方法, 使用预先定义的目标函数来评价模拟结果相对于历史观测值的模拟效果, 分析了预测的不确定性。预测模拟采用2,618个模型获取由CSG和煤矿开采所造成的最大水位降深值。结果表明, 5km深度以下某煤矿开采造成的地下水位下降不明显(最大降深<0.2 m)。总体上, 在CSG开采场地附近地下水位降低幅度较小, 但会围绕其进一步小幅降低。像GAB中常发现的那样, 表层含水层从较深的煤层中分离, 限制了CSG造成的地下水位下降到更接近地表的含水层中的传播。虽然该研究是针对Namoi地区进行的, 但该结果可能同样适用于较宽阔的GAB范围和其他类似水文地质条件的盆地。

Resumo

As bacias geológicas na Austrália, dentre as quais jaz a Grande Bacia Artesiana (GBA), contém importantes recursos de carvão e de gás de veio de carvão (GVC). A explotação destes recursos em bacias sedimentares profundas geralmente envolve o risco de impactar os recursos hídricos subterrâneos. A análise de predição dos possíveis impactos sobre às águas subterrâneas é importante para a análise de riscos de projetos de desenvolvimento de recursos. Um modelo numérico de águas subterrâneas em escala regional foi desenvolvido para avaliar a probabilidade de possíveis impactos sobre as águas subterrâneas devido ao desenvolvimento e explotação adicional de recursos de carvão de uma bacia sedimentar profunda subjacente à GBA. A simulação probabilística considerou a variabilidade plausível dos parâmetros do modelo e contabilizou incertezas. A análise preditiva da incerteza foi realizada utilizando um método de amostragem por rejeição após a varredura das execuções do modelo usando funções objetivas predefinidas para avaliar o desempenho das execuções do modelo em relação às observações históricas. As simulações preditivas foram realizadas utilizando 2618 execuções do modelo para obter o rebaixamento máximo do nível d’água causado pela mineração de carvão e extração de GVC. Os resultados mostraram que o rebaixamento do nível d’água de um único sítio de mineração de carvão se torna insignificante (diferença máxima no rebaixamento <0.2 m) a partir de 5 km. Em geral, foi produzido menos rebaixamento do lençol freático próximo ao local de extração do GVC, porém pequenas quantidades de rebaixamento se extenderam a maiores distâncias. A separação de aquíferos superficiais das formações de carvão mais profundas, como geralmente encontrada no GBA, limita a propagação do rebaixamento induzido por extração de GVC nos aquíferos mais próximos da superfície. Embora este estudo tenha sido realizado especificamente para a região de Namoi, resultados semelhantes poderiam ser esperados nas demais extensões da GBA e em outras bacias onde existam condições hidrogeológicas equivalentes.

Appendices

Appendix 1

Two sets of models were run to compare the difference in dmax predictions when confined and unconfined conditions (convertible layer) were assumed for the parts of the aquifer where water-table conditions exist. The predicted dmax in the model layer 6 corresponding to the Pilliga Sandstone aquifer were compared across these two sets (Fig. 16). The Pilliga Sandstone outcrops close to the areas of CRDP and the aquifer is unconfined close to the outcrop areas and has numerous receptors at which estimation of dmax is of prediction interest. The analysis indicates that the maximum drawdown simulated by both approaches are similar except that the confined model run slightly overestimates the dmax value. In the range of dmax simulated across all model runs, this slight overestimation is within practical observation errors and is not significant. The analysis also indicated that the confined assumption is conservative in purview of the objective of the modelling analysis, i.e. it does not underestimate drawdowns at vast majority of the points of interest and is suitable for the stated objective of the modelling study, which is to demarcate potentially impacted areas from gas development and mining.

Fig. 16

Comparison of the dmax predicted using two sets of model runs in which one set simulated unconfined conditions and the other used confined conditions, keeping all other model parameters constant

It is also noteworthy that, while a confined assumption was applied, the effect of this simplification on the model predictions is minimised by using storage values based on specific yield in areas where layers are outcropping. The specific yield parameters used for this are also included in the uncertainty analysis to explore prediction uncertainty caused by uncertainty of the specific yield parameters.

Appendix 2

As the modelling is only progressing until 2102 there will be situations where dmax, maximum difference in drawdown for one realisation within an ensemble of groundwater modelling runs, obtained by choosing the maximum of the time series of differences between two futures, has not been reached within this time. After the pumping associated with coal resource development has ceased, dmax at the well will have been reached but the cone of depression can still expand while the pressure is recovering at the well location. This can lead to dmax at a point away from the pumping, occurring well after the pumping has ceased. The analytical solution of Yeh and Wang (2009) allows us to investigate the impact of not running the model until dmax is reached:

$$ s\left(r,{t}_2\right)=\frac{s_0}{W\left(\frac{S{r}_w^2}{4T{t}_h}\right)}\left[W\left(\frac{S{r}^2}{4T\left({t}_h+{t}_2\right)}\right)-W\left(\frac{S{r}^2}{4T{t}_2}\right)\right],r\gg {r}_{\mathrm{w}} $$

(14)

where s(r,t) is the drawdown at a radial distance from the well r at time t, S is the storativity, T is the transmissivity, r_w is the radius of the well and W is the Theis well function. Figure 17 shows a solution of dmax and time to dmax as a function of distance from the extraction well for a case with a T/S of 254 m²/d (this is an example, not related to a specific bioregion). This shows that, close to the well, dmax is much greater than further from the well, whereas time to dmax occurs close to when the pumps are switched off for locations close to the well; however, time to dmax increases with increasing distance from the pumping well. There is a very clear negative correlation between dmax and time to dmax. For any model node where dmax has not occurred within the temporal domain of the model, dmax must be smaller than every point closer to the pumping well.

Fig. 17

Calculation of dmax and time to dmax as a function of distance from the pumping well for T/S = 254 m²/d using the analytical solution of Yeh and Wang (2009)