Contents lists available at ScienceDirect Forest Ecology and Management journal homepage: www.elsevier.com/locate/foreco A fine-scale state-space model to understand drivers of forest fires in the Himalayan foothills Karthik K. Murthy a, ⁎ , Samir Kumar Sinha b , Rahul Kaul b , Srinivas Vaidyanathan c a Centre for Ecological Sciences, Indian Institute of Science, Bangalore 560012, India b Wildlife Trust of India, F-13, Sector 8, National Capital Region (NCR), Noida 201301, India c Foundation for Ecological Research, Advocacy & Learning, Pondicherry Campus, 170/3 Morattandi, Auroville Post, Tamil Nadu, India ARTICLE INFO Keywords: Forest fire MODIS Himalayan terai Poisson regression Zero-inflated negative binomial models Anthropogenic influences Fire management ABSTRACT The tropical forests situated in the Himalayan foothills (terai) experience frequent wildfires which can alter the vegetation structure and composition, challenging tiger conservation efforts in this region. Hence, there is a need for better understanding of the drivers of forest fire to aid efficient management, but these efforts are hampered by the deficiency of spatial and temporal data on fire incidences. Advancement in remote sensing technology provides an opportunity to understand the spatial and temporal patterns of wildfires in relation to anthro- pogenic, ecological, and environmental drivers. We used MODIS fire data from 2001 to 2015 to understand fire incidences in Valmiki Tiger Reserve (VTR), an important tiger habitat area in the Himalayan terai region. We analyzed fire incidences to understand monthly and inter-annual variation of fire incidences at two spatial scales: first, using only climatic variables considering VTR as a single spatial unit and the second, to understand the fire dynamics at 1 km 2 spatial resolution using climatic, ecological, and anthropogenic variables. The results show that fire incidences occurred from January to May, 88% of which occurred in March and April. Overall, different variables affected fire incidences in March and April for both the temporal models. Precipitation had a sig- nificant negative effect on fire incidences in both March and April, but temperature had a positive effect only in March. Similarly, the fine scale temporal model showed that while ecological (litter load, NPP) and anthro- pogenic (distance to villages and roads) variables influenced fire incidences in March, altitude and village area surrounding the forest affected fires in April. Litter input, distance to nearest villages, and village area had a non- linear relationship with fire incidences indicating a few inconsistencies with the global patterns of fire with human activity. We show that the Sal dominated forests and terai grasslands at low altitudes (200 m), falling within a zone of 2.5–3 km from villages and with good road connectivity are more prone to fire. The fine-scale fire prediction map of VTR will be helpful to the Tiger Reserve management in developing appropriate strategies for the fire prone areas. 1. Introduction Quantitative estimates of burnt area are sporadic or non-existent for most tropical countries (Cochrane, 2003) and in India such datasets are sketchy and fragmented, and are often under-reported due to fear of accountability of forest managers (Satendra and Kaushik, 2014). With no reliable historical records of wildfires, spread, severity, and area burnt; developing a fire management plan is a major challenge. ‘Fire Management’ involves protection (early warning), preparedness, pre- vention, response and suppression, restoration/rehabilitation, and monitoring (FAO, 2007). It is crucial for the fire management agencies to get fire suppression resources while fires are small; otherwise, they can be overwhelmed and will be unable to take appropriate fire curbing actions, leading to spread of fire to vast areas (Flannigan et al., 2009). Thus, fire curbing activities require spatially explicit fire risk maps with fine scale information of timing, area of occurrence, intensity, and spread with associated drivers of fire incidences (Díaz-Avalos et al., 2016; Mann et al., 2016). Such fire risk maps are absent for forests in India and hence fire curbing activities are conducted in an ad-hoc and reactionary manner. Forest fires are influenced by fuel load, climate-weather, ignition agents, and people (Flannigan et al., 2009, 2005). Global patterns of wildfire are associated with climatic conditions including high tem- perature (> 28 °C), intermediate annual rainfall (350–1100 mm), and prolonged dry periods (Aldersley et al., 2011). Most wild fires in tro- pical regions are initiated by humans, and in India, fires are primarily https://doi.org/10.1016/j.foreco.2018.10.009 Received 29 May 2018; Received in revised form 30 September 2018; Accepted 2 October 2018 ⁎ Corresponding author. E-mail address: kartikmurthy@iisc.ac.in (K.K. Murthy). Forest Ecology and Management 432 (2019) 902–911 0378-1127/ © 2018 Elsevier B.V. All rights reserved. T